The Impact of Extension Services on Farm-level Income: An Instrumental Variable Approach to Combat Endogeneity Concerns

The Impact of Extension Services on Farm-level Income: An Instrumental Variable Approach to... Abstract Agricultural extension is an important policy instrument utilized to diffuse knowledge and increase profitability among farmers. However, analyses on impact are subject to endogeneity concerns, causing multiple biases. Failure to combat endogeneity can lead to false inferences on impact. This article addresses this issue by applying an instrumental variable approach with distance to local advisory office and a policy change chosen as instruments for extension participation. The results show that participation significantly increased farm income and that OLS estimates underestimated the impact. Therefore, a superior estimate of impact is achieved which can be leveraged to better support accurate policy making. Extension, farm income, endogeneity, instrumental variables, two-stage least squares estimation, panel data Agricultural extension builds the capabilities of clients through improved problem solving, decision making, and management (Vanclay and Leach 2011), and is a means of transferring specialist knowledge from emerging research or public policy arenas to farm level that is commonly adopted worldwide. Historically, most developed countries have had a form of extension service for rural land managers funded largely from general taxation and delivered by public organizations (Garforth et al. 2003). Indeed, governments actively promote extension services as a means of knowledge diffusion and technology adoption, and also to develop innovative solutions to emerging challenges (Läpple and Hennessy 2015). This form of public extension service has since been supplemented by the private sector with the common objective of improving the performance of farmers. There are many challenges for the agricultural sector, such as the need to strike a balance between increasing productivity to feed a growing global population and reducing negative environmental externalities, including climate change. Coupled with this responsibility is a financial challenge as global economies navigate turbulent macroeconomic cycles, and there is a renewed emphasis on “value for money” for public expenditure. Extension services are important in these circumstances as they can act as policy levers to change existing behavior, but must represent a worthy investment for policymakers. Furthermore, extension clients must also experience a value for money return from participation given the fee level paid, particularly given existing income challenges caused by economic volatility, and a strong reliance on subsidies (O’Donoghue and Hennessy 2015). Therefore, estimating the impact of extension participation on farm income is prudent and ensures a common outcome is measured regardless of the type of extension activity employed. In order to achieve this it is imperative that the estimation is robust and unbiased, as a plethora of confounding factors also influence this impact. In this paper this obstacle is overcome by instrumenting the extension variable to remove these influences. Many studies show that interaction with extension services affects income levels (Dercon et al. 2009; Davis et al. 2012; Läpple and Hennessy 2015). Anderson and Feder (2004) reviewed previous studies on extension impact, and whilst positive results are common, there is variability within these results and they warned that results should be treated with caution given the econometric challenges. Indeed, the inconsistency in research findings could be partly due to the broad range of extension methods and outcome measures (Läpple and Hennessy 2015) and therefore, the assessment of impact across disparate ranges is the major challenge for this type of research (Ragasa et al. 2016). However, from a policy perspective the impact of these services is the ultimate criterion (Birner et al. 2009) and thus the key objective of this research is to provide a robust causal estimation of the impact of extension participation on farm income. Providing an accurate and valid estimation of impact can assist the policy formulation process as policymakers can rely on the likely impact of an initiative with greater precision. In order to achieve this, it is critically important that the estimation addresses econometric concerns related to the issue of endogeneity. The econometric issue of endogeneity is typically caused by omitted variable bias, measurement error, and self-selection biases. Failure to combat endogeneity may lead to incorrect inferences on the causal relationship between extension participation and farm income (Akobundu et al. 2004; Abdallah et al. 2015). The omission of an innate ability variable among extension participants could overestimate the true effect of extension on farm income. Conversely, measurement error is likely to underestimate the true effect of this relationship (Card 1999). Similarly, self-selection biases are implicit in this research with the motivation to engage with extension being a critical element of estimating impact (Nordin and Höjgård 2016). For example, more capable farmers may be more motivated to engage with extension services to augment their performance, whereas weaker farmers may be less likely to participate. The Instrumental Variable (IV) approach is dependent on identifying suitable instruments that affect the endogenous variable in question but do not impact directly on the dependent variable. In our case, such an instrument must affect the decision to participate in extension services, but not impact farm family income directly. These instruments “purge” the endogenous regressors and allow consistent coefficient estimates (Gabel and Scheve 2007). Distance from the local advisory office and the introduction of a policy change in 2005 are chosen as appropriate instruments to meet these conditions. The former is expected to negatively affect the decision to participate but is uncorrelated with farm income, whereas the latter incentivized participation for assistance with subsidy applications decoupled from current production based on historical receipts across all farming systems. These instruments enable the evaluation of the impact of extension on farm level income over the 2000–2013 period. The remainder of the paper is structured as follows: initially the context for extension services in Ireland is outlined, followed by a review of the relevant literature. Next there is an overview of the methodology and data. Finally, the results are discussed, and the conclusion outlines the policy implications, along with some limitations and recommendations for future research. Context Agricultural extension programs are operated by multifunctional organizations that provide advisory services to the farming sector. Extension programs could be viewed as risk management devices from policy makers to mitigate issues in the rural economy, or as drivers of growth at the farm level to ensure that best practices are followed systematically. Läpple, Hennessy, and Newman (2013) summarized the definition and purpose of extension services as a program to improve farm performance and introduce new technologies to connect emerging research to on-farm practices. Thus, extension programs assist farmers to overcome barriers to achieving set goals due to a lack of knowledge, motivation, resources, insight, and power, or a combination of these (Van den Ban and Hawkins 1988). Extension programs aim to assist farmers on issues such as productivity, food safety, and the environment, among others (Boyle 2008). The extension service in Irish agriculture consists of both public and private consultants, with clients split relatively evenly among both types. Prager et al. (2016) estimated that approximately 250 private advizers (mainly represented by the Agricultural Consultants Association) were in operation alongside 250 Teagasc advizers. Teagasc is the main body for the public delivery of agricultural research, advice, and training since 1988; it is a unique organization that combines research, extension services, and education (Prager et al. 2016), and 30% of the operating budget of approximately €160 million annually is directed towards extension programs (Teagasc 2016). The research reported here uses data referring to Teagasc clients only for two reasons: first, the level of public expenditure allocated to this public extension system, and second, data constraints related to private extension organizations and their clients. Indeed, as the majority of private consultancies in Ireland consisted of just 1–3 advizers (Prager et al. 2016), that data is more likely to be fragmented as opposed to the superior data available from Teagasc extension. Therefore, farmers who engaged with private consultants are not captured in this research. Teagasc clients include any farmer who participates in any extension program under the various contracts offered. The clients’ specific farm systems and characteristics are included as control variables in the analysis. Furthermore, this analysis focuses on more recent Teagasc clients that became clients after the policy change in 2005 to reduce general self-selection biases further. The introduction of the Single Farm Payment scheme in 2005 increased the numbers of extension clients due to farmers’ need for assistance given bureaucratic challenges, and is applied as an instrument in the analysis. In other words, the sample is pooled to focus on farmers who were in the Teagasc National Farm Survey (NFS) sample prior to the policy change but only became Teagasc clients afterwards. This increases the comparability of these clients as they are more likely to have been motivated to engage in an extension program given the policy change as opposed to alternative diverse “traditional” motivations (Läpple, Hennessy, and Newman 2013). This hones the focus of the study by ensuring the analysis is conducted on a similar cohort of extension clients. Preliminary analysis on the impact of extension on this group showed a larger impact for those who joined after 2005. Ordinary Least Squares (OLS) estimation showed a positive impact, which increased from 9% to 15% for all clients pre- and post-2005, respectively. Similarly, a difference in difference estimation for clients who joined after the introduction of the policy change showed a positive and significant impact. These estimations are included in the appendix. Literature Review The dominant theme in previous studies on the impact of extension participation on farm level outcomes is that of variability across the results (Anderson and Feder 2004). Given the complexity of narrowing the focus of research, the variety in types of extension programs, the diversity of outcomes to measure, and the diversity of methodological options, this variability and inconsistency is not surprising (Läpple and Hennessy 2015). However, an IV approach to the impact of extension is not common in the literature given the challenge of identifying suitable instruments, a challenge which this study aims to address. Much of the literature analyses the relationship between extension participation and productivity with studies reporting a positive relationship between extension investment and output yields (Griliches 1964; Marsh, Pannell, and Linder 2004). Similarly, analysis of the impact of one-to-one consultations and productivity also identified positive relationships (Owens, Hoddinott, and Kinsey 2003) although in one case the levels diminished over time (Krishnan and Patnam 2014). Participatory extension techniques have also been studied in relation to farm yields, with positive outcomes reported in terms of increased crops (Davis et al. 2012) and milk yields per cow (Läpple and Hennessy 2015). In contrast, Hunt et al. (2014) found that the impact of extension services on productivity declined over time due to capacity constraints. Specifically in relation to extension participation and financial returns there are recent studies on participatory extension programs’ impact in terms of increased gross margin per hectare of €310 in the Irish dairy sector (Läpple, Hennessy, and Newman 2013), and an increase of €150 gross margin per dairy cow also in Ireland (Läpple and Hennessy 2015). The former study employed an endogenous switching model, whereas the latter utilized propensity score matching methods to address endogeneity using panel data sets. Davis et al. (2012) also found a positive effect using the propensity score matching method identifying income gains of 21% to 104% in selected African nations. These studies adopt extension participation as a binary variable based on participation. The switching model method relies on conditional expectations of impact based on non-participation in the extension program compared with the actual results. The main advantage of the switching model is that it combats self-selection biases but it is based on the hypothetical case of an outcome. The propensity score matching approach is effective but relies on a number of decisive steps based on the quality of data and underpinning assumptions often involving a trade-off between bias and efficiency (Caliendo and Kopeinig 2008). These studies utilized various econometric methodologies but endogeneity concerns were not central to the analysis and they did not employ the IV approach, which may lead to incorrect inferences due to uncontrolled biases that remain (Akobundu et al. 2004; Abdallah et al. 2015). Indeed, whereas some studies may prioritize a particular form of bias to control, the IV approach combats multiple forms of endogenous biases (Card 1999; Cawley and Meyerhoefer 2012; Howley, O’Neill, and Atkinson 2015). Thus, the IV approach is superior in that it removes the need to prioritize specific types of bias over others, but combats all forms, ensuring an improved estimation of extension impact on farms. IV estimation can consistently estimate coefficients that will almost certainly be close to the true coefficient value if the sample size is sufficiently large (Murray 2006). However, this approach is dependent on the precondition of identifying appropriate instruments, which often proves challenging. Accordingly, studies on the impact of extension engagement on farm-level income using an IV approach are less common but there are exceptions. Nordin and Höjgård (2016) applied an IV methodology to evaluate extension services in Sweden and found both a societal and farm-level benefit to participation in relation to nutrient management. The mean number of adviser visits to a farm was adopted as an instrument for the actual number of visits to a specific farm based on a positive relationship between the agent’s expertise and farm type. Nordin and Höjgård (2016) found that OLS estimates underestimated the benefits of the extension program compared to the corrected IV estimates. Akobundu et al. (2004) utilized this approach and found a U.S. extension program that incorporated both one-to-one and group-based activities increased net farm income, but the level of return relied on the intensity of participation. These researchers utilized the distance to local extension office, the intensity of participation captured through the number of adviser on-farm visits, and the level of household debt with higher debts classified as more likely to increase adviser visits, as suitable instruments. Similarly, Heanue and O’Donoghue (2014) found positive economic outcomes for farmers who participated in agricultural education in terms of farm income and the internal rate of return to investment, also using an IV approach. As instruments, they used the distance to local agricultural college and the introduction of a policy change. This study follows on from that work but instead focuses solely on Teagasc extension clients. Whilst an IV approach is not common in the agricultural economics literature it is important to note that it is widely used in other fields, with significant policy implications. For example, it has been utilized to study the impact of economic shocks on conflict where adverse weather conditions instrumented economic activity and showed an increase in the probability of civil conflict in the following year (Miguel et al. 2004). Further, IV estimation is viewed as a sensible addition to the analytical toolbox in health economics (Rassen et al. 2009). Indeed, in one such study regional cardiac catheterization rates were adopted as instruments to explore the relationship between specific treatments and mortality rates, with a 16% survival benefit being reported (Stukel et al. 2007). Similarly, genetic variation in weight was applied as an instrument in a study on the relationship between obesity and medical costs, which found that previous literature underestimated these costs (Cawley and Meyerhoefer 2012). IV estimation has also been applied to examine the relationship between education and subsequent income earnings, with distance to college used as an instrument also showing a previous underestimation in the literature (Card 1999). Therefore, the literature presents examples of downward biases that are subsequently corrected through instrumentation and this research queries whether this is also the case for the impact of extension on farm-level income. If so, existing estimations of impact may have underestimated the true effect on farms, and the effectiveness of extension services to act as policy instruments could be more pronounced. Accordingly, the purpose of this research is to measure the impact of extension programs on farm incomes. The IV approach ensures that endogeneity concerns are addressed, leading to a robust estimation of impact given the identification of suitable instruments. In this study it is assumed that farmers utilize extension programs primarily to improve profitability. Other beneficial outcomes are taken as secondary. Clients of extension programs are expected to achieve higher average farm income levels than farmers who do not use extension programs. Thus, a positive relationship is expected. Methodology Birkhaeuser, Evenson, and Feder (1991) identified a number of problems associated with assessing the impact of extension activities such as the phase of the farmers’ development cycle, policy and market influences, and information flows, but the predominant issue is that of endogeneity. An endogenous explanatory variable exists when the variable is correlated with the error term (Wooldridge 2013). This means that the estimated coefficient for that variable will be inconsistent and biased as its magnitude is somewhat determined by the error term. This may lead to inaccurate interpretations of the effects of findings unless appropriate action is undertaken (Abdallah et al. 2015). Endogeneity has three primary causes. Firstly, omitted variable bias causes an obvious problem for analysis as a farmer’s innate ability, effort, ambition, or motivation would have an effect on the impact of extension engagement, yet this data is unobserved. Secondly, self-selection bias is a methodological error caused by initial differences between participants and non-participants due to the conscious decision to enroll in an extension program or not (Imbens and Wooldridge 2009; Nordin and Höjgård 2016). Läpple, Hennessy, and Newman (2013) explain that higher-skilled producers may be more likely to adopt extension services given their capacity and motivation to enhance their enterprise, yet we do not have a variable to reflect this issue. Conversely, farmers with higher ability may not deem extension services necessary given their own capabilities on the farm. Similarly, farmers with lower ability may seek advisory assistance to bolster their performance, or alternatively they may feel investing in advisory services is not worthwhile. Akobundu et al. (2004) suggested that extension programs themselves may purposively target specific farmers, whether disadvantaged or eager to participate. This may be due to attempting to improve the performance of vulnerable producers in the first instance, who may avoid such opportunities without adequate incentives. Conversely, in the latter example, advisers may be more likely to target more motivated participants who may be more likely to disseminate the knowledge (Läpple, Hennessy, and Newman 2013). Conversely, advisers may avoid particular clients for various reasons such as location, personal characteristics, or due to time constraints. This indicates bidirectional causality with regard to self-selection bias. Therefore, farmers using the extension service are systematically different than those who do not (Tamini 2011; Hennessy and Heanue 2012). Finally, another form of bias may be related to measurement error. Given the endogenous variable for extension participation is imperfectly measured as a binary variable in our analysis, this is likely to cause a downward (attenuation) bias on the initial Ordinary Least Squares (OLS) estimation prior to instrumentation (Card 1999). The variety of extension programs options on offer range from annual visits to assist with scheme assistance to intensive technical assistance packages involving on-site visits, discussion groups, and seminars. The utilization of participation as a binary variable removes this variability for the purpose of providing an aggregated impact estimation regardless of the type of extension adopted; hence, it is imperfect but necessary for this analysis. Accordingly, it is important to note that the direction of the bias is not necessarily upward, given intuitive expectations on omitted ability or self-selection. There are a number of approaches to deal with endogeneity as discussed in the previous section. However, the IV approach is efficient as it accounts for all forms of bias listed above, provided suitable instruments are identified. The instrument must be correlated with the endogenous explanatory variable (relevant), but uncorrelated with the dependent variable and error term (valid) (Murray 2006; Burgess, Dudbridge, and Thompson 2016). If such an instrument can be found, then an unbiased consistent coefficient for the endogenous variable can be estimated (Gujarati 2003). The process involves a two-stage least squares regression, where firstly the instruments are regressed on the endogenous variable, and subsequently the predicted value of the now exogenous variable is inserted into the main structural equation and estimated. Accordingly, the primary challenge for IV analysis is identifying a suitable instrument. Murray (2006) noted that having at least as many instruments as endogenous variables is a necessary condition for identification and in most cases is sufficient. On this basis two instruments were identified. The distance to the local extension office and the policy change effect were chosen on the basis of previous literature. Geographic proximity was employed by Card (1993) and Callan and Harmon (1999) along with Heanue and O’Donoghue (2014), where the latter two also used a policy change as an exogenous shock. Callan and Harmon utilized the introduction of free schooling and Heanue and O’Donoghue adopted the introduction of the Stamp Duty Exemption Scheme that targeted young trained farmers as instruments. In both examples, the researchers argued that the instrument incentivized participation in education and agricultural training, but did not affect income levels directly. Both instruments were expected to affect the decision to participate in extension services independently of farmers’ personal characteristics and/or farm performance. These instruments were also interacted to examine the impact of both combined and improve the estimation approach. The rationale for the instruments is explained in more detail in the subsequent section. As noted above, a 2 Stage Least Squares (2SLS) IV approach is applied. Thus, our initial stage is to test for the exclusion restrictions of the instruments (Wooldridge 2013). In other words, we apply the first stage, which is the reduced-form equation for our endogenous regressor through the following reduced form equation for y2: y2= π0+π1z1+π2z2+ π3z1z2+ πX+ v2 (1) where y2 is our endogenous regressor (extension participation), πj is our estimated parameter coefficients, zk are our instruments, X is a vector of all other explanatory variables, and v2 is our error term. Partial correlation at least between zk and y2 is necessary to fulfil the requirement that the instruments affect the endogenous regressor. Therefore, we can apply our second stage and specify our structural equation as follows: y1= β0+β1y^2+βX+ u1 (2) where y1 is the unbiased estimation of our dependent variables, βj is our estimated parameter coefficients, y^2 is our “purged” endogenous variable, X is a vector of all other explanatory variables, and u1 is our error term. All models are presented with selected specification tests in the results. In the first stage, the multivariate Cragg-Donald Wald F test was conducted to measure whether the instruments affect the endogenous variable. Stock, Wright, and Yogo (2002) outlined a rule of thumb that the F statistic must exceed 10 to avoid instruments being classified as weak. For the IV models, the Sargan statistic is reported, which measures the null hypotheses that all instruments are valid. This is recognized as the standard overidentification test of the validity of the instruments and is a benefit of the 2SLS approach (Howley, O’Neill, and Atkinson 2015). Cawley and Meyerhoefer (2012) noted that rejecting the null hypothesis of