Abstract
The agri-environmental schemes (AESs) of the European Union employ the highest share of the public budget allocated to rural development programs. Our study applies a difference-indifferences propensity score matching estimator to perform a comparative analysis of the effects of these schemes on farmers’ performance across five E.U. member states. The effects of the AESs adoption largely depend on the share of the agri-environmental payment on farm revenue. If this share is larger than 5%, participation in AESs is effective in promoting greener farming practices in all countries but Spain, where a negative effect on farm income is also shown. (JEL Q15, Q18)
I. INTRODUCTION AND BACKGROUND
In recent years in developed countries there has been increasing concern about the relationship between agriculture and the environment. In the European Union, the Common Agricultural Policy has introduced several agri-environmental measures in order to discourage negative externalities and to promote positive externalities of agricultural activities. While negative externalities are sanctioned by a reduction in direct income payments if farmers do not comply with cross-compliance requirements, positive externalities are encouraged by some rural development measures that promote environmentally sustainable farming practices through payments that compensate farmers for the provision of environmental goods that the market does not reward. One of these measures are the agri-environmental schemes (AESs), which were introduced in the late 1980s as an option to be applied by member states and nowadays are the most important part of the rural development programs. The agri-environmental and animal welfare programs are the only measures that are compulsory in all rural development programs, and they absorb the highest share of E.U. funding for rural development both at the E.U. level and in most member states. In addition, the E.U. budget allocated to these measures has shown an upward trend over the years: in 2010 the E.U. money allocated to these schemes amounted to €3.03 billion compared to only €76 million in 1993, the year in which AESs were introduced into the Common Agricultural Policy framework. If we analyze the share of utilized agricultural area (UAA) under agri-environmental contracts in 2009, there are important differences across member states. While among the first 25 countries to join the European Union the average share was 25.2%, Luxembourg and Finland were the countries with the highest share (more than 90%), followed by Sweden (82.3%) and Austria (69.6%).
The literature on AESs is quite extensive. Most of it tries to analyze the factors affecting farmers’ participation in agri-environmental contracts (see, among others, Vanslembrouck, Van Huylenbroeck, and Verbeke 2002; De Francesco et al. 2007). Another widely studied topic is the analysis of the environmental effectiveness of farmers’ environmental practices; most of these studies outline the importance of accounting for farm heterogeneity by applying farm-specific measures (Aakkula et al. 2012). A few studies analyze the effects of AESs on farms’ practices and economic results. Sauer, Walsh, and Zilberman (2012, 6) argue that “only a few studies so far have attempted to empirically measure the actual impact of being subject to AESs on producer behavior at individual farm level using statistical or econometric tools.” The expectation is that farmers are heavily affected by participation in AESs, which may lead to a deep reorganization of the farm and to a change in the relative importance of farm income sources.
An expost tool recently applied to analyze the effects of agricultural policy measures on farms’ performances is propensity score matching (PSM). Propensity score analysis has been widely developed in the last 30 years as a program evaluation method based on observational data in a broad range of disciplines such as medicine, epidemiology, psychology, social sciences, and education. More recently it has also been applied to environmental economics and to the analyses of some measures of rural development programs. The most recent applications integrate PSM with a difference-in-differences (DID) estimator. Pufahl and Weiss (2009), Chabé-Ferret and Subervie (2013), and Udagawa, Hodge, and Reader (2014) apply such a DID-PSM estimator in order to evaluate the effect of AESs on farm choices. Pufahl and Weiss (2009) analyze the impact of AESs adoption on input use and output produced by a large sample of German farms observed over the period 2000–2005. Their work shows that farmers participating in AESs experience larger positive growth rates in sales, on-farm labor, area under cultivation, grassland, and rented land compared to the non-participating farmers; by contrast, they have larger negative growth rates in livestock den-sity and fertilizer and pesticide expenditure. Chabé-Ferret and Subervie (2013) investigate the windfall effect and the additional effect of seven agri-environmental measures implemented in France over the period 2000–2005, and they perform a cost-benefit analyses of each measure. An agri-environmental measure produces an additional effect if it pro-motes practices that the farmer would not have adopted in the absence of the scheme, even if those practices are not the specific target of that AES; by contrast, an agri-environ- mental measure generates a windfall effect if it encourages practices that would have been adopted even in the absence of the scheme.1 They find that AESs requiring small changes in the farm production plan suffer windfall effects and they are not cost-effective. By contrast, those schemes that require deeper changes in farmers’ practices, such as the option to convert to organic farming, produce higher additionality. The authors also provide a lower bound of the effect of AESs, and they test crossover effects and anticipation effects. Udagawa, Hodge, and Reader (2014) combine PSM with the DID estimator to investigate the impact of the adoption of AESs on income of a cereal farm sample in eastern England. They find a negative effect of participation in AESs on total farm business income but not on purely agricultural income. This provides evidence that participation in AESs in England is considered an alternative to nonagricultural activities. In addition, the negative effect diminishes over time, thus making the analyses of decreasing agri-environmental payments an interesting research topic.
