Abstract We conduct meta-analyses of the estimated impacts of corn ethanol on food and fuel prices, as well as greenhouse gases, and analyze the implications for the balance of trade. The meta-analyses suggest that corn ethanol has minor effects on greenhouse gas emissions and significant yet moderate effects on food and fuel prices. However, corn ethanol has a relatively significant impact on fuel security in terms of reductions in the import of oil to the U.S. and its overall effect on the U.S. balance of trade. U.S. policy makers have expressed concern about climate change, energy security, and balance of trade to promote corn as a sustainable fuel source. Introduction of the Energy Policy Act in 2005 and passage of the Energy Independence and Security Act of 2007 contributed to the increase in corn ethanol production from 1.65 billion gallons in 2000 used mostly as a gasoline additive, to 9 billion by 2008 and 13 billion by 2015. Indeed, by 2015, corn ethanol had become the largest biofuel feedstock utilized globally.1 Corn ethanol accounted for 9.8% and 9.9% of the U.S. motor gasoline in 2014 and 2015, respectively, and in 2016, the United States exported more than 1 billion gallons of corn ethanol, a 26% increase from 2015.2 However, over the past 10 years, the use of biofuels has also become a source of controversy, with a large body of literature debating the effect of biofuels on food prices and food security, including arguments that question the overall merit of using biofuels to mitigate climate change.3 Now that we have sufficient data and estimates regarding the impacts of biofuels, we can reassess the claims about biofuel impacts. This paper evaluates findings from several strands of the literature by conducting meta-analyses on the impacts of corn ethanol on food and fuel prices, greenhouse gases (GHGs), and indirect land use change (ILUC). Although the analysis surveys a vast body of literature, the statistical analyses focus on corn ethanol because of its importance, the large volume of research available, and the intense political debate regarding its continued use. The statistical analysis investigates the main reasons for the differences found in the literature that reflect the research heterogeneity in terms of the period covered, methods, elasticities, and regional coverage, among other factors. The statistical analyses are supplemented with production data, trade data, and conversion coefficients. The data are used to validate the conclusions of the meta-analyses, and simple calculations are used to understand the economic and environmental implications of those conclusions: while production and trade data are used to compute the effect of corn ethanol on the balance of trade, conversion factors are used to convert the direct and indirect GHG emissions reported in the literature to common units of analysis. The statistical analysis finds that, on average, the effects of biofuels on food and fuel prices are moderate and that corn ethanol results in the following, after correcting for publication bias: (a) an average increase in agricultural commodity prices of about 14%; and (b) a decrease in the average price of gasoline of about 5%. However, corn ethanol has had a relatively substantial impact on fuel security in terms of reductions in the import of oil from abroad and the overall balance of trade; the annual net gain of the balance of trade due to biofuels has been tens of billions of dollars. Our analysis also shows that the assumptions used in assessing the various effects of biofuel matter, and are a major source of differences in outcomes among studies. Overall, (a) biofuels have a smaller impact on prices in models that consider feedback effects among food commodities and petroleum-refining products; (b) the simulated price and emissions effects of biofuel policy are smaller in models that consider economic linkages among markets and regions; and (c) biofuels have greater price effects in studies that assume more inelastic demand and supply curves. Our meta-analysis evaluates the plausibility of a publication bias among studies investigating a given effect (i.e., the corn ethanol impacts on GHGs, food commodity prices, or fuel prices), and it also quantifies the heterogeneity among the aforementioned studies. We find evidence for publication bias among studies evaluating corn ethanol impacts on GHGs. When investigating heterogeneity among studies, the I2 statistic suggests low levels of heterogeneity among studies assessing GHG emissions and food commodity prices, but higher levels of heterogeneity among studies that evaluate the effect of corn ethanol on fuel prices. The next section presents the empirical approach used for estimating the importance of the various factors affecting the meta-analyses’ parameters. The following section describes the meta-data used, and the results of the meta-analyses for the three investigated effects are presented in the next section. Policy is discussed and concluding remarks are offered in the final section. The Empirical Approach The goal of our research design is to assess the impact of introducing biofuels on GHG, as well as food and fuel prices. We rely on findings in the literature to calculate the average effect of introducing corn ethanol on these factors, while exploring the effects of specific modeling features on the estimated impact. This study belongs to the literature that applies meta-analysis, aiming to integrate results of different studies, where each study has its own modeling framework and assumptions. The meta-regression analysis is an application of the concept of triangulation that has become popular in policy analysis and management (Cohen and Manion 2000; O’Donoghue and Punch 2003; Altrichter et al. 2008).4 Because the true model is not known, the use of results from multiple studies yields a more reliable outcome and a better understanding of the effect of interest. A vast body of literature investigates the impact of biofuels, the majority of which employ numerical methods to assess the effect of introducing corn ethanol in response to the renewable fuel standard (RFS). Usually, meta-analyses are used to glean the results of statistical studies (Loomis and White 1996; Stanley 2001; Nelson and Kennedy 2009; Stanley and Doucouliagos 2012; Havraneka et al. 2014; Palmer and Sterne 2016, among many others), not numerical studies. However, when the meta-analysis focuses on numerical studies (e.g., National Research Council 2011; Zhang et al. 2013; Persson 2015), it usually resorts to non-parametric tools. For example, most of the studies on the social cost of carbon employ non-parametric methods to better understand the results of numerical studies assessing the social cost of carbon (e.g., Tol 2008). One exception is Havranek et al. (2015), who employ a parametric approach such that the precision of the numerical estimates is derived based on the measurement of uncertainty reported in the various studies (only 267 of the 809 studies surveyed had such a measure). Another exception is Condon, Klemick, and Wolverton (2015), who review studies published between 2007 and 2014 that estimate the effects of U.S. corn ethanol policy on corn prices. These authors use a fixed-effect model to estimate the coefficients of covariates that vary within studies and a random-effect model to examine key attributes that only vary at the study level. We propose an alternative method of estimating a measure of precision for the various numerical estimates of interest. We propose using the variability among the estimates within each study as well as the variability between studies. The studies considered in the analysis each run several scenarios that are based on plausible values of exogenous parameters. Each of the scenarios produces estimates of the