Do Nutrient Management Plans Actually Manage Nutrients? Evidence from a Nationally-Representative Survey of Hog Producers

Do Nutrient Management Plans Actually Manage Nutrients? Evidence from a Nationally-Representative... Abstract A nutrient management plan (NMP) specifies recommended practices to match applied nutrients with crops’ uptake capacity. Because monitoring nutrient applications is difficult, regulators instead oversee NMP adoption. In this paper we examine whether having NMPs make hog farms more likely to adopt nutrient management practices. We estimate nutrient application and uptake rates to assess whether operations with NMPs are less likely to over-apply nutrients. Using an endogenous treatment effects model to control for potential confounding and selection bias, we find that NMPs are positively correlated with the adoption of nutrient management practices, as well as with the reduced application of excess nutrients. Livestock, nutrient management plans, regulation, CAFO, hogs The environmental regulation of manure application poses a particular challenge to regulators. Nutrients contained in manure, particularly nitrogen and phosphorus, can degrade water quality if they are over-applied to farmland and enter water resources through runoff or leaching. Because most livestock producers are not forced to internalize the social costs of their pollution, and because manure transportation is costly, they may have an incentive to over-apply nutrients. A main policy objective is therefore to prevent run-off by limiting nutrient application to agronomic rates. A regulatory challenge arises because even if limits are placed on application rates, in most situations regulators are unable to directly monitor compliance. Perhaps recognizing the costliness and general infeasibility of monitoring farmer behavior, state and federal policies instead encourage or require livestock operations to submit “nutrient management plans” (NMPs) stating how much manure and fertilizer they will apply to each acre of land. Regulators can observe these plans, thereby limiting their oversight costs. However, having a plan does not prove that a farmer is following appropriate nutrient management practices. If regulatory authorities only require these plans as proof of regulatory compliance and do not engage in further oversight, there will be limited incentive for livestock operators to change their behavior. A key policy question, therefore, is whether NMPs are useful predictors of recommended nutrient management practices and agronomic nutrient application rates. In this article we use information on NMP adoption, externally documentable nutrient management practices, and estimated nutrient application rates in order to ascertain whether there is a difference between planned and implemented activities. We use data from the 2004 and 2009 Agricultural Resource Management Survey (ARMS) for Hogs, a nationally-representative survey conducted by the USDA’s Economic Research Service and National Agricultural Statistics Service.1 The ARMS has information on NMP adoption, practices which are signals of NMP compliance (soil testing, adjustments to feed to alter manure nutrient content, incorporation of manure at application), animals by size and type, fertilizer expenditures, and crop yields. We first examine simple summary statistics on whether farms with NMPs are more likely to use certain nutrient management practices than those without. Perhaps a more pertinent measure of NMP implementation is whether operations match nutrient application to uptake. However, the ARMS hog survey instrument does not include questions on specific amounts of nutrients applied or used by crops or pasture. Instead, through an involved and detailed accounting process, we estimate the amounts of nutrients applied on farm in the form of manure or commercial fertilizer, as well as the nutrient absorptive capacity of the crops and land on the farm. We can therefore examine the average application rates of farms with and without NMPs. Simple summary statistics examining nutrient management behaviors and applications for operations with and without NMPs do not account for features that could be correlated with NMP adoption and behaviors. Such statistics also do not address potential selection bias that might occur if operations with better or worse nutrient management are more likely to voluntarily adopt NMPs.2 Therefore, we account for potential confounding and sample selection bias using an endogenous treatment effects model. Our basic empirical strategy uses cross-sectional variation in NMP status to identify its relationship to nutrient management practices, controlling for potential confounders that are associated with both NMP status and practices. Because we do not assign NMPs and because adoption is often voluntary, we model NMP adoption as endogenous. We estimate both adoption of NMP and the behavioral outcome variable jointly using maximum likelihood. Understanding the nutrient management behavior of farms with NMPs allows us to gain insight on the effectiveness of current policy, and whether there is a need to upgrade existing regulation or enforcement. The rest of this paper is organized as follows. Further background information is provided in the next section, while the two subsequent sections discuss the empirical strategy and data used in the analysis. The following section provides a discussion of the results, and the final section concludes. Background Nutrients are necessary for crop growth and are a natural by-product of livestock production, but may increase the risk for water pollution if they are applied at rates greater than what crop and pastureland can absorb. Trends in livestock production contribute to concerns of nutrient over-application. Farms increasingly specialize in either crop or livestock production, hence farm-level nutrient production may not match farm-level nutrient needs (Kellogg et al. 2000). Certain types of livestock agriculture have also become more geographically concentrated, occasionally in regions distant from the locus of crop production (McBride and Key 2013). Simultaneously, there has been continuous growth in the size of livestock operations; these larger operations often involve intensive production methods that result in the potential to apply more nutrients than needed (Johnson, Wheeler, and Christensen 1999). To avoid over-application, livestock facility operators with insufficient land on which to apply manure may ship manure off-farm. However, this practice is expensive and crop farmers’ willingness to pay for or even accept manure for free is often very low. Hence, manure has little value in many regions, creating an incentive for some livestock producers to treat it as a waste and apply it above agronomically-appropriate rates. Even in the absence of manure production, research suggests that many farmers apply more fertilizer than needed to ensure their optimum yield (Beegle, Carton, and Bailey 2000; Lawley, Lichtenberg, and Parker 2009). Applying excess nutrients that cannot be assimilated into the soil can contribute to nutrient run-off. This occurs when land-applied nutrients are carried to nearby water bodies via surface runoff of precipitation, resulting in the impairment of water quality and ecosystem resources (Copeland and Zinn 1998; Burkholder et al. 2007). In an effort to control nutrient discharge from farming operations, state and federal authorities have introduced NMPs, which have different requirements by jurisdiction, but are generally aimed at assuring that the amount of nutrients applied as manure and/or commercial fertilizer does not exceed the amount of nutrients that can be agronomically utilized by crop and pasture land. To this end, the NMP may require documentable actions, including nutrient testing of manure and soils and adjustments to feed rations. Nutrient testing of manure can inform how much manure to apply to cropland, and adjustments to feed rations can lower the nutrient content of manure (Albrecht 2003). These actions can be independently documented—at the very least through receipts. The least documentable action required by NMPs is for farms to match application rates to uptake. While NMPs generally provide prescriptions on the amount of nutrients that can be applied to individual fields, an independent inspector does not directly monitor applications. NMPs can be either voluntary or mandatory. Determining which operations are required to adopt NMPs and which do so voluntarily is difficult due to differing state, regional, and federal stipulations that may require a NMP. Operations that are deemed Concentrated Animal Feeding Operations (CAFOs) may be required to obtain an NMP as a condition to receiving a permit to operate, or may enact one as a stopgap against potential fines (Sneeringer 2016). In 2003, the EPA amended Clean Water Act CAFO rules originally adopted in the 1970s to provide some oversight of manure application; these 2003 rules were contested over the next several years, updated in 2008, and finalized in 2011. The 2003 rules began requiring NMPs for certain operations either designated or defined as CAFOs. Operations are defined to be large CAFOs if they have over a certain number of animals in inventory, if they raise animals in confinement for at least 45 days of the year, and do not grow vegetation in the location in which they raise animals. The qualifying threshold for “large CAFO” varies by animal type and size; early characterizations used 1,000 animal units as the threshold for a “large CAFO”. Operations with fewer animals may be designated to be CAFOs by the pertinent regulatory authority depending on the risk they pose to surface water pollution. The 2003 regulations stipulating that all CAFOs had to obtain permits (with NMPs) were contested. The final 2011 regulations required that only CAFOs with documented discharges needed to obtain permits (and therefore needed an NMP). Operations could also obtain an NMP as a measure to avoid fines should they be eventually found to have a documented discharge. Notably, the EPA devolves regulatory authority of the CWA CAFO rules to states after review of individual states’ implementation plans. States may adopt more stringent regulations requiring more farms to adopt NMPs.3 Enforcement may also vary between states. A second type of regulation that may require operations to adopt NMPs are Total Maximum Daily Loads (TMDLs). TMDLs are adopted at a regional level, and often focus on individual watersheds. Because TMDLs fall under the jurisdiction of a number of regulatory agencies, it is difficult to generalize about what TMDLs require, but at the very least their presence suggests a higher degree of oversight. NMPs can also be voluntary. The USDA’s National Resource Conservation Service (NRCS) as well as state and regional-level entities have been providing education and assistance for operations wishing to voluntarily adopt NMPs. The Environmental Quality Incentives Program (EQIP), run by NRCS in conjunction with state offices, offers financial and technical assistance to farmers to address a host of environmental concerns. A stipulation of receiving funds for any of a number of projects is often a NMP. Further, EQIP funds may also be used as a method to financially support the adoption of NMPs. Efficacy of NMPs in Reducing Nutrient Pollution While NMPs have been touted as one of the only methods of reducing nutrient run-off from livestock operations, they suffer, at least on the theoretical level, from problems of monitoring and enforcement. Innes (1999) points out that attributing nutrient pollution to individual livestock facilities is impossible without “a massive army of manure police patrolling a livestock producer’s surrounding crop fields to watch and limit the operator’s every manure application.” Noting that such a policy is too costly to be justified, Innes proposes alternative methods of nutrient pollution reduction such as fertilizer taxes, limits on the number of livestock in a region, manure use subsidies, and other mechanisms (Innes 1999, 2000). Why would farmers not implement their NMPs even if they adopt a plan? Noncompliance may be due to the lack of incentives for farmers to use NMPs and to costs of implementation. Studies suggest that farmers must have an understanding of the impact of their manure management practices on water quality in order for them to implement their NMPs (Shepard 2005; Ribaudo and Johansson 2007; Savage and Ribaudo 2013). Economic incentives through EQIP and fertilizer offsets may help counter costs of implementation but have been found to mainly benefit small farms (Ribaudo, Cattaneo, and Agapoff 2004). On the other hand, costs for abatement, plan design, and plan administration remain a disadvantage to small farms (McCann 2009). Other research supports the suggestion that NMPs are difficult to monitor or enforce. Anecdotal evidence of law evasion suggests that some farmers may actively mislead regulators (Perez 2011). Other studies have found that farmers may get a NMP but may not implement their plan on their entire operation, if at all (Shepard 2005; Genskow 2012). A lack of compliance, in combination with negligence and a lack of enforcement from regulatory authorities, has brought into question the ability of current regulations to reduce discharges (Centner and Newton 2008; Sneeringer 2016). A further question arises regarding the consistency of the information in NMPs. Inconsistencies in nutrient management are introduced when NMPs are developed by entities with different agendas. Studies have found that NMPs developed by private sector agents tend to have higher thresholds for the amount of nutrients that can be applied to the land than plans developed by public sector agents (Lawley, Lichtenberg, and Parker 2009; Perez 2011). Little research has been conducted to determine whether the adoption of NMPs improves environmental outcomes. The EPA conducted an ex ante study before the 2003 CAFO rules to predict improvements in environmental quality from the law, stemming largely from greater adoption of NMPs. The EPA estimated a 22% reduction in nutrient loadings from large and medium CAFOs from the updated law, based partly on assumptions about how producers with a NMP would behave (68 FR 7176-7274). As far as we are aware, there has been no ex post research showing how NMPs impact the end goal of environmental quality, or examining whether NMPs alter nutrient management behaviors. Data We use farm-level data from the 2004 and 2009 Agricultural Resource Management Surveys Phase III collected from hog operations in the United States to examine farm characteristics and nutrient management practices. The 2004 survey consists of 1,198 observations representing 40,940 operations; the 2009 survey consists of 1,208 observations representing 24,350 farms. Only farms with at least 25 hogs are surveyed, as anything less is considered to be raised for private consumption. In each year, the survey consists of data from 19 states, representing over 90% of hog production (Key et al. 2011; see appendix table 1 for a list of included states). The ARMS includes questions on our variable of interest—NMP adoption—as well as pertinent nutrient management behaviors, including nitrogen and phosphorus testing of manure, adjustments to feed for the purpose of altering manure nutrient content, and incorporation of manure at application. These are measures with the potential to be externally verified by an oversight authority; for example, farmers may report in the ARMS that they perform nutrient testing, but may have less incentive to lie about this, as testing labs can be questioned, or lab testing receipts or results can be requested. The ARMS also includes questions about potential confounding factors, including number of animals, crop acreage, the state in which the operation resides, whether the operation has an EQIP contract, whether the operation is farrow-to-finish or specialized for a specific portion of the hog life span, whether the operation is a contractor for an integrator, the operator’s years of farming experience, and whether the operator has a college degree. Using information on the sales and inventory of nursery pigs, feeder pigs, breeder hogs, and market hogs, we calculate the number of hog “animal units” (see the appendix for detail). “Animal units” is a method of standardizing across different animal weights and types; one animal unit is approximately 1,000 pounds of average live weight. From this we can characterize which operations have more than 1,000 hog animal units, our measure for whether an operation is a “large CAFO.” Prior literature notes the importance of enforcement in evaluating the efficacy of NMPs. Since we do not have a direct measure of enforcement (such as number of oversight visits from pertinent regulatory authorities), we proxy for enforcement with four variables. First, we control for large CAFO status. Second, we use the EPA’s 303d water quality reports to generate an indicator for whether a farm is in a county that has ever adopted a TMDL program for nutrients prior to the survey year. A TMDL is sometimes called a “pollution diet” and is implemented when traditional methods of pollution control fail to yield water quality goals. For livestock farmers, implementation generally means greater oversight of nutrient management practices in a watershed (e.g., Copeland 2012). Third, we include a measure for whether the operation has an Environmental Quality Incentives Program (EQIP) contract. EQIP contracts often require operators to obtain NMPs in order to access funds for more capital-intensive projects. Fourth, population land density at the county level is generated from the inter-censual county population estimates from the U.S. Census Bureau. Prior research has used population density to account for demands for environmental quality and restrictions in land application of manure due to odor complaints (Lyford and Hicks 2001; Ribaudo & Johansson 2007). In addition to evaluating the effects of NMPs on nutrient management practices, we also analyze whether having a NMP lowers the probability that a farm over-applies nutrients. The ARMS questionnaire does not directly collect information on nutrient application and uptake. Instead, we estimate these variables using ARMS questions and involved procedures partially developed from National Resource Conservation Service (NRCS) methods; see the appendices for details, as well as Kellogg et al. (2000) and Sneeringer (2016). We first estimate the total amount of manure nutrients generated on-farm and available for later application accounting for animal size, manure produced per animal pound, nutrients per unit of manure, nutrient loss through management or evaporation, and manure storage. We next estimate how much of these nutrients were applied on-farm by accounting for the amounts of manure that were either shipped off-farm or kept in storage and therefore not applied. We use total fertilizer expenditures along with data on crops planted and per unit fertilizer costs to estimate how much commercial fertilizer is applied. The manure and fertilizer applications provide us with an estimate of total nutrient applications. Finally, we use data on crop yields in 15 different commodities, as well as pasture to estimate the nutrient uptake capacity on the farm. Comparing the amount of nutrients applied to the uptake capacity, we can evaluate whether the operation applied excess nutrients. Notably, our measure of excess can be negative or positive; positive excess means more applications than uptake, while “negative excess” is the opposite. Summary Statistics Table 1 shows the summary statistics for the sample and compares farms with NMPs to those without. In the weighted samples, 30% of operations in 2004 and 55% of operations in 2009 had NMPs. These operations accounted for 61% and 82% of hog animal units in 2004 and 2009, respectively. On average, farms with NMPs had about 3.7 times more animal units in both years than those without, but only about 1.4 to 1.5 times the amount of crop acreage. Operations with NMPs were more likely to be specialized in a single phase of the hog life-cycle, have manure storage, and be contract operations. Operators with NMPs are, on average, slightly younger (52 versus 53), had slightly less experience and, in 2009, were more likely to have a college degree. Operations with NMPs were more likely to have EQIP contracts. In 2009, they were less likely to be in a county that ever had a TMDL. County-level population density is not statistically different between operations with and without NMPs in either year. Table 1 Operation-level Means of Relevant Independent Variables, by NMP Status 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Number of observations (unweighted) 1,196 541 655 1,283 810 473 Number of farms (weighted) 40,929 12,340 28,589 24,430 13,364 11,066 Proportion of farms 1.00 0.30 0.70 1.00 0.55 0.45 Proportion of hog animal units 1.00 0.61 0.39 1.00 0.82 0.18 Average number of hog animal units in inventory 294 598 163 11.08 545 814 219 13.04 (3,795) (3,683) (3,553) (4,180) (4,735) (2,121) Proportion with more than 1,000 hog animal units 0.07 0.16 0.03 7.52 0.16 0.24 0.05 10.41 (1.51) (1.77) (1.17) (1.59) (1.75) (1.09) Average total crop acreage 459 570 411 3.71 503 591 398 4.64 (3,998) (3,991) (3,965) (3,407) (3,685) (2,790) Proportion with no crop acreage 0.17 0.14 0.18 1.73 0.15 0.15 0.16 0.33 (2.18) (1.66) (2.52) (1.57) (1.44) (1.75) Proportion of operations that are farrow-to-finish 0.31 0.18 0.36 7.15 0.24 0.13 0.37 9.32 (2.70) (1.84) (3.17) (1.86) (1.37) (2.33) Proportion that are contract operations 0.28 0.53 0.17 13.53 0.49 0.67 0.28 14.66 (2.63) (2.39) (2.50) (2.18) (1.92) (2.17) Average operator experience (years) 25 24 25 2.21 28 27 29 2.06 (68) (57) (76) (56) (51) (63) Proportion of operators with college education 0.27 0.24 0.27 1.22 0.21 0.24 0.17 2.87 (2.58) (2.05) (2.95) (1.77) (1.73) (1.83) Proportion with an EQIP contract 0.02 0.04 0.003 4.52 0.04 0.07 0.02 4.93 (0.71) (0.98) (0.35) (0.89) (1.01) (0.59) Proportion with TMDL ever in county 0.16 0.15 0.16 0.43 0.53 0.50 0.57 2.51 (2.13) (1.71) (2.43) (2.18) (2.03) (2.40) Average population density in county 75 81 72 1.25 78 76 79 0.45 (786) (574) (924) (503) (474) (549) 