The PSM estimator has been also applied to compare voluntary and compulsory environmental measures in terms of their impact on farm production choices. Sauer, Walsh, and Zilberman (2012) find that voluntary agri-environmental measures affect farmers’ decisions more strongly than nonvoluntary measures. Their analysis on a sample of U.K. cereal farms shows that farmers participating in AESs do not reduce their efficiency, as they efficiently adjust their production plan to the new constraints, especially by becoming less specialized and more diversified. The use of fertilizers and chemicals decreases, as well as land and capital productivity, while labor productivity increases. Jaraitė and Kažukauskas (2012), applying a backward DID, show that farmers participating in voluntary agri-environmental measures reduce chemical pollution more compared to farmers not subject to voluntary agri-environmental programs, indicating a cross-positive effect between com pulsory and voluntary measures, while the share of farm direct payment on total farm revenue does not affect the degree of compliance with compulsory measures.
The PSM estimator has been also used to analyze some rural development measures in North America. In a recent study, Lawley and Towe (2014) apply the PSM estimator to investigate the impact of perpetual conservation easements on land value in the Manitoba region of Canada, which is characterized by intensive production. The results show that conservation programs are effective in protecting existing habitat vulnerable to conversion, while conservation easements lead to a drop in land values, which is adequately compensated by the payment for these easements. Liu and Lynch (2011) study the impact of development-right programs on preventing farmland loss in six Mid-Atlantic U.S. states, while Tamini (2011) analyzes the effect of extension advisory activities on farmers’ adoption of best management practices in Québec. His results show that such activities increase the environmental performance of farmers by increasing the rate of adoption of best management practices, and this increase is larger in the case of practices related to compulsory regulations.
Despite the increasing use of PSM methods in this area, to the best of our knowledge there are no studies that compare the effects of agri-environmental contracts on farmers’ choices and economic performance in different E.U. countries. In addition, none of the studies found in the literature focuses on the impact of AESs on farm economic variables such as farm income. The only exception is a study by Udagawa, Hodge, and Reader (2014) who confine their attention to the effect of AESs on farm income in eastern England, not comparing different member states and not considering environmental effectiveness.
This paper aims to fill this gap by applying a DID PSM estimator in order to perform comparative analyses of the effects of AESs on farmers’ practices and economic performance across five E.U. member states (United Kingdom, Spain, Italy, France, and Germany) over the period 2003–2006. The comparison among E.U. countries may indicate the areas where AESs adoption is more effective in promoting greener farming practices and where the effects on farm income are stronger. In addition, a comparison between the effects of AESs participation on farm income may shed light on the appropriateness of the level of the agri-environmental payment to compensate the potential income foregone.
II. AGRI-ENVIRONMENTAL SCHEMES
E.U. agri-environmental contracts are voluntary contracts of at least 5 years, stipulated between the farmer and the government; under these contracts, the farmer provides environmental goods that go beyond the minimum requirements of cross-compliance and of the European and national compulsory environmental regulations, and the farmer receives a fixed per hectare payment to compensate for the additional costs and the loss of income linked to these commitments. The main objective of AESs is to reduce agricultural pollution risks as well as protect biodiversity and the landscape. Agri-environmental payments are cofinanced by member states. The institutional setting of the AESs differs across member states; in some E.U. countries they are defined mainly at the national level, while in other countries they are defined and implemented at the regional or subregional level. This flexibility should allow taking into account the heterogeneity of the natural characteristics and agricultural systems throughout the E.U. member states. For example, in France the agri-environmental programs are set at the national level, whereas in the other countries considered in this study they are managed mainly at the regional level. In particular, they are defined at the level of states in the United Kingdom, of federal states in Germany, and of regions in Italy and in Spain; however, the latter has a set of mandatory measures that are applied countrywide.
Agri-environmental and animal welfare programs are the only measures that are compulsory in all E.U. rural development programs, and in the budget periods for both 2000–2006 and 2007–2013, they absorbed the highest share of E.U. funding for rural development. In the budget period 2000–2006 this share was 29%, followed by “less favored area and area with environmental restrictions” (17%) and “investment in farms” (8.7%) (KMC-IRDR 2012). If we consider the share of the rural development public budget devoted to AESs we find very different situations among member states; nonetheless, for the budget period 2000–2006, in 18 out of 25 E.U. countries, AESs and less-favored area measures represented the largest share. The E.U. countries with the highest share were Sweden (84.7%), Austria (65.4%), and Italy (50.9%), and in nine countries this share was larger than 40%.
AESs are expected to differently affect farmers in the five countries considered in this study, given the different climatic conditions, the different characteristics of agriculture, and the different national implementations of AESs. In addition, the differences in the participation rate and in the average per hectare payment may explain the differences in farm performance. In 2006, Germany was the country with the highest share of UAA covered by AESs (42.1%) followed by the United Kingdom (37%), France (24%), Spain (13.8%), and Italy (12.3%). The effect of the AESs’ adoption on farmers’ income depends also on the level of the agri-environmental payment and on the compliance costs: in the countries where the costs of compliance are low, a small payment may be enough to compensate the income foregone, while in the countries where the adoption of AESs requires higher compliance costs, the level of the payment should be higher. In 2005 (no homogeneous data for 2006 are available), farmers in the United Kingdom received the largest payment per agri-environmental contract (€9,258), while contracts in France were rewarded on average by only €1,728 ; the average payment per agri-environmental contract in Italy was €3,607, while German and Spanish farmers received €2,806 and €1,846 per contract, respectively. If we consider the average per hectare payment of land under AESs we find that Italy was the country with the highest payment (€176/ha) followed by the United Kingdom (€130/ha), Germany (€112/ha), Spain (€66/ha), and France (€55/ ha) (Directorate-General for Agriculture and Rural Development 2007). Other data that may explain the differences in the effects of AESs concerns the most adopted measure in each country. Data available from the Directorate-General for Agriculture and Rural Development (2007) provide a breakdown of AESs in broad categories: landscape and nature conservation measures were the most ap-plied in France (52%), Spain (58%), and the United Kingdom (87%), while in Germany and Italy the most adopted measures were extensification and input reduction.