outcomes of interest. For each study, we calculate the precision measure of the estimated outcomes while separating the within-study variability. Different studies have different mean outcomes, and the proposed method will associate each outcome with its own precision measure. In particular, we investigate the K variable of interest, one variable of interest at a time (i.e., GHG, corn commodity prices, and fuel prices, one at a time). The variable k∈K is evaluated in Jk studies and j∈Jk. A given study includes nj+1 simulations, a benchmark scenario, and nj alternative scenarios. We assume that each of the nj numerical simulations are an individual observation and that observations are clustered within studies. So we have Jk clusters with a total of ∑j=1Jknj observations to assess what determines variable of interest k. For simplicity, we drop the k index in what follows. For every variable of interest, two measures of impacts are calculated: the absolute impact yij (i.e., the calculated difference between the ith simulation in the jth study and the benchmark of study j) and the relative impact (i.e., 100×yij/benchmarkofstudyj). Therefore, we use two models: one to estimate the factors that affect absolute differences between simulated outcomes and the benchmark, and the other to explain the magnitude of the absolute difference relative to the benchmark level. Each observation includes H study-specific dummy variables, fj=f1,j,…,fH,j, and L observed covariates, xij, xij=1,x1,ij,…,xL,ij; let Xj=x1j,fj',…,xnjj,fj' and assume the absence of correlation among the study-specific dummy variables and the covariates. For example, when estimating both the absolute and relative price effects of corn ethanol on corn prices, the study-specific dummy variables include the modeling of gasohol markets (i.e., a dummy variable fixed across alternative scenarios within a given study but changing between studies) while the observed covariates include the demand and supply elasticities of corn ethanol (i.e., continuous variables that change value within studies). The data of the jth study includes (a) a vector of simulated outcomes Yj, (b) a variance-covariance matrix of the simulated outcomes Sj, and (c) a matrix of the study-specific dummy variables and observed covariates Xj. For the statistical analysis, assume that Yj∼Nuj,Sj where uj is unknown and is estimated statistically assuming that uj∼NβXj,Ω. The estimated coefficients are denoted by β=β0,β1,…,βL,βL+1,…,βL+H, ωnm∈Ω denote the between-studies variance-covariance, m,n∈Jk, and sbd,j∈Sj denote the within-study variance-covariance, b,d∈nj. The empirical model uses the data to estimate the regression coefficients, β, and the between-studies variance-covariance, Ω. These estimates are then used to investigate the impacts of interest. Our analysis builds on the fixed-effect meta-regression analysis literature (Davidson and MacKinnon 2004; Riley, Thompson, and Abrams 2008; Stanley and Doucouliagos 2012). It assumes that sii,j is the within variance, calculated using the population of simulations reported in study j, and defines the diagonal elements of the variance-covariance matrix Vii,j=sii,j+ωjj. To allow the variance to vary across groups, we follow Greene (2012), make the random-effect type assumptions used in that section, and calculate consistent estimates for the within and between variances. Because under these assumptions the pooled ordinary least squares (OLS) is consistent, we use it to provide a set of residuals that we employ to estimate Vii,j. On the other hand, because the residuals from the least squares dummy variable model are purged of the individual specific effect, these residuals are used to consistently estimate sii,j. The combination of the two yields an estimate for ωjj. The next step is to employ alternative statistical techniques to glean the results of the numerical studies. Our first approach, denoted as the Fixed-Effect approach, follows Hedges and Olkin (1985), and estimates the multivariate fixed-effect meta-regression model using a weighted least squares regression whereby the weights are the reciprocal of the variance (Cheung 2008; White 2009 and 2011; Palmer and Sterne 2015). The second approach, the Random-Effect approach, allows for individual effects. In the third approach, we estimate the parameters assuming that each cluster has its own standard deviation (Greene 2012, and references therein) and denote this approach the Cluster approach. Under the Cluster approach, the observations are independent across groups (clusters) but not necessarily within groups. The fourth approach is denoted the Frequency Weighted approach. This approach assumes a weighted regression, where the weights denote the inverse of the probability that the observation is the ith observation of jth study. While the Fixed- and Random-Effect approach aim to correct for the pernicious effect of omitted variable biases, the Cluster and Frequency-Weighted approaches aim to reflect the behavior of the random errors by allowing different weighting-fitting criteria when estimating the parameters of interest. Because of concerns that weights might underestimate studies with many scenarios, the outcome of the various regressions is compared to a simple OLS. Our study aims to investigate commonalities and regularities among studies, studies that begin with alternative baselines and each with its own modeling framework. Because the true numerical model is not known, the use of results from multiple studies yields a more reliable outcome and a better understanding of the effect of interest. When determining the empirical model, we employ Bayesian model averaging (BMA) techniques (Hoeting et al. 1999). We also employ backward elimination stepwise regression (elimination if p-value > 0.2). In the interest of space, most of the results of the robustness analysis are presented in the supplementary online material. Data The search for empirical studies uses the search engine Google Scholar and employs the following keywords: biofuels, ethanol, food prices, fuel prices, GHG, and ILUC. When collecting studies, the population is not limited to peer-reviewed studies but expanded to reports and unpublished papers, books, book chapters, and dissertations, that is, the “gray literature” (Loomis and White 1996). Some unpublished studies are not only newer and use newer data, but all studies, independent of their status, contribute to the statistical identification of factors responsible for the heterogeneity among papers (Stanley 2001; Stanley et al. 2013). The search concludes on January 2016. When choosing the regressions’ study-specific dummy variables and covariates, we included only those that had sufficient variability across observations. When coding the data, some of the study-specific dummy variables are common across all effects (i.e., peer reviewed, year study calibrated, period of analysis, computable general equilibrium (CGE) model), while others are effect-specific (e.g., corn price elasticity, fuel price elasticity). When assessing the GHG effect of biofuels, after coding the various parameters we modify the dataset in two ways. Because different studies use different economic, energy, and emissions parameters, we aggregate these variables to economic and to energy and pollution indexes; otherwise, any empirical model would result in missing data and thus many observations would be dropped. We aggregate the various parameters and create two indexes, where a higher value of the index denotes a higher GHG footprint of corn ethanol. The first index aggregates the economic parameters, while the second aggregates the emissions and energy parameters. The second modification relates to lifecycle analysis studies surveyed in Farrell et al. (2006). Because in many of these studies only one value is