no effect is impossible and therefore doubt will remain if an instrument is truly exogenous. However, the Sargan test of validity is applied on occasion and failure to reject implies the instruments cannot be argued as invalid, which bolsters the argument of their validity as instruments. If the computed chi-square exceeds the critical chi-square value, we reject the null hypothesis, which means at least one instrument is correlated with the error and therefore invalid (Gujarati 2003). Data Specific data was required to conduct this analysis. Firstly, it was important to observe participants and non-participants in extension programs. This was the basis of our endogenous variable. A binary variable was established based on extension participation with Teagasc coded with a value of 1 for clients, and 0 otherwise. Thus, non-participants were identified as all observations that are not recorded as Teagasc clients. While this binary variable is imperfect in the sense of not differentiating between types of engagement as noted previously, it does provide an initial aggregated value for extension in terms of extension participation that is testable, which is a common approach in the literature. Teagasc clients include annual contract holders as well as those that use the service for scheme assistance, defined as participation solely for the purpose of fulfilling bureaucratic requirements to receive subsidy payments as opposed to technical advice. Annual contracts involve “packages” of services that include combinations of events, discussion groups, farm walks, consultations, and news. Farm family income, which includes subsidies, was used as the dependent variable as it provides a useful barometer of performance, particularly over time, and reflects the total income accrued to the household related to farm-based activities. In addition, this measure also captures the effect of extension participation, regardless of which type of service was adopted. Furthermore, additional factors that influence farm income, such as the geographic location, farm system, and other characteristics were also included. These data were derived from the Teagasc National Farm Survey (NFS). The Teagasc NFS is an annual panel data source collected as part of the Farm Accountancy Data Network (FADN) required by the European Union. The NFS consists of approximately 1,100 farms per annum representing approximately 110,000 farms of the population in Ireland (Läpple and Hennessy 2015). This dataset determines the financial situation on Irish farms by measuring the level of gross output, margins, costs, income, investment, and indebtedness across the spectrum of farming systems, sizes and profiles in the various regions (Connolly et al. 2010). Panel data allows the tracking of the same farms over time, which enriches analysis of this type as some farms may opt in, opt out, avoid, return, or engage constantly with extension services. For the purpose of this analysis, and given that the introduction of the Single Farm Payment policy change in 2005 was chosen as an instrument, the sample selected prioritized farms that were in the sample prior to this period, but only became Teagasc clients subsequently. This refines the sample, and exploits a natural experiment by causing exogenous variation in an otherwise endogenous variable (Wooldridge 2010). This ensures we focus on a similar group of farmers who were likely to have been motivated to become clients in response to the policy change. Including the full sample increases the variation in the results due to the inclusion of “self-selecting” farmers who may have participated into extension services for reasons outlined above regardless of the policy change (Läpple and Hennessy 2015). Each observation employed an average population weight recorded on the NFS over the period to ensure the sample was representative of the general farming population (Buckley et al. 2016). Accordingly, the final sample size is 8,951 observations over the years 2000–2013 inclusive for the models estimated. The estimated coefficients are therefore an average estimation across the 14 years of the analysis. The dependent variable was family farm income, which was divided by utilizable agricultural hectares to obtain a more representative indicator to remove potential skewness from diverse farm sizes. This provided a per hectare based estimation for impact, at the expense of accommodating individual farm expansions. However, given the somewhat stifled land market with low levels of sales and within-family transfers and strong attachment to land ownership in Ireland (Hennessy and Rehman 2007; Leonard et al. 2017), this was not considered an obstruction to the analysis. Family farm income is defined as gross output less net expenses (direct and overhead costs) and includes subsidy receipts. It does not include off-farm income. Including subsidies gives a more inclusive portrayal of the overall farm income attributable to the enterprise, and also captures farmers who participate in extension for the primary purpose of scheme assistance, as well as those involved in more technical-based programs. Given that assistance with Single Farm Payment applications is an important extension service, including the subsidy effect in the dependent variable was prudent. Intuitively, a farmer receives a financial gain directly attributable to this process in a given year. However, as the payment was decoupled from production under the reform of the EU’s Common Agricultural Policy, this subsidy effect would not vary over time, and thus differences in farm income would be based on other factors associated with farm performance. Nonetheless, the inclusion of the subsidy is likely to make a significant contribution to farm income. The dependent variable was transformed using its natural logarithm to remove the influence of outliers in the sample, to smooth the distribution of the data, and to enable us to interpret our coefficients as percentages. The main independent or explanatory variable selected for the regression analysis was extension participation as a binary variable. This enabled the division of the sample into those that participated in extension over time and those that did not. Thus, participation was recorded with a value of 1 for any level of participation in Teagasc extension and 0 otherwise. This aggregated form of the variable ensured that a level of impact across all forms of participation was captured ensuring the public expenditure “value for money” element was assessed. Farm system, stocking density, soil type, land value, labor, age, off-farm employment and size were included as controls on the prior expectation that they would affect farm performance. These controls also strengthen the instruments’ exogeneity condition as each control reduces the effect of the error term on the instruments. For the first instrument, distance to the local Teagasc advisory office was expected to negatively influence the decision to participate in extension services (relevant), but not to affect personal characteristics of the farmer such as their innate ability or motivation that would affect farm family income (valid). Thus, it was expected to be correlated with the decision to participate in extension, but be exogenous to the omitted variables contained in the error term. For example, a farmer located a significant distance from a local office may choose not to participate with extension services, but this distance is unlikely to affect their farming capabilities or farm performance. Moreover, the location of farms in Ireland is largely due to inheritance, as opposed to a conscious choice of where to locate, and thus the distance to a local office is largely exogenous. Furthermore, Läpple and Hennessy (2015) argued that advisory offices may be strategically located in advantaged regions where impact is likely to be pronounced. For example, areas with suitable soil for intensive production may be targeted as ideal locations to base the provision of extension services, given dissemination benefits and likelihood of participation. Indeed, Anderson and Feder (2004) suggested that economies of scale were lost and coordination difficulties rose as advisory offices became decentralized from selected locations. However, in the case of Ireland the number and geographic spread of offices reduced this effect as farms were an average of 10.7 km from their nearest office, with a maximum distance recorded as 44.4 km pre 2009 and an average of 11.7 km with a maximum of 62.2 km thereafter when the number of Teagasc advisory offices were rationalized. This instrument was calculated by measuring the geographic distance from each observation to the nearest local advisory office. This was done by taking the geographic coordinates of each local office, and measuring the distance from the geo-referenced codes for each observation derived from the Teagasc NFS. The distance to office can be classified as random given the nationwide spread of offices as evident from the map in figure 1. Figure 1 View largeDownload slide Location of public extension offices in Ireland 2000-2013 Figure 1 View largeDownload slide Location of public extension offices in Ireland 2000-2013 The second instrument was the policy shock caused by the introduction of the Single Farm Payment Scheme in 2005, which replaced previous coupled payment schemes with a decoupled payment based on average subsidy receipts calculated over a historical reference period (2000-2002). In other words, the value of this annual flat payment was based on production decisions taken in that period (O’Donoghue and Hennessy 2015) and therefore did not vary on a current annual basis. Although additional schemes were included in the new policy that may have asymmetrically affected farm systems differently such as cross compliance, the increased number of clients was relatively symmetrical across all systems. This increase ranged from 9% to 12.5% across all systems after the new policy was introduced. Overall, Teagasc (2006) reported a 20% increase in client numbers credited to the complexities of the new scheme in 2005. Accordingly, the decision to participate in extension was influenced by the new scheme but did not directly affect the current farm performance. In addition, the timing of the introduction of the scheme was exogenous, as it was decided by policy makers and not farmers. Thus it also fulfils the exogeneity requirement for IV analysis. This binary variable was developed by assigning a value of 1 if the year was after 2005 and a 0 if before. The two instruments were also interacted to examine the impact of distance given the policy change. In other words, was the effect of distance on the decision to participate in extension less influential once the new policy was introduced? It is expected that it would remain a negative relationship but not as pronounced in magnitude given the increased numbers of clients. This addition helped to improve the functional form and robustness of the final model. Although it is important to be cautious with regard to the claim of validity with regard to instruments, formulating a strong argument in their favor is helpful (Murray 2006; Cawley and Meyerhoefer 2012). Thus, the instruments utilized here are based on the arguments above and these claims are further validated in the diagnostic results in terms of relevance and validity. Given these prerequisites, we have confidence that the instruments applied are effective in combating endogeneity in the results. Summary statistics of all variables are provided in table 1. Table 1 Data Description and Summary Statistics Variable Description Mean SD Min. Max. FFI/ha Family farm income (€) per ha 456.60 434.52 −1,798.38 3,572.29 Ln FFI/ha Log of family farm income per ha 5.92 0.94 −1.63 8.18 Advisory Participation = 1 if Teagasc client 0.54 0.50 0 1 Ln Land Value/ha Log of land value per ha −0.11 0.56 −4.80 2.61 Farm Size No. of utilisable hectares 38.03 34.55 2.80 1,116.58 Stocking Density Total Livestock Units per ha 1.36 0.63 0 4.80 Ln Labour Log of unpaid family labour −0.06 0.51 −4.61 1.43 Age Age of farmer 55.13 12.22 17 90 Years Agri ed = 0.5 if short course ; = 2 if ag cert; = 4 if ag university 0.68 0.98 0 4 System: Dairy = 1 if dairy 0.12 0.32 0 1 Dairy & Other = 1 if dairy & other 0.07 0.26 0 1 Cattle Rearing = 1 if cattle rearing 0.17 0.38 0 1 Cattle Other = 1 if cattle other 0.20 0.40 0 1 Mainly Sheep = 1 if mainly sheep 0.12 0.33 0 1 Tillage = 1 if tillage 0.05 0.21 0 1 Region: Border = 1 if farm is in the border region 0.20 0.40 0 1 Dublin = 1 if farm is in the Dublin region 0.01 0.10 0 1 East = 1 if farm is in the east region 0.09 0.28 0 1 Midlands = 1 if farm is in the midlands 0.11 0.31 0 1 Southwest = 1 if farm is in the southwest 0.10 0.30 0 1 Southeast = 1 if farm is in the southeast 0.14 0.35 0 1 South = 1 if farm is in the south region 0.18 0.38 0 1 Medium Soil 0.40 0.49 0 1 Poor Soil 0.12 0.33 0 1 Dist_advoff Distance to advisory office (km) 10.71 8.63 0 62.16 SFPyr = 1 if advisory client after 2005 0.66 0.48 0 1 SFPYR*Dist Interactive term for clients and distance 7.40 9.16 0 62.16 Variable Description Mean SD Min. Max. FFI/ha Family farm income (€) per ha 456.60 434.52 −1,798.38 3,572.29 Ln FFI/ha Log of family farm income per ha 5.92 0.94 −1.63 8.18 Advisory Participation = 1 if Teagasc client 0.54 0.50 0 1 Ln Land Value/ha Log of land value per ha −0.11 0.56 −4.80 2.61 Farm Size No. of utilisable hectares 38.03 34.55 2.80 1,116.58 Stocking Density Total Livestock Units per ha 1.36 0.63 0 4.80 Ln Labour Log of unpaid family labour −0.06 0.51 −4.61 1.43 Age Age of farmer 55.13 12.22 17 90 Years Agri ed = 0.5 if short course ; = 2 if ag cert; = 4 if ag university 0.68 0.98 0 4 System: Dairy = 1 if dairy 0.12 0.32 0 1 Dairy & Other = 1 if dairy & other 0.07 0.26 0 1 Cattle Rearing = 1 if cattle rearing 0.17 0.38 0 1 Cattle Other = 1 if cattle other 0.20 0.40 0 1 Mainly Sheep = 1 if mainly sheep 0.12 0.33 0 1 Tillage = 1 if tillage 0.05 0.21 0 1 Region: Border = 1 if farm is in the border region 0.20 0.40 0 1 Dublin = 1 if farm is in the Dublin region 0.01 0.10 0 1 East = 1 if farm is in the east region 0.09 0.28 0 1 Midlands = 1 if farm is in the midlands 0.11 0.31 0 1 Southwest = 1 if farm is in the southwest 0.10 0.30 0 1 Southeast = 1 if farm is in the southeast 0.14 0.35 0 1 South = 1 if farm is in the south region 0.18 0.38 0 1 Medium Soil 0.40 0.49 0 1 Poor Soil 0.12 0.33 0 1 Dist_advoff Distance to advisory office (km) 10.71 8.63 0 62.16 SFPyr = 1 if advisory client after 2005 0.66 0.48 0 1 SFPYR*Dist Interactive term for clients and distance 7.40 9.16 0 62.16 Note: All variables were recorded over the period 2000 to 2013. Given this data, IV models were estimated by analyzing the impact on farm family income. Table 1 Data Description and Summary Statistics Variable Description Mean SD Min. Max. FFI/ha Family farm income (€) per ha 456.60 434.52 −1,798.38 3,572.29 Ln FFI/ha Log of family farm income per ha 5.92 0.94 −1.63 8.18 Advisory Participation = 1 if Teagasc client 0.54 0.50 0 1 Ln Land Value/ha Log of land value per ha −0.11 0.56 −4.80 2.61 Farm Size No. of utilisable hectares 38.03 34.55 2.80 1,116.58 Stocking Density Total Livestock Units per ha 1.36 0.63 0 4.80 Ln Labour Log of unpaid family labour −0.06 0.51 −4.61 1.43 Age Age of farmer 55.13 12.22 17 90 Years Agri ed = 0.5 if short course ; = 2 if ag cert; = 4 if ag university 0.68 0.98 0 4 System: Dairy = 1 if dairy 0.12 0.32 0 1 Dairy & Other = 1 if dairy & other 0.07 0.26 0 1 Cattle Rearing = 1 if cattle rearing 0.17 0.38 0 1 Cattle Other = 1 if cattle other 0.20 0.40 0 1 Mainly Sheep = 1 if mainly sheep 0.12 0.33 0 1 Tillage = 1 if tillage 0.05 0.21 0 1 Region: Border = 1 if farm is in the border region 0.20 0.40 0 1 Dublin = 1 if farm is in the Dublin region 0.01 0.10 0 1 East = 1 if farm is in the east region 0.09 0.28 0 1 Midlands = 1 if farm is in the midlands 0.11 0.31 0 1 Southwest = 1 if farm is in the southwest 0.10 0.30 0 1 Southeast = 1 if farm is in the southeast 0.14 0.35 0 1 South = 1 if farm is in the south region 0.18 0.38 0 1 Medium Soil 0.40 0.49 0 1 Poor Soil 0.12 0.33 0 1 Dist_advoff Distance to advisory office (km) 10.71 8.63 0 62.16 SFPyr = 1 if advisory client after 2005 0.66 0.48 0 1 SFPYR*Dist Interactive term for clients and distance 7.40 9.16 0 62.16 Variable Description Mean SD Min. Max. FFI/ha Family farm income (€) per ha 456.60 434.52 −1,798.38 3,572.29 Ln FFI/ha Log of family farm income per ha 5.92 0.94 −1.63 8.18 Advisory Participation = 1 if Teagasc client 0.54 0.50 0 1 Ln Land Value/ha Log of land value per ha −0.11 0.56 −4.80 2.61 Farm Size No. of utilisable hectares 38.03 34.55 2.80 1,116.58 Stocking Density Total Livestock Units per ha 1.36 0.63 0 4.80 Ln Labour Log of unpaid family labour −0.06 0.51 −4.61 1.43 Age Age of farmer 55.13 12.22 17 90 Years Agri ed = 0.5 if short course ; = 2 if ag cert; = 4 if ag university 0.68 0.98 0 4 System: Dairy = 1 if dairy 0.12 0.32 0 1 Dairy & Other = 1 if dairy & other 0.07 0.26 0 1 Cattle Rearing = 1 if cattle rearing 0.17 0.38 0 1 Cattle Other = 1 if cattle other 0.20 0.40 0 1 Mainly Sheep = 1 if mainly sheep 0.12 0.33 0 1 Tillage = 1 if tillage 0.05 0.21 0 1 Region: Border = 1 if farm is in the border region 0.20 0.40 0 1 Dublin = 1 if farm is in the Dublin region 0.01 0.10 0 1 East = 1 if farm is in the east region 0.09 0.28 0 1 Midlands = 1 if farm is in the midlands 0.11 0.31 0 1 Southwest = 1 if farm is in the southwest 0.10 0.30 0 1 Southeast = 1 if farm is in the southeast 0.14 0.35 0 1 South = 1 if farm is in the south region 0.18 0.38 0 1 Medium Soil 0.40 0.49 0 1 Poor Soil 0.12 0.33 0 1 Dist_advoff Distance to advisory office (km) 10.71 8.63 0 62.16 SFPyr = 1 if advisory client after 2005 0.66 0.48 0 1 SFPYR*Dist Interactive term for clients and distance 7.40 9.16 0 62.16 Note: All variables were recorded over the period 2000 to 2013. Given this data, IV models were estimated by analyzing the impact on farm family income. Results Although we expect that OLS estimation would lead to bias in the results, these estimates are presented in addition to the IV estimates to illustrate the scale of the difference between both estimation methods when endogeneity concerns are addressed. Furthermore, the three instruments were inserted cumulatively to monitor each effect on the dependent variable and subsequent diagnostic statistics. Thus, in total four models were estimated, one using OLS, and the others using IV with one, two, and three instruments inserted, respectively. IV Results – First Stage Results The results of the first stage of the IV process are presented in table 2, confirming the relevance of the instruments on extension participation decisions. Table 2 First Stage Results of IV: Advisory Participation 1 Instrument 2 Instruments 3 Instruments Advisory Participation Coeff. (SE) p value Coeff. (SE) p value Coeff. (SE) p value SFP Policy Change 0.530 (0.010) .000 0.531 (0.010) .000 0.565 (0.016) .000 Dist. Adv Office −0.002 (0.001) .000 0.001 (0.001) .649 Interaction Term −0.004 (0.001) .004 CD Wald F Stat 2639.20 1331.86 891.43 1 Instrument 2 Instruments 3 Instruments Advisory Participation Coeff. (SE) p value Coeff. (SE) p value Coeff. (SE) p value SFP Policy Change 0.530 (0.010) .000 0.531 (0.010) .000 0.565 (0.016) .000 Dist. Adv Office −0.002 (0.001) .000 0.001 (0.001) .649 Interaction Term −0.004 (0.001) .004 CD Wald F Stat 2639.20 1331.86 891.43 Note: The n = 8,951, endogenous regressor (Advisory Participation), 3 instruments (Single Farm Payment year, Distance to advisory office and Interaction of both), additional explanatory variables included land value, farm system, labor, size, off-farm job, age, region, stocking density, and soil group. A P value <.01 indicates statistical significance at the 1% level, and the Cragg-Donald Wald F Stat measures relevancy of instruments Table 2 First Stage Results of IV: Advisory Participation 1 Instrument 2 Instruments 3 Instruments Advisory Participation Coeff. (SE) p value Coeff. (SE) p value Coeff. (SE) p value SFP Policy Change 0.530 (0.010) .000 0.531 (0.010) .000 0.565 (0.016) .000 Dist. Adv Office −0.002 (0.001) .000 0.001 (0.001) .649 Interaction Term −0.004 (0.001) .004 CD Wald F Stat 2639.20 1331.86 891.43 1 Instrument 2 Instruments 3 Instruments Advisory Participation Coeff. (SE) p value Coeff. (SE) p value Coeff. (SE) p value SFP Policy Change 0.530 (0.010) .000 0.531 (0.010) .000 0.565 (0.016) .000 Dist. Adv Office −0.002 (0.001) .000 0.001 (0.001) .649 Interaction Term −0.004 (0.001) .004 CD Wald F Stat 2639.20 1331.86 891.43 Note: The n = 8,951, endogenous regressor (Advisory Participation), 3 instruments (Single Farm Payment year, Distance to advisory office and Interaction of both), additional explanatory variables included land value, farm system, labor, size, off-farm job, age, region, stocking density, and soil group. A P value <.01 indicates statistical significance at the 1% level, and the Cragg-Donald Wald F Stat measures relevancy of instruments The above table shows there is a jointly significant relationship between the instruments and the endogenous regressor. Individually, when one instrument is applied, the policy change is a significant explanatory factor in the decision to participate in extension services. When the distance to local office is added, both instruments remain significant at the 1% level and the signs are as expected, with the policy change being positive and the distance being negative. However, the magnitude of the distance instrument is relatively small. This could be due to the fact that there were 90 local Teagasc offices in existence before a restructuring plan was introduced in 2009. Thus, the average distance increased after the office closures, as noted previously. Accordingly, the relative distances to local offices were not practically large. Once all three instruments are applied, the distance becomes positive and insignificant. However, as the interactive term is included, this variable becomes significant and negative as expected, showing the conditional influence of distance after the introduction of the policy change. The Cragg-Donald Wald F statistic illustrates the joint significance of the instruments and shows a strong positive relationship between all 3 instruments and the dependent variable in the first stage. In our case, Stock, Wright, and Yogo (2002) limit of 10 is easily exceeded and we can conclude that the instruments are relevant. IV Results – Second Stage Results The second stage of the IV process involved inserting the predicted values of the endogenous regressors (now exogenous) from the first stage into the main structural equation and applying them to the dependent variable. Accordingly, the results of the IV estimates for the main variable of interest are presented along with selected diagnostics in table 3 for clarity with the full table of results, including the control variable estimates available in appendix A. The OLS estimates are also included for comparison. Table 3 IV Parameter Estimates: Model of Log of Farm Family Income per Ha OLS IV – 1 Instrument IV – 2 Instruments IV – 3 Instruments Extension Participation 0.19*** (0.02) 0.35*** (0.04) 0.35*** (0.04) 0.35*** (0.04) R2 0.22 Centred R2 0.22 0.22 0.22 Sargan p value 0.00 0.81 0.30 OLS IV – 1 Instrument IV – 2 Instruments IV – 3 Instruments Extension Participation 0.19*** (0.02) 0.35*** (0.04) 0.35*** (0.04) 0.35*** (0.04) R2 0.22 Centred R2 0.22 0.22 0.22 Sargan p value 0.00 0.81 0.30 Note: The n = 8,951, additional explanatory variables included year, land value, farm system, labor, size, off-farm job, age, region, stocking density and soil group; standard errors appear in parenthesis; Asterisks represent statistical significance of p values: *** for 1% significance; ** for 5% significance; and * for 10% significance. Full tables of results available in appendix A; Sargan Overidentification P Value >.1 means that we fail to reject that the null of the instruments are valid, not applicable with 1 instrument as the equation is exactly identified. Table 3 IV Parameter Estimates: Model of Log of Farm Family Income per Ha OLS IV – 1 Instrument IV – 2 Instruments IV – 3 Instruments Extension Participation 0.19*** (0.02) 0.35*** (0.04) 0.35*** (0.04) 0.35*** (0.04) R2 0.22 Centred R2 0.22 0.22 0.22 Sargan p value 0.00 0.81 0.30 OLS IV – 1 Instrument IV – 2 Instruments IV – 3 Instruments Extension Participation 0.19*** (0.02) 0.35*** (0.04) 0.35*** (0.04) 0.35*** (0.04) R2 0.22 Centred R2 0.22 0.22 0.22 Sargan p value 0.00 0.81 0.30 Note: The n = 8,951, additional explanatory variables included year, land value, farm system, labor, size, off-farm job, age, region, stocking density and soil group; standard errors appear in parenthesis; Asterisks represent statistical significance of p values: *** for 1% significance; ** for 5% significance; and * for 10% significance. Full tables of results available in appendix A; Sargan Overidentification P Value >.1 means that we fail to reject that the null of the instruments are valid, not applicable with 1 instrument as the equation is exactly identified. The results presented here show that there are consistent positive returns to participating with extension services and all are significant across all models. The OLS results indicate a 19% increase in farm family income per hectare, ceteris paribus. However, as this variable suffers an endogeneity bias, the coefficients estimated for the IV models offer a more accurate prediction, and as evidenced in the table, this return is approximately 35% across the three models, with