III. METHODOLOGY
In applied research, matching analysis is generally applied to evaluate the impact of a treatment in a nonexperimental setting, where the treatment assignment is not random. Evaluating the effect of a treatment requires being able to observe at the same time the outcome of the same individual in both states, subject and not subject to the treatment (Smith and Todd 2005), such that the treatment effect would be the difference in the outcome between the two states. Obviously, for each individual only one state can be observed; hence, there is a missing data problem behind the evaluation problem. The use of all nontreated individuals as a counterfactual for the treated group causes a selection bias problem, since in a nonexperimental setting it is likely that the treatment is not randomly assigned and the treated and control groups differ not only with respect to the treatment status but also with respect to other characteristics. The selection bias issue is very likely to occur in our analysis, since farmers having the lowest costs of compliance for AESs are more likely to self-select into such schemes.
Matching estimators aim to overcome the selection bias on observables by matching each treated individual with one or more nontreated individuals that have similar observed characteristics, the covariates X, and interpreting the difference in their outcomes as the effect of the treatment (Smith and Todd 2005). Since conditioning on a large set of covariates X may be cumbersome, Rosenbaum and Rubin (1983) proposed the idea of conditioning on a function of X, the probability P(X) of being treated, such that the conditional distribution of X given P(X) is independent of the treatment assignment. If the matching is based on P(X) it is called PSM. Once the matching has been performed, the most common parameter used to evaluate the effect of an intervention is the “average treatment effect on the treated” (ATT), defined as the difference in the mean outcomes of the treated and the matched control group:
[1]
where Y0 and Y1 are the outcomes of an individual in the case of treatment and no treatment, respectively, and D is a dummy variable that takes the value of 1 for the treated individuals.
In order for the matching estimator of ATT to be unbiased, the conditional mean independence assumption and the common support condition must be satisfied. The conditional mean independent assumption states that after conditioning on the propensity score, the mean outcomes are independent of the treatment assignment (Rosenbaum and Rubin 1983), while the common support condition guarantees to find for each treated individual a potential matched untreated individual by restricting the probability of the treated to lower than 1. Matching can be performed by applying different matching algorithms (see Caliendo and Kopeinig 2008 for a review). Such algorithms differ for the weights given to each nontreated unit, and the choice of algorithm always presents a trade-off between variance and bias.
Cross-sectional matching methods correct for the selection bias on observables; however, there may be variables apart from X that are unobserved and that affect both the treatment status and the outcomes. Such variables may lead to selection bias on unobservables. To partially overcome this problem, Heckman, Ichimura, and Todd (1997) propose combining the PSM estimator with a DID estimator. The conditional DID estimator compares the conditional before-after outcome of the treated individuals with that of matched counterparts, and the matching is based on the propensity score:
[2]
where t′ is the pretreatment period, t is the posttreatment period, i identifies the treated individuals, j identifies the nontreated individuals, N is the number of units of the treated group falling in the region of common support, Wij indicates the weights (0≤Wij≤1), which depend on the distance between Pi and Pj, and S indicates the region of common support. The DID matching estimator allows for time-invariant differences in outcome levels between the treated and the control group; however, it requires that conditional on the propensity score, in absence of treatment, the average outcomes of the treated and control group would have followed parallel paths over time. Therefore, the conditional mean independence assumption is replaced by a weaker assumption, the DID mean independence, while the support condition must still be satisfied. The double difference of the DID estimator removes the bias due to time-invariant unobserved characteristics and the bias due to common time trends unrelated to the treatment. Another advantage of this technique is the possibility to use preprogram outcomes.
Thus, when the DID estimator is applied, the ATT measures the difference in the average growth of the outcome between the treated and the control group:
[3]
In order to analyze the farm-level effects of adoption of agri-environmental contracts, we have applied the DID PSM, since this estimator is suitable for our goal. First, our research problem is a program evaluation problem where the treatment is the participation in AESs. Second, the nonexperimental setting of the analyses likely leads to selection bias, as it is reasonable to think that the participating farms have different characteristics compared to the nonparticipating ones, and these characteristics may also affect the outcomes. Third, the use of a nonparametric analysis allows us to consider individual-specific effects as well as avoid functional form misspecification. Fourth, the availability of farm-level panel data allows the application of the DID estimator, which removes the bias due to time-invariant unobserved characteristics and allows the use of pretreatment outcomes in the matching procedure. The adoption of the DID PSM estimator is further motivated by the studies conducted by Smith and Todd (2005) and by Heckman, Ichimura, and Todd (1997), which show that DID matching estimators perform better compared to cross-sectional estimators. Finally, Heckman, Ichimura, and Todd (1997, 612) stated: “Placing comparison group members in the same economic environment and administering them the same questionnaire as participants substantially improves the performance of nonexperimental estimators.” With respect to this statement, the characteristics of the data used in this study (see below) guarantee the homogeneity of data collection between participants and nonparticipants, while the analyses conducted separately for each member state ensure a homogenous economic environment.