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Number of observations (unweighted) 1,196 541 655 1,283 810 473 Number of farms (weighted) 40,929 12,340 28,589 24,430 13,364 11,066 Proportion of farms 1.00 0.30 0.70 1.00 0.55 0.45 Proportion of hog animal units 1.00 0.61 0.39 1.00 0.82 0.18 Average number of hog animal units in inventory 294 598 163 11.08 545 814 219 13.04 (3,795) (3,683) (3,553) (4,180) (4,735) (2,121) Proportion with more than 1,000 hog animal units 0.07 0.16 0.03 7.52 0.16 0.24 0.05 10.41 (1.51) (1.77) (1.17) (1.59) (1.75) (1.09) Average total crop acreage 459 570 411 3.71 503 591 398 4.64 (3,998) (3,991) (3,965) (3,407) (3,685) (2,790) Proportion with no crop acreage 0.17 0.14 0.18 1.73 0.15 0.15 0.16 0.33 (2.18) (1.66) (2.52) (1.57) (1.44) (1.75) Proportion of operations that are farrow-to-finish 0.31 0.18 0.36 7.15 0.24 0.13 0.37 9.32 (2.70) (1.84) (3.17) (1.86) (1.37) (2.33) Proportion that are contract operations 0.28 0.53 0.17 13.53 0.49 0.67 0.28 14.66 (2.63) (2.39) (2.50) (2.18) (1.92) (2.17) Average operator experience (years) 25 24 25 2.21 28 27 29 2.06 (68) (57) (76) (56) (51) (63) Proportion of operators with college education 0.27 0.24 0.27 1.22 0.21 0.24 0.17 2.87 (2.58) (2.05) (2.95) (1.77) (1.73) (1.83) Proportion with an EQIP contract 0.02 0.04 0.003 4.52 0.04 0.07 0.02 4.93 (0.71) (0.98) (0.35) (0.89) (1.01) (0.59) Proportion with TMDL ever in county 0.16 0.15 0.16 0.43 0.53 0.50 0.57 2.51 (2.13) (1.71) (2.43) (2.18) (2.03) (2.40) Average population density in county 75 81 72 1.25 78 76 79 0.45 (786) (574) (924) (503) (474) (549) Table 1 Operation-level Means of Relevant Independent Variables, by NMP Status 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Number of observations (unweighted) 1,196 541 655 1,283 810 473 Number of farms (weighted) 40,929 12,340 28,589 24,430 13,364 11,066 Proportion of farms 1.00 0.30 0.70 1.00 0.55 0.45 Proportion of hog animal units 1.00 0.61 0.39 1.00 0.82 0.18 Average number of hog animal units in inventory 294 598 163 11.08 545 814 219 13.04 (3,795) (3,683) (3,553) (4,180) (4,735) (2,121) Proportion with more than 1,000 hog animal units 0.07 0.16 0.03 7.52 0.16 0.24 0.05 10.41 (1.51) (1.77) (1.17) (1.59) (1.75) (1.09) Average total crop acreage 459 570 411 3.71 503 591 398 4.64 (3,998) (3,991) (3,965) (3,407) (3,685) (2,790) Proportion with no crop acreage 0.17 0.14 0.18 1.73 0.15 0.15 0.16 0.33 (2.18) (1.66) (2.52) (1.57) (1.44) (1.75) Proportion of operations that are farrow-to-finish 0.31 0.18 0.36 7.15 0.24 0.13 0.37 9.32 (2.70) (1.84) (3.17) (1.86) (1.37) (2.33) Proportion that are contract operations 0.28 0.53 0.17 13.53 0.49 0.67 0.28 14.66 (2.63) (2.39) (2.50) (2.18) (1.92) (2.17) Average operator experience (years) 25 24 25 2.21 28 27 29 2.06 (68) (57) (76) (56) (51) (63) Proportion of operators with college education 0.27 0.24 0.27 1.22 0.21 0.24 0.17 2.87 (2.58) (2.05) (2.95) (1.77) (1.73) (1.83) Proportion with an EQIP contract 0.02 0.04 0.003 4.52 0.04 0.07 0.02 4.93 (0.71) (0.98) (0.35) (0.89) (1.01) (0.59) Proportion with TMDL ever in county 0.16 0.15 0.16 0.43 0.53 0.50 0.57 2.51 (2.13) (1.71) (2.43) (2.18) (2.03) (2.40) Average population density in county 75 81 72 1.25 78 76 79 0.45 (786) (574) (924) (503) (474) (549) 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Number of observations (unweighted) 1,196 541 655 1,283 810 473 Number of farms (weighted) 40,929 12,340 28,589 24,430 13,364 11,066 Proportion of farms 1.00 0.30 0.70 1.00 0.55 0.45 Proportion of hog animal units 1.00 0.61 0.39 1.00 0.82 0.18 Average number of hog animal units in inventory 294 598 163 11.08 545 814 219 13.04 (3,795) (3,683) (3,553) (4,180) (4,735) (2,121) Proportion with more than 1,000 hog animal units 0.07 0.16 0.03 7.52 0.16 0.24 0.05 10.41 (1.51) (1.77) (1.17) (1.59) (1.75) (1.09) Average total crop acreage 459 570 411 3.71 503 591 398 4.64 (3,998) (3,991) (3,965) (3,407) (3,685) (2,790) Proportion with no crop acreage 0.17 0.14 0.18 1.73 0.15 0.15 0.16 0.33 (2.18) (1.66) (2.52) (1.57) (1.44) (1.75) Proportion of operations that are farrow-to-finish 0.31 0.18 0.36 7.15 0.24 0.13 0.37 9.32 (2.70) (1.84) (3.17) (1.86) (1.37) (2.33) Proportion that are contract operations 0.28 0.53 0.17 13.53 0.49 0.67 0.28 14.66 (2.63) (2.39) (2.50) (2.18) (1.92) (2.17) Average operator experience (years) 25 24 25 2.21 28 27 29 2.06 (68) (57) (76) (56) (51) (63) Proportion of operators with college education 0.27 0.24 0.27 1.22 0.21 0.24 0.17 2.87 (2.58) (2.05) (2.95) (1.77) (1.73) (1.83) Proportion with an EQIP contract 0.02 0.04 0.003 4.52 0.04 0.07 0.02 4.93 (0.71) (0.98) (0.35) (0.89) (1.01) (0.59) Proportion with TMDL ever in county 0.16 0.15 0.16 0.43 0.53 0.50 0.57 2.51 (2.13) (1.71) (2.43) (2.18) (2.03) (2.40) Average population density in county 75 81 72 1.25 78 76 79 0.45 (786) (574) (924) (503) (474) (549) Farms with NMPs are more likely to engage in certain documentable nutrient management practices (table 2). Operations with NMPs are much more likely to test manure for nutrient content and to adjust feed levels to manage it. When they apply manure, farms with NMPs are more likely to incorporate it (a practice to reduce run-off); however, this difference is only statistically significant in 2004, not 2009. Table 2 Operation-level Means of Nutrient Management Behavior Variables, by Year and NMP Status 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Proportion testing manure for nitrogen content 0.24 0.56 0.10 18.88 0.38 0.58 0.13 19.36 (2.50) (2.37) (1.99) (2.11) (2.01) (1.62) Proportion using microbial phytase to reduce nutrient content of manure 0.13 0.25 0.08 7.69 0.24 0.34 0.11 10.29 (2.0) (2.07) (1.84) (1.86) (1.93) (1.54) Proportion adjusting schedule of feed formulations to adjust the nutrient content of manure 0.14 0.28 0.08 8.82 0.22 0.33 0.10 10.77 (2.04) (2.14) (1.82) (1.82) (1.91) (1.44) Proportion using other feed additives or formula adjustments to adjust the nutrient content of manure 0.05 0.14 0.01 8.48 0.08 0.13 0.02 7.88 (1.27) (1.66) (0.65) (1.20) (1.38) (0.73) Average proportion manure applied 0.78 0.80 0.77 1.33 0.74 0.75 0.72 1.48 (2.37) (1.79) (2.76) (1.85) (1.65) (2.14) Proportion applying any manure 0.80 0.85 0.78 3.40 0.77 0.80 0.73 2.83 (2.34) (1.69) (2.75) (1.83) (1.61) (2.14) Of those applying manure Proportion incorporating manure at application* 0.37 0.42 0.35 2.48 0.50 0.52 0.48 1.05 (2.82) (2.36) (3.17) (2.16) (1.98) (2.49) 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Proportion testing manure for nitrogen content 0.24 0.56 0.10 18.88 0.38 0.58 0.13 19.36 (2.50) (2.37) (1.99) (2.11) (2.01) (1.62) Proportion using microbial phytase to reduce nutrient content of manure 0.13 0.25 0.08 7.69 0.24 0.34 0.11 10.29 (2.0) (2.07) (1.84) (1.86) (1.93) (1.54) Proportion adjusting schedule of feed formulations to adjust the nutrient content of manure 0.14 0.28 0.08 8.82 0.22 0.33 0.10 10.77 (2.04) (2.14) (1.82) (1.82) (1.91) (1.44) Proportion using other feed additives or formula adjustments to adjust the nutrient content of manure 0.05 0.14 0.01 8.48 0.08 0.13 0.02 7.88 (1.27) (1.66) (0.65) (1.20) (1.38) (0.73) Average proportion manure applied 0.78 0.80 0.77 1.33 0.74 0.75 0.72 1.48 (2.37) (1.79) (2.76) (1.85) (1.65) (2.14) Proportion applying any manure 0.80 0.85 0.78 3.40 0.77 0.80 0.73 2.83 (2.34) (1.69) (2.75) (1.83) (1.61) (2.14) Of those applying manure Proportion incorporating manure at application* 0.37 0.42 0.35 2.48 0.50 0.52 0.48 1.05 (2.82) (2.36) (3.17) (2.16) (1.98) (2.49) Note: Asterisk * indicates that manure at application refers to manure applied in solid form and incorporated at application, or manure applied in slurry form that is injected or surface applied and incorporated. Table 2 Operation-level Means of Nutrient Management Behavior Variables, by Year and NMP Status 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Proportion testing manure for nitrogen content 0.24 0.56 0.10 18.88 0.38 0.58 0.13 19.36 (2.50) (2.37) (1.99) (2.11) (2.01) (1.62) Proportion using microbial phytase to reduce nutrient content of manure 0.13 0.25 0.08 7.69 0.24 0.34 0.11 10.29 (2.0) (2.07) (1.84) (1.86) (1.93) (1.54) Proportion adjusting schedule of feed formulations to adjust the nutrient content of manure 0.14 0.28 0.08 8.82 0.22 0.33 0.10 10.77 (2.04) (2.14) (1.82) (1.82) (1.91) (1.44) Proportion using other feed additives or formula adjustments to adjust the nutrient content of manure 0.05 0.14 0.01 8.48 0.08 0.13 0.02 7.88 (1.27) (1.66) (0.65) (1.20) (1.38) (0.73) Average proportion manure applied 0.78 0.80 0.77 1.33 0.74 0.75 0.72 1.48 (2.37) (1.79) (2.76) (1.85) (1.65) (2.14) Proportion applying any manure 0.80 0.85 0.78 3.40 0.77 0.80 0.73 2.83 (2.34) (1.69) (2.75) (1.83) (1.61) (2.14) Of those applying manure Proportion incorporating manure at application* 0.37 0.42 0.35 2.48 0.50 0.52 0.48 1.05 (2.82) (2.36) (3.17) (2.16) (1.98) (2.49) 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Proportion testing manure for nitrogen content 0.24 0.56 0.10 18.88 0.38 0.58 0.13 19.36 (2.50) (2.37) (1.99) (2.11) (2.01) (1.62) Proportion using microbial phytase to reduce nutrient content of manure 0.13 0.25 0.08 7.69 0.24 0.34 0.11 10.29 (2.0) (2.07) (1.84) (1.86) (1.93) (1.54) Proportion adjusting schedule of feed formulations to adjust the nutrient content of manure 0.14 0.28 0.08 8.82 0.22 0.33 0.10 10.77 (2.04) (2.14) (1.82) (1.82) (1.91) (1.44) Proportion using other feed additives or formula adjustments to adjust the nutrient content of manure 0.05 0.14 0.01 8.48 0.08 0.13 0.02 7.88 (1.27) (1.66) (0.65) (1.20) (1.38) (0.73) Average proportion manure applied 0.78 0.80 0.77 1.33 0.74 0.75 0.72 1.48 (2.37) (1.79) (2.76) (1.85) (1.65) (2.14) Proportion applying any manure 0.80 0.85 0.78 3.40 0.77 0.80 0.73 2.83 (2.34) (1.69) (2.75) (1.83) (1.61) (2.14) Of those applying manure Proportion incorporating manure at application* 0.37 0.42 0.35 2.48 0.50 0.52 0.48 1.05 (2.82) (2.36) (3.17) (2.16) (1.98) (2.49) Note: Asterisk * indicates that manure at application refers to manure applied in solid form and incorporated at application, or manure applied in slurry form that is injected or surface applied and incorporated. Turning to our estimates of nutrient application and uptake, we see a somewhat conflicting story compared to what we expect from the practice outcomes. Despite the higher ratio of animal units to crop acres at operations with NMPs, they apply an equal percentage of their manure on-farm (Table 2). The rest of the manure is either stored or shipped off-farm. As expected by their greater number of animal units, operations with NMPs generate more manure nutrients than operations without NMPs—4.6 times as much in 2004, and 3.8 times as much as in 2009 (see table 3). Accounting for both recoverable manure nutrients and estimated commercial fertilizer nutrients, operations with NMPs, on average, apply more than twice the nutrients over their entire operations than those without. Table 3 Operation-level Means of Nutrient Generation, Application, and Uptake Variables, by Year and NMP Status Year 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Recoverable manure nitrogen generated (lbs.) 10,398 22,879 5,011 11.22 19,254 28,908 7,595 11.24 (144,225) (161,220) (111,082) (171,391) (196,318) (88,890) Recoverable manure nitrogen applied (lbs.) 7,971 17,778 3,729 9.44 14,080 21,420 5,234 9.77 (132,058) (151,003) (102,064) (149,134) (173,762) (72,804) Fertilizer nitrogen applied (lbs.) 26,280 35,145 22,454 4.09 22,700 27,953 16,355 4.66 (285,973) (292,167) (277,174) (204,952) (222,454) (165,975) Fertilizer and manure nitrogen applied (lbs.) 34,240 52,923 26,175 7.13 36,759 49,320 21,588 8.69 (344,499) (362,032) (314,967) (275,468) (307,122) (185,937) Nitrogen uptake on crops and pasture (lbs.) 86,752 116,796 73,784 5.15 98,102 116,829 75,484 5.10 (749,400) (815,040) (673,264) (653,211) (679,469) (587,952) Excess nitrogen (applied minus uptake, lbs.) −52,304 −63,873 −47,301 2.77 −61,389 −67,592 −53,913 2.19 (529,417) (588,678) (471,388) (497,524) (521,105) (452,213) Proportion applying more fertilizer and manure nitrogen than uptake 0.12 0.23 0.07 7.73 0.13 0.18 0.06 6.57 (1.89) (2.01) (1.69) (1.44) (1.55) (1.17) For operations with non-zero crop acreage Nitrogen fertilizer and manure applied per crop acre (lbs./acre) 118 200 81 4.47 128 181 63 6.57 (1,933) (2,652) (925) (1,506) (1,801) (619) Nitrogen uptake per crop acre (lbs./acre) 180 182 179 0.64 184 192 174 3.59 (474) (413) (519) (349) (302) (413) Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre −61 18 −97 4.30 −56 −11 −111 5.25 (1,971) (2,653) (1,058) (1,574) (1,872) (742) Proportion applying more fertilizer and manure nitrogen than uptake per acre 0.12 0.40 0.18 7.62 0.13 0.36 0.11 10.30 (1.92) (0.49) (0.39) (1.45) (0.48) (0.31) Year 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Recoverable manure nitrogen generated (lbs.) 10,398 22,879 5,011 11.22 19,254 28,908 7,595 11.24 (144,225) (161,220) (111,082) (171,391) (196,318) (88,890) Recoverable manure nitrogen applied (lbs.) 7,971 17,778 3,729 9.44 14,080 21,420 5,234 9.77 (132,058) (151,003) (102,064) (149,134) (173,762) (72,804) Fertilizer nitrogen applied (lbs.) 26,280 35,145 22,454 4.09 22,700 27,953 16,355 4.66 (285,973) (292,167) (277,174) (204,952) (222,454) (165,975) Fertilizer and manure nitrogen applied (lbs.) 34,240 52,923 26,175 7.13 36,759 49,320 21,588 8.69 (344,499) (362,032) (314,967) (275,468) (307,122) (185,937) Nitrogen uptake on crops and pasture (lbs.) 86,752 116,796 73,784 5.15 98,102 116,829 75,484 5.10 (749,400) (815,040) (673,264) (653,211) (679,469) (587,952) Excess nitrogen (applied minus uptake, lbs.) −52,304 −63,873 −47,301 2.77 −61,389 −67,592 −53,913 2.19 (529,417) (588,678) (471,388) (497,524) (521,105) (452,213) Proportion applying more fertilizer and manure nitrogen than uptake 0.12 0.23 0.07 7.73 0.13 0.18 0.06 6.57 (1.89) (2.01) (1.69) (1.44) (1.55) (1.17) For operations with non-zero crop acreage Nitrogen fertilizer and manure applied per crop acre (lbs./acre) 118 200 81 4.47 128 181 63 6.57 (1,933) (2,652) (925) (1,506) (1,801) (619) Nitrogen uptake per crop acre (lbs./acre) 180 182 179 0.64 184 192 174 3.59 (474) (413) (519) (349) (302) (413) Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre −61 18 −97 4.30 −56 −11 −111 5.25 (1,971) (2,653) (1,058) (1,574) (1,872) (742) Proportion applying more fertilizer and manure nitrogen than uptake per acre 0.12 0.40 0.18 7.62 0.13 0.36 0.11 10.30 (1.92) (0.49) (0.39) (1.45) (0.48) (0.31) Table 3 Operation-level Means of Nutrient Generation, Application, and Uptake Variables, by Year and NMP Status Year 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Recoverable manure nitrogen generated (lbs.) 10,398 22,879 5,011 11.22 19,254 28,908 7,595 11.24 (144,225) (161,220) (111,082) (171,391) (196,318) (88,890) Recoverable manure nitrogen applied (lbs.) 7,971 17,778 3,729 9.44 14,080 21,420 5,234 9.77 (132,058) (151,003) (102,064) (149,134) (173,762) (72,804) Fertilizer nitrogen applied (lbs.) 26,280 35,145 22,454 4.09 22,700 27,953 16,355 4.66 (285,973) (292,167) (277,174) (204,952) (222,454) (165,975) Fertilizer and manure nitrogen applied (lbs.) 34,240 52,923 26,175 7.13 36,759 49,320 21,588 8.69 (344,499) (362,032) (314,967) (275,468) (307,122) (185,937) Nitrogen uptake on crops and pasture (lbs.) 86,752 116,796 73,784 5.15 98,102 116,829 75,484 5.10 (749,400) (815,040) (673,264) (653,211) (679,469) (587,952) Excess nitrogen (applied minus uptake, lbs.) −52,304 −63,873 −47,301 2.77 −61,389 −67,592 −53,913 2.19 (529,417) (588,678) (471,388) (497,524) (521,105) (452,213) Proportion applying more fertilizer and manure nitrogen than uptake 0.12 0.23 0.07 7.73 0.13 0.18 0.06 6.57 (1.89) (2.01) (1.69) (1.44) (1.55) (1.17) For operations with non-zero crop acreage Nitrogen fertilizer and manure applied per crop acre (lbs./acre) 118 200 81 4.47 128 181 63 6.57 (1,933) (2,652) (925) (1,506) (1,801) (619) Nitrogen uptake per crop acre (lbs./acre) 180 182 179 0.64 184 192 174 3.59 (474) (413) (519) (349) (302) (413) Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre −61 18 −97 4.30 −56 −11 −111 5.25 (1,971) (2,653) (1,058) (1,574) (1,872) (742) Proportion applying more fertilizer and manure nitrogen than uptake per acre 0.12 0.40 0.18 7.62 0.13 0.36 0.11 10.30 (1.92) (0.49) (0.39) (1.45) (0.48) (0.31) Year 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Recoverable manure nitrogen generated (lbs.) 10,398 22,879 5,011 11.22 19,254 28,908 7,595 11.24 (144,225) (161,220) (111,082) (171,391) (196,318) (88,890) Recoverable manure nitrogen applied (lbs.) 7,971 17,778 3,729 9.44 14,080 21,420 5,234 9.77 (132,058) (151,003) (102,064) (149,134) (173,762) (72,804) Fertilizer nitrogen applied (lbs.) 26,280 35,145 22,454 4.09 22,700 27,953 16,355 4.66 (285,973) (292,167) (277,174) (204,952) (222,454) (165,975) Fertilizer and manure nitrogen applied (lbs.) 34,240 52,923 26,175 7.13 36,759 49,320 21,588 8.69 (344,499) (362,032) (314,967) (275,468) (307,122) (185,937) Nitrogen uptake on crops and pasture (lbs.) 86,752 116,796 73,784 5.15 98,102 116,829 75,484 5.10 (749,400) (815,040) (673,264) (653,211) (679,469) (587,952) Excess nitrogen (applied minus uptake, lbs.) −52,304 −63,873 −47,301 2.77 −61,389 −67,592 −53,913 2.19 (529,417) (588,678) (471,388) (497,524) (521,105) (452,213) Proportion applying more fertilizer and manure nitrogen than uptake 0.12 0.23 0.07 7.73 0.13 0.18 0.06 6.57 (1.89) (2.01) (1.69) (1.44) (1.55) (1.17) For operations with non-zero crop acreage Nitrogen fertilizer and manure applied per crop acre (lbs./acre) 118 200 81 4.47 128 181 63 6.57 (1,933) (2,652) (925) (1,506) (1,801) (619) Nitrogen uptake per crop acre (lbs./acre) 180 182 179 0.64 184 192 174 3.59 (474) (413) (519) (349) (302) (413) Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre −61 18 −97 4.30 −56 −11 −111 5.25 (1,971) (2,653) (1,058) (1,574) (1,872) (742) Proportion applying more fertilizer and manure nitrogen than uptake per acre 0.12 0.40 0.18 7.62 0.13 0.36 0.11 10.30 (1.92) (0.49) (0.39) (1.45) (0.48) (0.31) These differences in application might not yield excess applications if the amount of crop nutrient uptake is comparably higher on operations with NMPs. While operations with NMPs do have more nutrient uptake capacity, the ratio between operations with NMPs and those without is about 1.6 in both years. This is lower than the 2 to 2.3 times as many nutrients applied. Like the comparisons of animal units to crop acreage, this again suggests that operations with NMPs may be more likely to over-apply nutrients. The two measures of excess applications (application of manure and fertilizer nutrients minus uptake) tell slightly different stories. Operations with NMPs have, on average, a higher amount of surplus uptake capacity (a more highly negative excess) than operations without NMPs. This is true for both 2004 and 2009, and suggests that on average, operations with NMPs apply fewer nutrients relative to their uptake capacities than operations without NMPs. However, operations with NMPs are approximately 3 times more likely to apply excess than those without (23% of operations with NMPs versus 7% of operations without in 2004; 18% versus 6%, respectively, in 2009). The different qualitative conclusions to be drawn from these two findings arise from a longer left-tailed distribution of excess for operations with NMPs compared to those without. While a greater percentage of NMP operations have excess nitrogen applications compared to non-NMP operations, NMP operations also have much lower values of excess (more negative) than those without NMPs. Table 4 shows the percentage of operations within each of six categories of excess nitrogen; while operations with NMPs are more likely than those without NMPs to be in the “most negative” category (i.e., they have the more uptake than applications), they are also more likely to have positive excess. Table 4 Distribution of Nitrogen Excess, by Year and NMP Status 2004 2009 Farms with NMP Farms without NMP Farms with NMP Farms without NMP Total nitrogen excess (applied minus uptake), lbs. Less than −100,000 26% 17% 29% 19% −100,000 to −50,000 14% 12% 14% 15% −50,000 to 0 37% 64% 39% 59% 1 to 50,000 19% 7% 14% 5% 50,001 to 100,000 3% 0.2% 3% 1% More than 100,000 1% 0.1% 1% 0.1% Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre More than −100 61% 60% 65% 70% −100 to −50 14% 22% 12% 14% −50 to 0 5% 13% 7% 12% 1 to 50 4% 2% 3% 2% 50−100 3% 1% 1% 0.3% More than 100 13% 2% 11% 2% 2004 2009 Farms with NMP Farms without NMP Farms with NMP Farms without NMP Total nitrogen excess (applied minus uptake), lbs. Less than −100,000 26% 17% 29% 19% −100,000 to −50,000 14% 12% 14% 15% −50,000 to 0 37% 64% 39% 59% 1 to 50,000 19% 7% 14% 5% 50,001 to 100,000 3% 0.2% 3% 1% More than 100,000 1% 0.1% 1% 0.1% Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre More than −100 61% 60% 65% 70% −100 to −50 14% 22% 12% 14% −50 to 0 5% 13% 7% 12% 1 to 50 4% 2% 3% 2% 50−100 3% 1% 1% 0.3% More than 100 13% 2% 11% 2% Table 4 Distribution of Nitrogen Excess, by Year and NMP Status 2004 2009 Farms with NMP Farms without NMP Farms with NMP Farms without NMP Total nitrogen excess (applied minus uptake), lbs. Less than −100,000 26% 17% 29% 19% −100,000 to −50,000 14% 12% 14% 15% −50,000 to 0 37% 64% 39% 59% 1 to 50,000 19% 7% 14% 5% 50,001 to 100,000 3% 0.2% 3% 1% More than 100,000 1% 0.1% 1% 0.1% Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre More than −100 61% 60% 65% 70% −100 to −50 14% 22% 12% 14% −50 to 0 5% 13% 7% 12% 1 to 50 4% 2% 3% 2% 50−100 3% 1% 1% 0.3% More than 100 13% 2% 11% 2% 2004 2009 Farms with NMP Farms without NMP Farms with NMP Farms without NMP Total nitrogen excess (applied minus uptake), lbs. Less than −100,000 26% 17% 29% 19% −100,000 to −50,000 14% 12% 14% 15% −50,000 to 0 37% 64% 39% 59% 1 to 50,000 19% 7% 14% 5% 50,001 to 100,000 3% 0.2% 3% 1% More than 100,000 1% 0.1% 1% 0.1% Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre More than −100 61% 60% 65% 70% −100 to −50 14% 22% 12% 14% −50 to 0 5% 13% 7% 12% 1 to 50 4% 2% 3% 2% 50−100 3% 1% 1% 0.3% More than 100 13% 2% 11% 2% For operations with any crop acreage, we can also examine the amount of nitrogen applied per acre, the uptake per acre, and the “excess” per acre. While looking at total amounts of nutrients applied versus uptake might provide us with an indication of nutrient management involving changes in number