instruments added cumulatively. Estimation with the distance to advisory office also showed a positive impact but its statistical significance was not strong due to the weaker relevance of the instrument. These findings are in line with previous findings in the literature (26% in Akobundu et al. 2004; 61% in Davis et al. 2012; 55% in O’Donoghue and Hennessy 2015). However, as this measure does not distinguish between the type of extension activity, the precision of individual impact is not captured. Furthermore, as the purpose of this study was to include all forms of extension participation, this measure includes subsidy receipts which, although an annual flat rate payment, contribute substantially to family farm income, particularly across all farm systems (Hanrahan et al. 2014). The consistency of the estimates justifies the instruments as being strong predictors of extension participation, and thus instrumenting it successfully to identify the causal impact of extension participation on farm family income per hectare. Interestingly, these effects are estimated alongside the control variables as noted above. Of those controls, farm system, education, stocking density, land quality, and off-farm employment were important predictors of family farm income, as expected. For example, dairy farmers showed an increase of 37% per hectare compared to non-dairy systems under the IV approach as opposed to 34% under OLS. Similarly, tillage systems increased their impact under the IV to 29% as opposed to 26% under OLS. Conversely, the effect of farm size was insignificant, possibly due to the per hectare basis of the dependent variable, and there was mixed evidence for different regions. Most other variables experienced a minimal change under both systems, but extension participation showed the largest increase. The full table of results for each model is provided in appendix A. Furthermore, the Sargan test statistics report p values exceeding significance values of 0.1, meaning that we fail to reject the null hypothesis that the instruments are valid. Indeed, as it is impossible to prove the null hypothesis of no effect based on the nature of the unobserved error term (Cawley and Meyerhoefer 2012), the Sargan statistic reverses the process and in this case the null is that the instruments have an effect, and we fail to reject that claim. In other words, these instruments can be argued as valid in addressing the endogeneity issue of extension participation, and thus our estimates provide a consistent and positive impact on farm family income per hectare. Given the results of the analyses we can infer that extension services had a positive impact on farm income. Discussion and Conclusion While much of the previous literature identified a positive relationship between participation in extension services and farm level outcomes, the IV approach presented here provides a robust estimation that comprehensively addresses endogeneity concerns. In line with previous literature, the IV estimates of the impact of extension services are uniformly higher than the OLS estimates (Card 1993; Cawley and Meyerhoefer 2012). Therefore, there is a clear indication of a net benefit to extension participation in the first instance. Identifying appropriate instruments is the critical challenge of the IV approach, particularly in terms of confirming exogeneity to the error term, and thus estimates should be interpreted with caution (Cawley and Meyerhoefer 2012). However, in this analysis the distance to the local advisory office and policy change instruments were relevant to the decision to participate in extension and valid in terms of uncorrelated to the dependent variable and error term based on both argument and the diagnostic tests, and thus, the results presented are superior estimates of the impact of extension services on farm income. In terms of policy implications, the results suggest a positive causal effect of participating in extension programs in terms of farm income. Furthermore, these findings are unbiased, implying that the impact is greater than previously thought and therefore the role of extension as a policy tool is effective. Thus, knowledge sharing of the benefits of participation should be more widely promoted among farmers to encourage increased engagement that can be leveraged to achieve policy goals. For example, if governments aim to grow agricultural output, an extension program could be targeted to provide technical assistance to bolster productivity through improved feed management to increase overall output levels. Conversely, farms could be targeted with environmental advice to curb greenhouse gas emissions as set out in international regulations. A more targeted approach that incentivizes deeper forms of engagement may be valuable (Akobundu et al. 2004), but this requires further analysis to identify the effect of different levels of extension engagement. Nonetheless, policy targets set for the agricultural sector by government should be adequately supported by a dynamic effective extension service. This study provides a robust estimation of the impact of extension services across all farming sectors on an aggregated basis in Ireland. However, a number of caveats should be considered when utilizing an IV approach, and further research is needed to enforce the findings provided here. Firstly, when applying the IV approach, the validity of the instruments is key. Although we have confidence and the results defend their validity, it is logical to assume there may be alternative instruments that could be used that are not available in this dataset. For example, the neighbor or peer effect could have been instrumented as farmers may be more likely to become clients based on their peers’ participation. Similarly, the availability of advisers could have been a useful instrument given the drop in numbers due to retirements over the same period reducing the availability of services. However, both of these instruments were not possible due to data limitations. Moreover, the distance to advisory office instrument could prove more effective if the offices were classified according to the types of services offered, as smaller offices may not have the facilities to address more intensive forms of extension participation. Similarly, the endogenous variable used here is a dummy and thus does not reflect the variability of extension programs available, and thus cannot rigorously distinguish different types of extension participation nor assess their impact. Distinguishing on the basis of the motivation to engage would enrich the analysis further. Indeed, the knowledge transfer effect is not homogenous, but more likely to be heterogeneous in nature (Coccia 2008). However, this research provides a foundation to build upon as the IV approach delivers a more precise and unbiased estimate of the positive effect of extension advice to farms, ensuring that the “value for money” objective for policy makers is met. This is an important first step in the evaluation of impact, and future research could aim to identify the impact of different types of extension. References Abdallah W. , Goergen M. , O’Sullivan N . 2015 . Endogeneity: How Failure to Correct for It Can Cause Wrong Inferences and Some Remedies . British Journal of Management 26 : 791 – 804 . Google Scholar CrossRef Search ADS Akobundu E. , Alwang J. , Essel A. , Norton G.W. , Tegene A . 2004 . Does Extension Work? Impacts of a Program to Assist Limited-Resource Farmers in Virginia . Applied Economic Perspectives and Policy 26 : 361 – 72 . Anderson J.R. , Feder G . 2004 . Agricultural Extension: Good Intentions and Hard Realities . The World Bank Research Observer 19 : 41 – 60 . Google Scholar CrossRef Search ADS Birkhaeuser D. , Evenson R.E. , Feder G . 1991 . The Economic Impact of Agricultural Extension: A Review . Economic Development and Cultural Change 39 : 607 – 50 . Google Scholar CrossRef Search ADS Birner R. , Davis K. , Pender J. , Nkonya E. , Anandajayasekeram P. , Ekboir J. , Mbabu A. , et al. 2009 . From Best Practice to Best Fit: A Framework for Designing and Analyzing Pluralistic Agricultural Advisory Services Worldwide . The Journal of Agricultural Education and Extension 15 : 341 – 55 . Google Scholar CrossRef Search ADS Boyle G. 2008 . A Future Driven by Knowledge and Innovation. In Growing Knowledge - Fifty Years of Research and Development in Irish Farming and Food , ed. Miley M. , 193 – 205 . Carlow : Teagasc Publications Office . Buckley C. , Wall D.P. , Moran B. , O’Neill S. , Murphy P.N.C . 2016 . Phosphorous Management on Irish Dairy Farms Post Controls Introduced under the EU Nitrates Directive . Agricultural System 142 : 1 – 8 . Google Scholar CrossRef Search ADS Burgess S. , Dudbridge F. , Thompson S.G . 2016 . Combining Information on Multiple Instrumental Variables in Mendelian Randomization: Comparison of Allele Score and Summarized Data Methods . Statistics in Medicine 35 : 1880 – 906 . Google Scholar CrossRef Search ADS PubMed Caliendo M. , Kopeinig S . 2008 . Some Practical Guidance for the Implementation of Propensity Score Matching . Journal of Economic Surveys 22 : 31 – 82 . Google Scholar CrossRef Search ADS Callan T. , Harmon C . 1999 . The Economic Return to Schooling in Ireland . Labour Economics 6 : 543 – 50 . Google Scholar CrossRef Search ADS Card D. 1993 . Using Geographic Variation in College Proximity to Estimate the Return to Schooling. NBER Working Paper no. 4483, National Bureau of Economic Research, Cambridge, MA. Card D. 1999 . The Causal Effect of Education on Earnings . Handbook of Labor Economics 3 : 1801 – 63 . Google Scholar CrossRef Search ADS Cawley J. , Meyerhoefer C . 2012 . The Medical Care Costs of Obesity: An Instrumental Variables Approach . Journal of Health Economics 31 : 219 – 30 . Google Scholar CrossRef Search ADS PubMed Coccia M. 2008 . Spatial Mobility of Knowledge Transfer and Absorptive Capacity: Analysis and Measurement of the Impact within the Geoeconomic Space . Journal of Technology Transfer 33 : 105 – 22 . Google Scholar CrossRef Search ADS Connolly L. , Kinsella A. , Quinlan A.C. , Moran B . 2010 . National Farm Survey. Carlow : Teagasc Publications Office . Davis K. , Nkonya E. , Kato E. , Mekonnen D.A. , Odendo M. , Miiro R. , Nkuba J . 2012 . Impact of Farmer Field Schools on Agricultural Productivity and Poverty in East Africa . World Development 40 : 402 – 13 . Google Scholar CrossRef Search ADS Dercon S. , Gilligan D. , Hoddinott J. , Woldehanna T . 2009 . The Impact of Agricultural Extension and Roads on Poverty and Consumption Growth in Fifteen Ethiopian Villages . American Journal of Agricultural Economics 91 : 1007 – 21 . Google Scholar CrossRef Search ADS Donnellan T. , Hanrahan K. , Hennessy T . 2008 . Study on the Functioning of Land Markets in the EU Member States under the Influence of Measures Applied under the Common Agricultural Policy, RMIS 5881. Carlow: Teagasc Publications Office. Gabel M. , Scheve K . 2007 . Estimating the Effect of Elite Communications on Public Opinion Using Instrumental Variables . American Journal of Political Science 51 : 1013 – 28 . Google Scholar CrossRef Search ADS Garforth C. , Angell B. , Archer J. , Green K . 2003 . Fragmentation or Creative Diversity? Options in the Provision of Land Management Advisory Services . Land Use Policy 20 : 323 – 33 . Google Scholar CrossRef Search ADS Griliches Z. 1964 . Research Expenditures, Education, and the Aggregate Agricultural Production Function . The American Economic Review 54 : 961 – 74 . Gujarati D.N. 2003 . Basic Econometrics . New York : McGraw Hill . Hanrahan K. , Hennessy T. , Kinsella A. , Moran B . 2014 . Teagasc National Farm Survey 2013 . Carlow : Teagasc Publications Office . Heanue K. , O'Donoghue C . 2014 . The Economic Returns to Formal Agricultural Education. Carlow : Teagasc Publications Office . Hennessy T. , Heanue K . 2012 . Quantifying the Effect of Discussion Group Membership on Technology Adoption and Farm Profit on Dairy Farms . The Journal of Agricultural Education and Extension 18 : 41 – 54 . Google Scholar CrossRef Search ADS Hennessy T. , Rehman T. 2007 . An Investigation into Factors Affecting the Occupational Choices of Nominated Farm Heirs in Ireland . Journal of Agricultural Economics 58 : 61 – 75 . Google Scholar CrossRef Search ADS Howley P. , O’Neill S. , Atkinson R . 2015 . Who Needs Good Neighbors? Environment and Planning A 47 : 939 – 56 . Google Scholar CrossRef Search ADS Hunt W. , Birch C. , Vanclay F. , Coutts J . 2014 , Recommendations Arising from an Analysis of Changes to the Australian Agricultural Research, Development and Extension System . Food Policy 44 : 129 – 41 . Google Scholar CrossRef Search ADS Imbens G.W. , Wooldridge J.M . 2009 . Recent Developments in the Econometrics of Program Evaluation . Journal of Economic Literature 47 : 5 – 86 . Google Scholar CrossRef Search ADS Krishnan P. , Patnam M . 2014 . Neighbours and Extension Agents in Ethiopia: Who Matters More for Technology Adoption? American Journal of Agricultural Economics 96 : 308 – 327 . Google Scholar CrossRef Search ADS Läpple D. , Hennessy T . 2015 . Assessing the Impact of Financial Incentives in Extension Programs: Evidence from Ireland . Journal of Agricultural Economics 66 : 781 – 95 . Google Scholar CrossRef Search ADS Läpple D. , Hennessy T. 2015 . Exploring the Role of Incentives in Agricultural Extension Programs . Applied Economic Perspectives and Policy 37 : 403 – 17 . Google Scholar CrossRef Search ADS Läpple D. , Hennessy T. , Newman C . 2013 . Quantifying the Economic Return to Participatory Extension Programs in Ireland: an Endogenous Switching Regression Analysis . Journal of Agricultural Economics 64 : 467 – 82 . Google Scholar CrossRef Search ADS Leonard B. , Mahon M. , Kinsella A. , O’Donoghue C. , Farrell M. , Curran T. , Hennessy T . 2017 . The Potential of Farm Partnerships to Facilitate Farm Succession and Inheritance International Journal of Agricultural Management 6 : 1 – 16 . Marsh S.P. , Pannell D.J. , Linder R.K . 2004 . Does Agricultural Extension Pay? Agricultural Economics 30 : 17 – 30 . Miguel E. , Satyanath S. , Sergenti E . 2004 . Economic Shocks and Civil Conflict: An Instrumental Variables Approach . Journal of political Economy 112 : 725 – 53 . Google Scholar CrossRef Search ADS Murray M.P. 2006 . Avoiding Invalid Instruments and Coping with Weak Instruments . The Journal of Economic Perspectives 20 : 111 – 32 . Google Scholar CrossRef Search ADS Nordin M. , Höjgård S. 2016 . An Evaluation of Extension Services in Sweden . Agricultural Economics 48 : 1 – 10 . O’Donoghue C. , Hennessy T . 2015 . Policy and Economic Change in the Agri-Food Sector in Ireland . The Economic and Social Review 46 : 315 – 37 . Owens T. , Hoddinott J. , Kinsey B . 2003 . The Impact of Agricultural Extension on Farm Production in Resettlement Areas of Zimbabwe . Economic Development and Cultural Change 51 : 337 – 57 . Google Scholar CrossRef Search ADS Prager K. , Labarthe P. , Caggiano P.M. , Lorenzo-Arribas A . 2016 . How Does Commercialisation Impact on the Provision of Farm Advisory Services? Evidence from Belgium, Italy, Ireland and the UK . Land Use Policy 52 : 329 – 44 . Google Scholar CrossRef Search ADS Ragasa C. , Ulimwengu J. , Randriamamonjy J. , Badibanga T. , 2016 . Factors Affecting Performance of Agricultural Extension: Evidence from Democratic Republic of Congo . The Journal of Agricultural Education and Extension 22 : 113 – 43 . Google Scholar CrossRef Search ADS Rassen J.A. , Brookhart M.A. , Glynn R.J. , Mittleman M.A. , Schneeweiss S . 2009 . Instrumental Variables I: Instrumental Variables Exploit Natural Variation in Non-experimental Data to Estimate Causal Relationships Journal of Clinical Epidemiology 62 : 1226 – 32 . Google Scholar CrossRef Search ADS PubMed Stock J.H. , Wright J.H. , Yogo M . 2002 . A Survey of Weak Instruments and Weak Identification in Generalized Method of Moments . Journal of Business & Economic Statistics 20 : 518 – 29 . Google Scholar CrossRef Search ADS Stukel T.A. , Fisher E.S. , Wennberg D.E. , Alter D.A. , Gottlieb D.J. , Vermuelen M.J . 2007 . Analysis of Observational Studies in the Presence of Treatment Selection Bias: Effects of Invasive Cardiac Management on AMI survival using Propensity Score and Instrumental Variable Methods . Journal of American Medical Association 297 : 278 – 85 . Google Scholar CrossRef Search ADS Tamini L.D. 2011 . A Nonparametric Analysis of the Impact of Agri-environmental Advisory Activities on Best Management Practice Adoption: A Case Study of Québec . Ecological Economics 70 : 1363 – 74 . Google Scholar CrossRef Search ADS Teagasc . 2006 . Annual Report 2005 and Financial Statements . Carlow : Teagasc Publications Office . Teagasc . 2016 . Our Organization. About Us. Available at: http://www.teagasc.ie/about/our-organization/. Van den Ban A.W. , Hawkins H.S . 1988 . Agricultural Extension . Essex : Longman Scientific & Technical . Vanclay F. , Leach G . 2011 . Enabling Change in Rural and Regional Australia. In Shaping Change: Natural Resource Management, Agriculture and the Role of Extension . Australasia Pacific Extension Network (APEN), ed. Jennings J. , Packham R. , Woodside D. , 6 – 11 . Australasia Pacific Extension Network . Wooldridge J.M. 2013 . Introductory Econometrics - A Modern Approach . South-Western : Cengage . Appendices Appendix A – Full table of OLS and IV estimates (2000-2013) Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Regression Instrumental Variable Regression – 1 Instrument Has Advisory Participation 0.19 0.02 0.00 0.35 0.41 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.34 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.09 0.04 0.03 −0.06 0.04 0.16 Cattle Rearing −0.19 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.19 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.08 0.03 0.02 −0.03 0.03 0.39 Tillage 0.26 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.15 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.13 0.01 0.01 0.08 Age Squared −0.00 0.00 0.11 −0.00 0.00 0.06 Has off farm employment −0.12 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.02 0.02 0.22 0.00 0.02 0.87 Has Forestry −0.05 0.04 0.22 −0.09 0.04 0.05 Completed Agricultural Short Course 0.22 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.20 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.25 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.17 0.04 0.00 −0.14 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.01 0.04 0.88 0.02 0.04 0.66 Cork, Kerry −0.15 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.98 0.00 0.03 0.95 Stocking Density 0.42 0.02 0.00 0.40 0.02 0.00 Medium Soil −0.07 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.01 0.04 0.77 −0.03 0.03 0.37 Constant 5.08 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (Weak Instrument) 2639.20 Sargan statistic p value (Overidentification Test) 0.000 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Regression Instrumental Variable Regression – 1 Instrument Has Advisory Participation 0.19 0.02 0.00 0.35 0.41 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.34 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.09 0.04 0.03 −0.06 0.04 0.16 Cattle Rearing −0.19 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.19 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.08 0.03 0.02 −0.03 0.03 0.39 Tillage 0.26 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.15 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.13 0.01 0.01 0.08 Age Squared −0.00 0.00 0.11 −0.00 0.00 0.06 Has off farm employment −0.12 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.02 0.02 0.22 0.00 0.02 0.87 Has Forestry −0.05 0.04 0.22 −0.09 0.04 0.05 Completed Agricultural Short Course 0.22 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.20 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.25 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.17 0.04 0.00 −0.14 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.01 0.04 0.88 0.02 0.04 0.66 Cork, Kerry −0.15 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.98 0.00 0.03 0.95 Stocking Density 0.42 0.02 0.00 0.40 0.02 0.00 Medium Soil −0.07 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.01 0.04 0.77 −0.03 0.03 0.37 Constant 5.08 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (Weak Instrument) 2639.20 Sargan statistic p value (Overidentification Test) 0.000 Note: The endogenous regressor is (Advisory Participation); 1 instrument (Single Farm Payment year policy change); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak; Sargan Statistic void due to equation being exactly identified. View Large Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Regression Instrumental Variable Regression – 1 Instrument Has Advisory Participation 0.19 0.02 0.00 0.35 0.41 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.34 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.09 0.04 0.03 −0.06 0.04 0.16 Cattle Rearing −0.19 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.19 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.08 0.03 0.02 −0.03 0.03 0.39 Tillage 0.26 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.15 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.13 0.01 0.01 0.08 Age Squared −0.00 0.00 0.11 −0.00 0.00 0.06 Has off farm employment −0.12 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.02 0.02 0.22 0.00 0.02 0.87 Has Forestry −0.05 0.04 0.22 −0.09 0.04 0.05 Completed Agricultural Short Course 0.22 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.20 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.25 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.17 0.04 0.00 −0.14 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.01 0.04 0.88 0.02 0.04 0.66 Cork, Kerry −0.15 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.98 0.00 0.03 0.95 Stocking Density 0.42 0.02 0.00 0.40 0.02 0.00 Medium Soil −0.07 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.01 0.04 0.77 −0.03 0.03 0.37 Constant 5.08 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (Weak Instrument) 2639.20 Sargan statistic p value (Overidentification Test) 0.000 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Regression Instrumental Variable Regression – 1 Instrument Has Advisory Participation 0.19 0.02 0.00 0.35 0.41 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.34 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.09 0.04 0.03 −0.06 0.04 0.16 Cattle Rearing −0.19 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.19 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.08 0.03 0.02 −0.03 0.03 0.39 Tillage 0.26 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.15 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.13 0.01 0.01 0.08 Age Squared −0.00 0.00 0.11 −0.00 0.00 0.06 Has off farm employment −0.12 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.02 0.02 0.22 0.00 0.02 0.87 Has Forestry −0.05 0.04 0.22 −0.09 0.04 0.05 Completed Agricultural Short Course 0.22 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.20 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.25 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.17 0.04 0.00 −0.14 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.01 0.04 0.88 0.02 0.04 0.66 Cork, Kerry −0.15 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.98 0.00 0.03 0.95 Stocking Density 0.42 0.02 0.00 0.40 0.02 0.00 Medium Soil −0.07 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.01 0.04 0.77 −0.03 0.03 0.37 Constant 5.08 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (Weak Instrument) 2639.20 Sargan statistic p value (Overidentification Test) 0.000 Note: The endogenous regressor is (Advisory Participation); 1 instrument (Single Farm Payment year policy change); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak; Sargan Statistic void due to equation being exactly identified. View Large Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – 2 Instruments Instrumental Variable Regression – 3 Instruments Has Advisory Participation 0.35 0.04 0.00 0.35 0.04 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.37 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.06 0.04 0.16 −0.06 0.04 0.16 Cattle Rearing −0.15 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.16 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.03 0.03 0.38 −0.03 0.03 0.37 Tillage 0.29 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.08 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.06 Has off farm employment −0.13 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.00 0.02 0.87 0.00 0.02 0.85 Has Forestry −0.09 0.04 0.05 −0.09 0.04 0.05 Completed Agricultural Short Course 0.21 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.24 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.15 0.04 0.00 −0.15 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.016 0.04 0.66 0.02 0.04 0.66 Cork, Kerry −0.17 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.95 0.00 0.03 0.95 Stocking Density 0.40 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.03 0.03 0.38 Constant 5.05 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 1331.86 891.43 Sargan statistic p value (Overidentification Test) 0.81 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – 2 Instruments Instrumental Variable Regression – 3 Instruments Has Advisory Participation 0.35 0.04 0.00 0.35 0.04 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.37 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.06 0.04 0.16 −0.06 0.04 0.16 Cattle Rearing −0.15 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.16 