IV. EMPIRICAL MODEL AND DATA
We conduct the analyses of the effects of AESs participation on farmers’ choices and performances separately for each country. This allows us to account for heterogeneous effects across countries and for country-specific characteristics and specific AESs. We start from a balanced panel of farms of each country observed over the period 2003–2006, and we apply the DID PSM estimator (Heckman, Ichimura, and Todd 1997). The pretreatment period considered is 2003, while the posttreatment period is 2006, the last year of the 2000–2006 E.U. budget and rural development program period. Farms of the country panel are assigned to the country treated group if they did not participate in AESs in 2003 but they did participate in 2006, and to the country control group if they did not participate for the whole period 2003–2006. Each country’s treated group is split into two subsamples according to the share of the agri-environmental payments on farm agricultural revenue, and the analysis of the effect of AESs adoption is carried out separately for each subsample. The cutoff point in each country’s treated sample is set at 5%, meaning that if the share of agri-environmental payments on farm revenue is lower than 5% then the farm belongs to subsample 1, whereas if the share is larger than 5% it belongs to subsample 2. We split the whole country treated sample into two subsamples because we expect different effects of AESs participation according to how much the agri-environmental payments contribute to farm revenue.
Table 1 shows some descriptive statistics for each country subsample. The mean share of agri-environmental payments on farm revenue in our subsamples varies from 1% in Germany to 2.8% in Spain for subsample 1, while in subsample 2 this range fluctuates between 7.5% in France and 15.5% in Italy. If we consider the share of total farm subsidies on farm revenue, Spain is the country whose farmers are more heavily dependent on subsidies, while Germany is the country with the lowest dependence. Despite the high dependence of Spanish farmers on subsidies, the share of agri-environmental payments on total farm subsidies is the lowest, thus indicating that the AESs in Spain play a marginal role. The highest share of agri-environmental payments on farm subsidies is found in subsample 1 in France and subsample 2 in Italy.
Selected Descriptive Statistics of the FADN Balanced Samples 2003–2006
The farm-level data used in this study come from the Farm Accountancy Data Network (FADN), the most widely used farm-level database in the European Union and the only source of microeconomic data harmonized at the E.U. level (European Commission 2008). The FADN data are yearly data on technical characteristics and economic results of a large sample of farms from each member state. Farms in the FADN database are representative of the population of E.U. farms above a minimum economic size (commercial farms). This threshold differs across member states, but the general rule is to represent farms that provide a level of income sufficient to support the farmers’ household. In order for the sample to be representative, the E.U. commercial farm population is stratified according to the region, the type of specialization, and the economic size, and farmers in the FADN sample are taken from each stratum. Finally, the data on macroeconomics indicators used in our study are taken from national official statistics and Eurostat.2
For each of the five countries, we consider all crop and livestock farms (excluding perennial crops) included in the FADN database, whose data are available for the whole period 2003–2006. Unfortunately, the AESs data included in the FADN database are aggregate data: no information is available either on which scheme has been applied by each farm or on the hectares of farmland committed to AESs.3 In addition, we do not distinguish the effects of AESs on crop and livestock farms, because this distinction would result in too small treated samples.
Propensity Score Matching
The first step of the matching procedure is to estimate a binary model of the participation in AESs, selecting the variables that affect simultaneously farmers’ participation and outcomes (Liu and Lynch 2011). We estimate a country-specific logit model for the pretreatment year 2003. The decision of how many variables to include in a propensity score binary model is a widely discussed issue in the literature. Bryson, Dorsett, and Purdon (2002) outlined that including too many variables in the binary model is not worthwhile because it may exacerbate the common support problem and it may increase the variance of the estimator. By contrast, according to Rubin and Thomas (1996), one should include all variables even if they are not statistically significant, with the exception of a few cases: if there is consensus that a variable does not affect the outcome, if the treated and full control mean values of a variable are “very close,” or if a variable is highly correlated with variables that are already in the model.
We start from a large set of variables derived from economic theory and the results of applied research (Vanslembrouck, Van Huylenbroeck, and Verbeke 2002), and we perform likelihood ratio tests on groups of variables in order to select the best specification of the logit models. In order to ensure comparability across countries, we choose the same specification for all country subsamples of the same type (subsample 1 or subsample 2), while we allow it to differ between the two. A given group of variables is kept in the model if not rejected by the likelihood ratio test at least in one of country, and variables of that group are included in the model even though they are not statistically significant.
The groups of variables considered in our model concern both farm technical and economic characteristics, their regional location, and altitude, as well as the macroeconomic environment. The groups of variables are age of the farmer, altitude of the farm location (dummy variables for hill and mountain), farm size (farm UAA), farm production intensity (total farm output value per hectare, share of grassland on farmland, and value of fixed assets per farm working units), use of chemical products (per hectare expenditure on fertilizers and crop protection products), farm labor (number of family working hours and hired labor working hours), farm profitability (farm income per hectare), and level of farm dependence on subsidies (farm direct income payment per hectare). In order to take into account the differences in the economic environment throughout the regions of a country we also include two macroeconomic indicators at the regional level: the gross domestic product per capita and the share of the agricultural value added over the total value added of the region. Finally, dummy variables are used to indicate farm type (livestock or arable crops) and geographical location.4 While for subsample 2 the group of variables concerning farm labor is withdrawn from the model, since it is not significant in any country; all the variables are kept in the model for subsample 1. The participation probabilities estimated from each country logit model are the predicted values of the farms’ propensity scores that are used in the matching procedure.5
We implement different matching algorithms, and the one that performs better is selected. Since we do not condition directly on the covariates but on the propensity scores, we check the ability of the matching procedure to balance the covariates between the treated and the matched control group. The matching quality of each algorithm is tested by three different indicators: the t-test on the differences of the covariates means of the two groups and the calculation of the standardized bias and of the pseudo R-squared before and after the matching (Caliendo and Kopeinig 2008).