of animals or amount of acreage (on the extensive margin), per-acre rates might provide an indication of nutrient management without adjusting farm size (on the intensive margin). The bottom panel of table 3 shows per-acre measures; operations with NMPs have higher average per-acre applications of nitrogen in both years. Uptake per acre is more similar across NMP status; “excess” per acre is also likely to be greater for NMP operations in both years. The bottom panel of table 4 shows the distributions of the excess-per-acre variable; this suggests that even controlling for acreage, operations with NMPs are more likely to apply excess per acre, and more excess per acre, than their non-NMP counterparts. These summary statistics suggest that operations with NMPs are more likely to perform certain documentable nutrient management behaviors (like testing, feed adjustments, and manure incorporation at application), but also may be more likely to apply excess nutrients. However, these cross-tabulations do not adjust for potential confounders that may influence both management behaviors and NMP adoption. We therefore address these as well as other empirical issues using econometric strategies. Empirical Strategy Two fundamental issues arise when trying to estimate the impact of NMP adoption on nutrient management behaviors (the outcomes). First, other factors may be associated with both NMP adoption and the outcomes. For example, the summary statistics reveal that operations with more hogs are more likely to have NMPs. Larger operations may also manage their manure differently from smaller operations based on size-related factors, not the NMP. Controlling for these observable variables that are correlated with both NMP status and the outcome can reduce the bias in the estimated effect of NMP status on the outcomes. A second factor that may bias estimated effects between NMP status and the outcomes is selection. As previously described, NMP adoption can be voluntarily or required. We are not able to identify in the ARMS data which farms have been required to adopt NMPs and which have done so voluntarily. The concern is that operations may voluntarily adopt NMPs because they already have good nutrient management protocols and it is easy to do so, or because they have worse behaviors and adopt an NMP to signal better behavior. Thus, the outcomes are influencing NMP adoption, rather than the reverse. This form of endogeneity can bias the estimated effect of the NMP on the outcomes. To confront both of these issues we employ an endogenous treatment effects model. The model assumes a joint normal distribution between the errors of the selection equation (the decision to obtain a NMP) and the outcome regression (the measure of nutrient management). This approach addresses potential unobserved correlation between the operator’s decision to adopt a NMP and the behavior (outcome) variable, thus reducing bias in the estimated impact of the NMP on the nutrient management outcomes. We estimate a linear model for the behavior with a constrained normal distribution to model the deviation from the assumption of conditional independence of NMP status using maximum likelihood. Begin with the equation explaining the management outcome yi: yi=α+δ1NMPi+Xi'β+ρs+ei. Here, i indexes operation while s indexes state; NMP is a binary variable indicating whether the operation has an NMP or not; δ1 is the coefficient of interest; X is a vector of independent variables, including size, operator characteristics, and farm characteristics.4 The summary statistics revealed that observable characteristics of operations varied across NMP status; these factors may be correlated with adoption on an NMP as well as the outcome variable, hence, we control for them to minimize bias in the estimated effect of NMPs; ρ is a vector of dummy variables (with coefficients) for state. These state fixed effects control for factors that affect all farms within the state that are correlated with both NMP adoption and the outcome variable.5 Notably, states may have different levels of regulatory stringency and environmental attitudes, impacting both adoption of NMP and the outcome variables.6 An empirical concern is potential endogeneity of the treatment variable. Including the variables in X can control for the observable influence of these variables on both the treatment and the outcomes, and the state fixed effects can control for the influence of unobservable state-level confounders. However, unobservable factors not captured in the state fixed effects might be correlated with NMP status and the outcome variables. Thus, the “treatment”—NMP status—is assumed to be a function of a latent variable NMPi*: NMPi*=Ziγ+θs+ui where the decision to obtain an NMP is made according to NMPi=1, NMPi*>00, otherwise. Here, Zi is a vector of operator, operation, and regional characteristics thought to influence NMP status, and θs refer to state fixed effects.7 The observable variables that serve as control include, most notably, the size variable for number of hogs; the state fixed effects here control for state-level variability in NMP adoption requirements. Unobservable factors may cause the error terms for the behavior regression (equation 1) and that for the treatment regression (equation 2) to be correlated, resulting in biased parameter estimates in our equation of interest (1). To address this issue, we assume a joint normal distribution for the errors and estimate the two equations together using maximum likelihood; see Maddala (1983) and StataCorp (2015) for the error matrix and log likelihood functions. In both the NMP adoption and the nutrient management regressions, we consider the size of the operation—defined in terms of animal units—to be exogenous. However, prior research suggests that some operations that are close to a regulatory threshold may adjust their size to remain just below the threshold (Sneeringer and Key 2011). We do not address this type of size adjustment here for several reasons. First, our sample size is too small to detect such adjustments; detection requires a large number of observations on either side of the regulatory size cut-off, which the ARMS does not provide. Second, we are examining the entire distribution of animal sizes, rather than just the small window of sizes around the threshold (as was done in prior research). Hence, the impact of any farm size changes near the threshold is likely to be limited in the overall results. Unlike number of animal units, we do not treat total crop acreage as exogenous. While we control for whether or not an operation has any crop acreage, and whether or not it has above the median value of crop acreage, our prior is that operations may abide by NMPs by adjusting the number of cultivated acres. Adjustments in acreage will affect the per-acre nutrient application and uptake rates. We analyze two types of outcome variables, ( y): (a) indicator variables for nutrient testing, feed adjustment for nutrient content of manure, manure incorporation at application, and excess application; and (b) continuous variables for amounts of nutrients applied, nutrient uptake capacity, amount of “excess” nutrients applied (which can be a negative number), applications per acre, uptake per acre, and excess per acre. For the continuous behavior variables, we model equation (1) as a linear regression. The choice of regression model to use for the dichotomous dependent variables is more complicated. A logit or probit is often preferred to a linear probability model (LPM) because the predicted probabilities are bounded by zero and one, and the nonlinear functional form may provide a less-biased estimate of the effect of the relevant variable, depending on the “true” data-generating process. Several complications arise in addressing the endogeneity of the treatment variable that leads us to use the LPM.8 The LPM is often used instead of the logit/probit model because the coefficients are much easier to interpret, particularly when interaction terms are included (Ai and Norton 2003; Norton, Wang, and Ai 2004). In addition, the LPM usually provides very similar estimates of the marginal effects to logits/probits. The non-linear functional form of the probit/logit is not necessarily better, particularly when there are supplementary concerns (like endogenous regressors).9 On the other hand, there are multiple concerns with the LPM. The LPM can generate heteroscedastic standard errors, but this can be surmounted by using robust standard errors. The LPM may yield predicted probabilities less than zero or greater than one, but this is only a problem if one is attempting to forecast probabilities; we are not trying to do this, hence this is not a concern here. As a check against the potential mis-specification bias in the marginal effect in the LPM, in the appendix we show results of the LPM of equation (1) without accounting for the endogeneity of the treatment variable as well as estimated marginal effects for probit regression (again, without accounting for the potential endogeneity of NMP status). If these estimates are fairly similar, we can be less concerned about potential bias generated by modeling the binary behavior variables as linear functions. We estimate results for each survey year separately; this amounts to a model that is fully-interacted by year and allows all parameters to vary between years. We do not estimate panel data models because there are only 138 observations that appear in both survey years, which strongly limits our ability to estimate panel models with precision. Results of Econometric Estimation Table 5 reports estimation results of the correlation between having an NMP and recorded nutrient management practices, controlling for potential confounders and potential endogeneity of NMP status. All models include fixed effects for state, and standard errors clustered by state. Clustering by state accounts for possible correlation in model errors for individual operations in the same state. The top panel shows results for 2004 while the bottom panel shows results for 2009.10 Table 5 Endogenous Treatment Effects Regression Results, 2004 and 2009 (maximum likelihood) Test manure for nitrogen content Added phytase to feed Adjusted schedule of feed formulations Used other feed additives or formula adjustments Incorporated manure at applicationb 2004 NMP 0.27*** 0.16*** 0.17*** 0.13*** 0.15 (0.03) (0.05) (0.04) (0.04) (0.22) ln(Hog animal units) 0.07*** 0.06*** 0.05* 0.02** 0.03 (0.01) (0.02) (0.03) (0.01) (0.04) Large CAFO −0.06 −0.00 −0.04 −0.01 −0.11 (0.06) (0.08) (0.15) (0.06) (0.07) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,195 967 Prob > chi2 0.05 0.04 0.75 0.02 0.58 2009 NMP 0.57 0.20*** 0.32* 0.14*** 0.07 (0.63) (0.08) (0.17) (0.05) (0.08) ln(Hog animal units) 0.06 0.07*** 0.02 0.01*** 0.00 (0.08) (0.01) (0.02) (0.00) (0.02) Large CAFO 0.03 −0.05 −0.04 −0.00 0.04 (0.06) (0.07) (0.06) (0.03) (0.04) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,282 1,017 Prob > chi2 0.56 0.10 0.16 0.15 0.28 Test manure for nitrogen content Added phytase to feed Adjusted schedule of feed formulations Used other feed additives or formula adjustments Incorporated manure at applicationb 2004 NMP 0.27*** 0.16*** 0.17*** 0.13*** 0.15 (0.03) (0.05) (0.04) (0.04) (0.22) ln(Hog animal units) 0.07*** 0.06*** 0.05* 0.02** 0.03 (0.01) (0.02) (0.03) (0.01) (0.04) Large CAFO −0.06 −0.00 −0.04 −0.01 −0.11 (0.06) (0.08) (0.15) (0.06) (0.07) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,195 967 Prob > chi2 0.05 0.04 0.75 0.02 0.58 2009 NMP 0.57 0.20*** 0.32* 0.14*** 0.07 (0.63) (0.08) (0.17) (0.05) (0.08) ln(Hog animal units) 0.06 0.07*** 0.02 0.01*** 0.00 (0.08) (0.01) (0.02) (0.00) (0.02) Large CAFO 0.03 −0.05 −0.04 −0.00 0.04 (0.06) (0.07) (0.06) (0.03) (0.04) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,282 1,017 Prob > chi2 0.56 0.10 0.16 0.15 0.28 Note: Superscript a indicates that other controls include whether the operation has any crop acreage, whether the operation is in the top half of crop acreage, whether the operation is farrow to finish, whether the operation has an EQIP contract, whether there has ever been a TMDL in the county before the survey year, whether the operation is a contractor, the operator's years of experience, whether the operator has a college degree, and the population density of the county. Superscript b indicates the sample consists of just those operations that applied manure in any form. Superscript C indicates a sample of just those operations with any crop acreage. These regressions do not include the independent variable of whether the operation had any crop acreage. Results of 24 regressions are shown. All outcome variables are indicator (0/1) variables. Asterisk * refers to confidence at the 10% level, ** refer to confidence at the 5% level, and *** refer to confidence at the 1% level. All regressions include state-level fixed effects. Prob > chi2 refers to a likelihood ratio test of whether we can reject the null hypothesis of no correlation between the treatment errors and the outcome errors. A probability of the chi-squared value less than 0.05 suggests we can reject the null hypothesis of no selection with 95% certainty. Table 5 Endogenous Treatment Effects Regression Results, 2004 and 2009 (maximum likelihood) Test manure for nitrogen content Added phytase to feed Adjusted schedule of feed formulations Used other feed additives or formula adjustments Incorporated manure at applicationb 2004 NMP 0.27*** 0.16*** 0.17*** 0.13*** 0.15 (0.03) (0.05) (0.04) (0.04) (0.22) ln(Hog animal units) 0.07*** 0.06*** 0.05* 0.02** 0.03 (0.01) (0.02) (0.03) (0.01) (0.04) Large CAFO −0.06 −0.00 −0.04 −0.01 −0.11 (0.06) (0.08) (0.15) (0.06) (0.07) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,195 967 Prob > chi2 0.05 0.04 0.75 0.02 0.58 2009 NMP 0.57 0.20*** 0.32* 0.14*** 0.07 (0.63) (0.08) (0.17) (0.05) (0.08) ln(Hog animal units) 0.06 0.07*** 0.02 0.01*** 0.00 (0.08) (0.01) (0.02) (0.00) (0.02) Large CAFO 0.03 −0.05 −0.04 −0.00 0.04 (0.06) (0.07) (0.06) (0.03) (0.04) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,282 1,017 Prob > chi2 0.56 0.10 0.16 0.15 0.28 Test manure for nitrogen content Added phytase to feed Adjusted schedule of feed formulations Used other feed additives or formula adjustments Incorporated manure at applicationb 2004 NMP 0.27*** 0.16*** 0.17*** 0.13*** 0.15 (0.03) (0.05) (0.04) (0.04) (0.22) ln(Hog animal units) 0.07*** 0.06*** 0.05* 0.02** 0.03 (0.01) (0.02) (0.03) (0.01) (0.04) Large CAFO −0.06 −0.00 −0.04 −0.01 −0.11 (0.06) (0.08) (0.15) (0.06) (0.07) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,195 967 Prob > chi2 0.05 0.04 0.75 0.02 0.58 2009 NMP 0.57 0.20*** 0.32* 0.14*** 0.07 (0.63) (0.08) (0.17) (0.05) (0.08) ln(Hog animal units) 0.06 0.07*** 0.02 0.01*** 0.00 (0.08) (0.01) (0.02) (0.00) (0.02) Large CAFO 0.03 −0.05 −0.04 −0.00 0.04 (0.06) (0.07) (0.06) (0.03) (0.04) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,282 1,017 Prob > chi2 0.56 0.10 0.16 0.15 0.28 Note: Superscript a indicates that other controls include whether the operation has any crop acreage, whether the operation is in the top half of crop acreage, whether the operation is farrow to finish, whether the operation has an EQIP contract, whether there has ever been a TMDL in the county before the survey year, whether the operation is a contractor, the operator's years of experience, whether the operator has a college degree, and the population density of the county. Superscript b indicates the sample consists of just those operations that applied manure in any form. Superscript C indicates a sample of just those operations with any crop acreage. These regressions do not include the independent variable of whether the operation had any crop acreage. Results of 24 regressions are shown. All outcome variables are indicator (0/1) variables. Asterisk * refers to confidence at the 10% level, ** refer to confidence at the 5% level, and *** refer to confidence at the 1% level. All regressions include state-level fixed effects. Prob > chi2 refers to a likelihood ratio test of whether we can reject the null hypothesis of no correlation between the treatment errors and the outcome errors. A probability of the chi-squared value less than 0.05 suggests we can reject the null hypothesis of no selection with 95% certainty. In 2004, the variable NMP is significant and positive at the 1% level in predicting whether an operator tests manure for nutrient content, as well as for the three measures of adjusting feed for manure nitrogen content; NMP is not a significant predictor of whether the operation incorporates manure at application to reduce potential run-off. Except for the nitrogen testing outcome, these results hold for 2009 as well. In 2004, operations with a NMP are 27 percentage points more likely to perform manure nutrient testing, and between 13 and 17 percentage points more likely to adjust feed to change manure nutrient content. In 2009, operations with NMPs are between 14 and 32 percentage points more likely to adjust feed.11,12 Table 5 also shows the test statistic for whether or not we can reject the null hypothesis of no correlation between the treatment regression errors and the behavior regression errors. Values less than 0.05 for this test statistic suggest that we can reject the null hypothesis of no correlation. Our results suggest that a correlation between the errors occurs for some but not all variables. Table 6 provides the results for the nutrient application, uptake, and excess outcomes. In 2004 having a NMP was not a significant predictor (in this empirical strategy) of whether the operation applied any excess, the amount of nitrogen applied, the total uptake, or the amount of excess applied. The presence of a NMP did significantly predict less nitrogen applied per acre and less excess per acre. This suggests that in 2004, operations with a NMP reduced their per-acre nutrient application rates but did not significantly change their planted acreage, nor did they alter their crop mix to generate higher uptake values.13 Table 6 Endogenous Treatment Effects Regression Results, 2004 and 2009 (maximum likelihood) Indicator (0/1) variable Continuous outcome variables Applied excess nitrogen Total nitrogen manure and fertilizer applied (lbs.) Total nitrogen uptake (lbs.) Excess nitrogen (applied minus uptake, lbs.) Applied per acre (lbs.)c Uptake per acre (lbs.)c Excess per acre (lbs.)c 2004 NMP 0.04 6,659 10,975 −4,084 −392.65*** 26.94 −380.89** (0.07) (4,330) (12,186) (9,719) (147.77) (67.08) (148.80) ln(Hog animal units) 0.03*** 7,307*** 14,885*** −7,572** 74.71*** 1.52 68.49** (0.01) (1,678) (3,575) (3,324) (27.19) (10.97) (26.97) Large CAFO −0.04 39,206*** 18,358 20,641 15.68 −16.57 33.34 (0.03) (14,492) (20,214) (15,450) (38.92) (13.95) (40.60) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,193 992 992 990 Prob > chi2 0.61 0.07 0.82 0.40 0.00 0.60 0.00 2009 NMP −0.30*** −75,766*** 29,303** −38,132** −389.33*** −54.74* −410.44*** (0.04) (11,730) (12,650) (17,042) (150.93) (32.69) (149.40) ln(Hog animal units) 0.08*** 18,027*** 14,136*** −4,278 78.25*** 13.93** 74.63*** (0.01) (2,835) (3,365) (3,162) (25.36) (5.59) (25.32) Large CAFO −0.00 20,972** −1,349 25,692*** −17.82 −25.79*** 1.77 (0.03) (9,716) (12,085) (9,799) (40.42) (8.24) (41.12) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,279 1,095 1,095 1,092 Prob > chi2 0.00 0.00 0.17 0.11 0.00 0.13 0.00 Indicator (0/1) variable Continuous outcome variables Applied excess nitrogen Total nitrogen manure and fertilizer applied (lbs.) Total nitrogen uptake (lbs.) Excess nitrogen (applied minus uptake, lbs.) Applied per acre (lbs.)c Uptake per acre (lbs.)c Excess per acre (lbs.)c 2004 NMP 0.04 6,659 10,975 −4,084 −392.65*** 26.94 −380.89** (0.07) (4,330) (12,186) (9,719) (147.77) (67.08) (148.80) ln(Hog animal units) 0.03*** 7,307*** 14,885*** −7,572** 74.71*** 1.52 68.49** (0.01) (1,678) (3,575) (3,324) (27.19) (10.97) (26.97) Large CAFO −0.04 39,206*** 18,358 20,641 15.68 −16.57 33.34 (0.03) (14,492) (20,214) (15,450) (38.92) (13.95) (40.60) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,193 992 992 990 Prob > chi2 0.61 0.07 0.82 0.40 0.00 0.60 0.00 2009 NMP −0.30*** −75,766*** 29,303** −38,132** −389.33*** −54.74* −410.44*** (0.04) (11,730) (12,650) (17,042) (150.93) (32.69) (149.40) ln(Hog animal units) 0.08*** 18,027*** 14,136*** −4,278 78.25*** 13.93** 74.63*** (0.01) (2,835) (3,365) (3,162) (25.36) (5.59) (25.32) Large CAFO −0.00 20,972** −1,349 25,692*** −17.82 −25.79*** 1.77 (0.03) (9,716) (12,085) (9,799) (40.42) (8.24) (41.12) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,279 1,095 1,095 1,092 Prob > chi2 0.00 0.00 0.17 0.11 0.00 0.13 0.00 Note: Superscript a indicates that other controls include whether the operation has any crop acreage, whether the operation is in the top half of crop acreage, whether the operation is farrow to finish, whether the operation has an EQIP contract, whether there has ever been a TMDL in the county before the survey year, whether the operation is a contractor, the operator's years of experience, whether the operator has a college degree, and the population density of the county. Superscript b indicate a sample of just those operations that applied manure in any form. Superscript C indicates a sample of just those operations with any crop acreage. These regressions do not include the independent variable of whether the operation had any crop acreage. Results of 24 regressions are shown. Asterisk * refers to confidence at the 10% level, ** refer to confidence at the 5% level, and *** refer to confidence at the 1% level. All regressions include state-level fixed effects. Prob > chi2 refers to a likelihood ratio test of whether we can reject the null hypothesis of no correlation between the treatment errors and the outcome errors. A probability of the chi-squared value less than 0.05 suggests we can reject the null hypothesis of no selection with 95% certainty. Table 6 Endogenous Treatment Effects Regression Results, 2004 and 2009 (maximum likelihood) Indicator (0/1) variable Continuous outcome variables Applied excess nitrogen Total nitrogen manure and fertilizer applied (lbs.) Total nitrogen uptake (lbs.) Excess nitrogen (applied minus uptake, lbs.) Applied per acre (lbs.)c Uptake per acre (lbs.)c Excess per acre (lbs.)c 2004 NMP 0.04 