reported, we use the methods employed in Farrell et al. (2006) to normalize this subgroup of studies to a common baseline and then estimate a common within standard error among these studies. When assessing the food commodity price effect, some papers do not report the price elasticity and/or amount of change in corn production because of the introduction of ethanol. Because these parameters are important, we impute the missing elasticity data. When presenting the results, we depict results both for the sample with no imputed data and with imputed data. Similarly, when assessing the fuel price effect, we impute data on fuel price demand and supply elasticities. We then present the results assuming the original dataset, as well as the imputed one. The statistical analysis focuses on the population of studies that offer a numerical analysis that evaluates the economic and environmental effects of corn ethanol. To this end, the collection of studies is screened and grouped based on the specific effects. The list of studies reviewed for each effect is presented in the supplementary material (appendix B). Results Below we evaluate the multiple criteria used to assess the biofuel effects, starting with GHG and ILUC. GHG and ILUC Effect of Corn Ethanol The meta-analysis suggests that change in GHG emissions due to the introduction of biofuels depends on the specific geographic location, biofuel feedstock, and type of land used to grow the crop (Fargione 2008), and that there is much variability among studies. A reference point to evaluate the GHG effect of biofuels is GHG emissions of gasoline (i.e., 3kg CO2 per liter). When comparing the simple average of corn ethanol in a liter of gasoline-equivalent units to this reference point (i.e., comparing ethanol to gasoline while holding energy content per liter constant), the literature suggests that, on average, the direct effect of corn ethanol is 66% that of the GHG emissions of gasoline, and that the indirect effect is 28% (the direct effect is emissions throughout the supply chain that are directly related to the production of corn ethanol; indirect emissions are the consequence of activities pertaining to the production of corn ethanol). However, one study in our sample suggests a substantially higher indirect effect than the rest. If we remove that study (i.e., remove Searchinger et al. 2008, from our sample), then the indirect effect drops to 23%, suggesting that corn ethanol reduces GHG emissions by 11% when compared to gasoline. We also observe a large decline in ILUC emission calculations over time. The literature recognizes the unintended effect of corn ethanol on land use and its ILUC in 2008 (Fargione 2008; Searchinger et al. 2008). However, since 2008, ILUC emission calculations have declined several-fold; while Searchinger et al.’s (2008) results suggest that ILUC emission is more than 2.2 kg CO2 per liter, Hertel et al.’s (2010) calculations indicate ILUC emission at 0.57 to 0.85 kg CO2 per liter. We suspect that over time the estimates of ILUC emissions have become more accurate because of better data, more detailed modeling of the economics and the environment (e.g., inclusion of market-mediated effects), and/or learning-by-doing within the ethanol industry, which has resulted in both cleaner production processes and more cost-efficiency. Table 1 depicts the various variables of interest, where 59 observations are collected from 15 studies (see appendix B.1). While the simple average of corn ethanol results in the numerical model saving, on average, −0.35 kg CO2 equivalence per liter of corn ethanol used, when correcting the analysis for publication bias and calculating the weighted average, the GHG saving becomes much smaller, albeit negative and significant (see table 7). Table 1. Descriptive Statistics of the GHG Effect Variable Observation Mean Standard Deviation Min. Max. kg CO2 per liter 59 −0.35 0.78 −2.52 2.71 Year calibrated 59 3.42 3.27 0 12 Period of analysis 59 2012.95 11.87 2000 2030 Peer reviewed 59 0.66 0.48 0 1 CGE model 59 0.71 0.46 0 1 United States explicitly modeled 59 0.78 0.42 0 1 Energy market dummy 59 0.68 0.47 0 1 Economic aggregate index 59 3.31 4.13 0 15 Emission and energy aggregate index 59 1.10 2.37 0 10 Variable Observation Mean Standard Deviation Min. Max. kg CO2 per liter 59 −0.35 0.78 −2.52 2.71 Year calibrated 59 3.42 3.27 0 12 Period of analysis 59 2012.95 11.87 2000 2030 Peer reviewed 59 0.66 0.48 0 1 CGE model 59 0.71 0.46 0 1 United States explicitly modeled 59 0.78 0.42 0 1 Energy market dummy 59 0.68 0.47 0 1 Economic aggregate index 59 3.31 4.13 0 15 Emission and energy aggregate index 59 1.10 2.37 0 10 Table 7. Summary of Results Obtained via Non-Parametric Trim and Fill Analysis Effect Pooled estimate 95% confidence interval Lower Upper GHG effect −0.23%*** −0.30 −0.16 Food commodity price 13.65%*** 11.21 16.09 Fuel price effect −4.48%*** −5.50 −3.49 Effect Pooled estimate 95% confidence interval Lower Upper GHG effect −0.23%*** −0.30 −0.16 Food commodity price 13.65%*** 11.21 16.09 Fuel price effect −4.48%*** −5.50 −3.49 Note: When calculating the average effect of corn ethanol on fuel prices, we used the dataset with the imputed values. Because theory cannot guide the choice of statistical model, we use BMA to account for model uncertainty. The analysis suggests that the empirical specification should not include the period of calibration and whether the paper is peer reviewed, while multicollinearity rules out period of analysis. Because of significant heterogeneity among the assumptions guiding the different models, we also test for heterogeneity. However, our heterogeneity statistic I2=100%×Q-dfQ=31%, where Q denotes Cochran’s Q (Higgins and Thompson 2002; Higgins et al. 2003) suggests low levels of heterogeneity.5 In addition, the Hausman test does not reject the null at a 10% significance level, suggesting that the random-effect coefficients are consistent. We use these steps to choose our statistical model and the preferred method of estimation. The meta-regression analysis outcomes are depicted in table 2. In models I through III of table 2, the dependent variable is the level of change in GHG emissions (in kg of CO2 per liter), while the last column of table 2 (i.e., model IV) depicts an alternative specification where the variable of interest is the percentage change in GHG emissions. Table 2. The Impact of Corn Ethanol on GHG Emissions: a Meta Regression Analysis Features of the analysis OLS (model I) Frequency-weighted approach (model II) Random-effect approach (model III) Random-effect approach / percentage (model IV) Model: Frequency approach X Random-effect approach X X Level of change as dependent variable X X X Percentage as dependent variable X Variables: Energy markets dummy −0.41 −0.52 −0.47 −18.09 0.33 0.37 0.45 13.65 United States explicitly modeled 1.3*** 1.27*** 0.69 38.34*** 0.38 0.42 0.44 14.14 CGE models −0.52** −0.94* −0.64* −30.56** 0.24 0.53 0.38 13.87 Economic index 0.03 0.006 0.07*** 1.43** 0.03 0.03 0.02 0.85 Emission index 0.14** 0.16* 0.14*** 5.43*** 0.06 0.09 0.05 1.93 Constant 0.96** −0.65 −0.23 −14.72 0.42 0.62 0.51 18.97 Statistics: N 59 59 59 59 R2 0.33 0.47 Wald chi2 24.00*** 23.83*** R2 within 0.32 0.23 between 0.35 0.47 overall 0.28 0.35 Features of the analysis OLS (model I) Frequency-weighted approach (model II) Random-effect approach (model III) Random-effect approach / percentage (model IV) Model: Frequency approach X Random-effect approach X X Level of change as dependent variable X X X Percentage as dependent variable X Variables: Energy markets dummy −0.41 −0.52 −0.47 −18.09 0.33 0.37 0.45 13.65 United States explicitly modeled 1.3*** 1.27*** 0.69 38.34*** 0.38 0.42 0.44 14.14 CGE models −0.52** −0.94* −0.64* −30.56** 0.24 0.53 0.38 13.87 Economic index 0.03 0.006 0.07*** 