After testing different matching algorithms and checking the covariates balancing property, we choose to implement nearest-neighbor matching with 10 neighbors, with replacement and with a caliper of 0.05.6 Thus, the 10 nearest-neighbor estimator matches each farmer participating in AESs in 2006 with 10 farmers who never participated in AESs over the whole 2003–2006 period, as long as his propensity score differs from the propensity score of the treated farmer less than 0.05. This matching algorithm is applied for both subsamples in each country in order to make the results comparable. The 10 nearest-neighbor estimators with caliper and with replacement result in good matching quality in all samples analyzed, always satisfying the covariates balancing property, reducing the standardized bias of most of the covariates, and showing a small pseudo R-squared of the logit model after matching.7 The use of the 10 nearest neighbors compared to the single nearest neighbor decreases the variance, because more individuals are used to construct the counterfactual for each treated individual, but it also increases the bias, because the additional individuals used are poorer matches by construction. The distribution of the propensity scores of the treated and untreated groups largely overlap, and the number of untreated individuals is much larger than the number of treated individuals; the 10 nearest-neighbors estimator allows a decrease in the variance of the ATT at a small cost in terms of bias (Lawley and Towe 2014). In addition, the bias is further reduced by setting a caliper of 0.05 and by allowing for replacement.8
Once the matching has been performed on 2003 data, the ATT of different outcomes is calculated using the DID estimators. In the DID framework, the before-after outcome differences for each participating farmer are compared to those of the matched nonparticipants. The outcomes analyzed as potentially affected by AESs adoption concern the farmers’ production choices and the farm economic performances. Among production choices, we consider those practices that are related to the environmental impact of agriculture, such as the per hectare expenditure on fertilizers and crop protection, the share of grassland, and the number of crops grown on the farm. In addition, a set of variables related to farm structure is analyzed, such as farm size, farm rented land, and family and hired labor working hours. The economic outcomes investigated are the farm output value per hectare, the variable costs per hectare, total farm income, and farm income per hectare, computed both including and not including agri-environmental payments. The results indicate for each outcome the mean difference in the 2003–2006 growth between farmers who adopted AESs after 2003 and farmers who never adopted AESs over the period 2003–2006. The DID estimator is applied on each country treated subsample and its matched control group; thus in total, the estimator has been applied 10 times. On these results, we implement a placebo test for the pretreatment period 2002–2003 (Chabé-Ferret and Subervie 2013), in order to test whether our data verify the DID mean independence assumption, meaning that in the absence of treatment the outcomes of the treated and matched control groups would have followed parallel path conditional on the observed variables.9
Different approaches are available to estimate the variance of the ATT estimator. Although the bootstrapping method is widely used in applied work, Abadie and Imbens (2008) show that bootstrap standard errors are not valid in the case of nearest-neighbor matching with replacement and with a fixed number of neighbors. Thus, we adopt the variance approximation approach proposed by Eichler and Lechner (2002) for the DID estimator.
V. RESULTS
In order to discuss the effects of AESs adoption on farm performance, we will focus our attention on the ATT estimator, which indicates the difference in the average growth of the outcomes between the treated and the matched control groups. The results are shown in Tables 2–6, where the mean difference indicates the average difference of the outcomes over the period 2003–2006 for both the treated and the matched control groups. A positive (negative) ATT indicates that the increase (decrease) in the outcome of the treated group is larger than the increase (decrease) of the control group or that the decrease (increase) in the outcome of the treated is smaller than the decrease (increase) of the controls. In order to allow the reader to calculate the percentage change of the outcomes over the period 2003–2006, we present the mean outcomes in 2003 in Appendix Tables A1 and A2.
We present the results distinguishing the effects of participation in AESs on farmers’ technical production choices and on farm economic results. Similarities and differences of these effects among countries are discussed.
Results of Farm Production Choices
The adoption of AESs is expected to affect farm technical decision variables, as they introduce new requirements/constraints in farm management. The effects differ across country samples as a consequence of differences in agricultural characteristics and heterogeneous implementation of AESs. As expected, the AESs affect more heavily the production choices of the treated subsample 2, the subsample for which the share of agri-environmental payments on farm income is larger than 5%, while in subsample 1 the impact is lower. We will consider the effects of AESs on subsample 2 first, then we will draw some conclusions for subsample 1. Our analyses will focus on farm practices that have a potential effect on the environment, since environmental preservation is the main goal of agri-environmental measures.
Changes in crop number, fertilizer, and crop protection expenditure per hectare and share of grassland may be considered indicators of the effectiveness of AESs in enhancing environmentally friendly practices at the farm level. Our study shows differences in the evolution of these variables across countries.
In the Spanish subsample 2 the adoption of AESs does not seem to affect any of the variables mentioned above, as their average change over the period 2003–2006 does not significantly differ between the treated and the control groups (Table 2). The only exception is fertilizer expenditure per hectare, which surprisingly increases for participant farmers by 3 .2%, compared to a decrease by 11.5% for nonparticipants. The most widespread agri-environmental measures in Spain until the new rural development framework started in 2007 were related to landscape protection (Peco et al. 2000); hence, we may explain the absence of effects on fertilizer and crop protection expenditure as well as on the share on grassland and on the number of crops by arguing that those measures were not targeted toward the sustainable practices considered in this paper. However, a more reliable explanation of this lack of effectiveness of Spanish agri-environmental measures may stem from the difficulties experienced by Spain in implementing AESs in the first programming periods, due to problems of cofinancing from the national budget (OECD 2004).