6,659 10,975 −4,084 −392.65*** 26.94 −380.89** (0.07) (4,330) (12,186) (9,719) (147.77) (67.08) (148.80) ln(Hog animal units) 0.03*** 7,307*** 14,885*** −7,572** 74.71*** 1.52 68.49** (0.01) (1,678) (3,575) (3,324) (27.19) (10.97) (26.97) Large CAFO −0.04 39,206*** 18,358 20,641 15.68 −16.57 33.34 (0.03) (14,492) (20,214) (15,450) (38.92) (13.95) (40.60) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,193 992 992 990 Prob > chi2 0.61 0.07 0.82 0.40 0.00 0.60 0.00 2009 NMP −0.30*** −75,766*** 29,303** −38,132** −389.33*** −54.74* −410.44*** (0.04) (11,730) (12,650) (17,042) (150.93) (32.69) (149.40) ln(Hog animal units) 0.08*** 18,027*** 14,136*** −4,278 78.25*** 13.93** 74.63*** (0.01) (2,835) (3,365) (3,162) (25.36) (5.59) (25.32) Large CAFO −0.00 20,972** −1,349 25,692*** −17.82 −25.79*** 1.77 (0.03) (9,716) (12,085) (9,799) (40.42) (8.24) (41.12) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,279 1,095 1,095 1,092 Prob > chi2 0.00 0.00 0.17 0.11 0.00 0.13 0.00 Indicator (0/1) variable Continuous outcome variables Applied excess nitrogen Total nitrogen manure and fertilizer applied (lbs.) Total nitrogen uptake (lbs.) Excess nitrogen (applied minus uptake, lbs.) Applied per acre (lbs.)c Uptake per acre (lbs.)c Excess per acre (lbs.)c 2004 NMP 0.04 6,659 10,975 −4,084 −392.65*** 26.94 −380.89** (0.07) (4,330) (12,186) (9,719) (147.77) (67.08) (148.80) ln(Hog animal units) 0.03*** 7,307*** 14,885*** −7,572** 74.71*** 1.52 68.49** (0.01) (1,678) (3,575) (3,324) (27.19) (10.97) (26.97) Large CAFO −0.04 39,206*** 18,358 20,641 15.68 −16.57 33.34 (0.03) (14,492) (20,214) (15,450) (38.92) (13.95) (40.60) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,193 992 992 990 Prob > chi2 0.61 0.07 0.82 0.40 0.00 0.60 0.00 2009 NMP −0.30*** −75,766*** 29,303** −38,132** −389.33*** −54.74* −410.44*** (0.04) (11,730) (12,650) (17,042) (150.93) (32.69) (149.40) ln(Hog animal units) 0.08*** 18,027*** 14,136*** −4,278 78.25*** 13.93** 74.63*** (0.01) (2,835) (3,365) (3,162) (25.36) (5.59) (25.32) Large CAFO −0.00 20,972** −1,349 25,692*** −17.82 −25.79*** 1.77 (0.03) (9,716) (12,085) (9,799) (40.42) (8.24) (41.12) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,279 1,095 1,095 1,092 Prob > chi2 0.00 0.00 0.17 0.11 0.00 0.13 0.00 Note: Superscript a indicates that other controls include whether the operation has any crop acreage, whether the operation is in the top half of crop acreage, whether the operation is farrow to finish, whether the operation has an EQIP contract, whether there has ever been a TMDL in the county before the survey year, whether the operation is a contractor, the operator's years of experience, whether the operator has a college degree, and the population density of the county. Superscript b indicate a sample of just those operations that applied manure in any form. Superscript C indicates a sample of just those operations with any crop acreage. These regressions do not include the independent variable of whether the operation had any crop acreage. Results of 24 regressions are shown. Asterisk * refers to confidence at the 10% level, ** refer to confidence at the 5% level, and *** refer to confidence at the 1% level. All regressions include state-level fixed effects. Prob > chi2 refers to a likelihood ratio test of whether we can reject the null hypothesis of no correlation between the treatment errors and the outcome errors. A probability of the chi-squared value less than 0.05 suggests we can reject the null hypothesis of no selection with 95% certainty. The predictive power of NMPs becomes greater in the 2009 survey. Operations with NMPs were 30 percentage points less likely to apply any excess nitrogen, controlling for confounders and selection. Such operations appear to have reduced the amount applied (by nearly 76,000 lbs. of nitrogen per operation), and increased their uptake (by 29,000 lbs. per operation). This resulted in a decline in total excess applied (of about 38,000 lbs. per operation). These results suggest that operations may have increased their crop acreage to more closely match crop uptake with nutrients applied. The per-acre outcomes suggest that operations with NMPs also adjusted their nutrient management practices at the intensive margin; they applied less per acre, resulting in less excess per acre. There is no evidence that farms with NMPs adjusted their crop mix to increase nitrogen uptake per acre. Comparison of the results in 2004 and 2009 suggests the effect of the NMP became stronger in terms of reducing nutrient applications. This might be explained by increasing enforcement and/or awareness among producers of better nutrient management practices. The summary statistics suggest different qualitative results from the econometric specifications for two outcomes. First, the summary statistics suggest that operations with NMPs are more likely to apply excess nutrients overall. Second, the summary statistics suggest that operations with NMPs, on average, applied more nutrients per acre than those without. In contrast, the results of the econometric specification suggest that in 2009, operations with NMPs were less likely to apply excess nutrients in 2009 (but not 2004), and applied less excess per acre in both years. Examination of a set of models (appendix table 9) without correcting for selection shows that clustering standard errors by state reduces the statistical significance of the impact of the NMP on these two outcomes, and the addition of covariates reduces the size of the estimated coefficients. The signs on the coefficients change when we control for selection. A model in which we only control for size of operation in the NMP prediction equation and the outcome equation shows statistically significant effects of NMP in predicting a lower likelihood of excess nutrient application, and less excess per acre applied. These checks suggest that both controlling for confounders and adjusting for selection are pertinent in estimating effects of NMPs. Finally, to examine whether enforcement of environmental regulations impacts the efficacy of NMPs, we interact NMP status with the four measures of enforcement (whether the county ever had a TMDL before the survey year, the population density of the county, whether the operation is a large CAFO, and whether the operation has an EQIP contract). These interaction terms are rarely statistically significant (results appear in appendix tables 10 through 17), suggesting that at least for these measures of enforcement, greater regulatory oversight does not yield better nutrient management outcomes via the NMPs. Conclusions The NMPs are a fundamental component of federal, state, and regional efforts to encourage livestock producers to apply nutrients to cropland at agronomic rates. Despite the important role that NMPs play in the environmental regulation of livestock farms, there is little information about their efficacy. This study is one of the first to examine whether having an NMP makes a farm more likely to use a set of recommended nutrient management practices and to apply manure nutrients at agronomic rates. Our results provide evidence to support the hypothesis that the use of NMPs encourages U.S. hog operations to implement manure management practices to reduce run-off into nearby waterways. After adjusting for operation size, location, and other potential confounders, and correcting for selection into NMP adoption, our empirical results suggest that farms with NMPs are significantly more likely to use practices consistent with careful nutrient management (manure nutrient testing and feed adjustments). The results of our empirical strategy suggest that in 2009, NMPs had their intended effect on excess nutrient applications (manure and commercial fertilizer nutrient applications minus nutrient uptake capacity). We find evidence that operations with NMPs (when accounting for confounding variables and selection) were statistically less likely to over-apply nutrients in 2009, but not 2004. This is potentially because farmers responded to stricter enforcement of the 2003 Clean Water Act amendments that were adopted and enforced gradually in the following years, or because of general knowledge dissemination of nutrient management. Our empirical results suggest that operations with NMPs reduce excess nutrient applications both by adjusting at the extensive margin (amount of land cropped) as well as the intensive margin (applications per unit of land). Stricter enforcement on NMPs, as indicated by having a TMDL in the county, being a large CAFO, or having an EQIP contract does not increase the efficacy of NMPs in encouraging nutrient management. Because our data were collected in 2004 and 2009, our findings may not be indicative of current practices of hog farmers. The change in predictive power between 2004 and 2009 in terms of the estimated relationships between NMPs and nutrient management applications suggest a trend toward more efficacy of NMPs. If this trend has continued or been maintained, then NMPs may still be effective in reducing potential run-off from hog operations. In this study we have examined the question of whether having a NMP is predictive of better nutrient management behaviors. We do not, however, address whether NMPs are the only policy necessary to obtain water quality goals. Since such goals vary by location, whether or not requiring NMPs will allow regulators to reach agricultural nutrient run-off reduction goals is a question answered only at the regional level. This paper suggests, however, that NMPs encourage better nutrient management behaviors, and can be one tool in promoting less agriculturally-based water pollution. Footnotes 1 These are the only two years of ARMS hogs data that contain information on NMPs. In an updated survey, completed in 2016, questions on manure management were removed. 2 For example, operations that already agronomically manage nutrients may adopt a NMP because they find it easy to do so. Alternatively, operations that do not agronomically manage nutrients may also be more likely to adopt a NMP if oversight is lax and they want to signal environmental stewardship to potential regulatory authorities or integrators. 3 In some years, some states have required all confined animal operations to obtain NMPs. For example, Maryland began requiring all operations to have NMPs by 2001 (Perez,2011). We examined the state laws for all 19 states in the sample, and none require that all operations (regardless of size, discharge status, or other feature) obtain NMPs. Notably, Maryland is not in our sample of states as it is not a major hog-producer. 4 The variables included in X are ln(number of hog animal units); whether the operation is a large CAFO with other 1,000 hog animal units; whether the operation has any crop acreage; whether the operation is in the top half of crop acreage distribution for the year; whether the operation is farrow to finish; whether the operation has an EQIP contract; whether there has ever been a TMDL in the county before the survey year; whether the operation is a contractor; the operator's years of experience; whether the operator has a college degree; and the population density of the county. Despite the subscript i on X, two of the variables included in this vector vary at the level of the county, not the operation. These two variables are county-level population density and whether or not the county has had a TMDL for nutrients before or in the survey year. We use the subscript i for brevity. 5 Notably, no state in either years has a 100% NMP adoption rate; see appendix table 1. This means that that there is still variation within states in NMP status. 6 Prior research has questioned whether operations that see less value in environmental stewardship selectively locate in states with lax enforcement of environmental regulation. In this were the case, we might see operations with worse nutrient management practices locate in places with lower enforcement. It is difficult to assess whether this would engender bias in the estimated effect of NMPs, or in what direction it would be. If operations with worse practices selectively locate in areas to avoid obtaining NMPs, then for those individual states we might see larger effects of NMPs than without selection in those states. However, relocation of poor performers would leave better environmental stewards in states with higher regulatory stringency. Comparing across NMP status in higher stringency states would yield smaller impacts of NMPs than without selection by location. The estimated overall effect of NMP status in the presence of such selection, inclusive of both types of states, could be the same, greater, or less than the estimated effect without such selection. 7 This vector includes the same variables as in X. 8 Complications arise in two main areas. First, obtaining maximum likelihood estimates requires convergence of the likelihood functions. In our application, we never obtain convergence when estimating the outcome equation as a probit, even with the most pared-down model. Second, we later consider interaction of NMP status with an enforcement variable. Interpretation of interaction terms in straightforward probits (i.e., those without endogenous binary independent variables) is complicated due to the nonlinearity of the functional form; estimates and statistical significance vary according to the levels of the included confounders. For these reasons we do not use the probit as the functional form for the outcome equation. 9 As Angrist and Pishchke note on their Mostly Harmless Econometrics blog, “[T]he LPM won’t give the true marginal effects from the right nonlinear model. But then, the same is true for the ‘wrong’ nonlinear model! The fact that we have a probit, logit and the LPM is just a statement to the fact that we don’t know what the ‘right’ model is.” Available at: http://www.mostlyharmlesseconometrics.com/2012/07/probit-better-than-lpm/. 10 Appendix table 2 shows results from the probit model to predict NMP status in each year. The strongest positive predictors for NMP adoption are the number of animal units and whether the operation has an EQIP contract. The operation being farrow-to-finish is a significant negative predictor in both years. The endogenous treatment effects model results in tables 5 and 6 estimates both the “first” and “second” stages simultaneously. We evaluate several outcome variables; hence, the estimated coefficients on the variables predicting NMP adoption are slightly different for each outcome variable. The estimates in appendix table 2 show the results of probit models estimated by themselves, rather than within the endogenous treatment model. The estimate effects are highly similar to those estimated in the “first stage” of the endogenous treatment model for each of the multiple outcome variables. 11 Appendix table 3 shows the results of the LPM without treating NMP status as endogenous. The marginal effects from probit regressions are shown in appendix table 4, again without treating NMP status as endogenous. The probit and LPM results are very similar in magnitude and statistical significance, particularly in 2009. This suggests that modeling the binary outcome behavioral regressions in the endogenous treatment model using a linear model is unlikely to result in heavily biased coefficients due to functional form. 12 While we do not show the estimated coefficients on all controls in the main text, the interested reader can find them in appendix tables 5 and 6. The strongest predictors, aside from NMP status and number of hog animal units, of the recorded nutrient management practices are whether the operation has a production contract and the state fixed effects. This is true for both years. 13 Appendix Tables 7 and 8 show the estimated coefficients on all included covariates. The strongest and most consistent observable predictors beyond those shown in the main text are whether the operation is a farrow-to-finish operation, whether the operation has a production contract, and the two measures of crop acreage. The state fixed effects are also strong predictors. References Ai C. , Norton E . 2003 . Interaction Terms in Logit and Probit Models . Economics Letters 80 : 123 – 29 . Google Scholar CrossRef Search ADS Albrecht J. 2003 . Reduction of Manure Nutrient Concentrations. Confined Animal Manure Managers Program. Swine Training Manual. Chapter 3b, last edit. Available at: http://www.clemson.edu.extension/camm/manuals/swine/sch3b_03.pdf. Beegle D. , Carton O. , Bailey J . 2000 . Nutrient Management Planning: Justification, Theory, Practice . Journal of Environmental Quality 29 ( 1 ): 72 – 9 . Google Scholar CrossRef Search ADS Burkholder J. , Libra B. , Weyer P. , Heathcote S. , Kolpin D. , Thorne P.S. , Wichman M . 2007 . Impacts of Waste from Concentrated Animal Feeding Operations on Water Quality . Environmental Health Perspectives 115 ( 2 ): 308 – 12 . Google Scholar CrossRef Search ADS PubMed Centner T.J. , Newton G.L . 2008 . Meeting Environmental Requirements for the Land Application of Manure . Journal of Animal Science 86 : 3228 – 34 . Google Scholar CrossRef Search ADS PubMed Copeland C. 2012 . Clean Water Act and Pollutant Total Maximum Daily Loads (TMDLs). Washington DC: Congressional Research Service Report 7-5700. Copeland C , Zinn J . 1998 . Animal Waste Management and the Environment: Background for Current Issues. Washington DC: Congressional Research Service Report 98-451. Genskow K.D. 2012 . Taking Stock of Voluntary Nutrient Management: Measuring and Tracking Change . Soil and Water Conservation 61 ( 1 ): 51 – 8 . Google Scholar CrossRef Search ADS Innes R. 1999 . Regulating Livestock Waste: An Economic Perspective . Choices 14 – 19 . Innes R. 2000 . The Economics of Livestock Waste and its Regulation . American Journal of Agricultural Economics 82 : 97 – 117 . Google Scholar CrossRef Search ADS Johnson R.S. , Wheeler W.J. , Christensen L.A . 1999 . EPA’s Approach to Controlling Pollution from Animal Feeding Operations: An Economics Analysis . American Journal of Agricultural Economics 81 ( 5 ): 1216–21 . Google Scholar CrossRef Search ADS Kellogg R.L. , Lander C.H. , Moffitt D.C. , Gollehon N . 2000 . Capacity of Cropland and Pastureland to Assimilate Nutrients: Spatial and Temporal Trends for the United States. Washington DC: U.S. Department of Agriculture, Natural Resources Conservation Service and Economic Research Service. Key N. , McBride W.D. , Ribaudo M. , Sneeringer S . 2011 . Trends and Developments in Hog Manure Management: 1998-2009. Washington DC: U.S. Department of Agriculture, Economic Research Service, Economic Information Bulletin No. 81. Lawley C. , Lichtenberg E. , Parker D . 2009 . Biases in Nutrient Management Planning . Land Economics 85 ( 1 ): 186 – 200 . Google Scholar CrossRef Search ADS Lyford C. , Hicks T . 2001 . The Environment and Pork Production: The Oklahoma Industry at a Crossroads . Review of Agricultural Economics 23 ( 1 ): 265 – 74 . Google Scholar CrossRef Search ADS Maddala G.S. 1983 . Limited-Dependent and Qualitative Variables in Econometrics . Cambridge, MA : Cambridge University Press . Google Scholar CrossRef Search ADS McBride W. , Key N . 2013 . U.S. Hog Production from 1992 to 2009: Technology, Restructuring, and Productivity Growth. U.S. Department of Agriculture, Economic Research Service, Economic Research Report No. 158. McCann L.M.J. 2009 . Transaction Costs of Environmental Policies and Returns to Scale: The Case of Comprehensive Nutrient Management Plans . Review of Agricultural Economics 31 ( 3 ): 561 – 73 . Google Scholar CrossRef Search ADS National Pollutant Discharge Elimination System Permit Regulation and Effluent Limitation Guidelines and Standards for Concentrated Animal Feeding Operations (CAFOs) . Federal Register 68 ( 29 ): 7176 –4. Norton E. , Wang H. , Ai C . 2004 . Computing Interaction Effects and Standard Errors in Logit and Probit Models . The Stata Journal 4 : 154 – 67 . Perez M. 2011 . Regulating Farmers: Lessons Learned from the Delmarva Peninsula . Choices 26 ( 3 ). Ribaudo M.O. , Cattaneo A. , Agapoff J . 2004 . Cost of Meeting Manure Nutrient Application Standards in Hog Production: The Roles of EQIP and Fertilizer Offsets . Review of Agricultural Economics 26 ( 4 ): 430 – 44 . Google Scholar CrossRef Search ADS Ribaudo M.O. , Johansson R.C . 2007 . Nutrient Management Use at the Rural-Urban Fringe: Does Demand for Environmental Quality Play a Role? Review of Agricultural Economics 29 ( 4 ): 689 – 99 . Google Scholar CrossRef Search ADS Ribuado M. , Key N. , Sneeringer S . 2016 . The Potential Role for a Nitrogen Compliance Policy in Mitigating Gulf Hypoxia . Applied Economic Perspectives and Policy 39 ( 3 ): 458 – 78 . Savage J.A. , Ribaudo M.O . 2013 . Impact of Environmental Policies on the Adoption of Manure Management Practices in the Chesapeake Bay Watershed . Journal of Environmental Management 129 : 143 – 48 . Google Scholar CrossRef Search ADS PubMed Shepard R. 2005 . Nutrient Management Planning: Is it the Answer to Better Management? Journal of Soil and Water Conservation 60 ( 4 ): 171 – 76 . Sneeringer S. , Key N . 2011 . Effects of Size-Based Environmental Regulations: Evidence of Regulatory Avoidance . American Journal of Agricultural Economics 93 ( 4 ): 1189 – 211 . Google Scholar CrossRef Search ADS Sneeringer S. 2016 . Comparing Participation in Nutrient Trading by Livestock Operations to Crop Producers in the Chesapeake Bay Watershed . Washington DC : U.S. Department of Agriculture, Economic Research Service , Economic Research Report No. 216. StataCorp . 2015 . Stata 14 Base Reference Manual. College Station, TX : Stata Press . Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association 2018. This work is written by US Government employees and is in the public domain in the US. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Applied Economic Perspectives and Policy Oxford University Press

Do Nutrient Management Plans Actually Manage Nutrients? Evidence from a Nationally-Representative Survey of Hog Producers

Loading next page...
 
/lp/ou_press/do-nutrient-management-plans-actually-manage-nutrients-evidence-from-a-FimYjTxkPr
Publisher
Oxford University Press
Copyright
Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association 2018. This work is written by US Government employees and is in the public domain in the US.