1.43** 0.03 0.03 0.02 0.85 Emission index 0.14** 0.16* 0.14*** 5.43*** 0.06 0.09 0.05 1.93 Constant 0.96** −0.65 −0.23 −14.72 0.42 0.62 0.51 18.97 Statistics: N 59 59 59 59 R2 0.33 0.47 Wald chi2 24.00*** 23.83*** R2 within 0.32 0.23 between 0.35 0.47 overall 0.28 0.35 Note: Asterisks *** denote p-value < 0.01, ** denotes p-value < 0.05, and * denotes p-value < 0.10. Our findings suggest that studies explicitly modeling US markets report results about 1.30 kg CO2 per liter higher than otherwise (models I and II, table 2). When employing the alternative specification, where the dependent variable is the percentage change, numerical models that explicitly model the U.S. yield estimates of GHG emissions that are 38% higher than otherwise (model IV, table 2). The analysis also suggests that the energy and emission characteristics of the system matter and that assuming natural gas boilers and dry grains results in much lower GHG footprints than the scenarios that assume coal-fired boilers and wet grains (models I through IV, table 2). Regarding the economic index (models III and IV, table 2), introducing ILUC yields higher GHG emissions, but assuming the expansion of corn production affects the intensive margins (not the extensive ones) will mitigate the GHG effect of ILUC. The economic index also decreases as demand becomes more elastic. Finally, energy markets yield a rebound effect whereby the expected environmental benefits are smaller than those calculated holding consumption constant (Lambin and Meyfroidt 2011) and thus result in a larger economic index and the larger GHG footprint of corn ethanol. Our analysis supports this conjecture. Finally, explicitly modeling economic linkages among markets through CGE models results in a lower GHG footprint for corn ethanol than otherwise (table 2). The results derived above are replicated in section A of appendix A while employing other techniques and methods. In addition, because of concern for the selective publication of studies, we evaluate the possibility of a small-study effect and find some support for publication bias.6 When testing for publication bias, and different than the meta-regression above, we aggregate the data to the study level. Using the aggregated data, we plot the contour-enhanced funnel plot—see figure 1 (Palmer et al. 2008), and use the Egger’s test (Egger et al. 1997). The Egger’s test rejects the hypothesis of no small-study effect at the 1% significance level. This suggests a publication bias whereby the reported GHG emission of corn ethanol is biased downward. When correcting for the bias, the weighted GHG footprint of corn ethanol is, on average, only marginally lower than that of gasoline (table 7). However, as the California regulation acknowledges and this work shows, assumptions on various economic, energy, and emissions parameters are key to calculating the GHG footprint of corn ethanol and different assumptions result in very different GHG footprints. Figure 1. View largeDownload slide The enhanced confunnel plot for GHG Figure 1. View largeDownload slide The enhanced confunnel plot for GHG Corn Commodity Prices Claims that corn ethanol is responsible for rising food prices through diverting grains that would have been consumed as food and feed to ethanol have raised serious questions regarding the use of corn ethanol (Chakravorty, Hubert, and Nøstbakken 2009). The understanding that corn ethanol does not have a substantial impact on the transportation sector’s GHG emissions has resulted in a demand to stop support for corn ethanol because of its negative effect on food consumption.7 Does the existing literature support these claims? We investigate the food vs. fuel debate and the claims that corn ethanol significantly strains corn prices. Our dependent variable is the impact of corn ethanol on corn prices, while the independent variables include data on the following: assumed demand and supply elasticity of various crops; whether the analysis is focused on the United States or also includes the rest of the world; the year to which data are calibrated; the period of analysis; a dummy that equals 1 for food commodity price inflation (2007/2008); the amount of corn ethanol introduced; the policy simulated (a dummy that equals 1 if a mandate is assumed); and assumptions on type of analysis (i.e., multimarket/general equilibrium vs. other structures). A summary of the various variables used in the analysis is depicted in table 3, with the studies employed presented in appendix B. In addition, henceforth, fuel and gasohol are considered the same, where gasohol is a mixture of gasoline and ethanol—ethyl alcohol—used as fuel in internal combustion engines. Table 3. Descriptive Statistics of the Corn Price Effect Variable Observation Mean Standard Deviation Min. Max. Change in corn prices 273 12.96 48.62 −0.85 439.76 Fuel market dummy 292 0.48 0.50 0 1 Rest of the World (ROW) dummy 292 0.63 0.48 0 1 Upstream oil market modeled 292 0.22 0.42 0 1 Elasticity of demand 107 −0.53 0.40 −1.67 −0.16 Elasticity of supply 61 0.24 0.12 0.14 0.50 Year calibrated 300 2006.51 4.03 2000 2012 Dummy for 2007/2008 311 0.21 0.41 0 1 Mandate policy 300 0.62 0.49 0 1 Period of analysis 300 10.53 10.17 0 50 Peer reviewed 292 0.46 0.50 0 1 Change in volume of corn supplied to the market (M Bushels) 101 1,261.50 1,298.34 −2,045.17 4,928.57 Variable Observation Mean Standard Deviation Min. Max. Change in corn prices 273 12.96 48.62 −0.85 439.76 Fuel market dummy 292 0.48 0.50 0 1 Rest of the World (ROW) dummy 292 0.63 0.48 0 1 Upstream oil market modeled 292 0.22 0.42 0 1 Elasticity of demand 107 −0.53 0.40 −1.67 −0.16 Elasticity of supply 61 0.24 0.12 0.14 0.50 Year calibrated 300 2006.51 4.03 2000 2012 Dummy for 2007/2008 311 0.21 0.41 0 1 Mandate policy 300 0.62 0.49 0 1 Period of analysis 300 10.53 10.17 0 50 Peer reviewed 292 0.46 0.50 0 1 Change in volume of corn supplied to the market (M Bushels) 101 1,261.50 1,298.34 −2,045.17 4,928.57 When addressing model uncertainty, we conclude that the ROW dummy variable should not be included in the final empirical model. In addition, multicollinearity results in dummy variables denoting the fuel commodity price inflation of 2007/2008 and peer-reviewed papers being dropped from the analysis. Regarding heterogeneity among studies, the analysis suggests that there is not much ( I2=46%). The data imputation for demand and supply elasticities and for change in corn production uses multivariate normal regression techniques.8 The regression results, assuming that the dependent variable is the level of change in corn prices, is depicted in table 2, models I and II. In these models, the dependent variable is the price difference in constant 2005 U.S. dollars relative to the baseline scenario. On the other hand, the outcome of the regression assuming the dependent variable is the percentage change in the price of corn is depicted in table 2, models III and IV, where the imputed data are used in model IV. We do not report the cluster outcomes because of data limitations resulting in the model not being of sufficient rank, and do not report the fixed-effect and random-effect models because neither is significant. Instead, we focus on the Frequency-Weighted approach models. Overall, the outcomes of the meta-regression suggest that modeling economic linkages reduces the food commodity price effect and results in smaller estimates of food commodity price fluctuations. More specifically, the estimated parameters suggest the following: First, economic linkages among markets matter, and their introduction yields smaller price effects. Modeling the oil/petroleum market allows petroleum producers to respond to prices and changes in demand and thus mitigates the effect of biofuels on demand for corn (table 4, upstream oil market). Second, the