Average Treatment Effect on the Treated (ATT) of AES Participation in Spain, 2003–2006
In the United Kingdom and Italy, subsample 2 agri-environmental measures seem to be effective in promoting sustainable practices (Tables 3 and 4). In the United Kingdom, participation in AESs has a positive effect on reducing the use of agrochemical products, as participant farmers increase the average fertilizer expenditure per hectare less than nonparticipants and they slightly decrease the expenditure in crop protection per hectare compared to a rise in the matched control group, with both these results being significant at the 5% level. Participation in AESs contributes to reduce the expenditure per hec-tare in fertilizer by 12.7%10 and in crop protection products by 16.0%; such decreases are in line with the findings of Sauer, Walsh, and Zilberman (2012). In the United Kingdom, AESs have been implemented since the 1980s; thus, the long history of these schemes and their large participation rate support their environmental effectiveness. In Italy, farms committed to AESs decrease both their fertilizer and crop protection expenditure by €10/ ha and €6/ha, respectively, which correspond to a 27.0% and 44.3% drop, respectively, compared to a rise for nonparticipants of 6.8% and 25%, respectively. This result is consistent with the fact that in Italy more than 50% of the area under AESs is committed to organic farming and integrated production. In addition, in both countries, agri-environmental measures promote farm crop diversification, as this variable shows a positive and significant ATT.
Average Treatment Effect on the Treated (ATT) of AES Participation in the United Kingdom, 2003–2006
Average Treatment Effect on the Treated (ATT) of AES Participation in Italy, 2003–2006
Many agri-environmental measures in France aim to promote extensive grazing, in particular by supporting grassland maintenance and conversion (Princé and Jiguet 2013), and this is confirmed in Franc’s subsample 2, where most of the participant farms are livestock farms. Since our results show that participant farmers show a statistically significant increase in the share of grassland as compared to the slight decline in the share for nonparticipants, we may say that AESs in France are effective in reaching their goal. In addition in France, participation in AESs contributes to a decrease in the expenditure for crop protection by 18.0% with respect to nonparticipation (Table 5).
Average Treatment Effect on the Treated (ATT) of AES Participation in France, 2003–2006
In Germany the only environment-related practice affected by AESs participation seems to be the per hectare expenditure for fertilizers, which is 88.9% lower than the expenditure participant farmers would afford if they did not participate in AESs (Table 6). This result is consistent with the findings of Pufahl and Weiss (2009), who, however, also found a significant effect on other sustainable farming practices, and it is consistent with the data provided by Germany’s Rural Development Network (2014), which indicates that in 2006 the most widespread agri-environmental measure was low-input agriculture (36.4% of the total number of contracts and 29.5% of UAA under AESs).
Average Treatment Effect on the Treated (ATT) of AES Participation in Germany, 2003–2006
In all countries but Spain, AESs do not seem to affect family labor use, since the dynamics follows the same negative growth in both the treated subsample 2 and the control groups. In Spain, participation in AESs enhances the negative trend of family labor use, which results in a negative and significant ATT. In Italy, the treated farms show a positive growth of family labor use, but the large standard error of the ATT does not lead to any statistically significant result. Hired labor is affected by participation in France and Italy, where AESs seem to promote agricultural employment, and in Spain, where it slows down the drop.
Finally, the participation in AESs results in a statistically significant increase in farm size in subsample 2 for both the U.K. and Italy, by 4.5% and 9.3%, respectively, compared to a reduction for the control group (-3.5% and - 2.7%, respectively). In Italy, where the most widespread measures are organic farming and low-input agriculture, the increase in farm size is likely due to the attempt to offset the decrease in the output value per hectare as a consequence of participation.11 Note that in both countries, the increase in the average farm size of the treated groups is mainly due to an increase in rented land.
If we consider the treated subsample 1 and the matched counterfactuals, we observe a few technical outcomes affected by AESs participation. The only country whose treated group has a statistically significant change toward more environmentally friendly practices is France, where the average fertilizer expenditure per hectare of participants increases less than the expenditure of nonparticipants. In Germany, the production variables affected by AESs are farm size and rented land, which increase as a consequence of participation; this is likely linked to the fact that extensification programs are the second most adopted measure in Germany. In France, AESs slow down the drop in family labor use. Crop protection expenditure, grassland share, number of crops, and hired labor use are not affected in any of the five countries. Hence, if the payment represents a small share of farm income, the production choices of the farm are only slightly affected and the AESs’ effectiveness in promoting sustainable practices is rather weak.
Results for Farm Economic Performance
AESs adoption does not affect only the farmer’s production choices but also the economic results of the farm. As in the previous section, we start by discussing the results of subsample 2, and later a short comment is provided for the subsample 1.
If we consider the value of total farm output per hectare as a measure of the degree of production intensity, we would expect a negative ATT as a consequence of extensification of farms participating in AESs. The corresponding ATT for subsample 2 is indeed negative and statistically different from zero in Germany, Spain, and Italy. In Germany and Italy the negative effect on farm output may be explained by the decrease in fertilizer expenditure due to participation, and in Italy it may also be the consequence of the drop in the expenditure for crop protection and of the introduction of new crops to promote farm diversification. In constrast, the negative ATT of farm output in Spain does not seem to be linked to more sustainable farm practices.