ISSN
2040-5790
eISSN
2040-5804
D.O.I.
10.1093/aepp/ppx057
Publisher site
See Article on Publisher Site

Abstract

Abstract A nutrient management plan (NMP) specifies recommended practices to match applied nutrients with crops’ uptake capacity. Because monitoring nutrient applications is difficult, regulators instead oversee NMP adoption. In this paper we examine whether having NMPs make hog farms more likely to adopt nutrient management practices. We estimate nutrient application and uptake rates to assess whether operations with NMPs are less likely to over-apply nutrients. Using an endogenous treatment effects model to control for potential confounding and selection bias, we find that NMPs are positively correlated with the adoption of nutrient management practices, as well as with the reduced application of excess nutrients. Livestock, nutrient management plans, regulation, CAFO, hogs The environmental regulation of manure application poses a particular challenge to regulators. Nutrients contained in manure, particularly nitrogen and phosphorus, can degrade water quality if they are over-applied to farmland and enter water resources through runoff or leaching. Because most livestock producers are not forced to internalize the social costs of their pollution, and because manure transportation is costly, they may have an incentive to over-apply nutrients. A main policy objective is therefore to prevent run-off by limiting nutrient application to agronomic rates. A regulatory challenge arises because even if limits are placed on application rates, in most situations regulators are unable to directly monitor compliance. Perhaps recognizing the costliness and general infeasibility of monitoring farmer behavior, state and federal policies instead encourage or require livestock operations to submit “nutrient management plans” (NMPs) stating how much manure and fertilizer they will apply to each acre of land. Regulators can observe these plans, thereby limiting their oversight costs. However, having a plan does not prove that a farmer is following appropriate nutrient management practices. If regulatory authorities only require these plans as proof of regulatory compliance and do not engage in further oversight, there will be limited incentive for livestock operators to change their behavior. A key policy question, therefore, is whether NMPs are useful predictors of recommended nutrient management practices and agronomic nutrient application rates. In this article we use information on NMP adoption, externally documentable nutrient management practices, and estimated nutrient application rates in order to ascertain whether there is a difference between planned and implemented activities. We use data from the 2004 and 2009 Agricultural Resource Management Survey (ARMS) for Hogs, a nationally-representative survey conducted by the USDA’s Economic Research Service and National Agricultural Statistics Service.1 The ARMS has information on NMP adoption, practices which are signals of NMP compliance (soil testing, adjustments to feed to alter manure nutrient content, incorporation of manure at application), animals by size and type, fertilizer expenditures, and crop yields. We first examine simple summary statistics on whether farms with NMPs are more likely to use certain nutrient management practices than those without. Perhaps a more pertinent measure of NMP implementation is whether operations match nutrient application to uptake. However, the ARMS hog survey instrument does not include questions on specific amounts of nutrients applied or used by crops or pasture. Instead, through an involved and detailed accounting process, we estimate the amounts of nutrients applied on farm in the form of manure or commercial fertilizer, as well as the nutrient absorptive capacity of the crops and land on the farm. We can therefore examine the average application rates of farms with and without NMPs. Simple summary statistics examining nutrient management behaviors and applications for operations with and without NMPs do not account for features that could be correlated with NMP adoption and behaviors. Such statistics also do not address potential selection bias that might occur if operations with better or worse nutrient management are more likely to voluntarily adopt NMPs.2 Therefore, we account for potential confounding and sample selection bias using an endogenous treatment effects model. Our basic empirical strategy uses cross-sectional variation in NMP status to identify its relationship to nutrient management practices, controlling for potential confounders that are associated with both NMP status and practices. Because we do not assign NMPs and because adoption is often voluntary, we model NMP adoption as endogenous. We estimate both adoption of NMP and the behavioral outcome variable jointly using maximum likelihood. Understanding the nutrient management behavior of farms with NMPs allows us to gain insight on the effectiveness of current policy, and whether there is a need to upgrade existing regulation or enforcement. The rest of this paper is organized as follows. Further background information is provided in the next section, while the two subsequent sections discuss the empirical strategy and data used in the analysis. The following section provides a discussion of the results, and the final section concludes. Background Nutrients are necessary for crop growth and are a natural by-product of livestock production, but may increase the risk for water pollution if they are applied at rates greater than what crop and pastureland can absorb. Trends in livestock production contribute to concerns of nutrient over-application. Farms increasingly specialize in either crop or livestock production, hence farm-level nutrient production may not match farm-level nutrient needs (Kellogg et al. 2000). Certain types of livestock agriculture have also become more geographically concentrated, occasionally in regions distant from the locus of crop production (McBride and Key 2013). Simultaneously, there has been continuous growth in the size of livestock operations; these larger operations often involve intensive production methods that result in the potential to apply more nutrients than needed (Johnson, Wheeler, and Christensen 1999). To avoid over-application, livestock facility operators with insufficient land on which to apply manure may ship manure off-farm. However, this practice is expensive and crop farmers’ willingness to pay for or even accept manure for free is often very low. Hence, manure has little value in many regions, creating an incentive for some livestock producers to treat it as a waste and apply it above agronomically-appropriate rates. Even in the absence of manure production, research suggests that many farmers apply more fertilizer than needed to ensure their optimum yield (Beegle, Carton, and Bailey 2000; Lawley, Lichtenberg, and Parker 2009). Applying excess nutrients that cannot be assimilated into the soil can contribute to nutrient run-off. This occurs when land-applied nutrients are carried to nearby water bodies via surface runoff of precipitation, resulting in the impairment of water quality and ecosystem resources (Copeland and Zinn 1998; Burkholder et al. 2007). In an effort to control nutrient discharge from farming operations, state and federal authorities have introduced NMPs, which have different requirements by jurisdiction, but are generally aimed at assuring that the amount of nutrients applied as manure and/or commercial fertilizer does not exceed the amount of nutrients that can be agronomically utilized by crop and pasture land. To this end, the NMP may require documentable actions, including nutrient testing of manure and soils and adjustments to feed rations. Nutrient testing of manure can inform how much manure to apply to cropland, and adjustments to feed rations can lower the nutrient content of manure (Albrecht 2003). These actions can be independently documented—at the very least through receipts. The least documentable action required by NMPs is for farms to match application rates to uptake. While NMPs generally provide prescriptions on the amount of nutrients that can be applied to individual fields, an independent inspector does not directly monitor applications. NMPs can be either voluntary or mandatory. Determining which operations are required to adopt NMPs and which do so voluntarily is difficult due to differing state, regional, and federal stipulations that may require a NMP. Operations that are deemed Concentrated Animal Feeding Operations (CAFOs) may be required to obtain an NMP as a condition to receiving a permit to operate, or may enact one as a stopgap against potential fines (Sneeringer 2016). In 2003, the EPA amended Clean Water Act CAFO rules originally adopted in the 1970s to provide some oversight of manure application; these 2003 rules were contested over the next several years, updated in 2008, and finalized in 2011. The 2003 rules began requiring NMPs for certain operations either designated or defined as CAFOs. Operations are defined to be large CAFOs if they have over a certain number of animals in inventory, if they raise animals in confinement for at least 45 days of the year, and do not grow vegetation in the location in which they raise animals. The qualifying threshold for “large CAFO” varies by animal type and size; early characterizations used 1,000 animal units as the threshold for a “large CAFO”. Operations with fewer animals may be designated to be CAFOs by the pertinent regulatory authority depending on the risk they pose to surface water pollution. The 2003 regulations stipulating that all CAFOs had to obtain permits (with NMPs) were contested. The final 2011 regulations required that only CAFOs with documented discharges needed to obtain permits (and therefore needed an NMP). Operations could also obtain an NMP as a measure to avoid fines should they be eventually found to have a documented discharge. Notably, the EPA devolves regulatory authority of the CWA CAFO rules to states after review of individual states’ implementation plans. States may adopt more stringent regulations requiring more farms to adopt NMPs.3 Enforcement may also vary between states. A second type of regulation that may require operations to adopt NMPs are Total Maximum Daily Loads (TMDLs). TMDLs are adopted at a regional level, and often focus on individual watersheds. Because TMDLs fall under the jurisdiction of a number of regulatory agencies, it is difficult to generalize about what TMDLs require, but at the very least their presence suggests a higher degree of oversight. NMPs can also be voluntary. The USDA’s National Resource Conservation Service (NRCS) as well as state and regional-level entities have been providing education and assistance for operations wishing to voluntarily adopt NMPs. The Environmental Quality Incentives Program (EQIP), run by NRCS in conjunction with state offices, offers financial and technical assistance to farmers to address a host of environmental concerns. A stipulation of receiving funds for any of a number of projects is often a NMP. Further, EQIP funds may also be used as a method to financially support the adoption of NMPs. Efficacy of NMPs in Reducing Nutrient Pollution While NMPs have been touted as one of the only methods of reducing nutrient run-off from livestock operations, they suffer, at least on the theoretical level, from problems of monitoring and enforcement. Innes (1999) points out that attributing nutrient pollution to individual livestock facilities is impossible without “a massive army of manure police patrolling a livestock producer’s surrounding crop fields to watch and limit the operator’s every manure application.” Noting that such a policy is too costly to be justified, Innes proposes alternative methods of nutrient pollution reduction such as fertilizer taxes, limits on the number of livestock in a region, manure use subsidies, and other mechanisms (Innes 1999, 2000). Why would farmers not implement their NMPs even if they adopt a plan? Noncompliance may be due to the lack of incentives for farmers to use NMPs and to costs of implementation. Studies suggest that farmers must have an understanding of the impact of their manure management practices on water quality in order for them to implement their NMPs (Shepard 2005; Ribaudo and Johansson 2007; Savage and Ribaudo 2013). Economic incentives through EQIP and fertilizer offsets may help counter costs of implementation but have been found to mainly benefit small farms (Ribaudo, Cattaneo, and Agapoff 2004). On the other hand, costs for abatement, plan design, and plan administration remain a disadvantage to small farms (McCann 2009). Other research supports the suggestion that NMPs are difficult to monitor or enforce. Anecdotal evidence of law evasion suggests that some farmers may actively mislead regulators (Perez 2011). Other studies have found that farmers may get a NMP but may not implement their plan on their entire operation, if at all (Shepard 2005; Genskow 2012). A lack of compliance, in combination with negligence and a lack of enforcement from regulatory authorities, has brought into question the ability of current regulations to reduce discharges (Centner and Newton 2008; Sneeringer 2016). A further question arises regarding the consistency of the information in NMPs. Inconsistencies in nutrient management are introduced when NMPs are developed by entities with different agendas. Studies have found that NMPs developed by private sector agents tend to have higher thresholds for the amount of nutrients that can be applied to the land than plans developed by public sector agents (Lawley, Lichtenberg, and Parker 2009; Perez 2011). Little research has been conducted to determine whether the adoption of NMPs improves environmental outcomes. The EPA conducted an ex ante study before the 2003 CAFO rules to predict improvements in environmental quality from the law, stemming largely from greater adoption of NMPs. The EPA estimated a 22% reduction in nutrient loadings from large and medium CAFOs from the updated law, based partly on assumptions about how producers with a NMP would behave (68 FR 7176-7274). As far as we are aware, there has been no ex post research showing how NMPs impact the end goal of environmental quality, or examining whether NMPs alter nutrient management behaviors. Data We use farm-level data from the 2004 and 2009 Agricultural Resource Management Surveys Phase III collected from hog operations in the United States to examine farm characteristics and nutrient management practices. The 2004 survey consists of 1,198 observations representing 40,940 operations; the 2009 survey consists of 1,208 observations representing 24,350 farms. Only farms with at least 25 hogs are surveyed, as anything less is considered to be raised for private consumption. In each year, the survey consists of data from 19 states, representing over 90% of hog production (Key et al. 2011; see appendix table 1 for a list of included states). The ARMS includes questions on our variable of interest—NMP adoption—as well as pertinent nutrient management behaviors, including nitrogen and phosphorus testing of manure, adjustments to feed for the purpose of altering manure nutrient content, and incorporation of manure at application. These are measures with the potential to be externally verified by an oversight authority; for example, farmers may report in the ARMS that they perform nutrient testing, but may have less incentive to lie about this, as testing labs can be questioned, or lab testing receipts or results can be requested. The ARMS also includes questions about potential confounding factors, including number of animals, crop acreage, the state in which the operation resides, whether the operation has an EQIP contract, whether the operation is farrow-to-finish or specialized for a specific portion of the hog life span, whether the operation is a contractor for an integrator, the operator’s years of farming experience, and whether the operator has a college degree. Using information on the sales and inventory of nursery pigs, feeder pigs, breeder hogs, and market hogs, we calculate the number of hog “animal units” (see the appendix for detail). “Animal units” is a method of standardizing across different animal weights and types; one animal unit is approximately 1,000 pounds of average live weight. From this we can characterize which operations have more than 1,000 hog animal units, our measure for whether an operation is a “large CAFO.” Prior literature notes the importance of enforcement in evaluating the efficacy of NMPs. Since we do not have a direct measure of enforcement (such as number of oversight visits from pertinent regulatory authorities), we proxy for enforcement with four variables. First, we control for large CAFO status. Second, we use the EPA’s 303d water quality reports to generate an indicator for whether a farm is in a county that has ever adopted a TMDL program for nutrients prior to the survey year. A TMDL is sometimes called a “pollution diet” and is implemented when traditional methods of pollution control fail to yield water quality goals. For livestock farmers, implementation generally means greater oversight of nutrient management practices in a watershed (e.g., Copeland 2012). Third, we include a measure for whether the operation has an Environmental Quality Incentives Program (EQIP) contract. EQIP contracts often require operators to obtain NMPs in order to access funds for more capital-intensive projects. Fourth, population land density at the county level is generated from the inter-censual county population estimates from the U.S. Census Bureau. Prior research has used population density to account for demands for environmental quality and restrictions in land application of manure due to odor complaints (Lyford and Hicks 2001; Ribaudo & Johansson 2007). In addition to evaluating the effects of NMPs on nutrient management practices, we also analyze whether having a NMP lowers the probability that a farm over-applies nutrients. The ARMS questionnaire does not directly collect information on nutrient application and uptake. Instead, we estimate these variables using ARMS questions and involved procedures partially developed from National Resource Conservation Service (NRCS) methods; see the appendices for details, as well as Kellogg et al. (2000) and Sneeringer (2016). We first estimate the total amount of manure nutrients generated on-farm and available for later application accounting for animal size, manure produced per animal pound, nutrients per unit of manure, nutrient loss through management or evaporation, and manure storage. We next estimate how much of these nutrients were applied on-farm by accounting for the amounts of manure that were either shipped off-farm or kept in storage and therefore not applied. We use total fertilizer expenditures along with data on crops planted and per unit fertilizer costs to estimate how much commercial fertilizer is applied. The manure and fertilizer applications provide us with an estimate of total nutrient applications. Finally, we use data on crop yields in 15 different commodities, as well as pasture to estimate the nutrient uptake capacity on the farm. Comparing the amount of nutrients applied to the uptake capacity, we can evaluate whether the operation applied excess nutrients. Notably, our measure of excess can be negative or positive; positive excess means more applications than uptake, while “negative excess” is the opposite. Summary Statistics Table 1 shows the summary statistics for the sample and compares farms with NMPs to those without. In the weighted samples, 30% of operations in 2004 and 55% of operations in 2009 had NMPs. These operations accounted for 61% and 82% of hog animal units in 2004 and 2009, respectively. On average, farms with NMPs had about 3.7 times more animal units in both years than those without, but only about 1.4 to 1.5 times the amount of crop acreage. Operations with NMPs were more likely to be specialized in a single phase of the hog life-cycle, have manure storage, and be contract operations. Operators with NMPs are, on average, slightly younger (52 versus 53), had slightly less experience and, in 2009, were more likely to have a college degree. Operations with NMPs were more likely to have EQIP contracts. In 2009, they were less likely to be in a county that ever had a TMDL. County-level population density is not statistically different between operations with and without NMPs in either year. Table 1 Operation-level Means of Relevant Independent Variables, by NMP Status 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Number of observations (unweighted) 1,196 541 655 1,283 810 473 Number of farms (weighted) 40,929 12,340 28,589 24,430 13,364 11,066 Proportion of farms 1.00 0.30 0.70 1.00 0.55 0.45 Proportion of hog animal units 1.00 0.61 0.39 1.00 0.82 0.18 Average number of hog animal units in inventory 294 598 163 11.08 545 814 219 13.04 (3,795) (3,683) (3,553) (4,180) (4,735) (2,121) Proportion with more than 1,000 hog animal units 0.07 0.16 0.03 7.52 0.16 0.24 0.05 10.41 (1.51) (1.77) (1.17) (1.59) (1.75) (1.09) Average total crop acreage 459 570 411 3.71 503 591 398 4.64 (3,998) (3,991) (3,965) (3,407) (3,685) (2,790) Proportion with no crop acreage 0.17 0.14 0.18 1.73 0.15 0.15 0.16 0.33 (2.18) (1.66) (2.52) (1.57) (1.44) (1.75) Proportion of operations that are farrow-to-finish 0.31 0.18 0.36 7.15 0.24 0.13 0.37 9.32 (2.70) (1.84) (3.17) (1.86) (1.37) (2.33) Proportion that are contract operations 0.28 0.53 0.17 13.53 0.49 0.67 0.28 14.66 (2.63) (2.39) (2.50) (2.18) (1.92) (2.17) Average operator experience (years) 25 24 25 2.21 28 27 29 2.06 (68) (57) (76) (56) (51) (63) Proportion of operators with college education 0.27 0.24 0.27 1.22 0.21 0.24 0.17 2.87 (2.58) (2.05) (2.95) (1.77) (1.73) (1.83) Proportion with an EQIP contract 0.02 0.04 0.003 4.52 0.04 0.07 0.02 4.93 (0.71) (0.98) (0.35) (0.89) (1.01) (0.59) Proportion with TMDL ever in county 0.16 0.15 0.16 0.43 0.53 0.50 0.57 2.51 (2.13) (1.71) (2.43) (2.18) (2.03) (2.40) Average population density in county 75 81 72 1.25 78 76 79 0.45 (786) (574) (924) (503) (474) (549) 