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.03 0.03 0.38 −0.03 0.03 0.37 Tillage 0.29 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.08 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.06 Has off farm employment −0.13 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.00 0.02 0.87 0.00 0.02 0.85 Has Forestry −0.09 0.04 0.05 −0.09 0.04 0.05 Completed Agricultural Short Course 0.21 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.24 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.15 0.04 0.00 −0.15 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.016 0.04 0.66 0.02 0.04 0.66 Cork, Kerry −0.17 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.95 0.00 0.03 0.95 Stocking Density 0.40 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.03 0.03 0.38 Constant 5.05 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 1331.86 891.43 Sargan statistic p value (Overidentification Test) 0.81 Note: The endogenous regressor is (Advisory Participation in both models); 2 instruments (Single Farm Payment policy change and Distance to advisory office); 3 Instruments (Single Farm Payment policy change, Distance to advisory office and Interaction of both); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors are adjusted for heterogeneity; a p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 considered weak; Sargan Statistic for overidentification p value > 0.1 fails to reject a null hypothesis of the instruments’ validity. View Large Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – 2 Instruments Instrumental Variable Regression – 3 Instruments Has Advisory Participation 0.35 0.04 0.00 0.35 0.04 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.37 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.06 0.04 0.16 −0.06 0.04 0.16 Cattle Rearing −0.15 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.16 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.03 0.03 0.38 −0.03 0.03 0.37 Tillage 0.29 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.08 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.06 Has off farm employment −0.13 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.00 0.02 0.87 0.00 0.02 0.85 Has Forestry −0.09 0.04 0.05 −0.09 0.04 0.05 Completed Agricultural Short Course 0.21 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.24 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.15 0.04 0.00 −0.15 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.016 0.04 0.66 0.02 0.04 0.66 Cork, Kerry −0.17 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.95 0.00 0.03 0.95 Stocking Density 0.40 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.03 0.03 0.38 Constant 5.05 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 1331.86 891.43 Sargan statistic p value (Overidentification Test) 0.81 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – 2 Instruments Instrumental Variable Regression – 3 Instruments Has Advisory Participation 0.35 0.04 0.00 0.35 0.04 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.37 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.06 0.04 0.16 −0.06 0.04 0.16 Cattle Rearing −0.15 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.16 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.03 0.03 0.38 −0.03 0.03 0.37 Tillage 0.29 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.08 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.06 Has off farm employment −0.13 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.00 0.02 0.87 0.00 0.02 0.85 Has Forestry −0.09 0.04 0.05 −0.09 0.04 0.05 Completed Agricultural Short Course 0.21 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.24 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.15 0.04 0.00 −0.15 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.016 0.04 0.66 0.02 0.04 0.66 Cork, Kerry −0.17 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.95 0.00 0.03 0.95 Stocking Density 0.40 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.03 0.03 0.38 Constant 5.05 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 1331.86 891.43 Sargan statistic p value (Overidentification Test) 0.81 Note: The endogenous regressor is (Advisory Participation in both models); 2 instruments (Single Farm Payment policy change and Distance to advisory office); 3 Instruments (Single Farm Payment policy change, Distance to advisory office and Interaction of both); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors are adjusted for heterogeneity; a p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 considered weak; Sargan Statistic for overidentification p value > 0.1 fails to reject a null hypothesis of the instruments’ validity. View Large Appendix B – Full table of IV estimates with 1 instrument only (2000-2013) Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – Sfpyr only Instrumental Variable Regression –1 Dist. only Has Advisory Participation 0.35 0.41 0.00 0.14 0.57 0.79 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.03 0.00 Dairy System 0.37 0.03 0.00 0.34 0.09 0.00 Dairy and Other System −0.06 0.04 0.16 −0.09 0.12 0.42 Cattle Rearing −0.15 0.03 0.00 −0.19 0.15 0.19 Cattle Other System −0.16 0.03 0.00 −0.20 0.14 0.13 Mainly Sheep −0.03 0.03 0.39 −0.09 0.17 0.59 Tillage 0.29 0.05 0.00 0.25 0.09 0.01 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.09 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.12 Has off farm employment −0.13 0.02 0.00 −0.12 0.03 0.00 Log Farm Size 0.00 0.02 0.87 0.03 0.07 0.71 Has Forestry −0.09 0.04 0.05 −0.05 0.12 0.71 Completed Agricultural Short Course 0.21 0.03 0.00 0.23 0.06 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.21 0.07 0.01 Completed Agricultural University 0.24 0.06 0.00 0.26 0.07 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.05 0.00 Clare, Limerick, Tipp. N.R. −0.14 0.04 0.00 −0.17 0.09 0.05 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.02 0.04 0.66 0.00 0.05 0.98 Cork, Kerry −0.17 0.04 0.00 −0.15 0.07 0.04 Galway, Mayo, Roscommon 0.00 0.03 0.95 −0.00 0.03 0.98 Stocking Density 0.40 0.02 0.00 0.42 0.05 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.00 0.08 0.96 Constant 5.05 0.17 0.00 4.97 0.21 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 2639.20 10.78 Sargan statistic p value (Overidentification Test) 0.000 0.000 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – Sfpyr only Instrumental Variable Regression –1 Dist. only Has Advisory Participation 0.35 0.41 0.00 0.14 0.57 0.79 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.03 0.00 Dairy System 0.37 0.03 0.00 0.34 0.09 0.00 Dairy and Other System −0.06 0.04 0.16 −0.09 0.12 0.42 Cattle Rearing −0.15 0.03 0.00 −0.19 0.15 0.19 Cattle Other System −0.16 0.03 0.00 −0.20 0.14 0.13 Mainly Sheep −0.03 0.03 0.39 −0.09 0.17 0.59 Tillage 0.29 0.05 0.00 0.25 0.09 0.01 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.09 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.12 Has off farm employment −0.13 0.02 0.00 −0.12 0.03 0.00 Log Farm Size 0.00 0.02 0.87 0.03 0.07 0.71 Has Forestry −0.09 0.04 0.05 −0.05 0.12 0.71 Completed Agricultural Short Course 0.21 0.03 0.00 0.23 0.06 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.21 0.07 0.01 Completed Agricultural University 0.24 0.06 0.00 0.26 0.07 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.05 0.00 Clare, Limerick, Tipp. N.R. −0.14 0.04 0.00 −0.17 0.09 0.05 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.02 0.04 0.66 0.00 0.05 0.98 Cork, Kerry −0.17 0.04 0.00 −0.15 0.07 0.04 Galway, Mayo, Roscommon 0.00 0.03 0.95 −0.00 0.03 0.98 Stocking Density 0.40 0.02 0.00 0.42 0.05 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.00 0.08 0.96 Constant 5.05 0.17 0.00 4.97 0.21 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 2639.20 10.78 Sargan statistic p value (Overidentification Test) 0.000 0.000 Note: The endogenous regressor is (Advisory Participation in both models); 1instrument in each (Single Farm Payment policy change and Distance to advisory office respectively); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors are adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak. View Large Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – Sfpyr only Instrumental Variable Regression –1 Dist. only Has Advisory Participation 0.35 0.41 0.00 0.14 0.57 0.79 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.03 0.00 Dairy System 0.37 0.03 0.00 0.34 0.09 0.00 Dairy and Other System −0.06 0.04 0.16 −0.09 0.12 0.42 Cattle Rearing −0.15 0.03 0.00 −0.19 0.15 0.19 Cattle Other System −0.16 0.03 0.00 −0.20 0.14 0.13 Mainly Sheep −0.03 0.03 0.39 −0.09 0.17 0.59 Tillage 0.29 0.05 0.00 0.25 0.09 0.01 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.09 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.12 Has off farm employment −0.13 0.02 0.00 −0.12 0.03 0.00 Log Farm Size 0.00 0.02 0.87 0.03 0.07 0.71 Has Forestry −0.09 0.04 0.05 −0.05 0.12 0.71 Completed Agricultural Short Course 0.21 0.03 0.00 0.23 0.06 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.21 0.07 0.01 Completed Agricultural University 0.24 0.06 0.00 0.26 0.07 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.05 0.00 Clare, Limerick, Tipp. N.R. −0.14 0.04 0.00 −0.17 0.09 0.05 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.02 0.04 0.66 0.00 0.05 0.98 Cork, Kerry −0.17 0.04 0.00 −0.15 0.07 0.04 Galway, Mayo, Roscommon 0.00 0.03 0.95 −0.00 0.03 0.98 Stocking Density 0.40 0.02 0.00 0.42 0.05 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.00 0.08 0.96 Constant 5.05 0.17 0.00 4.97 0.21 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 2639.20 10.78 Sargan statistic p value (Overidentification Test) 0.000 0.000 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – Sfpyr only Instrumental Variable Regression –1 Dist. only Has Advisory Participation 0.35 0.41 0.00 0.14 0.57 0.79 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.03 0.00 Dairy System 0.37 0.03 0.00 0.34 0.09 0.00 Dairy and Other System −0.06 0.04 0.16 −0.09 0.12 0.42 Cattle Rearing −0.15 0.03 0.00 −0.19 0.15 0.19 Cattle Other System −0.16 0.03 0.00 −0.20 0.14 0.13 Mainly Sheep −0.03 0.03 0.39 −0.09 0.17 0.59 Tillage 0.29 0.05 0.00 0.25 0.09 0.01 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.09 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.12 Has off farm employment −0.13 0.02 0.00 −0.12 0.03 0.00 Log Farm Size 0.00 0.02 0.87 0.03 0.07 0.71 Has Forestry −0.09 0.04 0.05 −0.05 0.12 0.71 Completed Agricultural Short Course 0.21 0.03 0.00 0.23 0.06 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.21 0.07 0.01 Completed Agricultural University 0.24 0.06 0.00 0.26 0.07 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.05 0.00 Clare, Limerick, Tipp. N.R. −0.14 0.04 0.00 −0.17 0.09 0.05 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.02 0.04 0.66 0.00 0.05 0.98 Cork, Kerry −0.17 0.04 0.00 −0.15 0.07 0.04 Galway, Mayo, Roscommon 0.00 0.03 0.95 −0.00 0.03 0.98 Stocking Density 0.40 0.02 0.00 0.42 0.05 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.00 0.08 0.96 Constant 5.05 0.17 0.00 4.97 0.21 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 2639.20 10.78 Sargan statistic p value (Overidentification Test) 0.000 0.000 Note: The endogenous regressor is (Advisory Participation in both models); 1instrument in each (Single Farm Payment policy change and Distance to advisory office respectively); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors are adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak. View Large Appendix C – OLS estimates (2000-2004; 2005-2013) Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Model pre 2005 OLS Model post 2005 Has Advisory Participation 0.09 0.02 0.00 0.15 0.02 0.00 Log Land Value per Ha 0.19 0.03 0.00 0.15 0.02 0.00 Dairy System 0.16 0.07 0.02 0.31 0.03 0.00 Dairy and Other System −0.29 0.07 0.00 −0.04 0.04 0.37 Cattle Rearing −0.38 0.06 0.00 −0.17 0.03 0.00 Cattle Other System −0.56 0.06 0.00 −0.14 0.03 0.00 Mainly Sheep −0.39 0.06 0.00 −0.09 0.03 0.01 Tillage omitted omitted omitted 0.28 0.05 0.00 Log of Family Labour (unpaid) −0.08 0.03 0.01 −0.12 0.02 0.00 Age −0.00 0.01 0.86 0.01 0.01 0.181 Age Squared 0.00 0.00 0.95 −0.00 0.00 0.16 Has off farm employment −0.13 0.03 0.00 −0.11 0.02 0.00 Log Farm Size 0.01 0.02 0.81 0.04 0.02 0.01 Has Forestry omitted omitted omitted −0.03 0.04 0.38 Completed Agricultural Short Course 0.27 0.04 0.00 0.16 0.03 0.00 Completed Agricultural Certificate 0.13 0.03 0.00 0.21 0.03 0.00 Completed Agricultural University −0.04 0.09 0.65 0.28 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.18 0.04 0.00 −0.15 0.03 0.00 Kildare, Meath, Wicklow −0.01 0.11 0.94 −0.31 0.09 0.00 Laois, Longford, Offaly, Westmeath −0.17 0.05 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.21 0.05 0.00 −0.19 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.03 0.04 0.48 −0.04 0.04 0.30 Cork, Kerry −0.37 0.05 0.00 −0.05 0.04 0.12 Galway, Mayo, Roscommon −0.15 0.04 0.00 −0.01 0.03 0.76 Stocking Density 0.46 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.17 0.03 0.00 −0.08 0.02 0.00 Poor Soil 0.03 0.04 0.47 0.01 0.03 0.77 Constant 5.69 0.21 0.00 5.07 0.17 0.00 Number of Observations 4,503 8,875 R2 0.30 0.19 F stat 76.61 75.19 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Model pre 2005 OLS Model post 2005 Has Advisory Participation 0.09 0.02 0.00 0.15 0.02 0.00 Log Land Value per Ha 0.19 0.03 0.00 0.15 0.02 0.00 Dairy System 0.16 0.07 0.02 0.31 0.03 0.00 Dairy and Other System −0.29 0.07 0.00 −0.04 0.04 0.37 Cattle Rearing −0.38 0.06 0.00 −0.17 0.03 0.00 Cattle Other System −0.56 0.06 0.00 −0.14 0.03 0.00 Mainly Sheep −0.39 0.06 0.00 −0.09 0.03 0.01 Tillage omitted omitted omitted 0.28 0.05 0.00 Log of Family Labour (unpaid) −0.08 0.03 0.01 −0.12 0.02 0.00 Age −0.00 0.01 0.86 0.01 0.01 0.181 Age Squared 0.00 0.00 0.95 −0.00 0.00 0.16 Has off farm employment −0.13 0.03 0.00 −0.11 0.02 0.00 Log Farm Size 0.01 0.02 0.81 0.04 0.02 0.01 Has Forestry omitted omitted omitted −0.03 0.04 0.38 Completed Agricultural Short Course 0.27 0.04 0.00 0.16 0.03 0.00 Completed Agricultural Certificate 0.13 0.03 0.00 0.21 0.03 0.00 Completed Agricultural University −0.04 0.09 0.65 0.28 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.18 0.04 0.00 −0.15 0.03 0.00 Kildare, Meath, Wicklow −0.01 0.11 0.94 −0.31 0.09 0.00 Laois, Longford, Offaly, Westmeath −0.17 0.05 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.21 0.05 0.00 −0.19 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.03 0.04 0.48 −0.04 0.04 0.30 Cork, Kerry −0.37 0.05 0.00 −0.05 0.04 0.12 Galway, Mayo, Roscommon −0.15 0.04 0.00 −0.01 0.03 0.76 Stocking Density 0.46 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.17 0.03 0.00 −0.08 0.02 0.00 Poor Soil 0.03 0.04 0.47 0.01 0.03 0.77 Constant 5.69 0.21 0.00 5.07 0.17 0.00 Number of Observations 4,503 8,875 R2 0.30 0.19 F stat 76.61 75.19 Note: Border region is omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level. View Large Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Model pre 2005 OLS Model post 2005 Has Advisory Participation 0.09 0.02 0.00 0.15 0.02 0.00 Log Land Value per Ha 0.19 0.03 0.00 0.15 0.02 0.00 Dairy System 0.16 0.07 0.02 0.31 0.03 0.00 Dairy and Other System −0.29 0.07 0.00 −0.04 0.04 0.37 Cattle Rearing −0.38 0.06 0.00 −0.17 0.03 0.00 Cattle Other System −0.56 0.06 0.00 −0.14 0.03 0.00 Mainly Sheep −0.39 0.06 0.00 −0.09 0.03 0.01 Tillage omitted omitted omitted 0.28 0.05 0.00 Log of Family Labour (unpaid) −0.08 0.03 0.01 −0.12 0.02 0.00 Age −0.00 0.01 0.86 0.01 0.01 0.181 Age Squared 0.00 0.00 0.95 −0.00 0.00 0.16 Has off farm employment −0.13 0.03 0.00 −0.11 0.02 0.00 Log Farm Size 0.01 0.02 0.81 0.04 0.02 0.01 Has Forestry omitted omitted omitted −0.03 0.04 0.38 Completed Agricultural Short Course 0.27 0.04 0.00 0.16 0.03 0.00 Completed Agricultural Certificate 0.13 0.03 0.00 0.21 0.03 0.00 Completed Agricultural University −0.04 0.09 0.65 0.28 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.18 0.04 0.00 −0.15 0.03 0.00 Kildare, Meath, Wicklow −0.01 0.11 0.94 −0.31 0.09 0.00 Laois, Longford, Offaly, Westmeath −0.17 0.05 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.21 0.05 0.00 −0.19 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.03 0.04 0.48 −0.04 0.04 0.30 Cork, Kerry −0.37 0.05 0.00 −0.05 0.04 0.12 Galway, Mayo, Roscommon −0.15 0.04 0.00 −0.01 0.03 0.76 Stocking Density 0.46 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.17 0.03 0.00 −0.08 0.02 0.00 Poor Soil 0.03 0.04 0.47 0.01 0.03 0.77 Constant 5.69 0.21 0.00 5.07 0.17 0.00 Number of Observations 4,503 8,875 R2 0.30 0.19 F stat 76.61 75.19 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Model pre 2005 OLS Model post 2005 Has Advisory Participation 0.09 0.02 0.00 0.15 0.02 0.00 Log Land Value per Ha 0.19 0.03 0.00 0.15 0.02 0.00 Dairy System 0.16 0.07 0.02 0.31 0.03 0.00 Dairy and Other System −0.29 0.07 0.00 −0.04 0.04 0.37 Cattle Rearing −0.38 0.06 0.00 −0.17 0.03 0.00 Cattle Other System −0.56 0.06 0.00 −0.14 0.03 0.00 Mainly Sheep −0.39 0.06 0.00 −0.09 0.03 0.01 Tillage omitted omitted omitted 0.28 0.05 0.00 Log of Family Labour (unpaid) −0.08 0.03 0.01 −0.12 0.02 0.00 Age −0.00 0.01 0.86 0.01 0.01 0.181 Age Squared 0.00 0.00 0.95 −0.00 0.00 0.16 Has off farm employment −0.13 0.03 0.00 −0.11 0.02 0.00 Log Farm Size 0.01 0.02 0.81 0.04 0.02 0.01 Has Forestry omitted omitted omitted −0.03 0.04 0.38 Completed Agricultural Short Course 0.27 0.04 0.00 0.16 0.03 0.00 Completed Agricultural Certificate 0.13 0.03 0.00 0.21 0.03 0.00 Completed Agricultural University −0.04 0.09 0.65 0.28 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.18 0.04 0.00 −0.15 0.03 0.00 Kildare, Meath, Wicklow −0.01 0.11 0.94 −0.31 0.09 0.00 Laois, Longford, Offaly, Westmeath −0.17 0.05 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.21 0.05 0.00 −0.19 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.03 0.04 0.48 −0.04 0.04 0.30 Cork, Kerry −0.37 0.05 0.00 −0.05 0.04 0.12 Galway, Mayo, Roscommon −0.15 0.04 0.00 −0.01 0.03 0.76 Stocking Density 0.46 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.17 0.03 0.00 −0.08 0.02 0.00 Poor Soil 0.03 0.04 0.47 0.01 0.03 0.77 Constant 5.69 0.21 0.00 5.07 0.17 0.00 Number of Observations 4,503 8,875 R2 0.30 0.19 F stat 76.61 75.19 Note: Border region is omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level. View Large Appendix D – Condensed model (2000-2013) Dependent Variable = Log of Farm Family Income per ha Coeff SE Z p CI CI Has Advisory Participation 0.31 0.04 7.95 0.00 0.23 0.38 Log Land Value per Ha 0.26 0.02 13.47 0.00 0.22 0.38 Age 0.03 0.01 4.74 0.00 0.02 0.04 Age Squared −0.00 0.00 −5.07 0.00 −0.00 −0.00 Completed Agricultural Short Course 0.26 0.03 8.58 0.00 0.20 0.32 Completed Agricultural Certificate 0.33 0.03 12.55 0.00 0.28 0.39 Completed Agricultural University 0.39 0.06 6.21 0.00 0.27 0.52 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.14 0.03 −4.65 0.00 −0.20 −0.08 Kildare, Meath, Wicklow −0.28 0.08 −3.35 0.00 −0.44 −0.12 Laois, Longford, Offaly, Westmeath −0.11 0.04 −2.61 0.01 −0.18 −0.03 Clare, Limerick, Tipp. N.R. −0.22 0.04 −5.50 0.00 −0.29 −0.14 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.06 0.04 −1.67 0.10 −0.13 0.01 Cork, Kerry −0.15 0.04 −3.99 0.00 −0.21 −0.07 Galway, Mayo, Roscommon 0.12 0.03 3.89 0.00 0.06 0.19 Medium Soil −0.17 0.02 −8.15 0.00 −0.22 −0.13 Poor Soil −0.16 0.03 −4.82 0.00 −0.22 −0.09 Constant 5.11 0.17 30.64 0.00 4.79 5.44 Number of Observations 8,951 Centred R2 0.12 Cragg-Donald Wald F Statistic (WeakInstrument) 1118.8 Sargan statistic p value (Overidentification Test) 0.19 Dependent Variable = Log of Farm Family Income per ha Coeff SE Z p CI CI Has Advisory Participation 0.31 0.04 7.95 0.00 0.23 0.38 Log Land Value per Ha 0.26 0.02 13.47 0.00 0.22 0.38 Age 0.03 0.01 4.74 0.00 0.02 0.04 Age Squared −0.00 0.00 −5.07 0.00 −0.00 −0.00 Completed Agricultural Short Course 0.26 0.03 8.58 0.00 0.20 0.32 Completed Agricultural Certificate 0.33 0.03 12.55 0.00 0.28 0.39 Completed Agricultural University 0.39 0.06 6.21 0.00 0.27 0.52 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.14 0.03 −4.65 0.00 −0.20 −0.08 Kildare, Meath, Wicklow −0.28 0.08 −3.35 0.00 −0.44 −0.12 Laois, Longford, Offaly, Westmeath −0.11 0.04 −2.61 0.01 −0.18 −0.03 Clare, Limerick, Tipp. N.R. −0.22 0.04 −5.50 0.00 −0.29 −0.14 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.06 0.04 −1.67 0.10 −0.13 0.01 Cork, Kerry −0.15 0.04 −3.99 0.00 −0.21 −0.07 Galway, Mayo, Roscommon 0.12 0.03 3.89 0.00 0.06 0.19 Medium Soil −0.17 0.02 −8.15 0.00 −0.22 −0.13 Poor Soil −0.16 0.03 −4.82 0.00 −0.22 −0.09 Constant 5.11 0.17 30.64 0.00 4.79 5.44 Number of Observations 8,951 Centred R2 0.12 Cragg-Donald Wald F Statistic (WeakInstrument) 1118.8 Sargan statistic p value (Overidentification Test) 0.19 Note: Endogenous regressor is (Advisory Participation); 3 Instruments (Single Farm Payment policy change, Distance to advisory office and Interaction of both); border region omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak. View Large Dependent Variable = Log of Farm Family Income per ha Coeff SE Z p CI CI Has Advisory Participation 0.31 0.04 7.95 0.00 0.23 0.38 Log Land Value per Ha 0.26 0.02 13.47 0.00 0.22 0.38 Age 0.03 0.01 4.74 0.00 0.02 0.04 Age Squared −0.00 0.00 −5.07 0.00 −0.00 −0.00 Completed Agricultural Short Course 0.26 0.03 8.58 0.00 0.20 0.32 Completed Agricultural Certificate 0.33 0.03 12.55 0.00 0.28 0.39 Completed Agricultural University 0.39 0.06 6.21 0.00 0.27 0.52 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.14 0.03 −4.65 0.00 −0.20 −0.08 Kildare, Meath, Wicklow −0.28 0.08 −3.35 0.00 −0.44 −0.12 Laois, Longford, Offaly, Westmeath −0.11 0.04 −2.61 0.01 −0.18 −0.03 Clare, Limerick, Tipp. N.R. −0.22 0.04 −5.50 0.00 −0.29 −0.14 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.06 0.04 −1.67 0.10 −0.13 0.01 Cork, Kerry −0.15 0.04 −3.99 0.00 −0.21 −0.07 Galway, Mayo, Roscommon 0.12 0.03 3.89 0.00 0.06 0.19 Medium Soil −0.17 0.02 −8.15 0.00 −0.22 −0.13 Poor Soil −0.16 0.03 −4.82 0.00 −0.22 −0.09 Constant 5.11 0.17 30.64 0.00 4.79 5.44 Number of Observations 8,951 Centred R2 0.12 Cragg-Donald Wald F Statistic (WeakInstrument) 1118.8 Sargan statistic p value (Overidentification Test) 0.19 Dependent Variable = Log of Farm Family Income per ha Coeff SE Z p CI CI Has Advisory Participation 0.31 0.04 7.95 0.00 0.23 0.38 Log Land Value per Ha 0.26 0.02 13.47 0.00 0.22 0.38 Age 0.03 0.01 4.74 0.00 0.02 0.04 Age Squared −0.00 0.00 −5.07 0.00 −0.00 −0.00 Completed Agricultural Short Course 0.26 0.03 8.58 0.00 0.20 0.32 Completed Agricultural Certificate 0.33 0.03 12.55 0.00 0.28 0.39 Completed Agricultural University 0.39 0.06 6.21 0.00 0.27 0.52 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.14 0.03 −4.65 0.00 −0.20 −0.08 Kildare, Meath, Wicklow −0.28 0.08 −3.35 0.00 −0.44 −0.12 Laois, Longford, Offaly, Westmeath −0.11 0.04 −2.61 0.01 −0.18 −0.03 Clare, Limerick, Tipp. N.R. −0.22 0.04 −5.50 0.00 −0.29 −0.14 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.06 0.04 −1.67 0.10 −0.13 0.01 Cork, Kerry −0.15 0.04 −3.99 0.00 −0.21 −0.07 Galway, Mayo, Roscommon 0.12 0.03 3.89 0.00 0.06 0.19 Medium Soil −0.17 0.02 −8.15 0.00 −0.22 −0.13 Poor Soil −0.16 0.03 −4.82 0.00 −0.22 −0.09 Constant 5.11 0.17 30.64 0.00 4.79 5.44 Number of Observations 8,951 Centred R2 0.12 Cragg-Donald Wald F Statistic (WeakInstrument) 1118.8 Sargan statistic p value (Overidentification Test) 0.19 Note: Endogenous regressor is (Advisory Participation); 3 Instruments (Single Farm Payment policy change, Distance to advisory office and Interaction of both); border region omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak. View Large Appendix E – Difference in difference model (2000-2013) Dependent Variable = Log of Farm Family Income per ha Coeff SE t p CI CI Single farm payment year −0.03 0.03 −1.21 0.23 −0.08 0.02 Treated group −0.28 0.03 −10.76 0.00 −0.33 −0.23 Interaction term SFP*Treated 0.28 0.03 7.85 0.00 0.21 0.34 Constant 6.10 0.02 366.6 0.00 6.06 6.13 Number of Observations 13,562 R2 0.01 Dependent Variable = Log of Farm Family Income per ha Coeff SE t p CI CI Single farm payment year −0.03 0.03 −1.21 0.23 −0.08 0.02 Treated group −0.28 0.03 −10.76 0.00 −0.33 −0.23 Interaction term SFP*Treated 0.28 0.03 7.85 0.00 0.21 0.34 Constant 6.10 0.02 366.6 0.00 6.06 6.13 Number of Observations 13,562 R2 0.01 Note: Advisory participation treated for those who only became clients after the introduction of the single farm payment policy change in 2005. View Large Dependent Variable = Log of Farm Family Income per ha Coeff SE t p CI CI Single farm payment year −0.03 0.03 −1.21 0.23 −0.08 0.02 Treated group −0.28 0.03 −10.76 0.00 −0.33 −0.23 Interaction term SFP*Treated 0.28 0.03 7.85 0.00 0.21 0.34 Constant 6.10 0.02 366.6 0.00 6.06 6.13 Number of Observations 13,562 R2 0.01 Dependent Variable = Log of Farm Family Income per ha Coeff SE t p CI CI Single farm payment year −0.03 0.03 −1.21 0.23 −0.08 0.02 Treated group −0.28 0.03 −10.76 0.00 −0.33 −0.23 Interaction term SFP*Treated 0.28 0.03 7.85 0.00 0.21 0.34 Constant 6.10 0.02 366.6 0.00 6.06 6.13 Number of Observations 13,562 R2 0.01 Note: Advisory participation treated for those who only became clients after the introduction of the single farm payment policy change in 2005. View Large © The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Applied Economic Perspectives and Policy Oxford University Press

The Impact of Extension Services on Farm-level Income: An Instrumental Variable Approach to Combat Endogeneity Concerns

Loading next page...
 
/lp/ou_press/the-impact-of-extension-services-on-farm-level-income-an-instrumental-L0O3EJfm8q
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com
ISSN
2040-5790
eISSN
2040-5804
D.O.I.