In the Spanish subsample 2, we observe a negative and significant ATT for farm income per hectare, both considering and not considering the agri-environmental payments. We may consider the farm income per hectare without agri-environmental payments as an indicator of the effect of AESs adoption on farm performance, while the evolution of the second variable may indicate the ability of the agri-environmental payments to compensate the income foregone. For participants, the average decrease of farm income without agri-environmental payments is €584 (− 76.7% compared to nonparticipation), compared to a rise of €417 for nonparticipants. Moreover, the agri-environmental payment is not sufficient to compensate the income foregone by Spanish adopters, since, considering the payment, we still observe a drop of 60.6%, which is statistically significant. The negative effects of AESs adoption on farm income in Spain may be related to the financial difficulties in implementing such schemes in the country, which may have led to setting too low levels of agri-environmental payments. Moreover, the share of livestock farms in Spanish subsample 2 is much smaller compared to that of the other countries. This may also explain the different performance of AESs in Spain compared to other countries: AESs on crop farms may be less effective in promoting environmentally friendly practices and may lead to higher income losses.
In fact, in none of the other countries did the adoption of AESs result in a negative effect on per hectare income of participants, both considering and not considering agri-environmental payments. This means that, on average, the change in farm practices does not result in a statistically significant income foregone, and the level of the agri-environmental payments is such that it does not overcompensate adopters. Only in Germany is a significant ATT found for the income per hectare without agri-environmental payments; however, the difference between the slight increase of adopters (€23/ha) and the sharp increase of nonadopters (€299/ha) is much smaller compared to Spain; moreover, this different trend is only barely significant, and once we include the agri-environmental payment the difference is no longer significant.
If we consider the treated subsample 1 we still notice a negative effect on income for participant farmers in Spain, since they experience a 48.2% drop, including the agri-environmental payment, which increases to 53.6% not considering the payment.12 In none of the other countries does the adoption of AESs produce any effect on the income per hectare. This may be explained by the marginal role of agri-environmental measures in the farm business. In Germany the treated group shows an increase of farm income higher compared to the control group, but this is the consequence of the increase in farm size due to participation. The significant ATT for farm income in the United Kingdom and Italy may be explained by a combination of the effects of income per hectare and farm size, which are not significant by themselves but only when they are combined.
VI. DISCUSSION AND CONCLUSIONS
The impact of AESs across E.U. countries may be strongly different, and this aspect has been seldom analyzed in the literature. In combining PSM and DID estimators, this paper has tried to fill this gap by analyzing the impact of AESs on balanced panels of FADN farms for five E.U. countries. When panel data are available, the combination of PSM with DID, compared to the simple PSM estimator, allows us to remove the bias due to timeinvariant unobserved variables. We have checked the robustness of our results by comparing our 10 nearest-neighbors estimators with other PSM estimators that satisfy the balancing property for the covariates. The results are similar, which supports the robustness of our findings.
Pufhal and Weiss (2009) remark on the importance of accounting for the heterogeneous response of treated individuals to the treatment. The different effects of agri-environmental participation in our two subsamples support the need to account for heterogeneity at an even more detailed level (i.e., livestock vs. arable crop farms). Unfortunately, we cannot split the samples further because of lack of sufficient observations. Moreover, the FADN database does not provide the disaggregation of payments by different AESs, which would allow us to draw more specific conclusions related to each specific measure, with more precise policy implications.
However, despite the above limitations, our study shows some interesting findings. First, the effects of AESs on farm production choices are stronger for farms in which agri-environmental payments represent a larger share of farm income. With the exception of Spain, participation in AESs seems to be effective in promoting more sustainable agricultural practices. For example, the fertilizer per hectare expenditure decreases in Germany, the United Kingdom, and Italy as a consequence of participation, crop protection per hectare expenditure decreases in France, the United Kingdom, and Italy, the share of grassland increases in France, and the number of crops grown on a farm increases in the United Kingdom and Italy. On the other hand, when the share of agri-environmental payments on farm income is smaller than 5%, farmers’ practices change only marginally. In terms of farm employment, the adoption of AESs does not seem to prevent the drop of family labor use in any country, while hired labor is positively affected in France, Spain, and Italy.
The effects of AES participation on farmers’ income are largely negative in Spain, where the level of agri-environmental payments does not seem to be sufficient to compensate farmers for the income foregone. Apart from Spain, Germany is the only country where participants’ farm income is affected; however, the drop due to participation is much smaller compared to Spain’s, and the agri-environmental payments compensate for the income foregone. These results may sound strange, since we may argue that a farmer is willing to participate in AESs only if he expects an increase in income. However, the time lag between decisions and results, as well as the specific implementation difficulties experienced by Spain in the 2000–2006 budget period, may explain the actual negative impact of AESs on farm income.
In conclusion, our study suggests that the AES design in Spain may be revised in order to produce environmental benefits and to fairly compensate participating farmers. In the other countries, the AESs seem to work better, since they promote environmentally friendly practices. However, this impact is relevant only for those farmers for which the share of agri-environmental payments on farm income is sufficiently large. Moreover, in several countries (France, United Kingdom, and Italy), AESs do not seem to impact per hectare farm income (without agri-environmental payments), despite their environmental constraints. Thus, the reorganization of production required by AESs seems to promote some efficiency gains in participating farms. Finally, our study suggests that some further measures for rural employment safeguards could be introduced, as AESs are not effective in contrasting the decline of family labor use on the farm.