effect of introducing biofuels on corn prices increases as the year used to calibrate the numerical model increases because the volume of corn ethanol introduced to the market is larger (table 4, year calibrated). Third, elasticities matter, and a more elastic demand curve results in a smaller price effect. Fourth, the mandate effect on food prices is smaller than that of a tax credit (see table 4, mandate dummy). Table 4. Effect of Corn Ethanol on Food Commodity Prices Features of the analysis OLS (model I) Frequency-weighted approach (model II) Frequency-weighted approach / percentage (model III) Frequency-weighted approach /percentage/imputed (model IV) Model: Frequency approach X X X Data imputed X Level of change as dependent variable X X Percentage as dependent variable X X Variables: Upstream oil markets −1.88*** −1.46** −47.42** −6.87 0.47 0.64 20.81 6.28 Fuel market 0.81 0.41 33.63 −6.89 0.50 0.60 20.05 4.66 Change in corn (M of bushels) −7.17 x 10−4*** 4.12 x 10−4 −0.02 1.85 x 10−3 1.95 x 10−4 3.74 x 10−4 0.01 1.38 x 10−3 Demand elasticity 2.09*** 1.77*** 56.00*** 4.95 0.60 0.37 10.64 5.47 Supply elasticity 0.20 −0.14 24.78 −9.26 1.32 0.61 16.38 18.80 Period of analysis 0.06 0.04* 1.96* 0.32** 0.04 0.02 1.09 0.14 Year calibrated 0.26*** 0.21** 9.63*** 1.38*** 0.06 0.08 2.63 0.43 Mandate dummy −1.21*** −0.88* −44.69** −9.75** 0.40 0.42 15.90 4.18 Constant −528.41*** −416.78** 19,268.03*** −2,728.59*** 129.51 159.27 5279.11 866.62 N 20 20 20 108 R2 0.78 0.68 0.74 0.22 Features of the analysis OLS (model I) Frequency-weighted approach (model II) Frequency-weighted approach / percentage (model III) Frequency-weighted approach /percentage/imputed (model IV) Model: Frequency approach X X X Data imputed X Level of change as dependent variable X X Percentage as dependent variable X X Variables: Upstream oil markets −1.88*** −1.46** −47.42** −6.87 0.47 0.64 20.81 6.28 Fuel market 0.81 0.41 33.63 −6.89 0.50 0.60 20.05 4.66 Change in corn (M of bushels) −7.17 x 10−4*** 4.12 x 10−4 −0.02 1.85 x 10−3 1.95 x 10−4 3.74 x 10−4 0.01 1.38 x 10−3 Demand elasticity 2.09*** 1.77*** 56.00*** 4.95 0.60 0.37 10.64 5.47 Supply elasticity 0.20 −0.14 24.78 −9.26 1.32 0.61 16.38 18.80 Period of analysis 0.06 0.04* 1.96* 0.32** 0.04 0.02 1.09 0.14 Year calibrated 0.26*** 0.21** 9.63*** 1.38*** 0.06 0.08 2.63 0.43 Mandate dummy −1.21*** −0.88* −44.69** −9.75** 0.40 0.42 15.90 4.18 Constant −528.41*** −416.78** 19,268.03*** −2,728.59*** 129.51 159.27 5279.11 866.62 N 20 20 20 108 R2 0.78 0.68 0.74 0.22 Note: Asterisks *** denote p-value < 0.01, ** denotes p-value < 0.05, and * denotes p-value < 0.10. The results are replicated using alternative specifications and depicted in section B of appendix A. We also evaluate the possibility of publication bias while focusing on the average effect per study; that is, different from the meta-regression above, when evaluating the plausibility of publication bias, we focus on the average per-study effect. When focusing on the percentage change analysis with data imputations, we plot a contour-enhanced funnel plot (figure 2). We also use both the Egger’s test and the Begg’s test. In both tests, the null hypothesis, which states that there is no small-study effect, cannot be rejected at the 10% significance level. Figure 2. View largeDownload slide The enhanced funnel plot for food Figure 2. View largeDownload slide The enhanced funnel plot for food Before concluding this section, we calculate a few basic statistics focusing on food prices as opposed to commodity food prices (i.e., corn commodity prices). While looking at all biofuel feedstocks surveyed in the literature and focusing on food prices, the literature indicates a small effect on U.S. consumers. The estimated effect of biofuels on the U.S. consumer price index averages less than 1 percentage point with not much variation among studies (Anderson et al. 2008; Collins 2008; Glauber 2008; Gecan, Johansson, and FitzGerald 2009). Measures such as the RFS divert resources of land away from food and feed, but the literature argues that their impact on the price of food commodities is moderate at best, with most of the effect washing away as we move through the supply chain, with minimal effect on the U.S. end-consumer. This, however, may change in the short run, as documented in the literature on the food commodity spike of 2007/2008 (McPhail and Babcock 2008). Fuel Prices The collection of data pertaining to the effect of introducing biofuels on fuel prices encompasses changes in fuel prices, fuel demand and supply elasticity, whether the study is peer reviewed, the year used to calibrate the model, whether a general equilibrium framework is employed, and whether the analysis focuses on the U.S. or also includes the rest of the world (table 5).9 Table 5. Descriptive Statistics: Fuel Prices Variable Observations Mean Standard Deviation Min. Max. Change in fuel prices 52 −0.12 0.10 −0.42 0.03 Demand elasticity of fuel 35 −1.12 1.61 −10 −0.2 Fuel supply elasticity 31 0.86 0.87 0.1 5 Model used 52 3.37 0.69 2 4 CGE model 52 0.48 0.50 0 1 Upstream oil markets 52 0.58 0.50 0 1 Corn prices do not affect fuel prices 52 0.04 0.19 0 1 ROW 52 0.40 0.50 0 1 Period of analysis 52 7.96 5.63 0 15 Year calibrated 52 2007.62 2.39 2000 2010 Variable Observations Mean Standard Deviation Min. Max. Change in fuel prices 52 −0.12 0.10 −0.42 0.03 Demand elasticity of fuel 35 −1.12 1.61 −10 −0.2 Fuel supply elasticity 31 0.86 0.87 0.1 5 Model used 52 3.37 0.69 2 4 CGE model 52 0.48 0.50 0 1 Upstream oil markets 52 0.58 0.50 0 1 Corn prices do not affect fuel prices 52 0.04 0.19 0 1 ROW 52 0.40 0.50 0 1 Period of analysis 52 7.96 5.63 0 15 Year calibrated 52 2007.62 2.39 2000 2010 The simple average of the change in fuel prices caused by the introduction of corn ethanol suggests that introducing corn ethanol yields a decline of 12 U.S. cents in the price of fuel in constant 2005 U.S. dollars. However, some studies suggest that corn ethanol results in fuel prices increasing with the introduction of biofuels (e.g., de Gorter and Just 2015) while others suggest it leads to a decline in fuel prices (e.g., Chen 2010): The average reported in Gallagher et al. (2003) is -29, and that reported in Chen et al. (2010) is -23; de Gorter and Drabik (2011) report an averge of 20 U.S. cents. We hypothesize that these variations among studies stem from modeling differences between food and fuel prices. While both Gallagher et al. (2003) and Chen et al. (2010) assume (implicitly) that corn ethanol affects fuel prices, de Gorter and Drabik (2011) assume that fuel prices affect corn prices but not the opposite. To better understand the implications of this modeling assumption, we add a dummy variable (the Corn-Fuel Price dummy) that equals 1 if corn prices follow fuel prices but not vice versa, and 0 otherwise.10 When selecting the model, the analysis suggests that the empirical model should not include the dummy variable ROW. In addition, and different from the outcome of the other two meta-regressions, we detect high levels of heterogeneity among studies ( I2=82%). When comparing the different methods used to estimate the parameters, the fixed-effect model is not significant at a 10% significance level, while the Chi statistic of the random-effect model equals 63.14 and is significant at the 1% significance level. The data imputation of demand and supply elasticities follows the techniques used above when estimating the effect of corn ethanol on corn prices. The outcome of the analysis is depicted in table 6, where we start with two different specifications of the dependent variable: models I through IV depict the outcome of the regression, assuming the dependent variable is the change in fuel prices in constant 2005 U.S. dollars; model V depicts the alternative specification, whereby the