10.1093/aepp/ppx062
Publisher site
See Article on Publisher Site

Abstract

Abstract Agricultural extension is an important policy instrument utilized to diffuse knowledge and increase profitability among farmers. However, analyses on impact are subject to endogeneity concerns, causing multiple biases. Failure to combat endogeneity can lead to false inferences on impact. This article addresses this issue by applying an instrumental variable approach with distance to local advisory office and a policy change chosen as instruments for extension participation. The results show that participation significantly increased farm income and that OLS estimates underestimated the impact. Therefore, a superior estimate of impact is achieved which can be leveraged to better support accurate policy making. Extension, farm income, endogeneity, instrumental variables, two-stage least squares estimation, panel data Agricultural extension builds the capabilities of clients through improved problem solving, decision making, and management (Vanclay and Leach 2011), and is a means of transferring specialist knowledge from emerging research or public policy arenas to farm level that is commonly adopted worldwide. Historically, most developed countries have had a form of extension service for rural land managers funded largely from general taxation and delivered by public organizations (Garforth et al. 2003). Indeed, governments actively promote extension services as a means of knowledge diffusion and technology adoption, and also to develop innovative solutions to emerging challenges (Läpple and Hennessy 2015). This form of public extension service has since been supplemented by the private sector with the common objective of improving the performance of farmers. There are many challenges for the agricultural sector, such as the need to strike a balance between increasing productivity to feed a growing global population and reducing negative environmental externalities, including climate change. Coupled with this responsibility is a financial challenge as global economies navigate turbulent macroeconomic cycles, and there is a renewed emphasis on “value for money” for public expenditure. Extension services are important in these circumstances as they can act as policy levers to change existing behavior, but must represent a worthy investment for policymakers. Furthermore, extension clients must also experience a value for money return from participation given the fee level paid, particularly given existing income challenges caused by economic volatility, and a strong reliance on subsidies (O’Donoghue and Hennessy 2015). Therefore, estimating the impact of extension participation on farm income is prudent and ensures a common outcome is measured regardless of the type of extension activity employed. In order to achieve this it is imperative that the estimation is robust and unbiased, as a plethora of confounding factors also influence this impact. In this paper this obstacle is overcome by instrumenting the extension variable to remove these influences. Many studies show that interaction with extension services affects income levels (Dercon et al. 2009; Davis et al. 2012; Läpple and Hennessy 2015). Anderson and Feder (2004) reviewed previous studies on extension impact, and whilst positive results are common, there is variability within these results and they warned that results should be treated with caution given the econometric challenges. Indeed, the inconsistency in research findings could be partly due to the broad range of extension methods and outcome measures (Läpple and Hennessy 2015) and therefore, the assessment of impact across disparate ranges is the major challenge for this type of research (Ragasa et al. 2016). However, from a policy perspective the impact of these services is the ultimate criterion (Birner et al. 2009) and thus the key objective of this research is to provide a robust causal estimation of the impact of extension participation on farm income. Providing an accurate and valid estimation of impact can assist the policy formulation process as policymakers can rely on the likely impact of an initiative with greater precision. In order to achieve this, it is critically important that the estimation addresses econometric concerns related to the issue of endogeneity. The econometric issue of endogeneity is typically caused by omitted variable bias, measurement error, and self-selection biases. Failure to combat endogeneity may lead to incorrect inferences on the causal relationship between extension participation and farm income (Akobundu et al. 2004; Abdallah et al. 2015). The omission of an innate ability variable among extension participants could overestimate the true effect of extension on farm income. Conversely, measurement error is likely to underestimate the true effect of this relationship (Card 1999). Similarly, self-selection biases are implicit in this research with the motivation to engage with extension being a critical element of estimating impact (Nordin and Höjgård 2016). For example, more capable farmers may be more motivated to engage with extension services to augment their performance, whereas weaker farmers may be less likely to participate. The Instrumental Variable (IV) approach is dependent on identifying suitable instruments that affect the endogenous variable in question but do not impact directly on the dependent variable. In our case, such an instrument must affect the decision to participate in extension services, but not impact farm family income directly. These instruments “purge” the endogenous regressors and allow consistent coefficient estimates (Gabel and Scheve 2007). Distance from the local advisory office and the introduction of a policy change in 2005 are chosen as appropriate instruments to meet these conditions. The former is expected to negatively affect the decision to participate but is uncorrelated with farm income, whereas the latter incentivized participation for assistance with subsidy applications decoupled from current production based on historical receipts across all farming systems. These instruments enable the evaluation of the impact of extension on farm level income over the 2000–2013 period. The remainder of the paper is structured as follows: initially the context for extension services in Ireland is outlined, followed by a review of the relevant literature. Next there is an overview of the methodology and data. Finally, the results are discussed, and the conclusion outlines the policy implications, along with some limitations and recommendations for future research. Context Agricultural extension programs are operated by multifunctional organizations that provide advisory services to the farming sector. Extension programs could be viewed as risk management devices from policy makers to mitigate issues in the rural economy, or as drivers of growth at the farm level to ensure that best practices are followed systematically. Läpple, Hennessy, and Newman (2013) summarized the definition and purpose of extension services as a program to improve farm performance and introduce new technologies to connect emerging research to on-farm practices. Thus, extension programs assist farmers to overcome barriers to achieving set goals due to a lack of knowledge, motivation, resources, insight, and power, or a combination of these (Van den Ban and Hawkins 1988). Extension programs aim to assist farmers on issues such as productivity, food safety, and the environment, among others (Boyle 2008). The extension service in Irish agriculture consists of both public and private consultants, with clients split relatively evenly among both types. Prager et al. (2016) estimated that approximately 250 private advizers (mainly represented by the Agricultural Consultants Association) were in operation alongside 250 Teagasc advizers. Teagasc is the main body for the public delivery of agricultural research, advice, and training since 1988; it is a unique organization that combines research, extension services, and education (Prager et al. 2016), and 30% of the operating budget of approximately €160 million annually is directed towards extension programs (Teagasc 2016). The research reported here uses data referring to Teagasc clients only for two reasons: first, the level of public expenditure allocated to this public extension system, and second, data constraints related to private extension organizations and their clients. Indeed, as the majority of private consultancies in Ireland consisted of just 1–3 advizers (Prager et al. 2016), that data is more likely to be fragmented as opposed to the superior data available from Teagasc extension. Therefore, farmers who engaged with private consultants are not captured in this research. Teagasc clients include any farmer who participates in any extension program under the various contracts offered. The clients’ specific farm systems and characteristics are included as control variables in the analysis. Furthermore, this analysis focuses on more recent Teagasc clients that became clients after the policy change in 2005 to reduce general self-selection biases further. The introduction of the Single Farm Payment scheme in 2005 increased the numbers of extension clients due to farmers’ need for assistance given bureaucratic challenges, and is applied as an instrument in the analysis. In other words, the sample is pooled to focus on farmers who were in the Teagasc National Farm Survey (NFS) sample prior to the policy change but only became Teagasc clients afterwards. This increases the comparability of these clients as they are more likely to have been motivated to engage in an extension program given the policy change as opposed to alternative diverse “traditional” motivations (Läpple, Hennessy, and Newman 2013). This hones the focus of the study by ensuring the analysis is conducted on a similar cohort of extension clients. Preliminary analysis on the impact of extension on this group showed a larger impact for those who joined after 2005. Ordinary Least Squares (OLS) estimation showed a positive impact, which increased from 9% to 15% for all clients pre- and post-2005, respectively. Similarly, a difference in difference estimation for clients who joined after the introduction of the policy change showed a positive and significant impact. These estimations are included in the appendix. Literature Review The dominant theme in previous studies on the impact of extension participation on farm level outcomes is that of variability across the results (Anderson and Feder 2004). Given the complexity of narrowing the focus of research, the variety in types of extension programs, the diversity of outcomes to measure, and the diversity of methodological options, this variability and inconsistency is not surprising (Läpple and Hennessy 2015). However, an IV approach to the impact of extension is not common in the literature given the challenge of identifying suitable instruments, a challenge which this study aims to address. Much of the literature analyses the relationship between extension participation and productivity with studies reporting a positive relationship between extension investment and output yields (Griliches 1964; Marsh, Pannell, and Linder 2004). Similarly, analysis of the impact of one-to-one consultations and productivity also identified positive relationships (Owens, Hoddinott, and Kinsey 2003) although in one case the levels diminished over time (Krishnan and Patnam 2014). Participatory extension techniques have also been studied in relation to farm yields, with positive outcomes reported in terms of increased crops (Davis et al. 2012) and milk yields per cow (Läpple and Hennessy 2015). In contrast, Hunt et al. (2014) found that the impact of extension services on productivity declined over time due to capacity constraints. Specifically in relation to extension participation and financial returns there are recent studies on participatory extension programs’ impact in terms of increased gross margin per hectare of €310 in the Irish dairy sector (Läpple, Hennessy, and Newman 2013), and an increase of €150 gross margin per dairy cow also in Ireland (Läpple and Hennessy 2015). The former study employed an endogenous switching model, whereas the latter utilized propensity score matching methods to address endogeneity using panel data sets. Davis et al. (2012) also found a positive effect using the propensity score matching method identifying income gains of 21% to 104% in selected African nations. These studies adopt extension participation as a binary variable based on participation. The switching model method relies on conditional expectations of impact based on non-participation in the extension program compared with the actual results. The main advantage of the switching model is that it combats self-selection biases but it is based on the hypothetical case of an outcome. The propensity score matching approach is effective but relies on a number of decisive steps based on the quality of data and underpinning assumptions often involving a trade-off between bias and efficiency (Caliendo and Kopeinig 2008). These studies utilized various econometric methodologies but endogeneity concerns were not central to the analysis and they did not employ the IV approach, which may lead to incorrect inferences due to uncontrolled biases that remain (Akobundu et al. 2004; Abdallah et al. 2015). Indeed, whereas some studies may prioritize a particular form of bias to control, the IV approach combats multiple forms of endogenous biases (Card 1999; Cawley and Meyerhoefer 2012; Howley, O’Neill, and Atkinson 2015). Thus, the IV approach is superior in that it removes the need to prioritize specific types of bias over others, but combats all forms, ensuring an improved estimation of extension impact on farms. IV estimation can consistently estimate coefficients that will almost certainly be close to the true coefficient value if the sample size is sufficiently large (Murray 2006). However, this approach is dependent on the precondition of identifying appropriate instruments, which often proves challenging. Accordingly, studies on the impact of extension engagement on farm-level income using an IV approach are less common but there are exceptions. Nordin and Höjgård (2016) applied an IV methodology to evaluate extension services in Sweden and found both a societal and farm-level benefit to participation in relation to nutrient management. The mean number of adviser visits to a farm was adopted as an instrument for the actual number of visits to a specific farm based on a positive relationship between the agent’s expertise and farm type. Nordin and Höjgård (2016) found that OLS estimates underestimated the benefits of the extension program compared to the corrected IV estimates. Akobundu et al. (2004) utilized this approach and found a U.S. extension program that incorporated both one-to-one and group-based activities increased net farm income, but the level of return relied on the intensity of participation. These researchers utilized the distance to local extension office, the intensity of participation captured through the number of adviser on-farm visits, and the level of household debt with higher debts classified as more likely to increase adviser visits, as suitable instruments. Similarly, Heanue and O’Donoghue (2014) found positive economic outcomes for farmers who participated in agricultural education in terms of farm income and the internal rate of return to investment, also using an IV approach. As instruments, they used the distance to local agricultural college and the introduction of a policy change. This study follows on from that work but instead focuses solely on Teagasc extension clients. Whilst an IV approach is not common in the agricultural economics literature it is important to note that it is widely used in other fields, with significant policy implications. For example, it has been utilized to study the impact of economic shocks on conflict where adverse weather conditions instrumented economic activity and showed an increase in the probability of civil conflict in the following year (Miguel et al. 2004). Further, IV estimation is viewed as a sensible addition to the analytical toolbox in health economics (Rassen et al. 2009). Indeed, in one such study regional cardiac catheterization rates were adopted as instruments to explore the relationship between specific treatments and mortality rates, with a 16% survival benefit being reported (Stukel et al. 2007). Similarly, genetic variation in weight was applied as an instrument in a study on the relationship between obesity and medical costs, which found that previous literature underestimated these costs (Cawley and Meyerhoefer 2012). IV estimation has also been applied to examine the relationship between education and subsequent income earnings, with distance to college used as an instrument also showing a previous underestimation in the literature (Card 1999). Therefore, the literature presents examples of downward biases that are subsequently corrected through instrumentation and this research queries whether this is also the case for the impact of extension on farm-level income. If so, existing estimations of impact may have underestimated the true effect on farms, and the effectiveness of extension services to act as policy instruments could be more pronounced. Accordingly, the purpose of this research is to measure the impact of extension programs on farm incomes. The IV approach ensures that endogeneity concerns are addressed, leading to a robust estimation of impact given the identification of suitable instruments. In this study it is assumed that farmers utilize extension programs primarily to improve profitability. Other beneficial outcomes are taken as secondary. Clients of extension programs are expected to achieve higher average farm income levels than farmers who do not use extension programs. Thus, a positive relationship is expected. Methodology Birkhaeuser, Evenson, and Feder (1991) identified a number of problems associated with assessing the impact of extension activities such as the phase of the farmers’ development cycle, policy and market influences, and information flows, but the predominant issue is that of endogeneity. An endogenous explanatory variable exists when the variable is correlated with the error term (Wooldridge 2013). This means that the estimated coefficient for that variable will be inconsistent and biased as its magnitude is somewhat determined by the error term. This may lead to inaccurate interpretations of the effects of findings unless appropriate action is undertaken (Abdallah et al. 2015). Endogeneity has three primary causes. Firstly, omitted variable bias causes an obvious problem for analysis as a farmer’s innate ability, effort, ambition, or motivation would have an effect on the impact of extension engagement, yet this data is unobserved. Secondly, self-selection bias is a methodological error caused by initial differences between participants and non-participants due to the conscious decision to enroll in an extension program or not (Imbens and Wooldridge 2009; Nordin and Höjgård 2016). Läpple, Hennessy, and Newman (2013) explain that higher-skilled producers may be more likely to adopt extension services given their capacity and motivation to enhance their enterprise, yet we do not have a variable to reflect this issue. Conversely, farmers with higher ability may not deem extension services necessary given their own capabilities on the farm. Similarly, farmers with lower ability may seek advisory assistance to bolster their performance, or alternatively they may feel investing in advisory services is not worthwhile. Akobundu et al. (2004) suggested that extension programs themselves may purposively target specific farmers, whether disadvantaged or eager to participate. This may be due to attempting to improve the performance of vulnerable producers in the first instance, who may avoid such opportunities without adequate incentives. Conversely, in the latter example, advisers may be more likely to target more motivated participants who may be more likely to disseminate the knowledge (Läpple, Hennessy, and Newman 2013). Conversely, advisers may avoid particular clients for various reasons such as location, personal characteristics, or due to time constraints. This indicates bidirectional causality with regard to self-selection bias. Therefore, farmers using the extension service are systematically different than those who do not (Tamini 2011; Hennessy and Heanue 2012). Finally, another form of bias may be related to measurement error. Given the endogenous variable for extension participation is imperfectly measured as a binary variable in our analysis, this is likely to cause a downward (attenuation) bias on the initial Ordinary Least Squares (OLS) estimation prior to instrumentation (Card 1999). The variety of extension programs options on offer range from annual visits to assist with scheme assistance to intensive technical assistance packages involving on-site visits, discussion groups, and seminars. The utilization of participation as a binary variable removes this variability for the purpose of providing an aggregated impact estimation regardless of the type of extension adopted; hence, it is imperfect but necessary for this analysis. Accordingly, it is important to note that the direction of the bias is not necessarily upward, given intuitive expectations on omitted ability or self-selection. There are a number of approaches to deal with endogeneity as discussed in the previous section. However, the IV approach is efficient as it accounts for all forms of bias listed above, provided suitable instruments are identified. The instrument must be correlated with the endogenous explanatory variable (relevant), but uncorrelated with the dependent variable and error term (valid) (Murray 2006; Burgess, Dudbridge, and Thompson 2016). If such an instrument can be found, then an unbiased consistent coefficient for the endogenous variable can be estimated (Gujarati 2003). The process involves a two-stage least squares regression, where firstly the instruments are regressed on the endogenous variable, and subsequently the predicted value of the now exogenous variable is inserted into the main structural equation and estimated. Accordingly, the primary challenge for IV analysis is identifying a suitable instrument. Murray (2006) noted that having at least as many instruments as endogenous variables is a necessary condition for identification and in most cases is sufficient. On this basis two instruments were identified. The distance to the local extension office and the policy change effect were chosen on the basis of previous literature. Geographic proximity was employed by Card (1993) and Callan and Harmon (1999) along with Heanue and O’Donoghue (2014), where the latter two also used a policy change as an exogenous shock. Callan and Harmon utilized the introduction of free schooling and Heanue and O’Donoghue adopted the introduction of the Stamp Duty Exemption Scheme that targeted young trained farmers as instruments. In both examples, the researchers argued that the instrument incentivized participation in education and agricultural training, but did not affect income levels directly. Both instruments were expected to affect the decision to participate in extension services independently of farmers’ personal characteristics and/or farm performance. These instruments were also interacted to examine the impact of both combined and improve the estimation approach. The rationale for the instruments is explained in more detail in the subsequent section. As noted above, a 2 Stage Least Squares (2SLS) IV approach is applied. Thus, our initial stage is to test for the exclusion restrictions of the instruments (Wooldridge 2013). In other words, we apply the first stage, which is the reduced-form equation for our endogenous regressor through the following reduced form equation for y2: y2= π0+π1z1+π2z2+ π3z1z2+ πX+ v2 (1) where y2 is our endogenous regressor (extension participation), πj is our estimated parameter coefficients, zk are our instruments, X is a vector of all other explanatory variables, and v2 is our error term. Partial correlation at least between zk and y2 is necessary to fulfil the requirement that the instruments affect the endogenous regressor. Therefore, we can apply our second stage and specify our structural equation as follows: y1= β0+β1y^2+βX+ u1 (2) where y1 is the unbiased estimation of our dependent variables, βj is our estimated parameter coefficients, y^2 is our “purged” endogenous variable, X is a vector of all other explanatory variables, and u1 is our error term. All models are presented with selected specification tests in the results. In the first stage, the multivariate Cragg-Donald Wald F test was conducted to measure whether the instruments affect the endogenous variable. Stock, Wright, and Yogo (2002) outlined a rule of thumb that the F statistic must exceed 10 to avoid instruments being classified as weak. For the IV models, the Sargan statistic is reported, which measures the null hypotheses that all instruments are valid. This is recognized as the standard overidentification test of the validity of the instruments and is a benefit of the 2SLS approach (Howley, O’Neill, and Atkinson 2015). Cawley and Meyerhoefer (2012) noted that rejecting the null hypothesis of no effect is impossible and therefore doubt will remain if an instrument is truly exogenous. However, the Sargan test of validity is applied on occasion and failure to reject implies the instruments cannot be argued as invalid, which bolsters the argument of their validity as instruments. If the computed chi-square exceeds the critical chi-square value, we reject the null hypothesis, which means at least one instrument is correlated with the error and therefore invalid (Gujarati 2003). Data Specific data was required to conduct this analysis. Firstly, it was important to observe participants and non-participants in extension programs. This was the basis of our endogenous variable. A binary variable was established based on extension participation with Teagasc coded with a value of 1 for clients, and 0 otherwise. Thus, non-participants were identified as all observations that are not recorded as Teagasc clients. While this binary variable is imperfect in the sense of not differentiating between types of engagement as noted previously, it does provide an initial aggregated value for extension in terms of extension participation that is testable, which is a common approach in the literature. Teagasc clients include annual contract holders as well as those that use the service for scheme assistance, defined as participation solely for the purpose of fulfilling bureaucratic requirements to receive subsidy payments as opposed to technical advice. Annual contracts involve “packages” of services that include combinations of events, discussion groups, farm walks, consultations, and news. Farm family income, which includes subsidies, was used as the dependent variable as it provides a useful barometer of performance, particularly over time, and reflects the total income accrued to the household related to farm-based activities. In addition, this measure also captures the effect of extension participation, regardless of which type of service was adopted. Furthermore, additional factors that influence farm income, such as the geographic location, farm system, and other characteristics were also included. These data were derived from the Teagasc National Farm Survey (NFS). The Teagasc NFS is an annual panel data source collected as part of the Farm Accountancy Data Network (FADN) required by the European Union. The NFS consists of approximately 1,100 farms per annum representing approximately 110,000 farms of the population in Ireland (Läpple and Hennessy 2015). This dataset determines the financial situation on Irish farms by measuring the level of gross output, margins, costs, income, investment, and indebtedness across the spectrum of farming systems, sizes and profiles in the various regions (Connolly et al. 2010). Panel data allows the tracking of the same farms over time, which enriches analysis of this type as some farms may opt in, opt out, avoid, return, or engage constantly with extension services. For the purpose of this analysis, and given that the introduction of the Single Farm Payment policy change in 2005 was chosen as an instrument, the sample selected prioritized farms that were in the sample prior to this period, but only became Teagasc clients subsequently. This refines the sample, and exploits a natural experiment by causing exogenous variation in an otherwise endogenous variable (Wooldridge 2010). This ensures we focus on a similar group of farmers who were likely to have been motivated to become clients in response to the policy change. Including the full sample increases the variation in the results due to the inclusion of “self-selecting” farmers who may have participated into extension services for reasons outlined above regardless of the policy change (Läpple and Hennessy 2015). Each observation employed an average population weight recorded on the NFS over the period to ensure the sample was representative of the general farming population (Buckley et al. 2016). Accordingly, the final sample size is 8,951 observations over the years 2000–2013 inclusive for the models estimated. The estimated coefficients are therefore an average estimation across the 14 years of the analysis. The dependent variable was family farm income, which was divided by utilizable agricultural hectares to obtain a more representative indicator to remove potential skewness from diverse farm sizes. This provided a per hectare based estimation for impact, at the expense of accommodating individual farm expansions. However, given the somewhat stifled land market with low levels of sales and within-family transfers and strong attachment to land ownership in Ireland (Hennessy and Rehman 2007; Leonard et al. 2017), this was not considered an obstruction to the analysis. Family farm income is defined as gross output less net expenses (direct and overhead costs) and includes subsidy receipts. It does not include off-farm income. Including subsidies gives a more inclusive portrayal of the overall farm income attributable to the enterprise, and also captures farmers who participate in extension for the primary purpose of scheme assistance, as well as those involved in more technical-based programs. Given that assistance with Single Farm Payment applications is an important extension service, including the subsidy effect in the dependent variable was prudent. Intuitively, a farmer receives a financial gain directly attributable to this process in a given year. However, as the payment was decoupled from production under the reform of the EU’s Common Agricultural Policy, this subsidy effect would not vary over time, and thus differences in farm income would be based on other factors associated with farm performance. Nonetheless, the inclusion of the subsidy is likely to make a significant contribution to farm income. The dependent variable was transformed using its natural logarithm to remove the influence of outliers in the sample, to smooth the distribution of the data, and to enable us to interpret our coefficients as percentages. The main independent or explanatory variable selected for the regression analysis was extension participation as a binary variable. This enabled the division of the sample into those that participated in extension over time and those that did not. Thus, participation was recorded with a value of 1 for any level of participation in Teagasc extension and 0 otherwise. This aggregated form of the variable ensured that a level of impact across all forms of participation was captured ensuring the public expenditure “value for money” element was assessed. Farm system, stocking density, soil type, land value, labor, age, off-farm employment and size were included as controls on the prior expectation that they would affect farm performance. These controls also strengthen the instruments’ exogeneity condition as each control reduces the effect of the error term on the instruments. For the first instrument, distance to the local Teagasc advisory office was expected to negatively influence the decision to participate in extension services (relevant), but not to affect personal characteristics of the farmer such as their innate ability or motivation that would affect farm family income (valid). Thus, it was expected to be correlated with the decision to participate in extension, but be exogenous to the omitted variables contained in the error term. For example, a farmer located a significant distance from a local office may choose not to participate with extension services, but this distance is unlikely to affect their farming capabilities or farm performance. Moreover, the location of farms in Ireland is largely due to inheritance, as opposed to a conscious choice of where to locate, and thus the distance to a local office is largely exogenous. Furthermore, Läpple and Hennessy (2015) argued that advisory offices may be strategically located in advantaged regions where impact is likely to be pronounced. For example, areas with suitable soil for intensive production may be targeted as ideal locations to base the provision of extension services, given dissemination benefits and likelihood of participation. Indeed, Anderson and Feder (2004) suggested that economies of scale were lost and coordination difficulties rose as advisory offices became decentralized from selected locations. However, in the case of Ireland the number and geographic spread of offices reduced this effect as farms were an average of 10.7 km from their nearest office, with a maximum distance recorded as 44.4 km pre 2009 and an average of 11.7 km with a maximum of 62.2 km thereafter when the number of Teagasc advisory offices were rationalized. This instrument was calculated by measuring the geographic distance from each observation to the nearest local advisory office. This was done by taking the geographic coordinates of each local office, and measuring the distance from the geo-referenced codes for each observation derived from the Teagasc NFS. The distance to office can be classified as random given the nationwide spread of offices as evident from the map in figure 1. Figure 1 View largeDownload slide Location of public extension offices in Ireland 2000-2013 Figure 1 View largeDownload slide Location of public extension offices in Ireland 2000-2013 The second instrument was the policy shock caused by the introduction of the Single Farm Payment Scheme in 2005, which replaced previous coupled payment schemes with a decoupled payment based on average subsidy receipts calculated over a historical reference period (2000-2002). In other words, the value of this annual flat payment was based on production decisions taken in that period (O’Donoghue and Hennessy 2015) and therefore did not vary on a current annual basis. Although additional schemes were included in the new policy that may have asymmetrically affected farm systems differently such as cross compliance, the increased number of clients was relatively symmetrical across all systems. This increase ranged from 9% to 12.5% across all systems after the new policy was introduced. Overall, Teagasc (2006) reported a 20% increase in client numbers credited to the complexities of the new scheme in 2005. Accordingly, the decision to participate in extension was influenced by the new scheme but did not directly affect the current farm performance. In addition, the timing of the introduction of the scheme was exogenous, as it was decided by policy makers and not farmers. Thus it also fulfils the exogeneity requirement for IV analysis. This binary variable was developed by assigning a value of 1 if the year was after 2005 and a 0 if before. The two instruments were also interacted to examine the impact of distance given the policy change. In other words, was the effect of distance on the decision to participate in extension less influential once the new policy was introduced? It is expected that it would remain a negative relationship but not as pronounced in magnitude given the increased numbers of clients. This addition helped to improve the functional form and robustness of the final model. Although it is important to be cautious with regard to the claim of validity with regard to instruments, formulating a strong argument in their favor is helpful (Murray 2006; Cawley and Meyerhoefer 2012). Thus, the instruments utilized here are based on the arguments above and these claims are further validated in the diagnostic results in terms of relevance and validity. Given these prerequisites, we have confidence that the instruments applied are effective in combating endogeneity in the results. Summary statistics of all variables are provided in table 1. Table 1 Data Description and Summary Statistics Variable Description Mean SD Min. Max. FFI/ha Family farm income (€) per ha 456.60 434.52 −1,798.38 3,572.29 Ln FFI/ha Log of family farm income per ha 5.92 0.94 −1.63 8.18 Advisory Participation = 1 if Teagasc client 0.54 0.50 0 1 Ln Land Value/ha Log of land value per ha −0.11 0.56 −4.80 2.61 Farm Size No. of utilisable hectares 38.03 34.55 2.80 1,116.58 Stocking Density Total Livestock Units per ha 1.36 0.63 0 4.80 Ln Labour Log of unpaid family labour −0.06 0.51 −4.61 1.43 Age Age of farmer 55.13 12.22 17 90 Years Agri ed = 0.5 if short course ; = 2 if ag cert; = 4 if ag university 0.68 0.98 0 4 System: Dairy = 1 if dairy 0.12 0.32 0 1 Dairy & Other = 1 if dairy & other 0.07 0.26 0 1 Cattle Rearing = 1 if cattle rearing 0.17 0.38 0 1 Cattle Other = 1 if cattle other 0.20 0.40 0 1 Mainly Sheep = 1 if mainly sheep 0.12 0.33 0 1 Tillage = 1 if tillage 0.05 0.21 0 1 Region: Border = 1 if farm is in the border region 0.20 0.40 0 1 Dublin = 1 if farm is in the Dublin region 0.01 0.10 0 1 East = 1 if farm is in the east region 0.09 0.28 0 1 Midlands = 1 if farm is in the midlands 0.11 0.31 0 1 Southwest = 1 if farm is in the southwest 0.10 0.30 0 1 Southeast = 1 if farm is in the southeast 0.14 0.35 0 1 South = 1 if farm is in the south region 0.18 0.38 0 1 Medium Soil 0.40 0.49 0 1 Poor Soil 0.12 0.33 0 1 Dist_advoff Distance to advisory office (km) 10.71 8.63 0 62.16 SFPyr = 1 if advisory client after 2005 0.66 0.48 0 1 SFPYR*Dist Interactive term for clients and distance 7.40 9.16 0 62.16 Variable Description Mean SD Min. Max. FFI/ha Family farm income (€) per ha 456.60 434.52 −1,798.38 3,572.29 Ln FFI/ha Log of family farm income per ha 5.92 0.94 −1.63 8.18 Advisory Participation = 1 if Teagasc client 0.54 0.50 0 1 Ln Land Value/ha Log of land value per ha −0.11 0.56 −4.80 2.61 Farm Size No. of utilisable hectares 38.03 34.55 2.80 1,116.58 Stocking Density Total Livestock Units per ha 1.36 0.63 0 4.80 Ln Labour Log of unpaid family labour −0.06 0.51 −4.61 1.43 Age Age of farmer 55.13 12.22 17 90 Years Agri ed = 0.5 if short course ; = 2 if ag cert; = 4 if ag university 0.68 0.98 0 4 System: Dairy = 1 if dairy 0.12 0.32 0 1 Dairy & Other = 1 if dairy & other 0.07 0.26 0 1 Cattle Rearing = 1 if cattle rearing 0.17 0.38 0 1 Cattle Other = 1 if cattle other 0.20 0.40 0 1 Mainly Sheep = 1 if mainly sheep 0.12 0.33 0 1 Tillage = 1 if tillage 0.05 0.21 0 1 Region: Border = 1 if farm is in the border region 0.20 0.40 0 1 Dublin = 1 if farm is in the Dublin region 0.01 0.10 0 1 East = 1 if farm is in the east region 0.09 0.28 0 1 Midlands = 1 if farm is in the midlands 0.11 0.31 0 1 Southwest = 1 if farm is in the southwest 0.10 0.30 0 1 Southeast = 1 if farm is in the southeast 0.14 0.35 0 1 South = 1 if farm is in the south region 0.18 0.38 0 1 Medium Soil 0.40 0.49 0 1 Poor Soil 0.12 0.33 0 1 Dist_advoff Distance to advisory office (km) 10.71 8.63 0 62.16 SFPyr = 1 if advisory client after 2005 0.66 0.48 0 1 SFPYR*Dist Interactive term for clients and distance 7.40 9.16 0 62.16 Note: All variables were recorded over the period 2000 to 2013. Given this data, IV models were estimated by analyzing the impact on farm family income. Table 1 Data Description and Summary Statistics Variable Description Mean SD Min. Max. FFI/ha Family farm income (€) per ha 456.60 434.52 −1,798.38 3,572.29 Ln FFI/ha Log of family farm income per ha 5.92 0.94 −1.63 8.18 Advisory Participation = 1 if Teagasc client 0.54 0.50 0 1 Ln Land Value/ha Log of land value per ha −0.11 0.56 −4.80 2.61 Farm Size No. of utilisable hectares 38.03 34.55 2.80 1,116.58 Stocking Density Total Livestock Units per ha 1.36 0.63 0 4.80 Ln Labour Log of unpaid family labour −0.06 0.51 −4.61 1.43 Age Age of farmer 55.13 12.22 17 90 Years Agri ed = 0.5 if short course ; = 2 if ag cert; = 4 if ag university 0.68 0.98 0 4 System: Dairy = 1 if dairy 0.12 0.32 0 1 Dairy & Other = 1 if dairy & other 0.07 0.26 0 1 Cattle Rearing = 1 if cattle rearing 0.17 0.38 0 1 Cattle Other = 1 if cattle other 0.20 0.40 0 1 Mainly Sheep = 1 if mainly sheep 0.12 0.33 0 1 Tillage = 1 if tillage 0.05 0.21 0 1 Region: Border = 1 if farm is in the border region 0.20 0.40 0 1 Dublin = 1 if farm is in the Dublin region 0.01 0.10 0 1 East = 1 if farm is in the east region 0.09 0.28 0 1 Midlands = 1 if farm is in the midlands 0.11 0.31 0 1 Southwest = 1 if farm is in the southwest 0.10 0.30 0 1 Southeast = 1 if farm is in the southeast 0.14 0.35 0 1 South = 1 if farm is in the south region 0.18 0.38 0 1 Medium Soil 0.40 0.49 0 1 Poor Soil 0.12 0.33 0 1 Dist_advoff Distance to advisory office (km) 10.71 8.63 0 62.16 SFPyr = 1 if advisory client after 2005 0.66 0.48 0 1 SFPYR*Dist Interactive term for clients and distance 7.40 9.16 0 62.16 Variable Description Mean SD Min. Max. FFI/ha Family farm income (€) per ha 456.60 434.52 −1,798.38 3,572.29 Ln FFI/ha Log of family farm income per ha 5.92 0.94 −1.63 8.18 Advisory Participation = 1 if Teagasc client 0.54 0.50 0 1 Ln Land Value/ha Log of land value per ha −0.11 0.56 −4.80 2.61 Farm Size No. of utilisable hectares 38.03 34.55 2.80 1,116.58 Stocking Density Total Livestock Units per ha 1.36 0.63 0 4.80 Ln Labour Log of unpaid family labour −0.06 0.51 −4.61 1.43 Age Age of farmer 55.13 12.22 17 90 Years Agri ed = 0.5 if short course ; = 2 if ag cert; = 4 if ag university 0.68 0.98 0 4 System: Dairy = 1 if dairy 0.12 0.32 0 1 Dairy & Other = 1 if dairy & other 0.07 0.26 0 1 Cattle Rearing = 1 if cattle rearing 0.17 0.38 0 1 Cattle Other = 1 if cattle other 0.20 0.40 0 1 Mainly Sheep = 1 if mainly sheep 0.12 0.33 0 1 Tillage = 1 if tillage 0.05 0.21 0 1 Region: Border = 1 if farm is in the border region 0.20 0.40 0 1 Dublin = 1 if farm is in the Dublin region 0.01 0.10 0 1 East = 1 if farm is in the east region 0.09 0.28 0 1 Midlands = 1 if farm is in the midlands 0.11 0.31 0 1 Southwest = 1 if farm is in the southwest 0.10 0.30 0 1 Southeast = 1 if farm is in the southeast 0.14 0.35 0 1 South = 1 if farm is in the south region 0.18 0.38 0 1 Medium Soil 0.40 0.49 0 1 Poor Soil 0.12 0.33 0 1 Dist_advoff Distance to advisory office (km) 10.71 8.63 0 62.16 SFPyr = 1 if advisory client after 2005 0.66 0.48 0 1 SFPYR*Dist Interactive term for clients and distance 7.40 9.16 0 62.16 Note: All variables were recorded over the period 2000 to 2013. Given this data, IV models were estimated by analyzing the impact on farm family income. Results Although we expect that OLS estimation would lead to bias in the results, these estimates are presented in addition to the IV estimates to illustrate the scale of the difference between both estimation methods when endogeneity concerns are addressed. Furthermore, the three instruments were inserted cumulatively to monitor each effect on the dependent variable and subsequent diagnostic statistics. Thus, in total four models were estimated, one using OLS, and the others using IV with one, two, and three instruments inserted, respectively. IV Results – First Stage Results The results of the first stage of the IV process are presented in table 2, confirming the relevance of the instruments on extension participation decisions. Table 2 First Stage Results of IV: Advisory Participation 1 Instrument 2 Instruments 3 Instruments Advisory Participation Coeff. (SE) p value Coeff. (SE) p value Coeff. (SE) p value SFP Policy Change 0.530 (0.010) .000 0.531 (0.010) .000 0.565 (0.016) .000 Dist. Adv Office −0.002 (0.001) .000 0.001 (0.001) .649 Interaction Term −0.004 (0.001) .004 CD Wald F Stat 2639.20 1331.86 891.43 1 Instrument 2 Instruments 3 Instruments Advisory Participation Coeff. (SE) p value Coeff. (SE) p value Coeff. (SE) p value SFP Policy Change 0.530 (0.010) .000 0.531 (0.010) .000 0.565 (0.016) .000 Dist. Adv Office −0.002 (0.001) .000 0.001 (0.001) .649 Interaction Term −0.004 (0.001) .004 CD Wald F Stat 2639.20 1331.86 891.43 Note: The n = 8,951, endogenous regressor (Advisory Participation), 3 instruments (Single Farm Payment year, Distance to advisory office and Interaction of both), additional explanatory variables included land value, farm system, labor, size, off-farm job, age, region, stocking density, and soil group. A P value <.01 indicates statistical significance at the 1% level, and the Cragg-Donald Wald F Stat measures relevancy of instruments Table 2 First Stage Results of IV: Advisory Participation 1 Instrument 2 Instruments 3 Instruments Advisory Participation Coeff. (SE) p value Coeff. (SE) p value Coeff. (SE) p value SFP Policy Change 0.530 (0.010) .000 0.531 (0.010) .000 0.565 (0.016) .000 Dist. Adv Office −0.002 (0.001) .000 0.001 (0.001) .649 Interaction Term −0.004 (0.001) .004 CD Wald F Stat 2639.20 1331.86 891.43 1 Instrument 2 Instruments 3 Instruments Advisory Participation Coeff. (SE) p value Coeff. (SE) p value Coeff. (SE) p value SFP Policy Change 0.530 (0.010) .000 0.531 (0.010) .000 0.565 (0.016) .000 Dist. Adv Office −0.002 (0.001) .000 0.001 (0.001) .649 Interaction Term −0.004 (0.001) .004 CD Wald F Stat 2639.20 1331.86 891.43 Note: The n = 8,951, endogenous regressor (Advisory Participation), 3 instruments (Single Farm Payment year, Distance to advisory office and Interaction of both), additional explanatory variables included land value, farm system, labor, size, off-farm job, age, region, stocking density, and soil group. A P value <.01 indicates statistical significance at the 1% level, and the Cragg-Donald Wald F Stat measures relevancy of instruments The above table shows there is a jointly significant relationship between the instruments and the endogenous regressor. Individually, when one instrument is applied, the policy change is a significant explanatory factor in the decision to participate in extension services. When the distance to local office is added, both instruments remain significant at the 1% level and the signs are as expected, with the policy change being positive and the distance being negative. However, the magnitude of the distance instrument is relatively small. This could be due to the fact that there were 90 local Teagasc offices in existence before a restructuring plan was introduced in 2009. Thus, the average distance increased after the office closures, as noted previously. Accordingly, the relative distances to local offices were not practically large. Once all three instruments are applied, the distance becomes positive and insignificant. However, as the interactive term is included, this variable becomes significant and negative as expected, showing the conditional influence of distance after the introduction of the policy change. The Cragg-Donald Wald F statistic illustrates the joint significance of the instruments and shows a strong positive relationship between all 3 instruments and the dependent variable in the first stage. In our case, Stock, Wright, and Yogo (2002) limit of 10 is easily exceeded and we can conclude that the instruments are relevant. IV Results – Second Stage Results The second stage of the IV process involved inserting the predicted values of the endogenous regressors (now exogenous) from the first stage into the main structural equation and applying them to the dependent variable. Accordingly, the results of the IV estimates for the main variable of interest are presented along with selected diagnostics in table 3 for clarity with the full table of results, including the control variable estimates available in appendix A. The OLS estimates are also included for comparison. Table 3 IV Parameter Estimates: Model of Log of Farm Family Income per Ha OLS IV – 1 Instrument IV – 2 Instruments IV – 3 Instruments Extension Participation 0.19*** (0.02) 0.35*** (0.04) 0.35*** (0.04) 0.35*** (0.04) R2 0.22 Centred R2 0.22 0.22 0.22 Sargan p value 0.00 0.81 0.30 OLS IV – 1 Instrument IV – 2 Instruments IV – 3 Instruments Extension Participation 0.19*** (0.02) 0.35*** (0.04) 0.35*** (0.04) 0.35*** (0.04) R2 0.22 Centred R2 0.22 0.22 0.22 Sargan p value 0.00 0.81 0.30 Note: The n = 8,951, additional explanatory variables included year, land value, farm system, labor, size, off-farm job, age, region, stocking density and soil group; standard errors appear in parenthesis; Asterisks represent statistical significance of p values: *** for 1% significance; ** for 5% significance; and * for 10% significance. Full tables of results available in appendix A; Sargan Overidentification P Value >.1 means that we fail to reject that the null of the instruments are valid, not applicable with 1 instrument as the equation is exactly identified. Table 3 IV Parameter Estimates: Model of Log of Farm Family Income per Ha OLS IV – 1 Instrument IV – 2 Instruments IV – 3 Instruments Extension Participation 0.19*** (0.02) 0.35*** (0.04) 0.35*** (0.04) 0.35*** (0.04) R2 0.22 Centred R2 0.22 0.22 0.22 Sargan p value 0.00 0.81 0.30 OLS IV – 1 Instrument IV – 2 Instruments IV – 3 Instruments Extension Participation 0.19*** (0.02) 0.35*** (0.04) 0.35*** (0.04) 0.35*** (0.04) R2 0.22 Centred R2 0.22 0.22 0.22 Sargan p value 0.00 0.81 0.30 Note: The n = 8,951, additional explanatory variables included year, land value, farm system, labor, size, off-farm job, age, region, stocking density and soil group; standard errors appear in parenthesis; Asterisks represent statistical significance of p values: *** for 1% significance; ** for 5% significance; and * for 10% significance. Full tables of results available in appendix A; Sargan Overidentification P Value >.1 means that we fail to reject that the null of the instruments are valid, not applicable with 1 instrument as the equation is exactly identified. The results presented here show that there are consistent positive returns to participating with extension services and all are significant across all models. The OLS results indicate a 19% increase in farm family income per hectare, ceteris paribus. However, as this variable suffers an endogeneity bias, the coefficients estimated for the IV models offer a more accurate prediction, and as evidenced in the table, this return is approximately 35% across the three models, with instruments added cumulatively. Estimation with the distance to advisory office also showed a positive impact but its statistical significance was not strong due to the weaker relevance of the instrument. These findings are in line with previous findings in the literature (26% in Akobundu et al. 2004; 61% in Davis et al. 2012; 55% in O’Donoghue and Hennessy 2015). However, as this measure does not distinguish between the type of extension activity, the precision of individual impact is not captured. Furthermore, as the purpose of this study was to include all forms of extension participation, this measure includes subsidy receipts which, although an annual flat rate payment, contribute substantially to family farm income, particularly across all farm systems (Hanrahan et al. 2014). The consistency of the estimates justifies the instruments as being strong predictors of extension participation, and thus instrumenting it successfully to identify the causal impact of extension participation on farm family income per hectare. Interestingly, these effects are estimated alongside the control variables as noted above. Of those controls, farm system, education, stocking density, land quality, and off-farm employment were important predictors of family farm income, as expected. For example, dairy farmers showed an increase of 37% per hectare compared to non-dairy systems under the IV approach as opposed to 34% under OLS. Similarly, tillage systems increased their impact under the IV to 29% as opposed to 26% under OLS. Conversely, the effect of farm size was insignificant, possibly due to the per hectare basis of the dependent variable, and there was mixed evidence for different regions. Most other variables experienced a minimal change under both systems, but extension participation showed the largest increase. The full table of results for each model is provided in appendix A. Furthermore, the Sargan test statistics report p values exceeding significance values of 0.1, meaning that we fail to reject the null hypothesis that the instruments are valid. Indeed, as it is impossible to prove the null hypothesis of no effect based on the nature of the unobserved error term (Cawley and Meyerhoefer 2012), the Sargan statistic reverses the process and in this case the null is that the instruments have an effect, and we fail to reject that claim. In other words, these instruments can be argued as valid in addressing the endogeneity issue of extension participation, and thus our estimates provide a consistent and positive impact on farm family income per hectare. Given the results of the analyses we can infer that extension services had a positive impact on farm income. Discussion and Conclusion While much of the previous literature identified a positive relationship between participation in extension services and farm level outcomes, the IV approach presented here provides a robust estimation that comprehensively addresses endogeneity concerns. In line with previous literature, the IV estimates of the impact of extension services are uniformly higher than the OLS estimates (Card 1993; Cawley and Meyerhoefer 2012). Therefore, there is a clear indication of a net benefit to extension participation in the first instance. Identifying appropriate instruments is the critical challenge of the IV approach, particularly in terms of confirming exogeneity to the error term, and thus estimates should be interpreted with caution (Cawley and Meyerhoefer 2012). However, in this analysis the distance to the local advisory office and policy change instruments were relevant to the decision to participate in extension and valid in terms of uncorrelated to the dependent variable and error term based on both argument and the diagnostic tests, and thus, the results presented are superior estimates of the impact of extension services on farm income. In terms of policy implications, the results suggest a positive causal effect of participating in extension programs in terms of farm income. Furthermore, these findings are unbiased, implying that the impact is greater than previously thought and therefore the role of extension as a policy tool is effective. Thus, knowledge sharing of the benefits of participation should be more widely promoted among farmers to encourage increased engagement that can be leveraged to achieve policy goals. For example, if governments aim to grow agricultural output, an extension program could be targeted to provide technical assistance to bolster productivity through improved feed management to increase overall output levels. Conversely, farms could be targeted with environmental advice to curb greenhouse gas emissions as set out in international regulations. A more targeted approach that incentivizes deeper forms of engagement may be valuable (Akobundu et al. 2004), but this requires further analysis to identify the effect of different levels of extension engagement. Nonetheless, policy targets set for the agricultural sector by government should be adequately supported by a dynamic effective extension service. This study provides a robust estimation of the impact of extension services across all farming sectors on an aggregated basis in Ireland. However, a number of caveats should be considered when utilizing an IV approach, and further research is needed to enforce the findings provided here. Firstly, when applying the IV approach, the validity of the instruments is key. Although we have confidence and the results defend their validity, it is logical to assume there may be alternative instruments that could be used that are not available in this dataset. For example, the neighbor or peer effect could have been instrumented as farmers may be more likely to become clients based on their peers’ participation. Similarly, the availability of advisers could have been a useful instrument given the drop in numbers due to retirements over the same period reducing the availability of services. However, both of these instruments were not possible due to data limitations. Moreover, the distance to advisory office instrument could prove more effective if the offices were classified according to the types of services offered, as smaller offices may not have the facilities to address more intensive forms of extension participation. Similarly, the endogenous variable used here is a dummy and thus does not reflect the variability of extension programs available, and thus cannot rigorously distinguish different types of extension participation nor assess their impact. Distinguishing on the basis of the motivation to engage would enrich the analysis further. Indeed, the knowledge transfer effect is not homogenous, but more likely to be heterogeneous in nature (Coccia 2008). However, this research provides a foundation to build upon as the IV approach delivers a more precise and unbiased estimate of the positive effect of extension advice to farms, ensuring that the “value for money” objective for policy makers is met. This is an important first step in the evaluation of impact, and future research could aim to identify the impact of different types of extension. References Abdallah W. , Goergen M. , O’Sullivan N . 2015 . Endogeneity: How Failure to Correct for It Can Cause Wrong Inferences and Some Remedies . British Journal of Management 26 : 791 – 804 . Google Scholar CrossRef Search ADS Akobundu E. , Alwang J. , Essel A. , Norton G.W. , Tegene A . 2004 . Does Extension Work? Impacts of a Program to Assist Limited-Resource Farmers in Virginia . Applied Economic Perspectives and Policy 26 : 361 – 72 . Anderson J.R. , Feder G . 2004 . Agricultural Extension: Good Intentions and Hard Realities . The World Bank Research Observer 19 : 41 – 60 . Google Scholar CrossRef Search ADS Birkhaeuser D. , Evenson R.E. , Feder G . 1991 . The Economic Impact of Agricultural Extension: A Review . Economic Development and Cultural Change 39 : 607 – 50 . Google Scholar CrossRef Search ADS Birner R. , Davis K. , Pender J. , Nkonya E. , Anandajayasekeram P. , Ekboir J. , Mbabu A. , et al. 2009 . From Best Practice to Best Fit: A Framework for Designing and Analyzing Pluralistic Agricultural Advisory Services Worldwide . The Journal of Agricultural Education and Extension 15 : 341 – 55 . Google Scholar CrossRef Search ADS Boyle G. 2008 . A Future Driven by Knowledge and Innovation. In Growing Knowledge - Fifty Years of Research and Development in Irish Farming and Food , ed. Miley M. , 193 – 205 . Carlow : Teagasc Publications Office . Buckley C. , Wall D.P. , Moran B. , O’Neill S. , Murphy P.N.C . 2016 . Phosphorous Management on Irish Dairy Farms Post Controls Introduced under the EU Nitrates Directive . Agricultural System 142 : 1 – 8 . Google Scholar CrossRef Search ADS Burgess S. , Dudbridge F. , Thompson S.G . 2016 . Combining Information on Multiple Instrumental Variables in Mendelian Randomization: Comparison of Allele Score and Summarized Data Methods . Statistics in Medicine 35 : 1880 – 906 . Google Scholar CrossRef Search ADS PubMed Caliendo M. , Kopeinig S . 2008 . Some Practical Guidance for the Implementation of Propensity Score Matching . Journal of Economic Surveys 22 : 31 – 82 . Google Scholar CrossRef Search ADS Callan T. , Harmon C . 1999 . The Economic Return to Schooling in Ireland . Labour Economics 6 : 543 – 50 . Google Scholar CrossRef Search ADS Card D. 1993 . Using Geographic Variation in College Proximity to Estimate the Return to Schooling. NBER Working Paper no. 4483, National Bureau of Economic Research, Cambridge, MA. Card D. 1999 . The Causal Effect of Education on Earnings . Handbook of Labor Economics 3 : 1801 – 63 . Google Scholar CrossRef Search ADS Cawley J. , Meyerhoefer C . 2012 . The Medical Care Costs of Obesity: An Instrumental Variables Approach . Journal of Health Economics 31 : 219 – 30 . Google Scholar CrossRef Search ADS PubMed Coccia M. 2008 . Spatial Mobility of Knowledge Transfer and Absorptive Capacity: Analysis and Measurement of the Impact within the Geoeconomic Space . Journal of Technology Transfer 33 : 105 – 22 . Google Scholar CrossRef Search ADS Connolly L. , Kinsella A. , Quinlan A.C. , Moran B . 2010 . National Farm Survey. Carlow : Teagasc Publications Office . Davis K. , Nkonya E. , Kato E. , Mekonnen D.A. , Odendo M. , Miiro R. , Nkuba J . 2012 . Impact of Farmer Field Schools on Agricultural Productivity and Poverty in East Africa . World Development 40 : 402 – 13 . Google Scholar CrossRef Search ADS Dercon S. , Gilligan D. , Hoddinott J. , Woldehanna T . 2009 . The Impact of Agricultural Extension and Roads on Poverty and Consumption Growth in Fifteen Ethiopian Villages . American Journal of Agricultural Economics 91 : 1007 – 21 . Google Scholar CrossRef Search ADS Donnellan T. , Hanrahan K. , Hennessy T . 2008 . Study on the Functioning of Land Markets in the EU Member States under the Influence of Measures Applied under the Common Agricultural Policy, RMIS 5881. Carlow: Teagasc Publications Office. Gabel M. , Scheve K . 2007 . Estimating the Effect of Elite Communications on Public Opinion Using Instrumental Variables . American Journal of Political Science 51 : 1013 – 28 . Google Scholar CrossRef Search ADS Garforth C. , Angell B. , Archer J. , Green K . 2003 . Fragmentation or Creative Diversity? Options in the Provision of Land Management Advisory Services . Land Use Policy 20 : 323 – 33 . Google Scholar CrossRef Search ADS Griliches Z. 1964 . Research Expenditures, Education, and the Aggregate Agricultural Production Function . The American Economic Review 54 : 961 – 74 . Gujarati D.N. 2003 . Basic Econometrics . New York : McGraw Hill . Hanrahan K. , Hennessy T. , Kinsella A. , Moran B . 2014 . Teagasc National Farm Survey 2013 . Carlow : Teagasc Publications Office . Heanue K. , O'Donoghue C . 2014 . The Economic Returns to Formal Agricultural Education. Carlow : Teagasc Publications Office . Hennessy T. , Heanue K . 2012 . Quantifying the Effect of Discussion Group Membership on Technology Adoption and Farm Profit on Dairy Farms . The Journal of Agricultural Education and Extension 18 : 41 – 54 . Google Scholar CrossRef Search ADS Hennessy T. , Rehman T. 2007 . An Investigation into Factors Affecting the Occupational Choices of Nominated Farm Heirs in Ireland . Journal of Agricultural Economics 58 : 61 – 75 . Google Scholar CrossRef Search ADS Howley P. , O’Neill S. , Atkinson R . 2015 . Who Needs Good Neighbors? Environment and Planning A 47 : 939 – 56 . Google Scholar CrossRef Search ADS Hunt W. , Birch C. , Vanclay F. , Coutts J . 2014 , Recommendations Arising from an Analysis of Changes to the Australian Agricultural Research, Development and Extension System . Food Policy 44 : 129 – 41 . Google Scholar CrossRef Search ADS Imbens G.W. , Wooldridge J.M . 2009 . Recent Developments in the Econometrics of Program Evaluation . Journal of Economic Literature 47 : 5 – 86 . Google Scholar CrossRef Search ADS Krishnan P. , Patnam M . 2014 . Neighbours and Extension Agents in Ethiopia: Who Matters More for Technology Adoption? American Journal of Agricultural Economics 96 : 308 – 327 . Google Scholar CrossRef Search ADS Läpple D. , Hennessy T . 2015 . Assessing the Impact of Financial Incentives in Extension Programs: Evidence from Ireland . Journal of Agricultural Economics 66 : 781 – 95 . Google Scholar CrossRef Search ADS Läpple D. , Hennessy T. 2015 . Exploring the Role of Incentives in Agricultural Extension Programs . Applied Economic Perspectives and Policy 37 : 403 – 17 . Google Scholar CrossRef Search ADS Läpple D. , Hennessy T. , Newman C . 2013 . Quantifying the Economic Return to Participatory Extension Programs in Ireland: an Endogenous Switching Regression Analysis . Journal of Agricultural Economics 64 : 467 – 82 . Google Scholar CrossRef Search ADS Leonard B. , Mahon M. , Kinsella A. , O’Donoghue C. , Farrell M. , Curran T. , Hennessy T . 2017 . The Potential of Farm Partnerships to Facilitate Farm Succession and Inheritance International Journal of Agricultural Management 6 : 1 – 16 . Marsh S.P. , Pannell D.J. , Linder R.K . 2004 . Does Agricultural Extension Pay? Agricultural Economics 30 : 17 – 30 . Miguel E. , Satyanath S. , Sergenti E . 2004 . Economic Shocks and Civil Conflict: An Instrumental Variables Approach . Journal of political Economy 112 : 725 – 53 . Google Scholar CrossRef Search ADS Murray M.P. 2006 . Avoiding Invalid Instruments and Coping with Weak Instruments . The Journal of Economic Perspectives 20 : 111 – 32 . Google Scholar CrossRef Search ADS Nordin M. , Höjgård S. 2016 . An Evaluation of Extension Services in Sweden . Agricultural Economics 48 : 1 – 10 . O’Donoghue C. , Hennessy T . 2015 . Policy and Economic Change in the Agri-Food Sector in Ireland . The Economic and Social Review 46 : 315 – 37 . Owens T. , Hoddinott J. , Kinsey B . 2003 . The Impact of Agricultural Extension on Farm Production in Resettlement Areas of Zimbabwe . Economic Development and Cultural Change 51 : 337 – 57 . Google Scholar CrossRef Search ADS Prager K. , Labarthe P. , Caggiano P.M. , Lorenzo-Arribas A . 2016 . How Does Commercialisation Impact on the Provision of Farm Advisory Services? Evidence from Belgium, Italy, Ireland and the UK . Land Use Policy 52 : 329 – 44 . Google Scholar CrossRef Search ADS Ragasa C. , Ulimwengu J. , Randriamamonjy J. , Badibanga T. , 2016 . Factors Affecting Performance of Agricultural Extension: Evidence from Democratic Republic of Congo . The Journal of Agricultural Education and Extension 22 : 113 – 43 . Google Scholar CrossRef Search ADS Rassen J.A. , Brookhart M.A. , Glynn R.J. , Mittleman M.A. , Schneeweiss S . 2009 . Instrumental Variables I: Instrumental Variables Exploit Natural Variation in Non-experimental Data to Estimate Causal Relationships Journal of Clinical Epidemiology 62 : 1226 – 32 . Google Scholar CrossRef Search ADS PubMed Stock J.H. , Wright J.H. , Yogo M . 2002 . A Survey of Weak Instruments and Weak Identification in Generalized Method of Moments . Journal of Business & Economic Statistics 20 : 518 – 29 . Google Scholar CrossRef Search ADS Stukel T.A. , Fisher E.S. , Wennberg D.E. , Alter D.A. , Gottlieb D.J. , Vermuelen M.J . 2007 . Analysis of Observational Studies in the Presence of Treatment Selection Bias: Effects of Invasive Cardiac Management on AMI survival using Propensity Score and Instrumental Variable Methods . Journal of American Medical Association 297 : 278 – 85 . Google Scholar CrossRef Search ADS Tamini L.D. 2011 . A Nonparametric Analysis of the Impact of Agri-environmental Advisory Activities on Best Management Practice Adoption: A Case Study of Québec . Ecological Economics 70 : 1363 – 74 . Google Scholar CrossRef Search ADS Teagasc . 2006 . Annual Report 2005 and Financial Statements . Carlow : Teagasc Publications Office . Teagasc . 2016 . Our Organization. About Us. Available at: http://www.teagasc.ie/about/our-organization/. Van den Ban A.W. , Hawkins H.S . 1988 . Agricultural Extension . Essex : Longman Scientific & Technical . Vanclay F. , Leach G . 2011 . Enabling Change in Rural and Regional Australia. In Shaping Change: Natural Resource Management, Agriculture and the Role of Extension . Australasia Pacific Extension Network (APEN), ed. Jennings J. , Packham R. , Woodside D. , 6 – 11 . Australasia Pacific Extension Network . Wooldridge J.M. 2013 . Introductory Econometrics - A Modern Approach . South-Western : Cengage . Appendices Appendix A – Full table of OLS and IV estimates (2000-2013) Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Regression Instrumental Variable Regression – 1 Instrument Has Advisory Participation 0.19 0.02 0.00 0.35 0.41 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.34 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.09 0.04 0.03 −0.06 0.04 0.16 Cattle Rearing −0.19 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.19 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.08 0.03 0.02 −0.03 0.03 0.39 Tillage 0.26 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.15 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.13 0.01 0.01 0.08 Age Squared −0.00 0.00 0.11 −0.00 0.00 0.06 Has off farm employment −0.12 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.02 0.02 0.22 0.00 0.02 0.87 Has Forestry −0.05 0.04 0.22 −0.09 0.04 0.05 Completed Agricultural Short Course 0.22 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.20 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.25 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.17 0.04 0.00 −0.14 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.01 0.04 0.88 0.02 0.04 0.66 Cork, Kerry −0.15 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.98 0.00 0.03 0.95 Stocking Density 0.42 0.02 0.00 0.40 0.02 0.00 Medium Soil −0.07 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.01 0.04 0.77 −0.03 0.03 0.37 Constant 5.08 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (Weak Instrument) 2639.20 Sargan statistic p value (Overidentification Test) 0.000 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Regression Instrumental Variable Regression – 1 Instrument Has Advisory Participation 0.19 0.02 0.00 0.35 0.41 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.34 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.09 0.04 0.03 −0.06 0.04 0.16 Cattle Rearing −0.19 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.19 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.08 0.03 0.02 −0.03 0.03 0.39 Tillage 0.26 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.15 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.13 0.01 0.01 0.08 Age Squared −0.00 0.00 0.11 −0.00 0.00 0.06 Has off farm employment −0.12 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.02 0.02 0.22 0.00 0.02 0.87 Has Forestry −0.05 0.04 0.22 −0.09 0.04 0.05 Completed Agricultural Short Course 0.22 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.20 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.25 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.17 0.04 0.00 −0.14 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.01 0.04 0.88 0.02 0.04 0.66 Cork, Kerry −0.15 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.98 0.00 0.03 0.95 Stocking Density 0.42 0.02 0.00 0.40 0.02 0.00 Medium Soil −0.07 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.01 0.04 0.77 −0.03 0.03 0.37 Constant 5.08 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (Weak Instrument) 2639.20 Sargan statistic p value (Overidentification Test) 0.000 Note: The endogenous regressor is (Advisory Participation); 1 instrument (Single Farm Payment year policy change); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak; Sargan Statistic void due to equation being exactly identified. View Large Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Regression Instrumental Variable Regression – 1 Instrument Has Advisory Participation 0.19 0.02 0.00 0.35 0.41 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.34 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.09 0.04 0.03 −0.06 0.04 0.16 Cattle Rearing −0.19 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.19 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.08 0.03 0.02 −0.03 0.03 0.39 Tillage 0.26 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.15 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.13 0.01 0.01 0.08 Age Squared −0.00 0.00 0.11 −0.00 0.00 0.06 Has off farm employment −0.12 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.02 0.02 0.22 0.00 0.02 0.87 Has Forestry −0.05 0.04 0.22 −0.09 0.04 0.05 Completed Agricultural Short Course 0.22 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.20 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.25 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.17 0.04 0.00 −0.14 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.01 0.04 0.88 0.02 0.04 0.66 Cork, Kerry −0.15 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.98 0.00 0.03 0.95 Stocking Density 0.42 0.02 0.00 0.40 0.02 0.00 Medium Soil −0.07 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.01 0.04 0.77 −0.03 0.03 0.37 Constant 5.08 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (Weak Instrument) 2639.20 Sargan statistic p value (Overidentification Test) 0.000 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Regression Instrumental Variable Regression – 1 Instrument Has Advisory Participation 0.19 0.02 0.00 0.35 0.41 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.34 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.09 0.04 0.03 −0.06 0.04 0.16 Cattle Rearing −0.19 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.19 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.08 0.03 0.02 −0.03 0.03 0.39 Tillage 0.26 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.15 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.13 0.01 0.01 0.08 Age Squared −0.00 0.00 0.11 −0.00 0.00 0.06 Has off farm employment −0.12 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.02 0.02 0.22 0.00 0.02 0.87 Has Forestry −0.05 0.04 0.22 −0.09 0.04 0.05 Completed Agricultural Short Course 0.22 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.20 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.25 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.17 0.04 0.00 −0.14 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.01 0.04 0.88 0.02 0.04 0.66 Cork, Kerry −0.15 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.98 0.00 0.03 0.95 Stocking Density 0.42 0.02 0.00 0.40 0.02 0.00 Medium Soil −0.07 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.01 0.04 0.77 −0.03 0.03 0.37 Constant 5.08 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (Weak Instrument) 2639.20 Sargan statistic p value (Overidentification Test) 0.000 Note: The endogenous regressor is (Advisory Participation); 1 instrument (Single Farm Payment year policy change); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak; Sargan Statistic void due to equation being exactly identified. View Large Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – 2 Instruments Instrumental Variable Regression – 3 Instruments Has Advisory Participation 0.35 0.04 0.00 0.35 0.04 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.37 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.06 0.04 0.16 −0.06 0.04 0.16 Cattle Rearing −0.15 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.16 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.03 0.03 0.38 −0.03 0.03 0.37 Tillage 0.29 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.08 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.06 Has off farm employment −0.13 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.00 0.02 0.87 0.00 0.02 0.85 Has Forestry −0.09 0.04 0.05 −0.09 0.04 0.05 Completed Agricultural Short Course 0.21 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.24 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.15 0.04 0.00 −0.15 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.016 0.04 0.66 0.02 0.04 0.66 Cork, Kerry −0.17 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.95 0.00 0.03 0.95 Stocking Density 0.40 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.03 0.03 0.38 Constant 5.05 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 1331.86 891.43 Sargan statistic p value (Overidentification Test) 0.81 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – 2 Instruments Instrumental Variable Regression – 3 Instruments Has Advisory Participation 0.35 0.04 0.00 0.35 0.04 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.37 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.06 0.04 0.16 −0.06 0.04 0.16 Cattle Rearing −0.15 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.16 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.03 0.03 0.38 −0.03 0.03 0.37 Tillage 0.29 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.08 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.06 Has off farm employment −0.13 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.00 0.02 0.87 0.00 0.02 0.85 Has Forestry −0.09 0.04 0.05 −0.09 0.04 0.05 Completed Agricultural Short Course 0.21 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.24 