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Number of observations (unweighted) 1,196 541 655 1,283 810 473 Number of farms (weighted) 40,929 12,340 28,589 24,430 13,364 11,066 Proportion of farms 1.00 0.30 0.70 1.00 0.55 0.45 Proportion of hog animal units 1.00 0.61 0.39 1.00 0.82 0.18 Average number of hog animal units in inventory 294 598 163 11.08 545 814 219 13.04 (3,795) (3,683) (3,553) (4,180) (4,735) (2,121) Proportion with more than 1,000 hog animal units 0.07 0.16 0.03 7.52 0.16 0.24 0.05 10.41 (1.51) (1.77) (1.17) (1.59) (1.75) (1.09) Average total crop acreage 459 570 411 3.71 503 591 398 4.64 (3,998) (3,991) (3,965) (3,407) (3,685) (2,790) Proportion with no crop acreage 0.17 0.14 0.18 1.73 0.15 0.15 0.16 0.33 (2.18) (1.66) (2.52) (1.57) (1.44) (1.75) Proportion of operations that are farrow-to-finish 0.31 0.18 0.36 7.15 0.24 0.13 0.37 9.32 (2.70) (1.84) (3.17) (1.86) (1.37) (2.33) Proportion that are contract operations 0.28 0.53 0.17 13.53 0.49 0.67 0.28 14.66 (2.63) (2.39) (2.50) (2.18) (1.92) (2.17) Average operator experience (years) 25 24 25 2.21 28 27 29 2.06 (68) (57) (76) (56) (51) (63) Proportion of operators with college education 0.27 0.24 0.27 1.22 0.21 0.24 0.17 2.87 (2.58) (2.05) (2.95) (1.77) (1.73) (1.83) Proportion with an EQIP contract 0.02 0.04 0.003 4.52 0.04 0.07 0.02 4.93 (0.71) (0.98) (0.35) (0.89) (1.01) (0.59) Proportion with TMDL ever in county 0.16 0.15 0.16 0.43 0.53 0.50 0.57 2.51 (2.13) (1.71) (2.43) (2.18) (2.03) (2.40) Average population density in county 75 81 72 1.25 78 76 79 0.45 (786) (574) (924) (503) (474) (549) Table 1 Operation-level Means of Relevant Independent Variables, by NMP Status 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Number of observations (unweighted) 1,196 541 655 1,283 810 473 Number of farms (weighted) 40,929 12,340 28,589 24,430 13,364 11,066 Proportion of farms 1.00 0.30 0.70 1.00 0.55 0.45 Proportion of hog animal units 1.00 0.61 0.39 1.00 0.82 0.18 Average number of hog animal units in inventory 294 598 163 11.08 545 814 219 13.04 (3,795) (3,683) (3,553) (4,180) (4,735) (2,121) Proportion with more than 1,000 hog animal units 0.07 0.16 0.03 7.52 0.16 0.24 0.05 10.41 (1.51) (1.77) (1.17) (1.59) (1.75) (1.09) Average total crop acreage 459 570 411 3.71 503 591 398 4.64 (3,998) (3,991) (3,965) (3,407) (3,685) (2,790) Proportion with no crop acreage 0.17 0.14 0.18 1.73 0.15 0.15 0.16 0.33 (2.18) (1.66) (2.52) (1.57) (1.44) (1.75) Proportion of operations that are farrow-to-finish 0.31 0.18 0.36 7.15 0.24 0.13 0.37 9.32 (2.70) (1.84) (3.17) (1.86) (1.37) (2.33) Proportion that are contract operations 0.28 0.53 0.17 13.53 0.49 0.67 0.28 14.66 (2.63) (2.39) (2.50) (2.18) (1.92) (2.17) Average operator experience (years) 25 24 25 2.21 28 27 29 2.06 (68) (57) (76) (56) (51) (63) Proportion of operators with college education 0.27 0.24 0.27 1.22 0.21 0.24 0.17 2.87 (2.58) (2.05) (2.95) (1.77) (1.73) (1.83) Proportion with an EQIP contract 0.02 0.04 0.003 4.52 0.04 0.07 0.02 4.93 (0.71) (0.98) (0.35) (0.89) (1.01) (0.59) Proportion with TMDL ever in county 0.16 0.15 0.16 0.43 0.53 0.50 0.57 2.51 (2.13) (1.71) (2.43) (2.18) (2.03) (2.40) Average population density in county 75 81 72 1.25 78 76 79 0.45 (786) (574) (924) (503) (474) (549) 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Number of observations (unweighted) 1,196 541 655 1,283 810 473 Number of farms (weighted) 40,929 12,340 28,589 24,430 13,364 11,066 Proportion of farms 1.00 0.30 0.70 1.00 0.55 0.45 Proportion of hog animal units 1.00 0.61 0.39 1.00 0.82 0.18 Average number of hog animal units in inventory 294 598 163 11.08 545 814 219 13.04 (3,795) (3,683) (3,553) (4,180) (4,735) (2,121) Proportion with more than 1,000 hog animal units 0.07 0.16 0.03 7.52 0.16 0.24 0.05 10.41 (1.51) (1.77) (1.17) (1.59) (1.75) (1.09) Average total crop acreage 459 570 411 3.71 503 591 398 4.64 (3,998) (3,991) (3,965) (3,407) (3,685) (2,790) Proportion with no crop acreage 0.17 0.14 0.18 1.73 0.15 0.15 0.16 0.33 (2.18) (1.66) (2.52) (1.57) (1.44) (1.75) Proportion of operations that are farrow-to-finish 0.31 0.18 0.36 7.15 0.24 0.13 0.37 9.32 (2.70) (1.84) (3.17) (1.86) (1.37) (2.33) Proportion that are contract operations 0.28 0.53 0.17 13.53 0.49 0.67 0.28 14.66 (2.63) (2.39) (2.50) (2.18) (1.92) (2.17) Average operator experience (years) 25 24 25 2.21 28 27 29 2.06 (68) (57) (76) (56) (51) (63) Proportion of operators with college education 0.27 0.24 0.27 1.22 0.21 0.24 0.17 2.87 (2.58) (2.05) (2.95) (1.77) (1.73) (1.83) Proportion with an EQIP contract 0.02 0.04 0.003 4.52 0.04 0.07 0.02 4.93 (0.71) (0.98) (0.35) (0.89) (1.01) (0.59) Proportion with TMDL ever in county 0.16 0.15 0.16 0.43 0.53 0.50 0.57 2.51 (2.13) (1.71) (2.43) (2.18) (2.03) (2.40) Average population density in county 75 81 72 1.25 78 76 79 0.45 (786) (574) (924) (503) (474) (549) Farms with NMPs are more likely to engage in certain documentable nutrient management practices (table 2). Operations with NMPs are much more likely to test manure for nutrient content and to adjust feed levels to manage it. When they apply manure, farms with NMPs are more likely to incorporate it (a practice to reduce run-off); however, this difference is only statistically significant in 2004, not 2009. Table 2 Operation-level Means of Nutrient Management Behavior Variables, by Year and NMP Status 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Proportion testing manure for nitrogen content 0.24 0.56 0.10 18.88 0.38 0.58 0.13 19.36 (2.50) (2.37) (1.99) (2.11) (2.01) (1.62) Proportion using microbial phytase to reduce nutrient content of manure 0.13 0.25 0.08 7.69 0.24 0.34 0.11 10.29 (2.0) (2.07) (1.84) (1.86) (1.93) (1.54) Proportion adjusting schedule of feed formulations to adjust the nutrient content of manure 0.14 0.28 0.08 8.82 0.22 0.33 0.10 10.77 (2.04) (2.14) (1.82) (1.82) (1.91) (1.44) Proportion using other feed additives or formula adjustments to adjust the nutrient content of manure 0.05 0.14 0.01 8.48 0.08 0.13 0.02 7.88 (1.27) (1.66) (0.65) (1.20) (1.38) (0.73) Average proportion manure applied 0.78 0.80 0.77 1.33 0.74 0.75 0.72 1.48 (2.37) (1.79) (2.76) (1.85) (1.65) (2.14) Proportion applying any manure 0.80 0.85 0.78 3.40 0.77 0.80 0.73 2.83 (2.34) (1.69) (2.75) (1.83) (1.61) (2.14) Of those applying manure Proportion incorporating manure at application* 0.37 0.42 0.35 2.48 0.50 0.52 0.48 1.05 (2.82) (2.36) (3.17) (2.16) (1.98) (2.49) 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Proportion testing manure for nitrogen content 0.24 0.56 0.10 18.88 0.38 0.58 0.13 19.36 (2.50) (2.37) (1.99) (2.11) (2.01) (1.62) Proportion using microbial phytase to reduce nutrient content of manure 0.13 0.25 0.08 7.69 0.24 0.34 0.11 10.29 (2.0) (2.07) (1.84) (1.86) (1.93) (1.54) Proportion adjusting schedule of feed formulations to adjust the nutrient content of manure 0.14 0.28 0.08 8.82 0.22 0.33 0.10 10.77 (2.04) (2.14) (1.82) (1.82) (1.91) (1.44) Proportion using other feed additives or formula adjustments to adjust the nutrient content of manure 0.05 0.14 0.01 8.48 0.08 0.13 0.02 7.88 (1.27) (1.66) (0.65) (1.20) (1.38) (0.73) Average proportion manure applied 0.78 0.80 0.77 1.33 0.74 0.75 0.72 1.48 (2.37) (1.79) (2.76) (1.85) (1.65) (2.14) Proportion applying any manure 0.80 0.85 0.78 3.40 0.77 0.80 0.73 2.83 (2.34) (1.69) (2.75) (1.83) (1.61) (2.14) Of those applying manure Proportion incorporating manure at application* 0.37 0.42 0.35 2.48 0.50 0.52 0.48 1.05 (2.82) (2.36) (3.17) (2.16) (1.98) (2.49) Note: Asterisk * indicates that manure at application refers to manure applied in solid form and incorporated at application, or manure applied in slurry form that is injected or surface applied and incorporated. Table 2 Operation-level Means of Nutrient Management Behavior Variables, by Year and NMP Status 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Proportion testing manure for nitrogen content 0.24 0.56 0.10 18.88 0.38 0.58 0.13 19.36 (2.50) (2.37) (1.99) (2.11) (2.01) (1.62) Proportion using microbial phytase to reduce nutrient content of manure 0.13 0.25 0.08 7.69 0.24 0.34 0.11 10.29 (2.0) (2.07) (1.84) (1.86) (1.93) (1.54) Proportion adjusting schedule of feed formulations to adjust the nutrient content of manure 0.14 0.28 0.08 8.82 0.22 0.33 0.10 10.77 (2.04) (2.14) (1.82) (1.82) (1.91) (1.44) Proportion using other feed additives or formula adjustments to adjust the nutrient content of manure 0.05 0.14 0.01 8.48 0.08 0.13 0.02 7.88 (1.27) (1.66) (0.65) (1.20) (1.38) (0.73) Average proportion manure applied 0.78 0.80 0.77 1.33 0.74 0.75 0.72 1.48 (2.37) (1.79) (2.76) (1.85) (1.65) (2.14) Proportion applying any manure 0.80 0.85 0.78 3.40 0.77 0.80 0.73 2.83 (2.34) (1.69) (2.75) (1.83) (1.61) (2.14) Of those applying manure Proportion incorporating manure at application* 0.37 0.42 0.35 2.48 0.50 0.52 0.48 1.05 (2.82) (2.36) (3.17) (2.16) (1.98) (2.49) 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Proportion testing manure for nitrogen content 0.24 0.56 0.10 18.88 0.38 0.58 0.13 19.36 (2.50) (2.37) (1.99) (2.11) (2.01) (1.62) Proportion using microbial phytase to reduce nutrient content of manure 0.13 0.25 0.08 7.69 0.24 0.34 0.11 10.29 (2.0) (2.07) (1.84) (1.86) (1.93) (1.54) Proportion adjusting schedule of feed formulations to adjust the nutrient content of manure 0.14 0.28 0.08 8.82 0.22 0.33 0.10 10.77 (2.04) (2.14) (1.82) (1.82) (1.91) (1.44) Proportion using other feed additives or formula adjustments to adjust the nutrient content of manure 0.05 0.14 0.01 8.48 0.08 0.13 0.02 7.88 (1.27) (1.66) (0.65) (1.20) (1.38) (0.73) Average proportion manure applied 0.78 0.80 0.77 1.33 0.74 0.75 0.72 1.48 (2.37) (1.79) (2.76) (1.85) (1.65) (2.14) Proportion applying any manure 0.80 0.85 0.78 3.40 0.77 0.80 0.73 2.83 (2.34) (1.69) (2.75) (1.83) (1.61) (2.14) Of those applying manure Proportion incorporating manure at application* 0.37 0.42 0.35 2.48 0.50 0.52 0.48 1.05 (2.82) (2.36) (3.17) (2.16) (1.98) (2.49) Note: Asterisk * indicates that manure at application refers to manure applied in solid form and incorporated at application, or manure applied in slurry form that is injected or surface applied and incorporated. Turning to our estimates of nutrient application and uptake, we see a somewhat conflicting story compared to what we expect from the practice outcomes. Despite the higher ratio of animal units to crop acres at operations with NMPs, they apply an equal percentage of their manure on-farm (Table 2). The rest of the manure is either stored or shipped off-farm. As expected by their greater number of animal units, operations with NMPs generate more manure nutrients than operations without NMPs—4.6 times as much in 2004, and 3.8 times as much as in 2009 (see table 3). Accounting for both recoverable manure nutrients and estimated commercial fertilizer nutrients, operations with NMPs, on average, apply more than twice the nutrients over their entire operations than those without. Table 3 Operation-level Means of Nutrient Generation, Application, and Uptake Variables, by Year and NMP Status Year 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Recoverable manure nitrogen generated (lbs.) 10,398 22,879 5,011 11.22 19,254 28,908 7,595 11.24 (144,225) (161,220) (111,082) (171,391) (196,318) (88,890) Recoverable manure nitrogen applied (lbs.) 7,971 17,778 3,729 9.44 14,080 21,420 5,234 9.77 (132,058) (151,003) (102,064) (149,134) (173,762) (72,804) Fertilizer nitrogen applied (lbs.) 26,280 35,145 22,454 4.09 22,700 27,953 16,355 4.66 (285,973) (292,167) (277,174) (204,952) (222,454) (165,975) Fertilizer and manure nitrogen applied (lbs.) 34,240 52,923 26,175 7.13 36,759 49,320 21,588 8.69 (344,499) (362,032) (314,967) (275,468) (307,122) (185,937) Nitrogen uptake on crops and pasture (lbs.) 86,752 116,796 73,784 5.15 98,102 116,829 75,484 5.10 (749,400) (815,040) (673,264) (653,211) (679,469) (587,952) Excess nitrogen (applied minus uptake, lbs.) −52,304 −63,873 −47,301 2.77 −61,389 −67,592 −53,913 2.19 (529,417) (588,678) (471,388) (497,524) (521,105) (452,213) Proportion applying more fertilizer and manure nitrogen than uptake 0.12 0.23 0.07 7.73 0.13 0.18 0.06 6.57 (1.89) (2.01) (1.69) (1.44) (1.55) (1.17) For operations with non-zero crop acreage Nitrogen fertilizer and manure applied per crop acre (lbs./acre) 118 200 81 4.47 128 181 63 6.57 (1,933) (2,652) (925) (1,506) (1,801) (619) Nitrogen uptake per crop acre (lbs./acre) 180 182 179 0.64 184 192 174 3.59 (474) (413) (519) (349) (302) (413) Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre −61 18 −97 4.30 −56 −11 −111 5.25 (1,971) (2,653) (1,058) (1,574) (1,872) (742) Proportion applying more fertilizer and manure nitrogen than uptake per acre 0.12 0.40 0.18 7.62 0.13 0.36 0.11 10.30 (1.92) (0.49) (0.39) (1.45) (0.48) (0.31) Year 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Recoverable manure nitrogen generated (lbs.) 10,398 22,879 5,011 11.22 19,254 28,908 7,595 11.24 (144,225) (161,220) (111,082) (171,391) (196,318) (88,890) Recoverable manure nitrogen applied (lbs.) 7,971 17,778 3,729 9.44 14,080 21,420 5,234 9.77 (132,058) (151,003) (102,064) (149,134) (173,762) (72,804) Fertilizer nitrogen applied (lbs.) 26,280 35,145 22,454 4.09 22,700 27,953 16,355 4.66 (285,973) (292,167) (277,174) (204,952) (222,454) (165,975) Fertilizer and manure nitrogen applied (lbs.) 34,240 52,923 26,175 7.13 36,759 49,320 21,588 8.69 (344,499) (362,032) (314,967) (275,468) (307,122) (185,937) Nitrogen uptake on crops and pasture (lbs.) 86,752 116,796 73,784 5.15 98,102 116,829 75,484 5.10 (749,400) (815,040) (673,264) (653,211) (679,469) (587,952) Excess nitrogen (applied minus uptake, lbs.) −52,304 −63,873 −47,301 2.77 −61,389 −67,592 −53,913 2.19 (529,417) (588,678) (471,388) (497,524) (521,105) (452,213) Proportion applying more fertilizer and manure nitrogen than uptake 0.12 0.23 0.07 7.73 0.13 0.18 0.06 6.57 (1.89) (2.01) (1.69) (1.44) (1.55) (1.17) For operations with non-zero crop acreage Nitrogen fertilizer and manure applied per crop acre (lbs./acre) 118 200 81 4.47 128 181 63 6.57 (1,933) (2,652) (925) (1,506) (1,801) (619) Nitrogen uptake per crop acre (lbs./acre) 180 182 179 0.64 184 192 174 3.59 (474) (413) (519) (349) (302) (413) Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre −61 18 −97 4.30 −56 −11 −111 5.25 (1,971) (2,653) (1,058) (1,574) (1,872) (742) Proportion applying more fertilizer and manure nitrogen than uptake per acre 0.12 0.40 0.18 7.62 0.13 0.36 0.11 10.30 (1.92) (0.49) (0.39) (1.45) (0.48) (0.31) Table 3 Operation-level Means of Nutrient Generation, Application, and Uptake Variables, by Year and NMP Status Year 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Recoverable manure nitrogen generated (lbs.) 10,398 22,879 5,011 11.22 19,254 28,908 7,595 11.24 (144,225) (161,220) (111,082) (171,391) (196,318) (88,890) Recoverable manure nitrogen applied (lbs.) 7,971 17,778 3,729 9.44 14,080 21,420 5,234 9.77 (132,058) (151,003) (102,064) (149,134) (173,762) (72,804) Fertilizer nitrogen applied (lbs.) 26,280 35,145 22,454 4.09 22,700 27,953 16,355 4.66 (285,973) (292,167) (277,174) (204,952) (222,454) (165,975) Fertilizer and manure nitrogen applied (lbs.) 34,240 52,923 26,175 7.13 36,759 49,320 21,588 8.69 (344,499) (362,032) (314,967) (275,468) (307,122) (185,937) Nitrogen uptake on crops and pasture (lbs.) 86,752 116,796 73,784 5.15 98,102 116,829 75,484 5.10 (749,400) (815,040) (673,264) (653,211) (679,469) (587,952) Excess nitrogen (applied minus uptake, lbs.) −52,304 −63,873 −47,301 2.77 −61,389 −67,592 −53,913 2.19 (529,417) (588,678) (471,388) (497,524) (521,105) (452,213) Proportion applying more fertilizer and manure nitrogen than uptake 0.12 0.23 0.07 7.73 0.13 0.18 0.06 6.57 (1.89) (2.01) (1.69) (1.44) (1.55) (1.17) For operations with non-zero crop acreage Nitrogen fertilizer and manure applied per crop acre (lbs./acre) 118 200 81 4.47 128 181 63 6.57 (1,933) (2,652) (925) (1,506) (1,801) (619) Nitrogen uptake per crop acre (lbs./acre) 180 182 179 0.64 184 192 174 3.59 (474) (413) (519) (349) (302) (413) Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre −61 18 −97 4.30 −56 −11 −111 5.25 (1,971) (2,653) (1,058) (1,574) (1,872) (742) Proportion applying more fertilizer and manure nitrogen than uptake per acre 0.12 0.40 0.18 7.62 0.13 0.36 0.11 10.30 (1.92) (0.49) (0.39) (1.45) (0.48) (0.31) Year 2004 2009 All Farms with NMP Farms without NMP t-statistic for difference between means All Farms with NMP Farms without NMP t-statistic for difference between means Recoverable manure nitrogen generated (lbs.) 10,398 22,879 5,011 11.22 19,254 28,908 7,595 11.24 (144,225) (161,220) (111,082) (171,391) (196,318) (88,890) Recoverable manure nitrogen applied (lbs.) 7,971 17,778 3,729 9.44 14,080 21,420 5,234 9.77 (132,058) (151,003) (102,064) (149,134) (173,762) (72,804) Fertilizer nitrogen applied (lbs.) 26,280 35,145 22,454 4.09 22,700 27,953 16,355 4.66 (285,973) (292,167) (277,174) (204,952) (222,454) (165,975) Fertilizer and manure nitrogen applied (lbs.) 34,240 52,923 26,175 7.13 36,759 49,320 21,588 8.69 (344,499) (362,032) (314,967) (275,468) (307,122) (185,937) Nitrogen uptake on crops and pasture (lbs.) 86,752 116,796 73,784 5.15 98,102 116,829 75,484 5.10 (749,400) (815,040) (673,264) (653,211) (679,469) (587,952) Excess nitrogen (applied minus uptake, lbs.) −52,304 −63,873 −47,301 2.77 −61,389 −67,592 −53,913 2.19 (529,417) (588,678) (471,388) (497,524) (521,105) (452,213) Proportion applying more fertilizer and manure nitrogen than uptake 0.12 0.23 0.07 7.73 0.13 0.18 0.06 6.57 (1.89) (2.01) (1.69) (1.44) (1.55) (1.17) For operations with non-zero crop acreage Nitrogen fertilizer and manure applied per crop acre (lbs./acre) 118 200 81 4.47 128 181 63 6.57 (1,933) (2,652) (925) (1,506) (1,801) (619) Nitrogen uptake per crop acre (lbs./acre) 180 182 179 0.64 184 192 174 3.59 (474) (413) (519) (349) (302) (413) Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre −61 18 −97 4.30 −56 −11 −111 5.25 (1,971) (2,653) (1,058) (1,574) (1,872) (742) Proportion applying more fertilizer and manure nitrogen than uptake per acre 0.12 0.40 0.18 7.62 0.13 0.36 0.11 10.30 (1.92) (0.49) (0.39) (1.45) (0.48) (0.31) These differences in application might not yield excess applications if the amount of crop nutrient uptake is comparably higher on operations with NMPs. While operations with NMPs do have more nutrient uptake capacity, the ratio between operations with NMPs and those without is about 1.6 in both years. This is lower than the 2 to 2.3 times as many nutrients applied. Like the comparisons of animal units to crop acreage, this again suggests that operations with NMPs may be more likely to over-apply nutrients. The two measures of excess applications (application of manure and fertilizer nutrients minus uptake) tell slightly different stories. Operations with NMPs have, on average, a higher amount of surplus uptake capacity (a more highly negative excess) than operations without NMPs. This is true for both 2004 and 2009, and suggests that on average, operations with NMPs apply fewer nutrients relative to their uptake capacities than operations without NMPs. However, operations with NMPs are approximately 3 times more likely to apply excess than those without (23% of operations with NMPs versus 7% of operations without in 2004; 18% versus 6%, respectively, in 2009). The different qualitative conclusions to be drawn from these two findings arise from a longer left-tailed distribution of excess for operations with NMPs compared to those without. While a greater percentage of NMP operations have excess nitrogen applications compared to non-NMP operations, NMP operations also have much lower values of excess (more negative) than those without NMPs. Table 4 shows the percentage of operations within each of six categories of excess nitrogen; while operations with NMPs are more likely than those without NMPs to be in the “most negative” category (i.e., they have the more uptake than applications), they are also more likely to have positive excess. Table 4 Distribution of Nitrogen Excess, by Year and NMP Status 2004 2009 Farms with NMP Farms without NMP Farms with NMP Farms without NMP Total nitrogen excess (applied minus uptake), lbs. Less than −100,000 26% 17% 29% 19% −100,000 to −50,000 14% 12% 14% 15% −50,000 to 0 37% 64% 39% 59% 1 to 50,000 19% 7% 14% 5% 50,001 to 100,000 3% 0.2% 3% 1% More than 100,000 1% 0.1% 1% 0.1% Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre More than −100 61% 60% 65% 70% −100 to −50 14% 22% 12% 14% −50 to 0 5% 13% 7% 12% 1 to 50 4% 2% 3% 2% 50−100 3% 1% 1% 0.3% More than 100 13% 2% 11% 2% 2004 2009 Farms with NMP Farms without NMP Farms with NMP Farms without NMP Total nitrogen excess (applied minus uptake), lbs. Less than −100,000 26% 17% 29% 19% −100,000 to −50,000 14% 12% 14% 15% −50,000 to 0 37% 64% 39% 59% 1 to 50,000 19% 7% 14% 5% 50,001 to 100,000 3% 0.2% 3% 1% More than 100,000 1% 0.1% 1% 0.1% Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre More than −100 61% 60% 65% 70% −100 to −50 14% 22% 12% 14% −50 to 0 5% 13% 7% 12% 1 to 50 4% 2% 3% 2% 50−100 3% 1% 1% 0.3% More than 100 13% 2% 11% 2% Table 4 Distribution of Nitrogen Excess, by Year and NMP Status 2004 2009 Farms with NMP Farms without NMP Farms with NMP Farms without NMP Total nitrogen excess (applied minus uptake), lbs. Less than −100,000 26% 17% 29% 19% −100,000 to −50,000 14% 12% 14% 15% −50,000 to 0 37% 64% 39% 59% 1 to 50,000 19% 7% 14% 5% 50,001 to 100,000 3% 0.2% 3% 1% More than 100,000 1% 0.1% 1% 0.1% Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre More than −100 61% 60% 65% 70% −100 to −50 14% 22% 12% 14% −50 to 0 5% 13% 7% 12% 1 to 50 4% 2% 3% 2% 50−100 3% 1% 1% 0.3% More than 100 13% 2% 11% 2% 2004 2009 Farms with NMP Farms without NMP Farms with NMP Farms without NMP Total nitrogen excess (applied minus uptake), lbs. Less than −100,000 26% 17% 29% 19% −100,000 to −50,000 14% 12% 14% 15% −50,000 to 0 37% 64% 39% 59% 1 to 50,000 19% 7% 14% 5% 50,001 to 100,000 3% 0.2% 3% 1% More than 100,000 1% 0.1% 1% 0.1% Nitrogen excess per crop acre (excess is applied minus uptake), lbs./acre More than −100 61% 60% 65% 70% −100 to −50 14% 22% 12% 14% −50 to 0 5% 13% 7% 12% 1 to 50 4% 2% 3% 2% 50−100 3% 1% 1% 0.3% More than 100 13% 2% 11% 2% For operations with any crop acreage, we can also examine the amount of