dependent variable is the percentage change in fuel prices. Table 6. Effect of Corn Ethanol on Fuel Prices Features of the analysis Frequency approach / price linkages (model I) Frequency approach / CGE modeling (model II) Frequency approach / CGE modeling / imputed (model III) Random-effect approach / CGE modeling (model IV) Frequency approach / CGE modeling / percentage (model V) Model: Frequency approach X X X X Random-effect approach X Data imputed X Level of change as dependent variable X X X X Percentage as dependent variable X Variables: Demand elasticity 0.12 0.08*** 0.04*** 0.09*** 3.36*** 0.11 0.03 0.1 0.02 1.25 Supply elasticity 0.2 0.13 0.06* 0.16*** 5.21 0.24 0.08 0.03 0.05 3.46 Period of analysis −0.012** −0.011*** −0.008*** −0.01*** −0.35*** 0.006 0.002 0.002 0.002 0.09 Year calibrated 0.02 −0.001 −0.02*** −0.0009 0.22 0.05 0.006 0.005 0.016 0.26 Upstream oil markets explicitly modeled 0.21 0.16*** 0.10*** 0.17*** 5.47** 0.16 0.06 0.03 0.05 2.51 Model used 0.09* 0.14 0.09 0.17*** 5.6 0.05 0.1 0.06 0.06 4.23 CGE model −0.15 −0.15* −0.19* −5.16 0.15 0.08 0.1 6.19 Corn prices do not affect fuel prices −0.012 0.28 Constant −48.22 2 32 −470.21 94.37 12.24 11.21 516.41 N 31 31 108 31 31 R2 0.7 0.71 0.22 0.68 R2 within 0.0021 between 0.8129 overall 0.7445 Features of the analysis Frequency approach / price linkages (model I) Frequency approach / CGE modeling (model II) Frequency approach / CGE modeling / imputed (model III) Random-effect approach / CGE modeling (model IV) Frequency approach / CGE modeling / percentage (model V) Model: Frequency approach X X X X Random-effect approach X Data imputed X Level of change as dependent variable X X X X Percentage as dependent variable X Variables: Demand elasticity 0.12 0.08*** 0.04*** 0.09*** 3.36*** 0.11 0.03 0.1 0.02 1.25 Supply elasticity 0.2 0.13 0.06* 0.16*** 5.21 0.24 0.08 0.03 0.05 3.46 Period of analysis −0.012** −0.011*** −0.008*** −0.01*** −0.35*** 0.006 0.002 0.002 0.002 0.09 Year calibrated 0.02 −0.001 −0.02*** −0.0009 0.22 0.05 0.006 0.005 0.016 0.26 Upstream oil markets explicitly modeled 0.21 0.16*** 0.10*** 0.17*** 5.47** 0.16 0.06 0.03 0.05 2.51 Model used 0.09* 0.14 0.09 0.17*** 5.6 0.05 0.1 0.06 0.06 4.23 CGE model −0.15 −0.15* −0.19* −5.16 0.15 0.08 0.1 6.19 Corn prices do not affect fuel prices −0.012 0.28 Constant −48.22 2 32 −470.21 94.37 12.24 11.21 516.41 N 31 31 108 31 31 R2 0.7 0.71 0.22 0.68 R2 within 0.0021 between 0.8129 overall 0.7445 Note: Asterisks *** denote p-value < 0.01, ** denotes p-value < 0.05, and * denotes p-value < 0.10. The analysis suggests that economic modeling matters and that the introduction of economic linkages leads to smaller price effects (table 6). More specifically, the estimated parameters suggest that modeling the energy system results in the introduction of corn ethanol having a smaller negative effect on the price of fuel. The difference in the effect of the introduction of corn ethanol on fuel prices is 0.10 to 0.17 U.S. dollars (table 6, models II through IV) and 5.47% (table 6, model V). Second, the effect of varying the elasticity of fuel demand is robust across models. Similar to the previous two meta-analyses, when evaluating the possibility of publication bias, we focus on the average effect per study and plot the contour-enhanced funnel plot for the percentage change in fuel prices using the imputed dataset (figure 3). We also employ the Egger’s test and the Begg’s test, but do not reject the null hypothesis, which states that there is no publication biases (at the 10% significance level). We also perform various sensitivity analyses but place these outputs in section C of appendix A. Figure 3. View largeDownload slide The enhanced funnel plot for fuel Figure 3. View largeDownload slide The enhanced funnel plot for fuel The analysis implies that the modeling of petroleum refineries reduces the effect of corn ethanol on fuel prices. This suggests that in the real world petroleum refineries are likely to respond and alleviate the effect of corn ethanol on fuel prices; see also Knittel and Smith (2015). The meta-analysis also suggests that although biofuel production increases, its effect on gasohol prices is relatively constant.11 However, the numbers imply that the refineries’ response is limited and that biofuels do make a dent in the refineries’ profit margins and negatively affect fuel prices (as well as crude oil prices). A recurring outcome of the meta-analysis is that a more detailed and explicit modeling of related markets positioned throughout the supply chain results in predicting smaller effects on prices; linkages among markets matter and mitigate the price effect of introducing corn ethanol. Here, the introduction of other petroleum markets has a smaller negative effect on fuel prices. Introducing related markets via the oil and petroleum markets results in corn ethanol having a smaller net effect on food commodity prices. Policy Discussions and Concluding Remarks We employ nonparametric trim and fill techniques (Steichen 2000) and depict in table 7 the random-effect of the pool estimates of the three effects of interest. Although policies promoting corn ethanol achieve only modest environmental improvements, if any, these policies do result in improvements to segments of the U.S. economy and its energy balance. On the one hand, the case studies presented in the literature looking at employment and economic activity generated by the introduction of biofuels in rural communities suggest that biofuels stimulate rural communities and create economic value. Although the case studies we surveyed conclude that biofuels result in a net economic benefit to rural communities, the introduction of biofuels negatively affects livestock farmers and reduces employment in conventional biomass industries (Remedio and Domac 2003; Domac, Richards, and Risovic 2005). On the other hand, simple calculations suggest that corn ethanol significantly affects the U.S. balance of trade, a parameter emphasized by U.S. Presidents Obama and Bush.12 Reducing the balance of trade deficit, specifically the difference in value between the United States’ imports and exports, has been a declared major policy objective of political leaders, and in 2011 corn ethanol substantially affected the US trade balance in the following manner.13,14 Even though the United States consumes 3.34 billion barrels of finished motor gasoline annually, in 2011 U.S. consumption of finished motor gasoline declines to 3.19 billion barrels annually, a decline of 4%. The amount of ethanol consumed in the United States in 2011 equals 67.25% of the decline of finished motor gasoline consumption from 2005 to 2011. On the other hand, production of U.S. gasoline in 2005 is 3.04 billion of barrels annually but increases to 3.31 in 2011, an increase of 9%. Using the July 2005 price per gallon of fuel of $2.333 and the July 2011 price of $3.705 suggests that the United States paid $29.40 (= 0.3×42×2.333) billion for fuel produced outside the United States in 2005 but received $18.67 ( =0.12×42×3.705) billion from foreign countries in 2011. The U.S. Census Bureau numbers suggest a U.S. trade deficit of $558 billion in 2011, where the changes documented above are equivalent to more than 8% of this deficit (i.e., 100×18.67+29.40/558=8.2%). The significant balance of trade effect of the Energy Acts of 2005 and 2007 is a major benefit from a political-economic perspective (see also Hochman, Barrow, and Zilberman 2013). The underlying considerations in the balance of trade objectives also support other elements of the U.S. energy policy. In particular, San et al. (2008) show that industrial production and employment are key determinants of the use of fossil fuels and renewables for energy consumption in the United States. Regarding the modeling of numerical models, a