Acknowledgments
This research was carried out as part of the FADNTOOL (Development of modeling tools based on Farm Accounting Data Network data adapted to assess the dynamic impacts of the Common Agricultural Policy) research project (Scientific Coordinator: Konstantinos Mattas), funded by the European Commission under the 7th Framework programme. The FADN data were provided in the context of that project.
APPENDIX
Mean Outcomes of the Treated and of the Matched Control Group in 2003 (Subsample 1)
Mean Outcomes of the Treated and of the Matched Control Group in 2003 (Subsample 2)
Footnotes
The authors are, respectively, postdoctoral fellow and professor, Department of Agri-food Economics, Università Cattolica del Sacro Cuore, Piacenza, Italy.
↵1 Among the additional effects, Chabé-Ferret and Subervie (2013) further distinguish between direct effects, those specifically targeted by each agri-environmental measure, and cross effects, those not targeted by that measure.
↵2 Data sources include Institut National de la Statistique et des études économiques (www.insee.fr/fr/themes/detail.asp?reg_id=99&ref_id=pib-va-reg, accessed June 2013), Statistische Ämter des Bundes und der Länder (www.vgrdl.de, accessed June 2013), U.K. National Statistics (www.ons.gov.uk/ons/index.html, assessed on June 2013), Instituto Nacional de Estadística (www.ine.es/jaxi/menu.do?type=pcaxis&path=%2Ft35%2Fp010&file=inebase&L=0, accessed June 2013), Istat (http://dati.istat.it/, accessed June 2013), and Eurostat (http://ec.europa.eu/eurostat/data/database accessed June 2013).
↵3 Given this data limitation problem, we cannot distinguish between the direct and the cross effects. In our paper, the ATT can be interpreted as the “additional” effects of all agri-environmental measures simultaneously, including both the direct and the cross effects defined by Chabé-Ferret and Subervie (2013).
↵4 The macroeconomic indicators are specified for each NUTS2 region in France, Spain, and Italy and for each NUTS1 region in Germany and the United Kingdom. The country geographical dummies are specified for NUTS1 regions or aggregations of them. (Nomenclature of Territorial Units for Statistics [NUTS] is a coding standard for country subdivisions in the European Union.)
↵5 Detailed estimation results are provided in the online appendix available at http://le.uwpress.org.
↵6 We have checked the robustness of our results by comparing the results of our 10 nearest-neighbors estimator with other PSM estimators that satisfy the balancing property for the covariates. Since results are similar, this supports the robustness of our findings. All these results are available from the authors upon request.
↵7 These results are presented in the online appendix available at http://le.uwpress.org.
↵8 As noted by Abadie and Imbens (2006, 240): “Matching with replacement produces matches of higher quality than matching without replacement by increasing the set of possible matches. In addition, matching with replacement has the advantage that it allows us to consider estimators that match all units, treated as well as controls, so that the estimand is identical to the population average treatment effect.”
↵9 The results of the placebo tests (see the online appendix available at http://le.uwpress.org) show that the ATT over the period 2002–2003 is significant for a few outcomes only (14% of the outcomes among all of subsample 1 and 7% of the outcomes among all of subsample 2), but in most of these cases its sign is opposite to the significant ATT over the period 2003–2006. This happens in subsample 2 for the crop protection expenditure per hectare in France and the United Kingdom and for the share of grassland in France. In these cases, we may argue that the effect of the participation in AESs over the period 2003–2006 is a lower bound, and the real effect is likely to be larger. The other significant ATT of the placebo test on the pretreatment period concerns some outcomes for subsample 1: farm size, output value per hectare, and income per hectare in France, crop protection and fertilizer expenditure per hectare in Spain, and farm size and rented land in the United Kingdom. These ATT values are positive in the placebo test, whereas they are not significant for the period 2003–2006; this indicates that participation in AESs has changed the trend of these outcomes, shifting from a nonparallel trend in the pretreatment period to a parallel trend in the treated period. Hence, a few outcomes in subsample 1 seem to be actually affected by AESs participation even if their ATT is not significant. Finally, the variable costs per hectare in Spain’s subsample 2 show a negative ATT in the placebo test, while in the 2003–2006 period this outcome is not significant. It was not possible to perform the placebo test on the Italian sample, since in 2003 a large number of new farms entered the FADN database, such that we could not follow the same sample over the years 2002–2003.
↵10 The percentage changes discussed in this section are computed between the mean outcome of the treated group in 2006 and the mean outcome the treated group would have experienced in 2006 without the treatment (i.e., the “windfall effect” as defined by Chabé-Ferret and Subervie [2013]). The latter is computed as the sum of the mean outcome of the treated group in 2003 and the mean difference of the matched control group.
↵11 A reviewer suggested that the increase in farm size may be linked to “less favored area” payments, which explicitly encourage extensification. We have checked, in each sample, the share of participant farmers obtaining such pay-ments compared to nonparticipants in both 2003 and 2006. It turns out that only for the Italian subsample 2 may the increase in the average farm size be explained by this factor, since in all the other cases the difference is not statistically significant.
↵12 Simulation analyses show that the average agri-environmental payment in Spain should increase by €130/ha of total UAA in subsample 1 and €561/ha in subsample 2 in order to compensate the income loss due to participation. The lack of data on the average number of hectares committed to AESs does not allow calculation of the required increase in the agri-environmental payment per hectare of land under AESs.