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.15 0.04 0.00 −0.15 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.016 0.04 0.66 0.02 0.04 0.66 Cork, Kerry −0.17 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.95 0.00 0.03 0.95 Stocking Density 0.40 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.03 0.03 0.38 Constant 5.05 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 1331.86 891.43 Sargan statistic p value (Overidentification Test) 0.81 Note: The endogenous regressor is (Advisory Participation in both models); 2 instruments (Single Farm Payment policy change and Distance to advisory office); 3 Instruments (Single Farm Payment policy change, Distance to advisory office and Interaction of both); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors are adjusted for heterogeneity; a p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 considered weak; Sargan Statistic for overidentification p value > 0.1 fails to reject a null hypothesis of the instruments’ validity. View Large Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – 2 Instruments Instrumental Variable Regression – 3 Instruments Has Advisory Participation 0.35 0.04 0.00 0.35 0.04 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.37 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.06 0.04 0.16 −0.06 0.04 0.16 Cattle Rearing −0.15 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.16 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.03 0.03 0.38 −0.03 0.03 0.37 Tillage 0.29 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.08 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.06 Has off farm employment −0.13 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.00 0.02 0.87 0.00 0.02 0.85 Has Forestry −0.09 0.04 0.05 −0.09 0.04 0.05 Completed Agricultural Short Course 0.21 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.24 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.15 0.04 0.00 −0.15 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.016 0.04 0.66 0.02 0.04 0.66 Cork, Kerry −0.17 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.95 0.00 0.03 0.95 Stocking Density 0.40 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.03 0.03 0.38 Constant 5.05 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 1331.86 891.43 Sargan statistic p value (Overidentification Test) 0.81 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – 2 Instruments Instrumental Variable Regression – 3 Instruments Has Advisory Participation 0.35 0.04 0.00 0.35 0.04 0.00 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.02 0.00 Dairy System 0.37 0.03 0.00 0.37 0.03 0.00 Dairy and Other System −0.06 0.04 0.16 −0.06 0.04 0.16 Cattle Rearing −0.15 0.03 0.00 −0.15 0.03 0.00 Cattle Other System −0.16 0.03 0.00 −0.16 0.03 0.00 Mainly Sheep −0.03 0.03 0.38 −0.03 0.03 0.37 Tillage 0.29 0.05 0.00 0.29 0.05 0.00 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.08 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.06 Has off farm employment −0.13 0.02 0.00 −0.13 0.02 0.00 Log Farm Size 0.00 0.02 0.87 0.00 0.02 0.85 Has Forestry −0.09 0.04 0.05 −0.09 0.04 0.05 Completed Agricultural Short Course 0.21 0.03 0.00 0.21 0.03 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.18 0.03 0.00 Completed Agricultural University 0.24 0.06 0.00 0.24 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.15 0.04 0.00 −0.15 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.016 0.04 0.66 0.02 0.04 0.66 Cork, Kerry −0.17 0.04 0.00 −0.17 0.04 0.00 Galway, Mayo, Roscommon 0.00 0.03 0.95 0.00 0.03 0.95 Stocking Density 0.40 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.03 0.03 0.38 Constant 5.05 0.17 0.00 5.05 0.17 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 1331.86 891.43 Sargan statistic p value (Overidentification Test) 0.81 Note: The endogenous regressor is (Advisory Participation in both models); 2 instruments (Single Farm Payment policy change and Distance to advisory office); 3 Instruments (Single Farm Payment policy change, Distance to advisory office and Interaction of both); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors are adjusted for heterogeneity; a p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 considered weak; Sargan Statistic for overidentification p value > 0.1 fails to reject a null hypothesis of the instruments’ validity. View Large Appendix B – Full table of IV estimates with 1 instrument only (2000-2013) Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – Sfpyr only Instrumental Variable Regression –1 Dist. only Has Advisory Participation 0.35 0.41 0.00 0.14 0.57 0.79 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.03 0.00 Dairy System 0.37 0.03 0.00 0.34 0.09 0.00 Dairy and Other System −0.06 0.04 0.16 −0.09 0.12 0.42 Cattle Rearing −0.15 0.03 0.00 −0.19 0.15 0.19 Cattle Other System −0.16 0.03 0.00 −0.20 0.14 0.13 Mainly Sheep −0.03 0.03 0.39 −0.09 0.17 0.59 Tillage 0.29 0.05 0.00 0.25 0.09 0.01 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.09 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.12 Has off farm employment −0.13 0.02 0.00 −0.12 0.03 0.00 Log Farm Size 0.00 0.02 0.87 0.03 0.07 0.71 Has Forestry −0.09 0.04 0.05 −0.05 0.12 0.71 Completed Agricultural Short Course 0.21 0.03 0.00 0.23 0.06 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.21 0.07 0.01 Completed Agricultural University 0.24 0.06 0.00 0.26 0.07 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.05 0.00 Clare, Limerick, Tipp. N.R. −0.14 0.04 0.00 −0.17 0.09 0.05 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.02 0.04 0.66 0.00 0.05 0.98 Cork, Kerry −0.17 0.04 0.00 −0.15 0.07 0.04 Galway, Mayo, Roscommon 0.00 0.03 0.95 −0.00 0.03 0.98 Stocking Density 0.40 0.02 0.00 0.42 0.05 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.00 0.08 0.96 Constant 5.05 0.17 0.00 4.97 0.21 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 2639.20 10.78 Sargan statistic p value (Overidentification Test) 0.000 0.000 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – Sfpyr only Instrumental Variable Regression –1 Dist. only Has Advisory Participation 0.35 0.41 0.00 0.14 0.57 0.79 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.03 0.00 Dairy System 0.37 0.03 0.00 0.34 0.09 0.00 Dairy and Other System −0.06 0.04 0.16 −0.09 0.12 0.42 Cattle Rearing −0.15 0.03 0.00 −0.19 0.15 0.19 Cattle Other System −0.16 0.03 0.00 −0.20 0.14 0.13 Mainly Sheep −0.03 0.03 0.39 −0.09 0.17 0.59 Tillage 0.29 0.05 0.00 0.25 0.09 0.01 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.09 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.12 Has off farm employment −0.13 0.02 0.00 −0.12 0.03 0.00 Log Farm Size 0.00 0.02 0.87 0.03 0.07 0.71 Has Forestry −0.09 0.04 0.05 −0.05 0.12 0.71 Completed Agricultural Short Course 0.21 0.03 0.00 0.23 0.06 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.21 0.07 0.01 Completed Agricultural University 0.24 0.06 0.00 0.26 0.07 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.05 0.00 Clare, Limerick, Tipp. N.R. −0.14 0.04 0.00 −0.17 0.09 0.05 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.02 0.04 0.66 0.00 0.05 0.98 Cork, Kerry −0.17 0.04 0.00 −0.15 0.07 0.04 Galway, Mayo, Roscommon 0.00 0.03 0.95 −0.00 0.03 0.98 Stocking Density 0.40 0.02 0.00 0.42 0.05 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.00 0.08 0.96 Constant 5.05 0.17 0.00 4.97 0.21 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 2639.20 10.78 Sargan statistic p value (Overidentification Test) 0.000 0.000 Note: The endogenous regressor is (Advisory Participation in both models); 1instrument in each (Single Farm Payment policy change and Distance to advisory office respectively); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors are adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak. View Large Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – Sfpyr only Instrumental Variable Regression –1 Dist. only Has Advisory Participation 0.35 0.41 0.00 0.14 0.57 0.79 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.03 0.00 Dairy System 0.37 0.03 0.00 0.34 0.09 0.00 Dairy and Other System −0.06 0.04 0.16 −0.09 0.12 0.42 Cattle Rearing −0.15 0.03 0.00 −0.19 0.15 0.19 Cattle Other System −0.16 0.03 0.00 −0.20 0.14 0.13 Mainly Sheep −0.03 0.03 0.39 −0.09 0.17 0.59 Tillage 0.29 0.05 0.00 0.25 0.09 0.01 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.09 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.12 Has off farm employment −0.13 0.02 0.00 −0.12 0.03 0.00 Log Farm Size 0.00 0.02 0.87 0.03 0.07 0.71 Has Forestry −0.09 0.04 0.05 −0.05 0.12 0.71 Completed Agricultural Short Course 0.21 0.03 0.00 0.23 0.06 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.21 0.07 0.01 Completed Agricultural University 0.24 0.06 0.00 0.26 0.07 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.05 0.00 Clare, Limerick, Tipp. N.R. −0.14 0.04 0.00 −0.17 0.09 0.05 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.02 0.04 0.66 0.00 0.05 0.98 Cork, Kerry −0.17 0.04 0.00 −0.15 0.07 0.04 Galway, Mayo, Roscommon 0.00 0.03 0.95 −0.00 0.03 0.98 Stocking Density 0.40 0.02 0.00 0.42 0.05 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.00 0.08 0.96 Constant 5.05 0.17 0.00 4.97 0.21 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 2639.20 10.78 Sargan statistic p value (Overidentification Test) 0.000 0.000 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha Instrumental Variable Regression – Sfpyr only Instrumental Variable Regression –1 Dist. only Has Advisory Participation 0.35 0.41 0.00 0.14 0.57 0.79 Log Land Value per Ha 0.15 0.02 0.00 0.15 0.03 0.00 Dairy System 0.37 0.03 0.00 0.34 0.09 0.00 Dairy and Other System −0.06 0.04 0.16 −0.09 0.12 0.42 Cattle Rearing −0.15 0.03 0.00 −0.19 0.15 0.19 Cattle Other System −0.16 0.03 0.00 −0.20 0.14 0.13 Mainly Sheep −0.03 0.03 0.39 −0.09 0.17 0.59 Tillage 0.29 0.05 0.00 0.25 0.09 0.01 Log of Family Labour (unpaid) −0.14 0.02 0.00 −0.14 0.02 0.00 Age 0.01 0.01 0.08 0.01 0.01 0.09 Age Squared −0.00 0.00 0.06 −0.00 0.00 0.12 Has off farm employment −0.13 0.02 0.00 −0.12 0.03 0.00 Log Farm Size 0.00 0.02 0.87 0.03 0.07 0.71 Has Forestry −0.09 0.04 0.05 −0.05 0.12 0.71 Completed Agricultural Short Course 0.21 0.03 0.00 0.23 0.06 0.00 Completed Agricultural Certificate 0.18 0.03 0.00 0.21 0.07 0.01 Completed Agricultural University 0.24 0.06 0.00 0.26 0.07 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.13 0.03 0.00 −0.13 0.03 0.00 Kildare, Meath, Wicklow −0.21 0.08 0.01 −0.21 0.08 0.01 Laois, Longford, Offaly, Westmeath −0.18 0.04 0.00 −0.18 0.05 0.00 Clare, Limerick, Tipp. N.R. −0.14 0.04 0.00 −0.17 0.09 0.05 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford 0.02 0.04 0.66 0.00 0.05 0.98 Cork, Kerry −0.17 0.04 0.00 −0.15 0.07 0.04 Galway, Mayo, Roscommon 0.00 0.03 0.95 −0.00 0.03 0.98 Stocking Density 0.40 0.02 0.00 0.42 0.05 0.00 Medium Soil −0.08 0.02 0.00 −0.08 0.02 0.00 Poor Soil −0.03 0.03 0.37 −0.00 0.08 0.96 Constant 5.05 0.17 0.00 4.97 0.21 0.00 Number of Observations 8,951 8,951 Centred R2 0.22 0.22 Cragg-Donald Wald F Statistic (WeakInstrument) 2639.20 10.78 Sargan statistic p value (Overidentification Test) 0.000 0.000 Note: The endogenous regressor is (Advisory Participation in both models); 1instrument in each (Single Farm Payment policy change and Distance to advisory office respectively); border region omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors are adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak. View Large Appendix C – OLS estimates (2000-2004; 2005-2013) Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Model pre 2005 OLS Model post 2005 Has Advisory Participation 0.09 0.02 0.00 0.15 0.02 0.00 Log Land Value per Ha 0.19 0.03 0.00 0.15 0.02 0.00 Dairy System 0.16 0.07 0.02 0.31 0.03 0.00 Dairy and Other System −0.29 0.07 0.00 −0.04 0.04 0.37 Cattle Rearing −0.38 0.06 0.00 −0.17 0.03 0.00 Cattle Other System −0.56 0.06 0.00 −0.14 0.03 0.00 Mainly Sheep −0.39 0.06 0.00 −0.09 0.03 0.01 Tillage omitted omitted omitted 0.28 0.05 0.00 Log of Family Labour (unpaid) −0.08 0.03 0.01 −0.12 0.02 0.00 Age −0.00 0.01 0.86 0.01 0.01 0.181 Age Squared 0.00 0.00 0.95 −0.00 0.00 0.16 Has off farm employment −0.13 0.03 0.00 −0.11 0.02 0.00 Log Farm Size 0.01 0.02 0.81 0.04 0.02 0.01 Has Forestry omitted omitted omitted −0.03 0.04 0.38 Completed Agricultural Short Course 0.27 0.04 0.00 0.16 0.03 0.00 Completed Agricultural Certificate 0.13 0.03 0.00 0.21 0.03 0.00 Completed Agricultural University −0.04 0.09 0.65 0.28 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.18 0.04 0.00 −0.15 0.03 0.00 Kildare, Meath, Wicklow −0.01 0.11 0.94 −0.31 0.09 0.00 Laois, Longford, Offaly, Westmeath −0.17 0.05 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.21 0.05 0.00 −0.19 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.03 0.04 0.48 −0.04 0.04 0.30 Cork, Kerry −0.37 0.05 0.00 −0.05 0.04 0.12 Galway, Mayo, Roscommon −0.15 0.04 0.00 −0.01 0.03 0.76 Stocking Density 0.46 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.17 0.03 0.00 −0.08 0.02 0.00 Poor Soil 0.03 0.04 0.47 0.01 0.03 0.77 Constant 5.69 0.21 0.00 5.07 0.17 0.00 Number of Observations 4,503 8,875 R2 0.30 0.19 F stat 76.61 75.19 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Model pre 2005 OLS Model post 2005 Has Advisory Participation 0.09 0.02 0.00 0.15 0.02 0.00 Log Land Value per Ha 0.19 0.03 0.00 0.15 0.02 0.00 Dairy System 0.16 0.07 0.02 0.31 0.03 0.00 Dairy and Other System −0.29 0.07 0.00 −0.04 0.04 0.37 Cattle Rearing −0.38 0.06 0.00 −0.17 0.03 0.00 Cattle Other System −0.56 0.06 0.00 −0.14 0.03 0.00 Mainly Sheep −0.39 0.06 0.00 −0.09 0.03 0.01 Tillage omitted omitted omitted 0.28 0.05 0.00 Log of Family Labour (unpaid) −0.08 0.03 0.01 −0.12 0.02 0.00 Age −0.00 0.01 0.86 0.01 0.01 0.181 Age Squared 0.00 0.00 0.95 −0.00 0.00 0.16 Has off farm employment −0.13 0.03 0.00 −0.11 0.02 0.00 Log Farm Size 0.01 0.02 0.81 0.04 0.02 0.01 Has Forestry omitted omitted omitted −0.03 0.04 0.38 Completed Agricultural Short Course 0.27 0.04 0.00 0.16 0.03 0.00 Completed Agricultural Certificate 0.13 0.03 0.00 0.21 0.03 0.00 Completed Agricultural University −0.04 0.09 0.65 0.28 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.18 0.04 0.00 −0.15 0.03 0.00 Kildare, Meath, Wicklow −0.01 0.11 0.94 −0.31 0.09 0.00 Laois, Longford, Offaly, Westmeath −0.17 0.05 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.21 0.05 0.00 −0.19 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.03 0.04 0.48 −0.04 0.04 0.30 Cork, Kerry −0.37 0.05 0.00 −0.05 0.04 0.12 Galway, Mayo, Roscommon −0.15 0.04 0.00 −0.01 0.03 0.76 Stocking Density 0.46 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.17 0.03 0.00 −0.08 0.02 0.00 Poor Soil 0.03 0.04 0.47 0.01 0.03 0.77 Constant 5.69 0.21 0.00 5.07 0.17 0.00 Number of Observations 4,503 8,875 R2 0.30 0.19 F stat 76.61 75.19 Note: Border region is omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level. View Large Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Model pre 2005 OLS Model post 2005 Has Advisory Participation 0.09 0.02 0.00 0.15 0.02 0.00 Log Land Value per Ha 0.19 0.03 0.00 0.15 0.02 0.00 Dairy System 0.16 0.07 0.02 0.31 0.03 0.00 Dairy and Other System −0.29 0.07 0.00 −0.04 0.04 0.37 Cattle Rearing −0.38 0.06 0.00 −0.17 0.03 0.00 Cattle Other System −0.56 0.06 0.00 −0.14 0.03 0.00 Mainly Sheep −0.39 0.06 0.00 −0.09 0.03 0.01 Tillage omitted omitted omitted 0.28 0.05 0.00 Log of Family Labour (unpaid) −0.08 0.03 0.01 −0.12 0.02 0.00 Age −0.00 0.01 0.86 0.01 0.01 0.181 Age Squared 0.00 0.00 0.95 −0.00 0.00 0.16 Has off farm employment −0.13 0.03 0.00 −0.11 0.02 0.00 Log Farm Size 0.01 0.02 0.81 0.04 0.02 0.01 Has Forestry omitted omitted omitted −0.03 0.04 0.38 Completed Agricultural Short Course 0.27 0.04 0.00 0.16 0.03 0.00 Completed Agricultural Certificate 0.13 0.03 0.00 0.21 0.03 0.00 Completed Agricultural University −0.04 0.09 0.65 0.28 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.18 0.04 0.00 −0.15 0.03 0.00 Kildare, Meath, Wicklow −0.01 0.11 0.94 −0.31 0.09 0.00 Laois, Longford, Offaly, Westmeath −0.17 0.05 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.21 0.05 0.00 −0.19 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.03 0.04 0.48 −0.04 0.04 0.30 Cork, Kerry −0.37 0.05 0.00 −0.05 0.04 0.12 Galway, Mayo, Roscommon −0.15 0.04 0.00 −0.01 0.03 0.76 Stocking Density 0.46 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.17 0.03 0.00 −0.08 0.02 0.00 Poor Soil 0.03 0.04 0.47 0.01 0.03 0.77 Constant 5.69 0.21 0.00 5.07 0.17 0.00 Number of Observations 4,503 8,875 R2 0.30 0.19 F stat 76.61 75.19 Coeff SE p Coeff SE p Dependent Variable = Log of Farm Family Income per ha OLS Model pre 2005 OLS Model post 2005 Has Advisory Participation 0.09 0.02 0.00 0.15 0.02 0.00 Log Land Value per Ha 0.19 0.03 0.00 0.15 0.02 0.00 Dairy System 0.16 0.07 0.02 0.31 0.03 0.00 Dairy and Other System −0.29 0.07 0.00 −0.04 0.04 0.37 Cattle Rearing −0.38 0.06 0.00 −0.17 0.03 0.00 Cattle Other System −0.56 0.06 0.00 −0.14 0.03 0.00 Mainly Sheep −0.39 0.06 0.00 −0.09 0.03 0.01 Tillage omitted omitted omitted 0.28 0.05 0.00 Log of Family Labour (unpaid) −0.08 0.03 0.01 −0.12 0.02 0.00 Age −0.00 0.01 0.86 0.01 0.01 0.181 Age Squared 0.00 0.00 0.95 −0.00 0.00 0.16 Has off farm employment −0.13 0.03 0.00 −0.11 0.02 0.00 Log Farm Size 0.01 0.02 0.81 0.04 0.02 0.01 Has Forestry omitted omitted omitted −0.03 0.04 0.38 Completed Agricultural Short Course 0.27 0.04 0.00 0.16 0.03 0.00 Completed Agricultural Certificate 0.13 0.03 0.00 0.21 0.03 0.00 Completed Agricultural University −0.04 0.09 0.65 0.28 0.06 0.00 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.18 0.04 0.00 −0.15 0.03 0.00 Kildare, Meath, Wicklow −0.01 0.11 0.94 −0.31 0.09 0.00 Laois, Longford, Offaly, Westmeath −0.17 0.05 0.00 −0.18 0.04 0.00 Clare, Limerick, Tipp. N.R. −0.21 0.05 0.00 −0.19 0.04 0.00 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.03 0.04 0.48 −0.04 0.04 0.30 Cork, Kerry −0.37 0.05 0.00 −0.05 0.04 0.12 Galway, Mayo, Roscommon −0.15 0.04 0.00 −0.01 0.03 0.76 Stocking Density 0.46 0.02 0.00 0.41 0.02 0.00 Medium Soil −0.17 0.03 0.00 −0.08 0.02 0.00 Poor Soil 0.03 0.04 0.47 0.01 0.03 0.77 Constant 5.69 0.21 0.00 5.07 0.17 0.00 Number of Observations 4,503 8,875 R2 0.30 0.19 F stat 76.61 75.19 Note: Border region is omitted for collinearity; dairy system omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level. View Large Appendix D – Condensed model (2000-2013) Dependent Variable = Log of Farm Family Income per ha Coeff SE Z p CI CI Has Advisory Participation 0.31 0.04 7.95 0.00 0.23 0.38 Log Land Value per Ha 0.26 0.02 13.47 0.00 0.22 0.38 Age 0.03 0.01 4.74 0.00 0.02 0.04 Age Squared −0.00 0.00 −5.07 0.00 −0.00 −0.00 Completed Agricultural Short Course 0.26 0.03 8.58 0.00 0.20 0.32 Completed Agricultural Certificate 0.33 0.03 12.55 0.00 0.28 0.39 Completed Agricultural University 0.39 0.06 6.21 0.00 0.27 0.52 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.14 0.03 −4.65 0.00 −0.20 −0.08 Kildare, Meath, Wicklow −0.28 0.08 −3.35 0.00 −0.44 −0.12 Laois, Longford, Offaly, Westmeath −0.11 0.04 −2.61 0.01 −0.18 −0.03 Clare, Limerick, Tipp. N.R. −0.22 0.04 −5.50 0.00 −0.29 −0.14 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.06 0.04 −1.67 0.10 −0.13 0.01 Cork, Kerry −0.15 0.04 −3.99 0.00 −0.21 −0.07 Galway, Mayo, Roscommon 0.12 0.03 3.89 0.00 0.06 0.19 Medium Soil −0.17 0.02 −8.15 0.00 −0.22 −0.13 Poor Soil −0.16 0.03 −4.82 0.00 −0.22 −0.09 Constant 5.11 0.17 30.64 0.00 4.79 5.44 Number of Observations 8,951 Centred R2 0.12 Cragg-Donald Wald F Statistic (WeakInstrument) 1118.8 Sargan statistic p value (Overidentification Test) 0.19 Dependent Variable = Log of Farm Family Income per ha Coeff SE Z p CI CI Has Advisory Participation 0.31 0.04 7.95 0.00 0.23 0.38 Log Land Value per Ha 0.26 0.02 13.47 0.00 0.22 0.38 Age 0.03 0.01 4.74 0.00 0.02 0.04 Age Squared −0.00 0.00 −5.07 0.00 −0.00 −0.00 Completed Agricultural Short Course 0.26 0.03 8.58 0.00 0.20 0.32 Completed Agricultural Certificate 0.33 0.03 12.55 0.00 0.28 0.39 Completed Agricultural University 0.39 0.06 6.21 0.00 0.27 0.52 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.14 0.03 −4.65 0.00 −0.20 −0.08 Kildare, Meath, Wicklow −0.28 0.08 −3.35 0.00 −0.44 −0.12 Laois, Longford, Offaly, Westmeath −0.11 0.04 −2.61 0.01 −0.18 −0.03 Clare, Limerick, Tipp. N.R. −0.22 0.04 −5.50 0.00 −0.29 −0.14 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.06 0.04 −1.67 0.10 −0.13 0.01 Cork, Kerry −0.15 0.04 −3.99 0.00 −0.21 −0.07 Galway, Mayo, Roscommon 0.12 0.03 3.89 0.00 0.06 0.19 Medium Soil −0.17 0.02 −8.15 0.00 −0.22 −0.13 Poor Soil −0.16 0.03 −4.82 0.00 −0.22 −0.09 Constant 5.11 0.17 30.64 0.00 4.79 5.44 Number of Observations 8,951 Centred R2 0.12 Cragg-Donald Wald F Statistic (WeakInstrument) 1118.8 Sargan statistic p value (Overidentification Test) 0.19 Note: Endogenous regressor is (Advisory Participation); 3 Instruments (Single Farm Payment policy change, Distance to advisory office and Interaction of both); border region omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak. View Large Dependent Variable = Log of Farm Family Income per ha Coeff SE Z p CI CI Has Advisory Participation 0.31 0.04 7.95 0.00 0.23 0.38 Log Land Value per Ha 0.26 0.02 13.47 0.00 0.22 0.38 Age 0.03 0.01 4.74 0.00 0.02 0.04 Age Squared −0.00 0.00 −5.07 0.00 −0.00 −0.00 Completed Agricultural Short Course 0.26 0.03 8.58 0.00 0.20 0.32 Completed Agricultural Certificate 0.33 0.03 12.55 0.00 0.28 0.39 Completed Agricultural University 0.39 0.06 6.21 0.00 0.27 0.52 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.14 0.03 −4.65 0.00 −0.20 −0.08 Kildare, Meath, Wicklow −0.28 0.08 −3.35 0.00 −0.44 −0.12 Laois, Longford, Offaly, Westmeath −0.11 0.04 −2.61 0.01 −0.18 −0.03 Clare, Limerick, Tipp. N.R. −0.22 0.04 −5.50 0.00 −0.29 −0.14 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.06 0.04 −1.67 0.10 −0.13 0.01 Cork, Kerry −0.15 0.04 −3.99 0.00 −0.21 −0.07 Galway, Mayo, Roscommon 0.12 0.03 3.89 0.00 0.06 0.19 Medium Soil −0.17 0.02 −8.15 0.00 −0.22 −0.13 Poor Soil −0.16 0.03 −4.82 0.00 −0.22 −0.09 Constant 5.11 0.17 30.64 0.00 4.79 5.44 Number of Observations 8,951 Centred R2 0.12 Cragg-Donald Wald F Statistic (WeakInstrument) 1118.8 Sargan statistic p value (Overidentification Test) 0.19 Dependent Variable = Log of Farm Family Income per ha Coeff SE Z p CI CI Has Advisory Participation 0.31 0.04 7.95 0.00 0.23 0.38 Log Land Value per Ha 0.26 0.02 13.47 0.00 0.22 0.38 Age 0.03 0.01 4.74 0.00 0.02 0.04 Age Squared −0.00 0.00 −5.07 0.00 −0.00 −0.00 Completed Agricultural Short Course 0.26 0.03 8.58 0.00 0.20 0.32 Completed Agricultural Certificate 0.33 0.03 12.55 0.00 0.28 0.39 Completed Agricultural University 0.39 0.06 6.21 0.00 0.27 0.52 Donegal, Leitrim, Sligo, Cavan, Monaghan, Louth omitted omitted omitted omitted omitted omitted Dublin −0.14 0.03 −4.65 0.00 −0.20 −0.08 Kildare, Meath, Wicklow −0.28 0.08 −3.35 0.00 −0.44 −0.12 Laois, Longford, Offaly, Westmeath −0.11 0.04 −2.61 0.01 −0.18 −0.03 Clare, Limerick, Tipp. N.R. −0.22 0.04 −5.50 0.00 −0.29 −0.14 Carlow, Kilkenny, Wexford, Tipp S.R., Waterford −0.06 0.04 −1.67 0.10 −0.13 0.01 Cork, Kerry −0.15 0.04 −3.99 0.00 −0.21 −0.07 Galway, Mayo, Roscommon 0.12 0.03 3.89 0.00 0.06 0.19 Medium Soil −0.17 0.02 −8.15 0.00 −0.22 −0.13 Poor Soil −0.16 0.03 −4.82 0.00 −0.22 −0.09 Constant 5.11 0.17 30.64 0.00 4.79 5.44 Number of Observations 8,951 Centred R2 0.12 Cragg-Donald Wald F Statistic (WeakInstrument) 1118.8 Sargan statistic p value (Overidentification Test) 0.19 Note: Endogenous regressor is (Advisory Participation); 3 Instruments (Single Farm Payment policy change, Distance to advisory office and Interaction of both); border region omitted for collinearity; good soil omitted for collinearity; standard errors adjusted for heterogeneity; p value > 0.1 denotes significance at the 10% level, p value > 0.05 at the 5% level, and p value > 0.01 at the 1% level; Stock, Wright, and Yogo (2002) argue that a Wald F Statistic <10 is considered weak. View Large Appendix E – Difference in difference model (2000-2013) Dependent Variable = Log of Farm Family Income per ha Coeff SE t p CI CI Single farm payment year −0.03 0.03 −1.21 0.23 −0.08 0.02 Treated group −0.28 0.03 −10.76 0.00 −0.33 −0.23 Interaction term SFP*Treated 0.28 0.03 7.85 0.00 0.21 0.34 Constant 6.10 0.02 366.6 0.00 6.06 6.13 Number of Observations 13,562 R2 0.01 Dependent Variable = Log of Farm Family Income per ha Coeff SE t p CI CI Single farm payment year −0.03 0.03 −1.21 0.23 −0.08 0.02 Treated group −0.28 0.03 −10.76 0.00 −0.33 −0.23 Interaction term SFP*Treated 0.28 0.03 7.85 0.00 0.21 0.34 Constant 6.10 0.02 366.6 0.00 6.06 6.13 Number of Observations 13,562 R2 0.01 Note: Advisory participation treated for those who only became clients after the introduction of the single farm payment policy change in 2005. View Large Dependent Variable = Log of Farm Family Income per ha Coeff SE t p CI CI Single farm payment year −0.03 0.03 −1.21 0.23 −0.08 0.02 Treated group −0.28 0.03 −10.76 0.00 −0.33 −0.23 Interaction term SFP*Treated 0.28 0.03 7.85 0.00 0.21 0.34 Constant 6.10 0.02 366.6 0.00 6.06 6.13 Number of Observations 13,562 R2 0.01 Dependent Variable = Log of Farm Family Income per ha Coeff SE t p CI CI Single farm payment year −0.03 0.03 −1.21 0.23 −0.08 0.02 Treated group −0.28 0.03 −10.76 0.00 −0.33 −0.23 Interaction term SFP*Treated 0.28 0.03 7.85 0.00 0.21 0.34 Constant 6.10 0.02 366.6 0.00 6.06 6.13 Number of Observations 13,562 R2 0.01 Note: Advisory participation treated for those who only became clients after the introduction of the single farm payment policy change in 2005. View Large © The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com

Journal

Applied Economic Perspectives and PolicyOxford University Press

Published: Feb 20, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off