recurring outcome of the meta-analysis is that a more detailed and explicit modeling of related markets positioned throughout the supply chain results in the analysis predicting smaller effects on prices; economic linkages matter and reduce the price effects attributed to the introduction of corn ethanol. The introduction of other petroleum markets results in a smaller negative effect of corn ethanol on fuel prices, while introducing the oil and petroleum markets results in corn ethanol having a smaller positive net effect on food commodity prices. GHG emission savings are also smaller if economic models are used. These conclusions emphasize the importance of estimating demand and supply parameters of major agricultural and energy commodities, and of updating these estimates over time. As the literature on biofuels expands and more ex post studies are performed, the estimated effects of biofuels should be revisited. Future research should also employ similar techniques to analyze other ethanol and biodiesel feedstock in other regions of the world. Supplementary Material Supplementary material are available at American Journal of Agricultural Economics online. Footnotes 1 Data available at: http://www.eia.gov/renewable/. 2 Although the U.S. also imports ethanol, the amounts are relatively small. In 2016, ethanol imports declined on an annual basis by 60% to 36 million gallons (see http://aginfotoday.com/News/2016-US-Ethanol-Exports-Rise-2nd-Highest-Level-on-Record-2017-03-17/15770). 3 Reference to indirect land use change provoked sharp disagreement in Working Group III of the Intergovernmental Panel on Climate Change and led the group to agree to disagree. See article in Scientific American by Tiffany Stecker and ClimateWire on April 14, 2014, available at: http://www.scientificamerican.com/article/controversy-over-biofuels-and-land-cut-from-ipcc-summary/. 4 The notion of triangulation suggests that using different reference points gives decision makers a better understanding of reality. 5 The classical measure of heterogeneity is Cochran’s Q, which is calculated as the weighted sum of squared differences between individual study effects and the pooled effect across studies, with the weights being those used in the pooling method. Cochran’s Q has low power as a comprehensive test of heterogeneity (Gavaghan, Moore, and McQay 2000) when the number of studies is small, but too much power if the number of studies is large (Higgins et al. 2003). The I2 statistic, on the other hand, describes the percentage of variation across studies that is due to heterogeneity (Higgins and Thompson 2002; Higgins et al. 2003). Unlike Cochran’s Q, the I2 statistic does not depend on the number of studies considered. 6 The small study effect captures the phenomenon that studies with fewer observations show a lower GHG footprint than larger studies. 7 For example, see https://www.technologyreview.com/s/424050/do-biofuels-reduce-greenhouse-gases/. 8 These techniques, programed in STATA, use an iterative Markov chain Monte Carlo (MCMC) method, where the imputations are obtained from an iterative MCMC method that draws from the posterior predictive distribution of the missing data given the observed data. When running the MCMC, we assume 1,000 burn-ins and 10,000 iterations. For the analysis, the last set imputed is employed. 9 Because there is a difference in whether petroleum products other than fuel are introduced or food commodity markets are explicitly modeled, we think it better to explicitly introduce both into the analysis. 10 The empirical literature regarding the short-run interaction among fuel, corn ethanol, and corn prices is inconclusive (Serra and Zilberman 2013). 11 When including all data points (i.e., results of statistical as well as numerical studies), the negative effect of introducing biofuels on gasoline prices increases over time. However, when we exclude Due and Hayes’ (2008) study, there is no change over time, with biofuels, on average, resulting in fuel prices declining by 4.5%. 12 For example, when President Obama listed the ways in which the economy had recovered, he included American exports and thus the improvement in the balance of trade: “We now sell more products made in America to the rest of the world than ever before,” (Obama, August 6, 2013). Article in Politifact by Jon Greenberg, Tuesday, August 20, 2013. The article is available at http://www.politifact.com/truth-o-meter/statements/2013/aug/20/barack-obama/obama-says-exports-us-goods-all-time-high/. 13 President Obama launched the National Export Initiative to promote jobs and economic growth during his State of the Union in January 2010. President Obama committed to doubling U.S. exports in the next five years and creating 2 million well-paying jobs (Griswold 2010). President Bush in 1991 stated that he made exports one of his top priorities (Worcester Telegram and Gazette 1991), and when visiting Australia in 1992, despite protests by angry Australian farmers, President Bush said that the United States would continue subsidizing agricultural exports (Daily News of Los Angeles 1992). 14 More broadly, from a behavioral perspective, a trade deficit is viewed as a bad thing (Buffett 2003) and politicians take pride in reducing these deficits (e.g., President Obama’s State of the Union in January 2015). References Altrichter H., Feldman A., Posch P., Somekh B.. 2008. Teachers Investigate Their Work; An Introduction to Action Research across the Professions. NYC, NY: Routledge. Anderson D.P., Outlaw J.L., Bryant H.L., Richardson J.W., Ernstes D.P., Raulston J.M., Allison M.S.. 2008. The Effects of Ethanol on Texas Food and Feed. Agricultural and Food Policy Center. Texas A&M University. Berger J.O., Wolpert R.L., Bayarri M.J., DeGroot M.H., Hill B.M., Lane D.A., LeCam L.. 1988. The Likelihood Principle. Lecture Notes - Monograph Series 6: III-199. Institute of mathematical statistics. Buffet W.E., Loomis C.J.. 2003. America's Growing Trade Deficit is Selling the Nation out from under Us. Here's a Way to Fix the Problem – and We Need to Do it Now. Fortune . November 10. Chakravorty U., Hubert M.H., Nøstbakken L.. 2009. Fuel versus Food. Resource Economics 1: 645– 63. Cohen L., Manion L.. 2000. Research Methods in Education. NYC, NY: Routledge. Google Scholar CrossRef Search ADS Collins K. 2008. The Role of Biofuels and Other Factors in Increasing Farm and Food Prices. Supporting Material for a Review Conducted by Kraft Food Global. Condon N., Klemick H., Wolverton A.. 2015. Impacts of Ethanol Policy on Corn Prices: A Review and Meta-Analysis of Recent Evidence. Food Policy 51: 63– 73. Google Scholar CrossRef Search ADS Domac J., Richards K., Risovic S.. 2005. Socio-Economic Drivers in Implementing Bioenergy Projects. Biomass and Bioenergy 28 2: 97– 106. http://dx.doi.org/10.1016/j.biombioe.2004.08.002 Google Scholar CrossRef Search ADS David B.D. 2001. Commentary: Practical Advantages of Bayesian Analysis of Epidemiologic Data. American Journal of Epidemiology 153 12: 1222– 6. http://dx.doi.org/10.1093/aje/153.12.1222 Google Scholar CrossRef Search ADS PubMed Daily News of Los Angeles. 1992. Bush Backs Farm Export Subsidies – Australian Protests Say Result in Lost Jobs. Daily News of Los Angeles. January 2, Section: News, N1. Farrell A. E., Plevin R. J., Turner B. T., Jones A. D., O’hare M., Kammen D. M.. 2006. Ethanol can contribute to energy and environmental goals. Science 311 5760: 506– 508. Google Scholar CrossRef Search ADS PubMed Fargione J., Hill J., Tilman D., Polasky S., Hawthorne P.. 2008. Land Clearing and the Biofuel Carbon Debt. Science 319 5867: 1235– 8. Google Scholar CrossRef Search ADS PubMed Gavaghan D.J., Moore A.R., McQay H.J.. 2000. An Evaluation of Homogeneity Tests in Meta-Aanalysis in Pain Using Simulations of Patient Data. Pain 85: 415– 24. Google Scholar CrossRef Search ADS PubMed Gecan R., Johansson R., FitzGerald K.. 2009. The Impact of Ethanol Use on Food Prices and Greenhouse-Gas Emissions. Pub. No. 3155. Washington DC: Congress of the United States, Congressional Budget Office. Glauber J. 2008. Statement of Joseph Glauber, Chief Economist, Before the Committee on Energy and Natural Resources. United States Senate 12. Greene W.H. 2012. Econometric Analysis . 