nitrogen applied per acre, the uptake per acre, and the “excess” per acre. While looking at total amounts of nutrients applied versus uptake might provide us with an indication of nutrient management involving changes in number of animals or amount of acreage (on the extensive margin), per-acre rates might provide an indication of nutrient management without adjusting farm size (on the intensive margin). The bottom panel of table 3 shows per-acre measures; operations with NMPs have higher average per-acre applications of nitrogen in both years. Uptake per acre is more similar across NMP status; “excess” per acre is also likely to be greater for NMP operations in both years. The bottom panel of table 4 shows the distributions of the excess-per-acre variable; this suggests that even controlling for acreage, operations with NMPs are more likely to apply excess per acre, and more excess per acre, than their non-NMP counterparts. These summary statistics suggest that operations with NMPs are more likely to perform certain documentable nutrient management behaviors (like testing, feed adjustments, and manure incorporation at application), but also may be more likely to apply excess nutrients. However, these cross-tabulations do not adjust for potential confounders that may influence both management behaviors and NMP adoption. We therefore address these as well as other empirical issues using econometric strategies. Empirical Strategy Two fundamental issues arise when trying to estimate the impact of NMP adoption on nutrient management behaviors (the outcomes). First, other factors may be associated with both NMP adoption and the outcomes. For example, the summary statistics reveal that operations with more hogs are more likely to have NMPs. Larger operations may also manage their manure differently from smaller operations based on size-related factors, not the NMP. Controlling for these observable variables that are correlated with both NMP status and the outcome can reduce the bias in the estimated effect of NMP status on the outcomes. A second factor that may bias estimated effects between NMP status and the outcomes is selection. As previously described, NMP adoption can be voluntarily or required. We are not able to identify in the ARMS data which farms have been required to adopt NMPs and which have done so voluntarily. The concern is that operations may voluntarily adopt NMPs because they already have good nutrient management protocols and it is easy to do so, or because they have worse behaviors and adopt an NMP to signal better behavior. Thus, the outcomes are influencing NMP adoption, rather than the reverse. This form of endogeneity can bias the estimated effect of the NMP on the outcomes. To confront both of these issues we employ an endogenous treatment effects model. The model assumes a joint normal distribution between the errors of the selection equation (the decision to obtain a NMP) and the outcome regression (the measure of nutrient management). This approach addresses potential unobserved correlation between the operator’s decision to adopt a NMP and the behavior (outcome) variable, thus reducing bias in the estimated impact of the NMP on the nutrient management outcomes. We estimate a linear model for the behavior with a constrained normal distribution to model the deviation from the assumption of conditional independence of NMP status using maximum likelihood. Begin with the equation explaining the management outcome yi: yi=α+δ1NMPi+Xi'β+ρs+ei. Here, i indexes operation while s indexes state; NMP is a binary variable indicating whether the operation has an NMP or not; δ1 is the coefficient of interest; X is a vector of independent variables, including size, operator characteristics, and farm characteristics.4 The summary statistics revealed that observable characteristics of operations varied across NMP status; these factors may be correlated with adoption on an NMP as well as the outcome variable, hence, we control for them to minimize bias in the estimated effect of NMPs; ρ is a vector of dummy variables (with coefficients) for state. These state fixed effects control for factors that affect all farms within the state that are correlated with both NMP adoption and the outcome variable.5 Notably, states may have different levels of regulatory stringency and environmental attitudes, impacting both adoption of NMP and the outcome variables.6 An empirical concern is potential endogeneity of the treatment variable. Including the variables in X can control for the observable influence of these variables on both the treatment and the outcomes, and the state fixed effects can control for the influence of unobservable state-level confounders. However, unobservable factors not captured in the state fixed effects might be correlated with NMP status and the outcome variables. Thus, the “treatment”—NMP status—is assumed to be a function of a latent variable NMPi*: NMPi*=Ziγ+θs+ui where the decision to obtain an NMP is made according to NMPi=1, NMPi*>00, otherwise. Here, Zi is a vector of operator, operation, and regional characteristics thought to influence NMP status, and θs refer to state fixed effects.7 The observable variables that serve as control include, most notably, the size variable for number of hogs; the state fixed effects here control for state-level variability in NMP adoption requirements. Unobservable factors may cause the error terms for the behavior regression (equation 1) and that for the treatment regression (equation 2) to be correlated, resulting in biased parameter estimates in our equation of interest (1). To address this issue, we assume a joint normal distribution for the errors and estimate the two equations together using maximum likelihood; see Maddala (1983) and StataCorp (2015) for the error matrix and log likelihood functions. In both the NMP adoption and the nutrient management regressions, we consider the size of the operation—defined in terms of animal units—to be exogenous. However, prior research suggests that some operations that are close to a regulatory threshold may adjust their size to remain just below the threshold (Sneeringer and Key 2011). We do not address this type of size adjustment here for several reasons. First, our sample size is too small to detect such adjustments; detection requires a large number of observations on either side of the regulatory size cut-off, which the ARMS does not provide. Second, we are examining the entire distribution of animal sizes, rather than just the small window of sizes around the threshold (as was done in prior research). Hence, the impact of any farm size changes near the threshold is likely to be limited in the overall results. Unlike number of animal units, we do not treat total crop acreage as exogenous. While we control for whether or not an operation has any crop acreage, and whether or not it has above the median value of crop acreage, our prior is that operations may abide by NMPs by adjusting the number of cultivated acres. Adjustments in acreage will affect the per-acre nutrient application and uptake rates. We analyze two types of outcome variables, ( y): (a) indicator variables for nutrient testing, feed adjustment for nutrient content of manure, manure incorporation at application, and excess application; and (b) continuous variables for amounts of nutrients applied, nutrient uptake capacity, amount of “excess” nutrients applied (which can be a negative number), applications per acre, uptake per acre, and excess per acre. For the continuous behavior variables, we model equation (1) as a linear regression. The choice of regression model to use for the dichotomous dependent variables is more complicated. A logit or probit is often preferred to a linear probability model (LPM) because the predicted probabilities are bounded by zero and one, and the nonlinear functional form may provide a less-biased estimate of the effect of the relevant variable, depending on the “true” data-generating process. Several complications arise in addressing the endogeneity of the treatment variable that leads us to use the LPM.8 The LPM is often used instead of the logit/probit model because the coefficients are much easier to interpret, particularly when interaction terms are included (Ai and Norton 2003; Norton, Wang, and Ai 2004). In addition, the LPM usually provides very similar estimates of the marginal effects to logits/probits. The non-linear functional form of the probit/logit is not necessarily better, particularly when there are supplementary concerns (like endogenous regressors).9 On the other hand, there are multiple concerns with the LPM. The LPM can generate heteroscedastic standard errors, but this can be surmounted by using robust standard errors. The LPM may yield predicted probabilities less than zero or greater than one, but this is only a problem if one is attempting to forecast probabilities; we are not trying to do this, hence this is not a concern here. As a check against the potential mis-specification bias in the marginal effect in the LPM, in the appendix we show results of the LPM of equation (1) without accounting for the endogeneity of the treatment variable as well as estimated marginal effects for probit regression (again, without accounting for the potential endogeneity of NMP status). If these estimates are fairly similar, we can be less concerned about potential bias generated by modeling the binary behavior variables as linear functions. We estimate results for each survey year separately; this amounts to a model that is fully-interacted by year and allows all parameters to vary between years. We do not estimate panel data models because there are only 138 observations that appear in both survey years, which strongly limits our ability to estimate panel models with precision. Results of Econometric Estimation Table 5 reports estimation results of the correlation between having an NMP and recorded nutrient management practices, controlling for potential confounders and potential endogeneity of NMP status. All models include fixed effects for state, and standard errors clustered by state. Clustering by state accounts for possible correlation in model errors for individual operations in the same state. The top panel shows results for 2004 while the bottom panel shows results for 2009.10 Table 5 Endogenous Treatment Effects Regression Results, 2004 and 2009 (maximum likelihood) Test manure for nitrogen content Added phytase to feed Adjusted schedule of feed formulations Used other feed additives or formula adjustments Incorporated manure at applicationb 2004 NMP 0.27*** 0.16*** 0.17*** 0.13*** 0.15 (0.03) (0.05) (0.04) (0.04) (0.22) ln(Hog animal units) 0.07*** 0.06*** 0.05* 0.02** 0.03 (0.01) (0.02) (0.03) (0.01) (0.04) Large CAFO −0.06 −0.00 −0.04 −0.01 −0.11 (0.06) (0.08) (0.15) (0.06) (0.07) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,195 967 Prob > chi2 0.05 0.04 0.75 0.02 0.58 2009 NMP 0.57 0.20*** 0.32* 0.14*** 0.07 (0.63) (0.08) (0.17) (0.05) (0.08) ln(Hog animal units) 0.06 0.07*** 0.02 0.01*** 0.00 (0.08) (0.01) (0.02) (0.00) (0.02) Large CAFO 0.03 −0.05 −0.04 −0.00 0.04 (0.06) (0.07) (0.06) (0.03) (0.04) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,282 1,017 Prob > chi2 0.56 0.10 0.16 0.15 0.28 Test manure for nitrogen content Added phytase to feed Adjusted schedule of feed formulations Used other feed additives or formula adjustments Incorporated manure at applicationb 2004 NMP 0.27*** 0.16*** 0.17*** 0.13*** 0.15 (0.03) (0.05) (0.04) (0.04) (0.22) ln(Hog animal units) 0.07*** 0.06*** 0.05* 0.02** 0.03 (0.01) (0.02) (0.03) (0.01) (0.04) Large CAFO −0.06 −0.00 −0.04 −0.01 −0.11 (0.06) (0.08) (0.15) (0.06) (0.07) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,195 967 Prob > chi2 0.05 0.04 0.75 0.02 0.58 2009 NMP 0.57 0.20*** 0.32* 0.14*** 0.07 (0.63) (0.08) (0.17) (0.05) (0.08) ln(Hog animal units) 0.06 0.07*** 0.02 0.01*** 0.00 (0.08) (0.01) (0.02) (0.00) (0.02) Large CAFO 0.03 −0.05 −0.04 −0.00 0.04 (0.06) (0.07) (0.06) (0.03) (0.04) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,282 1,017 Prob > chi2 0.56 0.10 0.16 0.15 0.28 Note: Superscript a indicates that other controls include whether the operation has any crop acreage, whether the operation is in the top half of crop acreage, whether the operation is farrow to finish, whether the operation has an EQIP contract, whether there has ever been a TMDL in the county before the survey year, whether the operation is a contractor, the operator's years of experience, whether the operator has a college degree, and the population density of the county. Superscript b indicates the sample consists of just those operations that applied manure in any form. Superscript C indicates a sample of just those operations with any crop acreage. These regressions do not include the independent variable of whether the operation had any crop acreage. Results of 24 regressions are shown. All outcome variables are indicator (0/1) variables. Asterisk * refers to confidence at the 10% level, ** refer to confidence at the 5% level, and *** refer to confidence at the 1% level. All regressions include state-level fixed effects. Prob > chi2 refers to a likelihood ratio test of whether we can reject the null hypothesis of no correlation between the treatment errors and the outcome errors. A probability of the chi-squared value less than 0.05 suggests we can reject the null hypothesis of no selection with 95% certainty. Table 5 Endogenous Treatment Effects Regression Results, 2004 and 2009 (maximum likelihood) Test manure for nitrogen content Added phytase to feed Adjusted schedule of feed formulations Used other feed additives or formula adjustments Incorporated manure at applicationb 2004 NMP 0.27*** 0.16*** 0.17*** 0.13*** 0.15 (0.03) (0.05) (0.04) (0.04) (0.22) ln(Hog animal units) 0.07*** 0.06*** 0.05* 0.02** 0.03 (0.01) (0.02) (0.03) (0.01) (0.04) Large CAFO −0.06 −0.00 −0.04 −0.01 −0.11 (0.06) (0.08) (0.15) (0.06) (0.07) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,195 967 Prob > chi2 0.05 0.04 0.75 0.02 0.58 2009 NMP 0.57 0.20*** 0.32* 0.14*** 0.07 (0.63) (0.08) (0.17) (0.05) (0.08) ln(Hog animal units) 0.06 0.07*** 0.02 0.01*** 0.00 (0.08) (0.01) (0.02) (0.00) (0.02) Large CAFO 0.03 −0.05 −0.04 −0.00 0.04 (0.06) (0.07) (0.06) (0.03) (0.04) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,282 1,017 Prob > chi2 0.56 0.10 0.16 0.15 0.28 Test manure for nitrogen content Added phytase to feed Adjusted schedule of feed formulations Used other feed additives or formula adjustments Incorporated manure at applicationb 2004 NMP 0.27*** 0.16*** 0.17*** 0.13*** 0.15 (0.03) (0.05) (0.04) (0.04) (0.22) ln(Hog animal units) 0.07*** 0.06*** 0.05* 0.02** 0.03 (0.01) (0.02) (0.03) (0.01) (0.04) Large CAFO −0.06 −0.00 −0.04 −0.01 −0.11 (0.06) (0.08) (0.15) (0.06) (0.07) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,195 967 Prob > chi2 0.05 0.04 0.75 0.02 0.58 2009 NMP 0.57 0.20*** 0.32* 0.14*** 0.07 (0.63) (0.08) (0.17) (0.05) (0.08) ln(Hog animal units) 0.06 0.07*** 0.02 0.01*** 0.00 (0.08) (0.01) (0.02) (0.00) (0.02) Large CAFO 0.03 −0.05 −0.04 −0.00 0.04 (0.06) (0.07) (0.06) (0.03) (0.04) Other controls included?a Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,282 1,017 Prob > chi2 0.56 0.10 0.16 0.15 0.28 Note: Superscript a indicates that other controls include whether the operation has any crop acreage, whether the operation is in the top half of crop acreage, whether the operation is farrow to finish, whether the operation has an EQIP contract, whether there has ever been a TMDL in the county before the survey year, whether the operation is a contractor, the operator's years of experience, whether the operator has a college degree, and the population density of the county. Superscript b indicates the sample consists of just those operations that applied manure in any form. Superscript C indicates a sample of just those operations with any crop acreage. These regressions do not include the independent variable of whether the operation had any crop acreage. Results of 24 regressions are shown. All outcome variables are indicator (0/1) variables. Asterisk * refers to confidence at the 10% level, ** refer to confidence at the 5% level, and *** refer to confidence at the 1% level. All regressions include state-level fixed effects. Prob > chi2 refers to a likelihood ratio test of whether we can reject the null hypothesis of no correlation between the treatment errors and the outcome errors. A probability of the chi-squared value less than 0.05 suggests we can reject the null hypothesis of no selection with 95% certainty. In 2004, the variable NMP is significant and positive at the 1% level in predicting whether an operator tests manure for nutrient content, as well as for the three measures of adjusting feed for manure nitrogen content; NMP is not a significant predictor of whether the operation incorporates manure at application to reduce potential run-off. Except for the nitrogen testing outcome, these results hold for 2009 as well. In 2004, operations with a NMP are 27 percentage points more likely to perform manure nutrient testing, and between 13 and 17 percentage points more likely to adjust feed to change manure nutrient content. In 2009, operations with NMPs are between 14 and 32 percentage points more likely to adjust feed.11,12 Table 5 also shows the test statistic for whether or not we can reject the null hypothesis of no correlation between the treatment regression errors and the behavior regression errors. Values less than 0.05 for this test statistic suggest that we can reject the null hypothesis of no correlation. Our results suggest that a correlation between the errors occurs for some but not all variables. Table 6 provides the results for the nutrient application, uptake, and excess outcomes. In 2004 having a NMP was not a significant predictor (in this empirical strategy) of whether the operation applied any excess, the amount of nitrogen applied, the total uptake, or the amount of excess applied. The presence of a NMP did significantly predict less nitrogen applied per acre and less excess per acre. This suggests that in 2004, operations with a NMP reduced their per-acre nutrient application rates but did not significantly change their planted acreage, nor did they alter their crop mix to generate higher uptake values.13 Table 6 Endogenous Treatment Effects Regression Results, 2004 and 2009 (maximum likelihood) Indicator (0/1) variable Continuous outcome variables Applied excess nitrogen Total nitrogen manure and fertilizer applied (lbs.) Total nitrogen uptake (lbs.) Excess nitrogen (applied minus uptake, lbs.) Applied per acre (lbs.)c Uptake per acre (lbs.)c Excess per acre (lbs.)c 2004 NMP 0.04 6,659 10,975 −4,084 −392.65*** 26.94 −380.89** (0.07) (4,330) (12,186) (9,719) (147.77) (67.08) (148.80) ln(Hog animal units) 0.03*** 7,307*** 14,885*** −7,572** 74.71*** 1.52 68.49** (0.01) (1,678) (3,575) (3,324) (27.19) (10.97) (26.97) Large CAFO −0.04 39,206*** 18,358 20,641 15.68 −16.57 33.34 (0.03) (14,492) (20,214) (15,450) (38.92) (13.95) (40.60) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,193 992 992 990 Prob > chi2 0.61 0.07 0.82 0.40 0.00 0.60 0.00 2009 NMP −0.30*** −75,766*** 29,303** −38,132** −389.33*** −54.74* −410.44*** (0.04) (11,730) (12,650) (17,042) (150.93) (32.69) (149.40) ln(Hog animal units) 0.08*** 18,027*** 14,136*** −4,278 78.25*** 13.93** 74.63*** (0.01) (2,835) (3,365) (3,162) (25.36) (5.59) (25.32) Large CAFO −0.00 20,972** −1,349 25,692*** −17.82 −25.79*** 1.77 (0.03) (9,716) (12,085) (9,799) (40.42) (8.24) (41.12) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,279 1,095 1,095 1,092 Prob > chi2 0.00 0.00 0.17 0.11 0.00 0.13 0.00 Indicator (0/1) variable Continuous outcome variables Applied excess nitrogen Total nitrogen manure and fertilizer applied (lbs.) Total nitrogen uptake (lbs.) Excess nitrogen (applied minus uptake, lbs.) Applied per acre (lbs.)c Uptake per acre (lbs.)c Excess per acre (lbs.)c 2004 NMP 0.04 6,659 10,975 −4,084 −392.65*** 26.94 −380.89** (0.07) (4,330) (12,186) (9,719) (147.77) (67.08) (148.80) ln(Hog animal units) 0.03*** 7,307*** 14,885*** −7,572** 74.71*** 1.52 68.49** (0.01) (1,678) (3,575) (3,324) (27.19) (10.97) (26.97) Large CAFO −0.04 39,206*** 18,358 20,641 15.68 −16.57 33.34 (0.03) (14,492) (20,214) (15,450) (38.92) (13.95) (40.60) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,193 992 992 990 Prob > chi2 0.61 0.07 0.82 0.40 0.00 0.60 0.00 2009 NMP −0.30*** −75,766*** 29,303** −38,132** −389.33*** −54.74* −410.44*** (0.04) (11,730) (12,650) (17,042) (150.93) (32.69) (149.40) ln(Hog animal units) 0.08*** 18,027*** 14,136*** −4,278 78.25*** 13.93** 74.63*** (0.01) (2,835) (3,365) (3,162) (25.36) (5.59) (25.32) Large CAFO −0.00 20,972** −1,349 25,692*** −17.82 −25.79*** 1.77 (0.03) (9,716) (12,085) (9,799) (40.42) (8.24) (41.12) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,279 1,095 1,095 1,092 Prob > chi2 0.00 0.00 0.17 0.11 0.00 0.13 0.00 Note: Superscript a indicates that other controls include whether the operation has any crop acreage, whether the operation is in the top half of crop acreage, whether the operation is farrow to finish, whether the operation has an EQIP contract, whether there has ever been a TMDL in the county before the survey year, whether the operation is a contractor, the operator's years of experience, whether the operator has a college degree, and the population density of the county. Superscript b indicate a sample of just those operations that applied manure in any form. Superscript C indicates a sample of just those operations with any crop acreage. These regressions do not include the independent variable of whether the operation had any crop acreage. Results of 24 regressions are shown. Asterisk * refers to confidence at the 10% level, ** refer to confidence at the 5% level, and *** refer to confidence at the 1% level. All regressions include state-level fixed effects. Prob > chi2 refers to a likelihood ratio test of whether we can reject the null hypothesis of no correlation between the treatment errors and the outcome errors. A probability of the chi-squared value less than 0.05 suggests we can reject the null hypothesis of no selection with 95% certainty. Table 6 Endogenous Treatment Effects Regression Results, 2004 and 2009 (maximum likelihood) Indicator (0/1) variable Continuous outcome variables Applied excess nitrogen Total nitrogen manure and fertilizer applied (lbs.) Total nitrogen uptake (lbs.) Excess nitrogen (applied minus uptake, lbs.) Applied per acre (lbs.)c Uptake per acre (lbs.)c Excess per acre (lbs.)c 2004 NMP 0.04 6,659 10,975 −4,084 −392.65*** 26.94 −380.89** (0.07) (4,330) (12,186) (9,719) (147.77) (67.08) (148.80) ln(Hog animal units) 0.03*** 7,307*** 14,885*** −7,572** 74.71*** 1.52 68.49** (0.01) (1,678) (3,575) (3,324) (27.19) (10.97) (26.97) Large CAFO −0.04 39,206*** 18,358 20,641 15.68 −16.57 33.34 (0.03) (14,492) (20,214) (15,450) (38.92) (13.95) (40.60) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,193 992 992 990 Prob > chi2 0.61 0.07 0.82 0.40 0.00 0.60 0.00 2009 NMP −0.30*** −75,766*** 29,303** −38,132** −389.33*** −54.74* −410.44*** (0.04) (11,730) (12,650) (17,042) (150.93) (32.69) (149.40) ln(Hog animal units) 0.08*** 18,027*** 14,136*** −4,278 78.25*** 13.93** 74.63*** (0.01) (2,835) (3,365) (3,162) (25.36) (5.59) (25.32) Large CAFO −0.00 20,972** −1,349 25,692*** −17.82 −25.79*** 1.77 (0.03) (9,716) (12,085) (9,799) (40.42) (8.24) (41.12) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,279 1,095 1,095 1,092 Prob > chi2 0.00 0.00 0.17 0.11 0.00 0.13 0.00 Indicator (0/1) variable Continuous outcome variables Applied excess nitrogen Total nitrogen manure and fertilizer applied (lbs.) Total nitrogen uptake (lbs.) Excess nitrogen (applied minus uptake, lbs.) Applied per acre (lbs.)c Uptake per acre (lbs.)c Excess per acre (lbs.)c 2004 NMP 0.04 6,659 10,975 −4,084 −392.65*** 26.94 −380.89** (0.07) (4,330) (12,186) (9,719) (147.77) (67.08) (148.80) ln(Hog animal units) 0.03*** 7,307*** 14,885*** −7,572** 74.71*** 1.52 68.49** (0.01) (1,678) (3,575) (3,324) (27.19) (10.97) (26.97) Large CAFO −0.04 39,206*** 18,358 20,641 15.68 −16.57 33.34 (0.03) (14,492) (20,214) (15,450) (38.92) (13.95) (40.60) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,195 1,195 1,195 1,193 992 992 990 Prob > chi2 0.61 0.07 0.82 0.40 0.00 0.60 0.00 2009 NMP −0.30*** −75,766*** 29,303** −38,132** −389.33*** −54.74* −410.44*** (0.04) (11,730) (12,650) (17,042) (150.93) (32.69) (149.40) ln(Hog animal units) 0.08*** 18,027*** 14,136*** −4,278 78.25*** 13.93** 74.63*** (0.01) (2,835) (3,365) (3,162) (25.36) (5.59) (25.32) Large CAFO −0.00 20,972** −1,349 25,692*** −17.82 −25.79*** 1.77 (0.03) (9,716) (12,085) (9,799) (40.42) (8.24) (41.12) Other controls included?a Yes Yes Yes Yes Yes Yes Yes Number of observations 1,282 1,282 1,282 1,279 1,095 1,095 1,092 Prob > chi2 0.00 0.00 0.17 0.11 0.00 0.13 0.00 Note: Superscript a indicates that other controls include whether the operation has any crop acreage, whether the operation is in the top half of crop acreage, whether the operation is farrow to finish, whether the operation has an EQIP contract, whether there has ever been a TMDL in the county before the survey year, whether the operation is a contractor, the operator's years of experience, whether the operator has a college degree, and the population density of the county. Superscript b indicate a sample of just those operations that applied manure in any form. Superscript C indicates a sample of just those operations with any crop acreage. These regressions do not include the independent variable of whether the operation had any crop acreage. Results of 24 regressions are shown. Asterisk * refers to confidence at the 10% level, ** refer to confidence at the 5% level, and *** refer to confidence at the 1% level. All regressions include state-level fixed effects. Prob > chi2 refers to a likelihood ratio test of whether we can reject the null hypothesis of no correlation between the treatment errors and the outcome errors. A probability of the chi-squared value less than 0.05 suggests we can reject the null hypothesis of no selection with 95% certainty. The predictive power of NMPs becomes greater in the 2009 survey. Operations with NMPs were 30 percentage points less likely to apply any excess nitrogen, controlling for confounders and selection. Such operations appear to have reduced the amount applied (by nearly 76,000 lbs. of nitrogen per operation), and increased their uptake (by 29,000 lbs. per operation). This resulted in a decline in total excess applied (of about 38,000 lbs. per operation). These results suggest that operations may have increased their crop acreage to more closely match crop uptake with nutrients applied. The per-acre outcomes suggest that operations with NMPs also adjusted their nutrient management practices at the intensive margin; they applied less per acre, resulting in less excess per acre. There is no evidence that farms with NMPs adjusted their crop mix to increase nitrogen uptake per acre. Comparison of the results in 2004 and 2009 suggests the effect of the NMP became stronger in terms of reducing nutrient applications. This might be explained by increasing enforcement and/or awareness among producers of better nutrient management practices. The summary statistics suggest different qualitative results from the econometric specifications for two outcomes. First, the summary statistics suggest that operations with NMPs are more likely to apply excess nutrients overall. Second, the summary statistics suggest that operations with NMPs, on average, applied more nutrients per acre than those without. In contrast, the results of the econometric specification suggest that in 2009, operations with NMPs were less likely to apply excess nutrients in 2009 (but not 2004), and applied less excess per acre in both years. Examination of a set of models (appendix table 9) without correcting for selection shows that clustering standard errors by state reduces the statistical significance of the impact of the NMP on these two outcomes, and the addition of covariates reduces the size of the estimated coefficients. The signs on the coefficients change when we control for selection. A model in which we only control for size of operation in the NMP prediction equation and the outcome equation shows statistically significant effects of NMP in predicting a lower likelihood of excess nutrient application, and less excess per acre applied. These checks suggest that both controlling for confounders and adjusting for selection are pertinent in estimating effects of NMPs. Finally, to examine whether enforcement of environmental regulations impacts the efficacy of NMPs, we interact NMP status with the four measures of enforcement (whether the county ever had a TMDL before the survey year, the population density of the county, whether the operation is a large CAFO, and whether the operation has an EQIP contract). These interaction terms are rarely statistically significant (results appear in appendix tables 10 through 17), suggesting that at least for these measures of enforcement, greater regulatory oversight does not yield better nutrient management outcomes via the NMPs. Conclusions The NMPs are a fundamental component of federal, state, and regional efforts to encourage livestock producers to apply nutrients to cropland at agronomic rates. Despite the important role that NMPs play in the environmental regulation of livestock farms, there is little information about their efficacy. This study is one of the first to examine whether having an NMP makes a farm more likely to use a set of recommended nutrient management practices and to apply manure nutrients at agronomic rates. Our results provide evidence to support the hypothesis that the use of NMPs encourages U.S. hog operations to implement manure management practices to reduce run-off into nearby waterways. After adjusting for operation size, location, and other potential confounders, and correcting for selection into NMP adoption, our empirical results suggest that farms with NMPs are significantly more likely to use practices consistent with careful nutrient management (manure nutrient testing and feed adjustments). The results of our empirical strategy suggest that in 2009, NMPs had their intended effect on excess nutrient applications (manure and commercial fertilizer nutrient applications minus nutrient uptake capacity). We find evidence that operations with NMPs (when accounting for confounding variables and selection) were statistically less likely to over-apply nutrients in 2009, but not 2004. This is potentially because farmers responded to stricter enforcement of the 2003 Clean Water Act amendments that were adopted and enforced gradually in the following years, or because of general knowledge dissemination of nutrient management. Our empirical results suggest that operations with NMPs reduce excess nutrient applications both by adjusting at the extensive margin (amount of land cropped) as well as the intensive margin (applications per unit of land). Stricter enforcement on NMPs, as indicated by having a TMDL in the county, being a large CAFO, or having an EQIP contract does not increase the efficacy of NMPs in encouraging nutrient management. Because our data were collected in 2004 and 2009, our findings may not be indicative of current practices of hog farmers. The change in predictive power between 2004 and 2009 in terms of the estimated relationships between NMPs and nutrient management applications suggest a trend toward more efficacy of NMPs. If this trend has continued or been maintained, then NMPs may still be effective in reducing potential run-off from hog operations. In this study we have examined the question of whether having a NMP is predictive of better nutrient management behaviors. We do not, however, address whether NMPs are the only policy necessary to obtain water quality goals. Since such goals vary by location, whether or not requiring NMPs will allow regulators to reach agricultural nutrient run-off reduction goals is a question answered only at the regional level. This paper suggests, however, that NMPs encourage better nutrient management behaviors, and can be one tool in promoting less agriculturally-based water pollution. Footnotes 1 These are the only two years of ARMS hogs data that contain information on NMPs. In an updated survey, completed in 2016, questions on manure management were removed. 2 For example, operations that already agronomically manage nutrients may adopt a NMP because they find it easy to do so. Alternatively, operations that do not agronomically manage nutrients may also be more likely to adopt a NMP if oversight is lax and they want to signal environmental stewardship to potential regulatory authorities or integrators. 3 In some years, some states have required all confined animal operations to obtain NMPs. For example, Maryland began requiring all operations to have NMPs by 2001 (Perez,2011). We examined the state laws for all 19 states in the sample, and none require that all operations (regardless of size, discharge status, or other feature) obtain NMPs. Notably, Maryland is not in our sample of states as it is not a major hog-producer. 4 The variables included in X are ln(number of hog animal units); whether the operation is a large CAFO with other 1,000 hog animal units; whether the operation has any crop acreage; whether the operation is in the top half of crop acreage distribution for the year; whether the operation is farrow to finish; whether the operation has an EQIP contract; whether there has ever been a TMDL in the county before the survey year; whether the operation is a contractor; the operator's years of experience; whether the operator has a college degree; and the population density of the county. Despite the subscript i on X, two of the variables included in this vector vary at the level of the county, not the operation. These two variables are county-level population density and whether or not the county has had a TMDL for nutrients before or in the survey year. We use the subscript i for brevity. 5 Notably, no state in either years has a 100% NMP adoption rate; see appendix table 1. This means that that there is still variation within states in NMP status. 6 Prior research has questioned whether operations that see less value in environmental stewardship selectively locate in states with lax enforcement of environmental regulation. In this were the case, we might see operations with worse nutrient management practices locate in places with lower enforcement. It is difficult to assess whether this would engender bias in the estimated effect of NMPs, or in what direction it would be. If operations with worse practices selectively locate in areas to avoid obtaining NMPs, then for those individual states we might see larger effects of NMPs than without selection in those states. However, relocation of poor performers would leave better environmental stewards in states with higher regulatory stringency. Comparing across NMP status in higher stringency states would yield smaller impacts of NMPs than without selection by location. The estimated overall effect of NMP status in the presence of such selection, inclusive of both types of states, could be the same, greater, or less than the estimated effect without such selection. 7 This vector includes the same variables as in X. 8 Complications arise in two main areas. First, obtaining maximum likelihood estimates requires convergence of the likelihood functions. In our application, we never obtain convergence when estimating the outcome equation as a probit, even with the most pared-down model. Second, we later consider interaction of NMP status with an enforcement variable. Interpretation of interaction terms in straightforward probits (i.e., those without endogenous binary independent variables) is complicated due to the nonlinearity of the functional form; estimates and statistical significance vary according to the levels of the included confounders. For these reasons we do not use the probit as the functional form for the outcome equation. 9 As Angrist and Pishchke note on their Mostly Harmless Econometrics blog, “[T]he LPM won’t give the true marginal effects from the right nonlinear model. But then, the same is true for the ‘wrong’ nonlinear model! The fact that we have a probit, logit and the LPM is just a statement to the fact that we don’t know what the ‘right’ model is.” Available at: http://www.mostlyharmlesseconometrics.com/2012/07/probit-better-than-lpm/. 10 Appendix table 2 shows results from the probit model to predict NMP status in each year. The strongest positive predictors for NMP adoption are the number of animal units and whether the operation has an EQIP contract. The operation being farrow-to-finish is a significant negative predictor in both years. The endogenous treatment effects model results in tables 5 and 6 estimates both the “first” and “second” stages simultaneously. We evaluate several outcome variables; hence, the estimated coefficients on the variables predicting NMP adoption are slightly different for each outcome variable. The estimates in appendix table 2 show the results of probit models estimated by themselves, rather than within the endogenous treatment model. The estimate effects are highly similar to those estimated in the “first stage” of the endogenous treatment model for each of the multiple outcome variables. 11 Appendix table 3 shows the results of the LPM without treating NMP status as endogenous. The marginal effects from probit regressions are shown in appendix table 4, again without treating NMP status as endogenous. The probit and LPM results are very similar in magnitude and statistical significance, particularly in 2009. This suggests that modeling the binary outcome behavioral regressions in the endogenous treatment model using a linear model is unlikely to result in heavily biased coefficients due to functional form. 12 While we do not show the estimated coefficients on all controls in the main text, the interested reader can find them in appendix tables 5 and 6. The strongest predictors, aside from NMP status and number of hog animal units, of the recorded nutrient management practices are whether the operation has a production contract and the state fixed effects. This is true for both years. 13 Appendix Tables 7 and 8 show the estimated coefficients on all included covariates. The strongest and most consistent observable predictors beyond those shown in the main text are whether the operation is a farrow-to-finish operation, whether the operation has a production contract, and the two measures of crop acreage. The state fixed effects are also strong predictors. References Ai C. , Norton E . 2003 . Interaction Terms in Logit and Probit Models . Economics Letters 80 : 123 – 29 . Google Scholar CrossRef Search ADS Albrecht J. 2003 . Reduction of Manure Nutrient Concentrations. Confined Animal Manure Managers Program. Swine Training Manual. Chapter 3b, last edit. Available at: http://www.clemson.edu.extension/camm/manuals/swine/sch3b_03.pdf. Beegle D. , Carton O. , Bailey J . 2000 . Nutrient Management Planning: Justification, Theory, Practice . Journal of Environmental Quality 29 ( 1 ): 72 – 9 . Google Scholar CrossRef Search ADS Burkholder J. , Libra B. , Weyer P. , Heathcote S. , Kolpin D. , Thorne P.S. , Wichman M . 2007 . Impacts of Waste from Concentrated Animal Feeding Operations on Water Quality . Environmental Health Perspectives 115 ( 2 ): 308 – 12 . Google Scholar CrossRef Search ADS PubMed Centner T.J. , Newton G.L . 2008 . Meeting Environmental Requirements for the Land Application of Manure . Journal of Animal Science 86 : 3228 – 34 . Google Scholar CrossRef Search ADS PubMed Copeland C. 2012 . Clean Water Act and Pollutant Total Maximum Daily Loads (TMDLs). Washington DC: Congressional Research Service Report 7-5700. Copeland C , Zinn J . 1998 . Animal Waste Management and the Environment: Background for Current Issues. Washington DC: Congressional Research Service Report 98-451. Genskow K.D. 2012 . Taking Stock of Voluntary Nutrient Management: Measuring and Tracking Change . Soil and Water Conservation 61 ( 1 ): 51 – 8 . Google Scholar CrossRef Search ADS Innes R. 1999 . Regulating Livestock Waste: An Economic Perspective . Choices 14 – 19 . Innes R. 2000 . The Economics of Livestock Waste and its Regulation . American Journal of Agricultural Economics 82 : 97 – 117 . Google Scholar CrossRef Search ADS Johnson R.S. , Wheeler W.J. , Christensen L.A . 1999 . EPA’s Approach to Controlling Pollution from Animal Feeding Operations: An Economics Analysis . American Journal of Agricultural Economics 81 ( 5 ): 1216–21 . Google Scholar CrossRef Search ADS Kellogg R.L. , Lander C.H. , Moffitt D.C. , Gollehon N . 2000 . Capacity of Cropland and Pastureland to Assimilate Nutrients: Spatial and Temporal Trends for the United States. Washington DC: U.S. Department of Agriculture, Natural Resources Conservation Service and Economic Research Service. Key N. , McBride W.D. , Ribaudo M. , Sneeringer S . 2011 . Trends and Developments in Hog Manure Management: 1998-2009. Washington DC: U.S. Department of Agriculture, Economic Research Service, Economic Information Bulletin No. 81. Lawley C. , Lichtenberg E. , Parker D . 2009 . Biases in Nutrient Management Planning . Land Economics 85 ( 1 ): 186 – 200 . Google Scholar CrossRef Search ADS Lyford C. , Hicks T . 2001 . The Environment and Pork Production: The Oklahoma Industry at a Crossroads . Review of Agricultural Economics 23 ( 1 ): 265 – 74 . Google Scholar CrossRef Search ADS Maddala G.S. 1983 . Limited-Dependent and Qualitative Variables in Econometrics . Cambridge, MA : Cambridge University Press . Google Scholar CrossRef Search ADS McBride W. , Key N . 2013 . U.S. Hog Production from 1992 to 2009: Technology, Restructuring, and Productivity Growth. U.S. Department of Agriculture, Economic Research Service, Economic Research Report No. 158. McCann L.M.J. 2009 . Transaction Costs of Environmental Policies and Returns to Scale: The Case of Comprehensive Nutrient Management Plans . Review of Agricultural Economics 31 ( 3 ): 561 – 73 . Google Scholar CrossRef Search ADS National Pollutant Discharge Elimination System Permit Regulation and Effluent Limitation Guidelines and Standards for Concentrated Animal Feeding Operations (CAFOs) . Federal Register 68 ( 29 ): 7176 –4. Norton E. , Wang H. , Ai C . 2004 . Computing Interaction Effects and Standard Errors in Logit and Probit Models . The Stata Journal 4 : 154 – 67 . Perez M. 2011 . Regulating Farmers: Lessons Learned from the Delmarva Peninsula . Choices 26 ( 3 ). Ribaudo M.O. , Cattaneo A. , Agapoff J . 2004 . Cost of Meeting Manure Nutrient Application Standards in Hog Production: The Roles of EQIP and Fertilizer Offsets . Review of Agricultural Economics 26 ( 4 ): 430 – 44 . Google Scholar CrossRef Search ADS Ribaudo M.O. , Johansson R.C . 2007 . Nutrient Management Use at the Rural-Urban Fringe: Does Demand for Environmental Quality Play a Role? Review of Agricultural Economics 29 ( 4 ): 689 – 99 . Google Scholar CrossRef Search ADS Ribuado M. , Key N. , Sneeringer S . 2016 . The Potential Role for a Nitrogen Compliance Policy in Mitigating Gulf Hypoxia . Applied Economic Perspectives and Policy 39 ( 3 ): 458 – 78 . Savage J.A. , Ribaudo M.O . 2013 . Impact of Environmental Policies on the Adoption of Manure Management Practices in the Chesapeake Bay Watershed . Journal of Environmental Management 129 : 143 – 48 . Google Scholar CrossRef Search ADS PubMed Shepard R. 2005 . Nutrient Management Planning: Is it the Answer to Better Management? Journal of Soil and Water Conservation 60 ( 4 ): 171 – 76 . Sneeringer S. , Key N . 2011 . Effects of Size-Based Environmental Regulations: Evidence of Regulatory Avoidance . American Journal of Agricultural Economics 93 ( 4 ): 1189 – 211 . Google Scholar CrossRef Search ADS Sneeringer S. 2016 . Comparing Participation in Nutrient Trading by Livestock Operations to Crop Producers in the Chesapeake Bay Watershed . Washington DC : U.S. Department of Agriculture, Economic Research Service , Economic Research Report No. 216. StataCorp . 2015 . Stata 14 Base Reference Manual. College Station, TX : Stata Press . Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association 2018. This work is written by US Government employees and is in the public domain in the US.

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