7th Edition. NJ: Pearson. Griswold D. 2010. A Free-Market ‘5-Year Plan’ to Boost U.S. Exports. Washington Times. April 27, Section: Nation A05. Havranek T., Irsova Z., Janda K., Zilberman D.. 2015. Selective Reporting and the Social Cost of Carbon. Energy Economics 51: 394– 406. Google Scholar CrossRef Search ADS Hertel T.W., Golub A.A., Jones A.D., O'Hare M., Plevin R.J., Kammen D.M.. 2010. Effects of US Maize Ethanol on Global Land Use and Greenhouse Gas Emissions: Estimating Market-Mediated Responses. BioScience 60 3: 223– 31. Google Scholar CrossRef Search ADS Higgins J.P.T., Green S., eds. 2011. Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0. NYC, NY: The Cochrane Collaboration. Higgins J.P.T., Thompson S.G.. 2002. Quantifying Heterogeneity in a Meta-Analysis. Statistics in Medicine 21: 1539– 58. Google Scholar CrossRef Search ADS PubMed Higgins J.P.T., Thompson S.G., Deeks J.J., Altman D.G.. 2003. Measuring Inconsistency in Meta-Analyses. British Medical Journal 327: 557– 60. Google Scholar CrossRef Search ADS PubMed Hochman G., Rajagopal D., Zilberman D.. 2011. The Effect of Biofuels on the International Oil Market. Applied Economic Perspectives and Policy 33 3: 402– 27. http://dx.doi.org/10.1093/aepp/ppr016 Google Scholar CrossRef Search ADS Hochman G., Barrow G., Zilberman D.. 2013. U.S. Biofuels Policy: Few Environmental Benefits but Large Trade Gains. ARE Update 17 2. November/December. Available at: http://giannini.ucop.edu/publications/are-update/issues/2013/17/2/us-biofuels-policy/. Hoeting J.A., Madigan D., Raftery A.E., Volinsky C.T.. 1999. Bayesian Model Averaging: A Tutorial. Statistical Science 382– 401. Kaltschmitt M., Reinhardt G.A., Stelzer T.. 1997. Life Cycle Analysis of Biofuels under Different Environmental Aspects. Biomass and Bioenergy 12 2: 121– 34. http://dx.doi.org/10.1016/S0961-9534(96)00071-2 Google Scholar CrossRef Search ADS Knittel C.R., Smith A.. 2014. Ethanol Production and Gasoline Prices: A Spurious Correlation. Energy Journal 36 1: 73– 113. Kruschke J.K. 2013. Bayesian Estimation Supersedes the T Test. Journal of Experimental Psychology: General 142 2: 573– 603. http://dx.doi.org/10.1037/a0029146 Google Scholar CrossRef Search ADS PubMed Lambin E.F., Meyfroidt P.. 2011. Global Land Use Change, Economic Globalization, and the Looming Land Scarcity. Proceedings of the National Academy of Sciences 108 9: 3465– 72. http://dx.doi.org/10.1073/pnas.1100480108 Google Scholar CrossRef Search ADS Loomis J.B., White D.S.. 1996. Economic Benefits of Rare and Endangered Species: Summary and Meta-Analysis. Ecological Economics 18 3: 197– 206. http://dx.doi.org/10.1016/0921-8009(96)00029-8 Google Scholar CrossRef Search ADS McPhail L.L., Babcock B.A.. 2008. Short-Run Price and Welfare Impacts of Federal Ethanol Policies. CARD Working Papers . 484. https://lib.dr.iastate.edu/card_workingpapers/484 Nakata T., Tonetti C.. 2015. Small Sample Properties of Bayesian Estimators of Labor Income Processes. Journal of Applied Economics 18 1: 121– 48. http://dx.doi.org/10.1016/S1514-0326(15)30006-4 Google Scholar CrossRef Search ADS National Research Council, Committee on Economic and Environmental Impacts of Increasing Biofuels Production. 2011. Renewable Fuel Standard: Potential Economic and Environmental Effects of U.S. Biofuel Policy. Washington DC: National Academies of Sciences. Nelson J.P., Kennedy P.E.. 2009. The Use (and Abuse) of Meta-Analysis in Environmental and Natural Resource Economics: An Assessment. Environmental and Resource Economics 42 3: 345– 77. http://dx.doi.org/10.1007/s10640-008-9253-5 Google Scholar CrossRef Search ADS O'Donoghue T., Punch K.. 2003. Qualitative Educational Research in Action: Doing and Reflecting. Routledge, Taylor & Francis Group. Owen A. B. 2017. Statistically efficient thinning of a Markov chain sampler. Journal of Computational and Graphical Statistics 26 3: 738– 744. Persson U.M. 2015. The Impact of Biofuel Demand on Agricultural Commodity Prices: a Systematic Review. WIREs Energy and Environ 4: 410– 28. Google Scholar CrossRef Search ADS Rajagopal D. 2014. Consequential Life Cycle Assessment of Policy Vulnerability to Price Effects. Journal of Industrial Ecology 18 2: 164– 75. http://dx.doi.org/10.1111/jiec.12058 Google Scholar CrossRef Search ADS Remedio E.M., Domac J.U.. 2003. Socio-Economic Analysis of Bioenergy Systems: A Focus on Employment . FAO Forestry Department, Wood Energy Programme. Rome: FAO. Robert C.P. 2001. The Bayesian Choice: From Decision Theoretic Foundations to Computational Implementation. New York: Springer-Verlag. Sari R., Ewing B.T., Soytas U.. 2008. The Relationship between Disaggregate Energy Consumption and Industrial Production in the United States: An ARDL Approach. Energy Economics 30 5: 2302– 13. http://dx.doi.org/10.1016/j.eneco.2007.10.002 Google Scholar CrossRef Search ADS Schultz T.W. 1964. Transforming Traditional Agriculture . New Haven: Yale University Press. Searchinger T., Heimlich R., Houghton R.A., Dong F., Elobeid A., Fabiosa J., Tokgoz S., Hayes D., Yu T.H.. 2008. Use of U.S. Croplands for Biofuels Increases Greenhouse Gases through Emissions from Land-Use Change. Science 319 5867: 1238– 40. Google Scholar CrossRef Search ADS PubMed Serra T., Zilberman D.. 2013. Biofuel-Related Price Transmission Literature: A Review. Energy Economics 37: 141– 51. Google Scholar CrossRef Search ADS Stanley T.D. 2001. Wheat from Chaff: Meta-Analysis as Quantitative Literature Review. Journal of Economic Perspectives 15 3: 131– 50. http://dx.doi.org/10.1257/jep.15.3.131 Google Scholar CrossRef Search ADS Stanley T.D., Doucouliagos H., Giles M., Heckemeyer J.H., Johnston R.J., Laroche P., Rosenberger R.S.. 2013. Meta-Analysis of Economics Research Reporting Guidelines. Journal of Economic Surveys 27 2: 390– 4. Google Scholar CrossRef Search ADS StataCorp. 2015. Stata Bayesian Analysis Reference Manual: Release 14. College Station, TX: StataCorp LP. Steichen T. 2000. Nonparametric Trim and Fill Analysis of Publication Bias in Meta-Analysis. Stata Technical Bulletin , STB57, sbe39, 8– 14. Stoft S. 2010. Renewable Fuel and the Global Rebound Effect. Global Energy Policy Center Research Paper (10–06). Thompson J. 2014. Bayesian Analysis with Stata . Stata Press. Tol R.S.J. 2009. The Economic Effect of Climate Change. Journal of Economic Perspectives 23 2: 29– 51. http://dx.doi.org/10.1257/jep.23.2.29 Google Scholar CrossRef Search ADS Tyner W.E., Taheripour F., Zhuang Q., Birur D., Baldos U.. 2010. Land Use Changes and Consequent CO2 Emissions Due to U.S. Corn Ethanol Production: A Comprehensive Analysis. Department of Agricultural Economics, Purdue University. Worcester Telegram, and Gazette. 1991. Bush: Free Trade Will Bring Jobs to State. Worcester Telegram and Gazette May 25, Section: News 1. Zhang W., Yu E., Rozelle S., Yang J., Msangi S.. 2013. The Impact of Biofuel Growth on Agriculture: Why Is the Range of Estimates so Wide? Food Policy 38: 227– 39. Google Scholar CrossRef Search ADS © The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For Permissions, please email: email@example.com
American Journal of Agricultural Economics – Oxford University Press
Published: Mar 1, 2018
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
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
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.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera