Abstract
This study uses a structural econometric model to evaluate the impacts of support from a European Union agri-environmental program designed to reduce nutrient pollution from agricultural land. Drawing on a representative sample of individual grain farms, we first quantify the effects of agri-environmental payments on farms’ decisions on land allocation and on fertilizer use. We then combine the predicted land allocation and fertilizer use with environmental production functions to quantify the impact on nutrient loading. Finally, we assess the monetary value of reduced nutrient pollution, drawing on a recent valuation study. (JEL Q53, Q58)
I. Introduction
The past two decades have seen a substantial increase in farm subsidies provided under agri-environmental programs (AEPs) by the world’s dominant agricultural producers, the European Union (EU) and the United States (US). The programs are designed to reduce agriculturally produced pollution and to encourage provision of agriculture’s nonmarket benefits. In the EU, national AEPs have been compulsory for the member states since the 1992 reform of the Common Agricultural Policy (CAP). While the policy changes reflect increasing demands for environmental quality, other driving factors were the need to reduce agricultural overproduction and demands from the World Trade Organization for a reduction in trade-distorting measures (Hanley and Oglethorpe 1999; Buller, Wilson and Höll 2000; Baylis et al. 2011). By 2002, 25% of the agricultural area in the EU (EU15) was registered in one or more AEPs, and the annual EU budget spending on AEPs was on the order of €2,000 million (European Commission 2005).
Little is known about whether participation in the EU AEPs actually improves farms’ environmental performance. At the level of the individual farmer, participation in an AEP is voluntary. Program payments are designed to provide incentives for reducing farm-source pollution and for adopting conservation measures, but the requirements for participating farms tend to be quite general in nature, and payments are conditioned on environmentally benign practices rather than measurable outcomes. Rigorous empirical studies on program impacts are rare. To our knowledge, Pufahl and Weiss (2009) and Chabé-Ferret and Subervie (2013) provide the only econometric analyses that explicitly investigate the effects of EU AEPs on observed farm production decisions.1 Pufahl and Weiss (2009) evaluate the effect of German AEPs on farms’ input use, using difference-in-difference propensity score matching. They find that AEPs increased both the area under cultivation and grassland, and decreased the use of agrichemicals. Chabé-Ferret and Subervie (2013), also applying difference-in-difference matching, estimate the effects of five agri-environmental schemes in France. They find that two of the schemes may well be socially efficient, whereas others have had only limited success.
The impacts of agri-environmental payments in the US have been examined by Wu et al. (2004), for example. Analyzing microlevel data, they find that while payments for conservation practices increased the use of these practices, the overall environmental benefits were small. Goodwin and Smith (2003) use US county-level data to analyze the impacts of the Conservation Reserve Program (CRP). Their results suggest that the CRP has been effective in reducing erosion, but that part of the reduction has been offset by other income support programs.
The limited knowledge of EU AEP impacts is a significant shortcoming for two reasons. First, where environmental policy is concerned, it is important that we be able to ascertain whether programs are actually fulfilling their promise as policy measures by reducing environmental damage or enhancing the positive effects attributable to agriculture vis-à-vis no policy. Second, in terms of trade policy, we need further information on whether programs are actually compensating farmers for nonmarket production activities or just greenwashing production subsidies; this is an issue that has led to considerable disagreement between the EU and the US—and between these two trading powers and developing countries—in the now-stalled Doha Round of trade liberalization talks (Hanrahan and Schnepf 2013; Baylis et al. 2011).
The present paper sheds light on these issues and augments the empirical literature on the impacts of EU AEPs by analyzing Finland’s implementation of the EU agri-environmental mandate, the Finnish Agri-environmental Program (FAEP). We investigate how crop producers have responded to the FAEP by analyzing a structural econometric model of production decisions, drawing on a representative sample of individual grain farms over the period 1996-2005. We use the variation in compensatory payment rates across regions and over time to identify the impact of payments on farms’ decisions on land allocation and agrichemical input use. To assess the impact of the FAEP payments on farm-source pollution, we use the estimated land allocation and input demand functions to predict farms’ land allocation and chemical use under two scenarios: a “factual” case, where program payments are set to their historical values, and a counterfactual one, where agrienvironmental payments are set to zero. As the FAEP’s main focus is on reducing agriculturally produced nutrient pollution, we combine the predicted fertilizer intensity and land allocation with environmental production functions to predict nutrient loading under each scenario. Comparison of the outcomes using the factual baseline and the counterfactual reveals the effect of the agrienvironmental payments. Finally, relying on a valuation study measuring the willingness to pay for reducing Finland’s nutrient loads, we compute the monetary value of environmental benefits attributable to the FAEP in our sample and compare them to the costs represented by the agri-environmental payments.
The structural approach taken here is different from the treatment effect approach by Pufahl and Weiss (2009) and Chabé-Ferret and Subervie (2013) in that we explicitly model farms’ production decisions. That is, we use an economic model to recover fundamental parameters describing farms’ production choices. Under the assumption that the parameters remain unchanged, the approach enables consistent predictions of farms’ responses to policy changes, which can serve to both evaluate the impact of present AEP payments and forecast the impacts of alternative policy interventions, such as taxes on polluting inputs. Of course, the choice of methodology is also dictated by the policy to be evaluated. We examine an AEP setting that is very different from the German and French systems analyzed by Pufahl and Weiss (2009) and Chabé-Ferret and Subervie (2013). Germany and France have numerous agri-environmental schemes, and only a limited proportion of agricultural land is managed under AEPs. In Finland, participation in the single, nationwide AEP is almost universal. Thus, the treatment effect approach, which requires a “treatment group” of participants and a “comparison group” of nonparticipants, is not applicable. In the structural approach, functional form and support conditions substitute for the lack of a comparison group (see, e.g., Heckman and Vytlacil 2007; Heckman 2010; Keane 2010; Nevo and Whinston 2010).
II. Background
Water pollution caused by agriculture, in particular nutrient enrichment of surface waters, is viewed as a major environmental problem in Finland. The adjacent Baltic Sea suffers from severe nutrient-related degradation of water quality, with intensive agriculture the largest source of nutrients (e.g., Helcom 2011). Launched upon Finland’s accession to the EU in 1995, the FAEP emphasizes pollution control, although it includes measures targeting biodiversity and landscape protection. The program provides payments to support environmentally beneficial farming practices on all, not just environmentally sensitive, land. The overall participation rate was 84% in 1995-1999 and 90% in 2000-2006 (MAF 2004).
The FAEP is divided into general and special subprograms. Farms opting to participate in the general subprogram agree to follow a set of environmental management measures on working lands. For grain production, the major form of crop production in Finland, the measures impose limits on fertilizer use and require construction of field margins and vegetative filter strips along waterways.2 Farms are compensated through an area-based payment, where the per hectare payment rate is uniform for all farms within a support region. The special subprogram provides support for clearly defined conservation measures, such as converting some cropland to riparian zones or wetlands.3 Other subsidy payments available to grain farms (in addition to the FAEP), also proportional to land area, include CAP arable-area and less-favored-area payments and national aid for crop production. Instead of planting grains, farms may leave arable land fallow, known as set-aside. Set-aside receives lower unit support payments and produces lower nutrient losses than land in grain production. Set-aside has only been entitled to subsidy payments through the FAEP in the period 1995-1999, and the unit FAEP subsidies for set-aside were lower than those for grain areas.
The per hectare agricultural support payment rates, including agri-environmental support, are graded over seven support regions. The regions were delineated at the time of Finland’s accession to the EU in 1995 and reflect regional climatic conditions. We focus on the support regions labeled A to C2 (see Figure 1). These support regions contain 98% of Finland’s grain production.4 Variation in the per hectare compensatory payment rates across support regions and over time allows us to identify the impact of the subsidies on production decisions in the subsequent econometric analysis. Variation in the per hectare payment rates arises for several reasons: CAP arable-area payment rates in each support region are determined on the basis of regional historical reference yields, general agri-environmental support rates are calculated based on the regional historical average costs of implementing the required changes in farming practices; support region A was initially not eligible for EU less-favored-area support (the support was extended to all of Finland in 2000); and at the time of accession, Finland bargained for and was granted the right to pay additional aid to areas north of the 62nd parallel (support regions C1 and C2 in our study area). Part of the northern aid is paid in conjunction with the agri-environmental support. Changes in the per hectare payment rates over time have also been asymmetric across support regions.5
Study Area and Delineation of Agricultural Support Regions within the Study Area
The enforcement of the FAEP fertilizer use constraints is weak. Approximately 5% of participating farms are audited each year and sanctions are mild. Violations of the limits on fertilizer application, for example, result in at most a 9% cut in the current year’s payments (ARA 2011). Nitrogen fertilization rates for the sample of grain farms included in our empirical analysis reflect the weak enforcement. Figure 2 shows the distribution of deviations from the FAEP imposed limit on nitrogen use. A significant proportion of farms that received agri-environmental payments appear to have violated the constraint on nitrogen application.6
III. Behavioral Model: Farms’ Decisions On Land Allocation and Input Use
This section presents the microeconomic behavioral model, which comprises farms’ decisions on land allocation and agrichemical use. The behavioral model and estimation methodology follow a dual profit approach proposed by Lacroix and Thomas (2011). A similar approach has been adopted for example by Arnade and Kelch (2007) and Fezzi and Bateman (2011). All these studies build on the dual profit function approach with fixed allocatable inputs introduced by Chambers and Just (1989).7
Farms maximize total profits over a set of crops.8 We assume that they consider input and output prices and unit agricultural support payment rates (per hectare) to be exogenous, including agri-environmental support through the FAEP general subprogram. Finland’s overall cereal production amounted to 1% to 2% of the EU total in the years 1997-2007,9 and price feedbacks are likely to be minor. Support payment rates, uniform across a support region, are determined in negotiations between Finland and the EU and are based on regional historical yields, regional historical average costs of agri-environmental measures, and geographic location. In the estimation stage, we control for province and farm fixed effects, which may be correlated with the regional historical reference yields and average costs.
Difference between Farms’ Nitrogen Use and Finnish Agri-environmental Program (FAEP) Imposed Nitrogen Limit for Sample Farms Registered in the FAEP, 1996-2005
The farms in the sample produce grain crops (barley, wheat, oats, and rye). We aggregate across the grains in the analysis to follow. The main reason for the aggregation is substantial colinearity between the prices of the four grains in our data, as well as colinearity between the unit agricultural support payment rates for the four crops. Furthermore, our data contain information only on each farm’s total expenditure on inputs, not crop-specific input use. The average proportion of land allocated to each grain has remained relatively stable over the study period, and the four grains are similar in terms of the use of agrichemicals as well as environmental impacts.10 For these reasons, we do not expect the aggregation to significantly affect predictions of farm-source nutrient pollution.
Let L denote the total arable land area of the farm.11 A farm engaged in grain production decides how to allocate arable land to grains and set-aside based on their relative profitability. Once this decision is taken, the farm determines the profit-maximizing output level. By assumption, a farm considers only the private net benefits of farming, ignoring any environmental impacts. Let lg denote the land allocated to grains, lf set-aside, pg grain price, qg per hectare grain yield, wk the kth component of the input vector, rk the corresponding input price, and sg and sf per hectare subsidy rates for grains and for set-aside. The unit subsidy rates sg and sf are the sums of the FAEP general subprogram, CAP arable-area and less-favored-area, and Finnish national crop production subsidy rates to each land use. In addition to arable land, a farm may have land converted permanently (at least five years) to a specific conservation practice under the FAEP special subprogram; let lsp denote the land area in permanent conservation use, and ssp the FAEP special subprogram subsidy rate. Farm profit is given by
[1]
The representative farm is assumed to maximize annual profits under the constraint on total arable land, lg + lf = L. While the FAEP imposes limits on fertilizer use, we assume that farms do not consider this limit as a constraint in their input decision. If the limits on fertilizer use were considered binding, we would expect a disproportionally large number of farms to apply the amount allowed by the fertilizer limit. There is no bunching at the limit on fertilizer use (Figure 2).
Maximizing the profit function yields optimal input demand and output supply functions. We use the dual representation of farm profit maximizing behavior. While a primal approach would require specifying a production function for grains, the dual approach is based on the specification of a flexible indirect profit function, Π(p,r,s,L). The factor demand and output supply functions can be derived from the indirect profit function using Hotelling’s lemma and will be functions of exogenous variables. The dual approach will be convenient for examining how a particular agri-environmental support policy affects farms’ input demand: we can investigate how the policy affects the prices farms face and then see how those changes in prices affect the input demand functions.
Any well-behaved profit function must satisfy the following regularity conditions: homogeneity of degree one in prices, convexity in prices, monotonicity, and symmetry. The assumption of a given total arable land area available for crop production imposes an additional land adding-up condition:
[2]
[3]
where sj denotes the jth component of the subsidy vector.
IV. Model Specification and Estimation
We specify a quadratic profit function, which provides a flexible approximation of the true profit function. We normalize the profit function by dividing the profit, prices, and subsidies by the price of one input, labor. Conditions [2] and [3], as well as homogeneity of profit with respect to prices, are then easily imposed. The quadratic profit function is
[4]
where J indexes the subsidy rates (for grains and set-aside) and K the inputs (fertilizers, pesticides, labor), and the upper bar indicates normalized profit, price, and subsidy variables. That is,
, where rK is the price of the numeraire, here labor.
By Hotelling’s lemma, differentiating the profit function with respect to prices and the arable land-related payment rates yields
[5]
where lgqg is the total grain supply;
[6]
where lg and ls are land allocated to grains and set-aside, respectively; and
[7]
where Wf and Wp are input demand functions for fertilizers and pesticides, respectively. We only estimate land allocation equations for grains (lg) and set-aside (lf), since only this land allocation decision can be made annually. Furthermore, our data do not specify the land area converted to conservation practices under the special subprogram (lsp), but only the total special subprogram payments received (lspssp). Consistent estimation of farms’ land allocation and input use decisions requires controlling for the fact that only a proportion of farms receive the FAEP special subprogram subsidy, and for the amount received, since these may have direct implications for farms’ input use, land allocation, and grain yield. We address this econometric consideration with more detail below.
Land allocation and output are also influenced by factors that are unobservable to the analyst. These factors can be either period specific (e.g., weather and pests) or farm specific (e.g., soil quality and farmer skills) (Wu et al. 2004; Lacroix and Thomas 2011). Using panel data allows us to partly compensate for the lack of farm-level soil and weather information, and facilitates control of unobserved individual heterogeneity.
From equation [6], condition [2] implies the following parameter constraints:
[8]
A further constraint is imposed by the CAP mandatory set-aside mechanism, which requires farms to leave a proportion of land fallow each year.12 Set-aside in the land allocation equation corresponds to set-aside area exceeding the mandatory area, hereafter referred to as voluntary set-aside.
We estimate the profit function simultaneously with the demand functions for fertilizers and pesticides, the equations for land allocated to grains and voluntary set-aside, and total grain output, all subject to the parameter constraints [8]. Estimating a system of equations improves the efficiency of parameter estimates when there is some correlation between errors across equations. The system of equations for farm i in year t is written as follows:
[9]
The terms u1,it to u6,it are idiosyncratic error terms, possibly correlated across equations, and by assumption of mean zero. The terms μ1i to μ6i represent farm-specific unobserved effects and are assumed to be fixed parameters. In order to control for possible correlation between unobserved farmer-specific effects and some of the explanatory variables, we apply the Within transformation to all variables and estimate a Seemingly Unrelated Regression Equations (SURE) model. The Within transformation, which deviates variables from their individual means, cancels out time-constant unobserved individual effects.
Two issues have to be addressed in estimating the system. The first is selection. Reflecting the almost universal participation rate, all farms in the sample are registered in the FAEP general subprogram and receive general subprogram subsidies embedded in sg and sf. Only a subset of farms has contracted to devote land to specific conservation practices and receives payments through the FAEP special subprogram (lspssp). Because our primary purpose is to analyze the impact of the FAEP on farms’ production decisions, it is important to control for the fact that only a proportion of farms receive the FAEP special subprogram subsidy and for the amount received, since these may have direct implications for farms’ input use, land allocation, and grain yield. In order to control for a possible selection bias due to only some farms registering in the FAEP special subprogram, we run a first-stage random-effects tobit regression with the amount of the special subprogram subsidies received by the farm as the dependent variable.13 The amount of the special subprogram subsidies predicted from the estimation of the tobit model is then incorporated into the right-hand side of each equation in the system.
The second issue is that some farms do not have any voluntary set-aside (i.e., lf,it = 0). In order to deal with the problem of corner solutions for set-aside land, we follow the approach presented by Shonkwiler and Yen (1999), which allows one to estimate a system of equations when some of the dependent variables are censored. Details are provided in Appendix A.
V. Data
This study uses farm-level data on physical and financial variables for agricultural production, obtained from bookkeeping records that provide the Finnish data for the European Commission’s Farm Accountancy Data Network (FADN). The records are collected annually and contain information on crop areas, crop yield, total expenditures on fertilizers and pesticides, work hours, and compensatory payments received, including agri-environmental payments. While the data distinguish payments made through the FAEP general and special subprograms, they do not specify the agri-environmental measures adopted. The analysis spans the years 1996-2005, that is, from Finland’s second year in the EU to the last year when crop-specific CAP arable-area payments were used.14 The final sample used in the analysis includes farms that are located in support region A, B, C1, or C2; that had some land allocated to grain crops; and that attributed a maximum of 30% of their total variable costs to animal production. The resulting data comprise 343 farms and 1,564 observations (an unbalanced panel).
Full-time farm enterprises are overrepresented in the bookkeeping data, whereby the average farm size is larger than the national average. This feature is also present in our sample. Otherwise, the sample is representative of grain farms in Finland. The patterns of farm size across time and geographic location are similar for the sample and for national data. Average farm size increases over the period 1996-2005, and decreases from south to north (Appendix Tables B1 and B2). In terms of the geographic location of the farms, southern Finland (support regions A and B) is somewhat overrepresented in our sample (Appendix Tables B3 and B4). Fertilizer purchases in proportion to cultivated land in the sample show a pattern that is similar to national statistics (Appendix Table B5).
Mean Levels of Grain and Set-aside Areas and Fertilizer Use in the Sample
The farm data were complemented with average national crop prices collected from Finnish Agriculture and Rural Industries, published by MTT Agrifood Research annually, and with price indices for fertilizers and pesticides (100 = 2005) and labor prices, obtained from Statistics Finland (Appendix Table C1). The average hourly wage for forest maintenance work was used as a proxy for the labor price, as statistics on agricultural wages are not available. The same input price indices were applied to all farms. For output price we computed region-specific values using the farm-level data on revenues, yields, and areas used for each grain crop (barley, wheat, oats, and rye), complemented by the national statistics on crop prices. Total grain output for each farm was calculated as the ratio of grain revenue to the region-specific grain price.15 The per-hectare subsidy rates for grains and set-aside used in the analysis are the sums of the CAP arable-area, CAP less-favored-area, national crop production, northern aid, and FAEP general subprogram support rates applicable to each land use and each support region. The subsidy rates were ascertained from Finnish Agriculture and Rural Industries.16
Summary statistics (Table 1) show that on average the grain area per farm increased over the study period. This trend is similar to the one observed for Finland as a whole (see, e.g., Niemi and Ahlstedt 2005). The intensity of production in terms of fertilizer use decreased on average, also reflecting the national trend (Appendix Table B5). The explanation for the decrease in fertilizer use may be the strong increasing trend in fertilizer price over the same period (Appendix Table C1). There is indeed a negative correlation between fertilizer price and fertilizer use (-0.78). The use of pesticides, in turn, increased over the same time period. Again, the explanation may be the strong decreasing trend in pesticide price (Appendix Table C1), with a negative correlation between pesticide price and pesticide use (-0.82).
As shown in Table 2, overall the total subsidies in proportion to land area in our sample increased over the study period. The subsidies paid for areas in grain production increased on average, while the subsidies for set-aside decreased. The general FAEP payments in proportion to land area decreased slightly on average. The proportion of farms participating in the FAEP special subprogram increased over the study period, whereas the average amount of special environmental subsidies in proportion to farm area among special subprogram participants decreased.
Mean Levels of Subsidies Received in Proportion to Land Area (2005 Euros) and Percentages of Farms Receiving General and Special FAEP Payments in the Sample Mean FAEP
VI. Estimation Results and Simulations
The six-equation system described in [9] was estimated on a total of 1,564 observations; the standard errors and t-statistics were obtained using nonparametric bootstrap with 500 replications.17 Chi-squared tests indicate overall significance in the case of each of the six equations. Our main interest is the impact of area-based grain and set-aside subsidies (which include the FAEP general subprogram subsidies) and FAEP special subprogram subsidies on land allocated to grain and voluntary set-aside as well as on the application of fertilizers and pesticides. The estimated coefficients for the corresponding four equations are shown in Appendix Table D1. The parameters of the profit equation (to preserve space, not included in Appendix Table D1) are consistent with expectations.18 The coefficient of the squared grain price is not statistically different from zero, satisfying the regularity condition of convexity in prices.19
Table 3 presents the median elasticities calculated on the basis of the estimated coefficients. All subsidy elasticities of grain and set-aside areas and input use are statistically significant. Area-based subsidies for grains and set-aside both had a fairly small impact on the grain-producing area. The median elasticity of grain area with respect to the grain subsidies is 0.15, which is close to the estimate reported by Lacroix and Thomas (2011). Using individual farm data from France, those authors found an elasticity of 0.16 for land planted with cereals with respect to area-based subsidies. By contrast, area-based subsidies for set-aside area had a large impact on the voluntary set-aside area in our sample: the median elasticity is 1.52, whereas Lacroix and Thomas report an elasticity of 0.12.20 The relatively large elasticity of voluntary set-aside is not surprising given that the entire field area has to be in either grain production or set-aside, and voluntary set-aside averages 9% of total arable land in the sample. Converting a unit of land from grain production to set-aside (or vice versa) results in a far larger percentage change in set-aside than in grain area. Lacroix and Thomas consider four groups of crops, so changes in one crop group’s subsidy rate may result in allocating more land into another crop group or to set-aside.
Elasticities of Land Allocation and Agrichemical Input Use
Area-based subsidies for grains also increased total use of fertilizers and pesticides, although the impacts were small (elasticities of 0.01 and 0.04). Adjustments in land allocation in response to area-based subsidies are small, and subsidies for grains amount to 1.4 times grain revenue on our sample. It seems plausible that farms do not change their input mix notably in response to small changes in subsidy rates.
The subsidies provided through the FAEP special subprogram had a positive but very small impact on land planted with grains, and a negative impact on voluntary set-aside for the farms that participate in the special subprogram. The most common conservation measure receiving support through the special subprogram and applicable to conventionally producing grain farms is riparian zones along field edges (Grӧnroos and Koikkalainen 2010). As riparian zones have to be adjacent to cropland, payments for riparian zones may spur farms to increase the area under cultivation. The signs of the effects on grain and set-aside areas thus are as one would expect. However, land eligible for support as riparian zones is restricted to areas bordering waterways, farms have to provide authorities with a detailed plan, and payments are granted only upon approval of the plan. These factors may explain the small magnitude of the effect. The special subsidies decreased total fertilizer use and increased total pesticide use, but these effects were small in magnitude: a 1% increase in the special subprogram subsidy resulted in only a 0.05% decrease in total fertilizer use and a 0.06% increase in total pesticide use. The signs and magnitudes of these effects are also as one would expect. Fertilizers are not used on riparian zones or wetlands. Increased pesticide use is plausible in that farms may perceive riparian zones with permanent vegetation to increase the need for weed control. As the effects on land use are small, one would expect the effects on farms’ total input use to also remain small.
To check the consistency of our estimates, we also calculated own price elasticities, which were all statistically significant (at the 1% level of significance) and plausible in terms of sign and magnitude. The median elasticity of grain area to grain price is 0.30, whereas the own price elasticities of the demand for fertilizers and pesticides are -0.91 and -1.96, respectively. The relatively small effect of grain price on grain area is explained by the significant role of subsidies in farm income.
Land Allocation and Fertilizer Use in the Sample under the Prevailing Policy and under a Counterfactual Scenario
To assess the impact of the FAEP payments on nutrient pollution from agricultural land, we apply the estimated land allocation and fertilizer demand functions to simulate two policy scenarios: (1) a factual scenario, where the FAEP payments are set at their historical values for 1996-2005, and (2) a counterfactual no-policy scenario, where the FAEP payments (both general and special subprograms) are set at zero. All other variables, including compensatory payments through the CAP and national crop production aid, remain at their actual historical values under both scenarios in order to identify the effects of the FAEP payments. Comparing the factual simulation with the counterfactual allows us to isolate the effects of the FAEP payments on land allocation and input use, assuming all else has remained constant. The counterfactual scenario changes the per hectare subsidy for grains by 10% to 45% and the per hectare subsidy for set-aside by 16% to 30% on the sample.
The results in Table 4 indicate that the impacts of the FAEP payments on land allocation and fertilizer use in our sample were minor. In terms of land allocation, the impact is counterproductive in that the payments increase the grain area and reduce set-aside, which, other things being equal, would increase nutrient loading. The land-use effect is due to the FAEP payments raising incentives for crop production. Set-aside was eligible for FAEP subsidies in 1996-1999, but the per hectare payment rates were markedly lower than those for land planted with grains. From the year 2000 onward, set-aside has not been entitled to FAEP payments. The simulation results confirm what behavioral models of farm decision-making predict: raising the profitability of cropland relative to set-aside increases the area under cultivation. Our findings are also in line with previous empirical results. Pufahl and Weiss (2009) found that the area under cultivation for participants in the German AEPs grew by 7.7% on average from 2000 to 2005, while the growth rate was only 4.2% for farmers not participating in an AEP.21
As fertilizers are applied on land for grains but not on set-aside, the counterproductive land use effect of the FAEP payments could even have increased total fertilizer use. However, the subsidies provided through the FAEP special subprogram promoted a reduction in fertilizer use, which resulted in the overall net effect of the FAEP payments going in the desired direction: the payments produced a 1.5% reduction in fertilizer use in the sample. This result is also in line with the findings of Pufahl and Weiss (2009) for Germany.
VII. Benefits and Costs of The Agri-Environmental Payments
Changes in land allocation and fertilizer use will affect nutrient pollution from agricultural land. We now proceed to assess the impact of FAEP payments on this environmental outcome in our sample of grain farms. Specifically, we use the predicted land allocation and fertilizer intensity under the factual baseline and the no-policy counterfactual scenario as inputs in environmental production functions in order to quantify the impact of program payments on nutrient pollution. Nutrient loading is also affected by management measures designed to filter nutrients or reduce runoff. Among the key measures that the FAEP imposes on grain farms are vegetative filter strips along waterways and cover crops in the winter.22 The FAEP-imposed vegetative filter strips and cover crops are removed in the no-policy scenario. Finally, to evaluate program-induced reductions in nutrient pollution in monetary terms, we couple the simulated nutrient load reductions with results from a valuation study assessing the benefits of reducing nutrient loads from Finland to the Baltic Sea.
Environmental Production Functions
Degradation of the quality of surface water in the Baltic Sea, the main recipient of nutrient loads from agriculture in Finland, is governed by the joint presence of nitrogen and phosphorus (see, e.g., Tamminen and Andersen 2007). We predict nutrient loads using environmental production functions, one for nitrogen and two for the focal forms of phosphorus, dissolved and particulate. These functions relate fertilization intensity to nutrient loading using coefficients that capture the impacts of crop choice, tillage practice, soil and field characteristics, and climatic factors. As we do not have information on the environmental characteristics of the farms, we apply a parameterization corresponding to the average soil and field characteristics and climatic conditions in southern Finland as an approximation (from Helin, Laukkanen, and Koikkalainen 2006).23 A potential limitation of the environmental production function approach is that it does not incorporate annual or seasonal variation in hydrology. However, such variation is unlikely to have had a significant effect on nutrient loading in the study region over the period 1996-2005: a study of 16 agricultural catchments in southern and western Finland over the period 1990-2004 did not detect statistically significant annual or seasonal trends in daily discharge or monthly cumulative runoff in any of the catchments. The environmental production functions predict edge-of-field nutrient loads, and the predicted loads conform to findings for different land uses and fertilizer intensities in Finnish field experiments.24
The notation in the environmental production functions is as follows: indexes N, DP, and PP refer to nitrogen, dissolved phosphorus, and particulate phosphorus; ϕm,j summarizes the impact of land use and tillage j (grains, grains with wintertime vegetation, set-aside) and local soil, field, and climatic conditions on the loss of nutrient m, with m = N,DP,PP; sm and dm are the shares of nutrient m loss carried through surface and drainage flow; B denotes the share of land allocated to vegetative filter strips; bm measures the effect of such strips on the loss of nutrient m;
is a reference nitrogen fertilization level; and θ is the soil phosphorus level. We consider a compound fertilizer with 20% nitrogen and 3% phosphorus content.25 Given a predicted fertilizer quantity x, the amounts of nitrogen and phosphorus applied are xN = 0.20x and xP = 0.03x.
Nutrient Load Parameters
The nitrogen load (kg/ha) under land use j is given by26
[10]
the dissolved phosphorus loss (kg/ha) by
[11]
and the particulate phosphorus loss (kg/ha) by
[12]
Table 5 shows the nutrient load parameters ϕm,j and the reference nitrogen fertilization level
for each crop. Since we consider aggregate grain production, we use a weighted average of the parameters in Table 5 to describe nutrient loading from grain areas.
The other parameter values used for equations [10],equations [11],[12] are as follows: B was set at 0.29% for 1996–1999 and 0.40% in 2000–2005 in the baseline factual scenario, based on averages for farms participating in the FAEP (MAF 2004), and at 0.04% in the nopolicy scenario (CAP field margin). The parameter values bN = 0.2, bDP = 1.3, and bPP = 0.3 are estimates obtained from Lankoski, Ollikainen, and Uusitalo (2006). The parameter values sN = 0.5, sDP = 0.7 and sPP = 0.7 are average values from Turtola and Paajanen (1995). The soil phosphorus levels θ are soil test averages for each province for the periods 1996–2000 and 2001–2005, obtained from Viljavuuspalvelu Oy (Soil Testing Service Ltd.). Unfortunately, our data do not record the extent of vegetative filter strips and wintertime cover crops on the sample farms. As an approximation, we apply the proportion of the total field area covered by vegetative filter strips and wintertime cover crops for all the farms participating in the FAEP. The share of field area under wintertime cover crops was set at 30% in support regions A and B and 0% elsewhere in the years 1996–1999, an approximation based on the FAEP requirements for that period. In the years 2000–2005 the share was set at 14.8%, the average on farms participating in the FAEP in the period (MAF 2004). The most common way to maintain cover crops (crop stubble) in the winter has been reduced tillage (MAF 2004), and we apply ϕm,j parameters corresponding to reduced tillage on the area covered by wintertime cover crops.
Monetary Benefits of Reduced Nutrient Pollution
Valuation studies assessing the benefits of reducing nutrient loads to the Baltic Sea generally address water quality improvements that are attributable to the combined effect of reductions in nitrogen and phosphorus loads (e.g., Sӧderqvist 1996, 1998; Markowska and Zylicz 1999; Kosenius 2010). Accordingly, what is needed for assessing the benefits of reduced nutrient pollution on the basis of valuation studies is a composite measure of nitrogen and phosphorus loads. We consider a weighted sum of nitrogen and phosphorus as such a composite measure. Thus, environmental damage is connected to a composite nutrient load ZNP, defined as
[13]
where ZN is the nitrogen load and zp is the sum of dissolved and particulate phosphorus loads. We consider two alternative parameterizations: one where nitrogen and phosphorus have equal weights, with a = 1, and one where phosphorus is given the weight a = 7.2 to reflect the prevalence of nitrogen-fixing blue-green algae in the Baltic Sea.27
Kosenius (2010) conducted a choice experiment to assess Finns’ willingness to pay (WTP) for water quality improvements associated with reducing nitrogen and phosphorus loads to the Baltic Sea. The estimated annual WTP for reducing Finland’s nitrogen and phosphorus loads by 7,986 and 525 tons per year, relative to the 1997-2002 levels, ranged from €652 million for a multinomial logit to €945 million for a random parameters logit model, with 95% confidence intervals of (€602–€702) and (€891–€998) million. We computed a constant marginal benefit by dividing the annual national WTp by the annual nutrient load reduction underlying the choice experiment, measured in terms of a composite nutrient load reduction. Table 6 displays the marginal benefit measures corresponding to the multinomial logit and random parameters logit models (in 2005 prices).
Constant Marginal Benefit of Reducing Nutrient Loads from Finland
Benefit-Cost Comparison
The impact of agri-environmental payments on social welfare comprises changes in consumer surplus and producer surplus. To achieve transparent comparisons with a previous EU AEP study, that by Chabé-Ferret and Subervie (2013), we focus on consumer surplus, measured by the benefits in terms of nutrient load reductions attributable to FAEP payments. The impact of the FAEP payments on total profits in our sample is also small, amounting to an increase of 2.7% over 1996–2005. On the cost side, we consider the direct costs of the FAEP payments, administrative costs, and the opportunity cost of public funds. Based on National Audit Office of Finland (2008), the administrative costs of the FAEP amount to approximately 10% of all program payments. A plausible value for the opportunity cost of public funds in Finland is 1.15.28
The estimated effect of the FAEP payments was to reduce nitrogen loading from the sample of grain farms by 11%, and phosphorus loading by 13% over 1996-2005.29 The consequences of these changes for environmental damage related to the surface water quality of the Baltic Sea are reported in Table 7. The damage measures have been calculated on the basis of the farm-source nutrient loads estimated to reach the Baltic Sea.30 Part A in the table displays the estimated damage from nutrient loading originating from the sample farms for each scenario and for alternative damage parameterizations. Part B shows the overall change in damage and the benefit-cost ratio for the FAEP payments to the sample farms. Overall, the effect of the FAEP payments was to reduce the damage from grain production by 11% to 12%. These estimates combine the reduction in the per-hectare nutrient load from grain areas stemming from decreased fertilizer use as well as the increase in grain area and decrease in set-aside area that are attributable to the agri-environmental payments.
Benefits and Costs of the Nutrient Load Reductions Attributable to Agri-environmental Payments on the Sample of Grain Farms, 1996-2005
The total damage avoided in the sample is robust to the choice of composite nutrient load measure but quite sensitive to the model used to obtain the underlying WTP estimates. For the multinomial logit specification, the benefits clearly fall short of the costs, with benefit-cost ratios estimated at 0.68 to 0.73. For the random parameters logit specification, the benefits are approximately on par with the costs, with benefit-cost ratios estimated at 0.99 to 1.05. The superiority of one WTP model over another is not straightforward. Kosenius (2010) found support for heterogeneous preferences for water quality attributes, which speaks for the use of the random parameters logit model over the multinomial logit.
It should be noted that our assessment of environmental benefits focuses on surface water quality in the Baltic Sea. Additional benefits may be attributable to improvements in water quality in Finland’s inland waters and reductions in the application of pesticides. We have not been able to evaluate these changes as there are no empirical studies available on the relevant nonmarket benefits. In terms of biodiversity the direction of the overall FAEP impact is not clear. Riparian zones and set-aside have been shown to promote biodiversity (e.g., Kuussaari et al. 2007; Ma, Tarmi, and Helenius 2002). The FAEP payments provided incentives for establishing riparian zones but reduced set-aside. Again, there are no empirical studies available that would allow us to quantify the overall biodiversity effects of the increase in riparian zones and the reduction in set-aside associated with the FAEP payments.
VIII. Conclusion
This study presents a structural econometric model designed to evaluate the consequences of payments through the FAEP—one of the most extensive AEPs in the EU—for nutrient pollution originating from grain production. We estimated farms’ land allocation and input decisions under a system of compensatory payments, including agri-environmental subsidies. The estimated land allocation and input demand functions were then used to predict the impact of the agrienvironmental payments on grain and set-aside areas and fertilization intensity. We next combined the predicted land allocation and input use with environmental production functions to assess the impact of program payments on water protection.
The econometric and simulation analyses indicate that the agri-environmental payments to grain farms had a fairly small effect on fertilizer use and area of land used for grain production over the period 1996–2005. At –1.5% the impact on fertilizer use is smaller than that indicated by Pufahl and Weiss (2009) for German AEPs (average treatment effect of – 9.5%). The FAEP general subprogram payments are proportional to area in grain production. By raising the relative profitability of land used to produce grain, the payments had the counterproductive impact of reducing the amount of land in set-aside. Accounting for FAEP-imposed specific water protection measures, vegetative filter strips and wintertime vegetation, the preventive impact of the payments over the period 1996–2005 was to reduce nitrogen loading by 11% and phosphorus loading by 13% relative to what would have happened without agri-environmental subsidies. Combined with monetary estimates for damage from nutrient pollution, the results indicate that the agri-environmental payments reduced the damage from grain production by 11% to 12%, with the estimated ratio of benefits produced by program payments to costs ranging from 0.68 to 1.05.
Overall the agri-environmental payments have reduced nutrient pollution from farms. This finding is consistent with results by Chabé-Ferret and Subervie (2013) for the impact of French agri-environmental schemes targeted at reducing nitrogen loading. However, the load reductions achieved fall short of Finland’s water protection targets and the reductions needed to significantly improve water quality in the Baltic Sea. This suggests that more specifically targeted policies would be needed to further reduce farm-source nutrient loading, such as taxing fertilizers and emphasizing payments for devoting land to specific water protection measures. The empirical modeling framework presented in this study could be applied to examine the effects of fertilizer taxes, alternative agri-environmental payment rates, and the relative emphases to be placed on payments provided through general and more specifically targeted agri-environmental schemes. We leave comparing the cost-effectiveness of different types of policies as a topic for future study.
The predictive ability of our modeling approach regarding grain and set-aside areas and fertilizer use is strong. Prediction of the effects of agri-environmental payments on nutrient loading is less reliable given that the environmental production functions have been calibrated for average field characteristics and hydrological conditions in southern Finland and thus provide only an approximation for the study area as a whole. An ideal dataset would include farm-level information on field characteristics and hydrological factors as well as the precise location of the farm. The accuracy of nutrient load predictions could then be improved by integrating the land allocation and fertilizer intensities simulated with the economic model with catchment-scale physical models simulating nutrient loads from different land uses and fertilizer intensities (e.g., INCA, Wade et al. 2002; and SWAT, Arnold and Fohrer 2005). Our estimate of the benefits from the FAEP is based on reductions in farm-source nutrient pollution, as the program’s main focus is on water protection objectives. The program also seeks to reduce the risks associated with the use of pesticides and to maintain biodiversity and rural landscapes. Possible benefits produced in terms of these additional objectives are not included in our benefit estimate. That estimate was also derived under the assumption of constant marginal damage from nutrient loading, which is a simplification. However, as the predicted changes in nutrient loading are not very large, constant marginal damage provides a reasonable approximation.
Appendix A Description Of Shonkwiler and Yen’S Approach To Control For Censoring of Land Set-Aside
In our model, the nonconditional expectation of land set-aside for farm i at time t (lf,it) can be written as follows:
[A1]
Let us denote by dit the variable taking the value 1 if lf,it > 0 and 0 otherwise. We assume the following specifications for the corresponding (unobserved) latent variables:
[A2]
with β3 and α vectors of unknown parameters and xit the vector of explanatory variables, which includes the price of output, the prices of fertilizer and pesticides, the set of subsidies and total land. The vector
contains explanatory variables that are assumed to influence the farm’s decision to have some set-aside (xit and
can have variables in common), and εit and vit are random errors assumed to follow a bivariate normal distribution with cov(εit, vit) = δ. In addition, we have the following relationships:
[A3]
The unconditional mean of lf,it is (Shonkwiler and Yen 1999):
[A4]
where ϕ(·) and Φ(·) are the standard normal probability density function and cumulative distribution function, respectively. Hence, the equation for land set-aside to be estimated in the system is as follows:
[A5]
The estimation procedure involves two steps. In the first step, estimates
of α are obtained from the estimation of a random-effect probit model using the binary decision to set aside land (dit = 1,0).31 We use as independent variables the proportion of land planted with grain in the previous period; the prices of fertilizer, pesticides, and labor; the per-hectare area-based subsidies, including CAP arable-area and less-favored-area and the FAEP general subprogram subsidy rates; the per-hectare subsidy for set-aside; the price of grass; and dummies for support regions. The estimates
of α are then used to calculate
.
In the second stage, the system is estimated with the following equation for set-aside:
[A6]
Estimation results of the first-stage random-effect probit model are not shown here but are available upon request.
Appendix B Summary Statistics For The Sample And The Population of Finnish Grain Farms
Average Farm Size (ha) by Year and Agricultural Support Region, Sample
Average Farm Size (ha) by Year and Agricultural Support Region, All Finnish Grain Farms
Proportion of Farms Located in Each Agricultural Support Region by Year and over the Entire Study Period, Sample
Proportion of Farms Located in Each Agricultural Support Region by Year and over the Entire Study Period, All Finnish Grain Farms
Fertilizer Purchases in Proportion to Cultivated Area (kg/ha)
Appendix C Price Statistics
Mean Grain Price, Mean Labor Price, and Fertilizer and Pesticide Price Indices (Base 100 in 2005)
Appendix D Estimation Results (Four Equations Out of Six)
Within-SUR (Seemingly Unrelated Regressions) Estimation Results (Main Equations of Interest), 1,564 Observations
Footnotes
The authors are, respectively, senior economist, Government Institute for Economic Research, and adjunct professor, Department of Economics, University of Helsinki, Helsinki, Finland; and Australian Research Council Future Fellow, School of Economics, University of Queensland, Brisbane, Australia.
↵1. There is a substantial ecological literature that investigates AEP impacts. While not an exhaustive list, examples of studies dealing with European programs include those by Ekholm et al. (2007), on Finland; Marggraf (2003), on German states; and Primdahl et al. (2003), on nine EU member states and Switzerland. The ecological literature focuses on trends in environmental indicators or the effect that stated goals, in terms of changes in agricultural practices, would have on the environment. This literature has not attempted to disentangle the effects of AEPs from other factors affecting production decisions such as input and output prices and other agricultural support policies.
↵2. The general scheme also lists a number of mandatory, albeit loosely defined, environmentally beneficial practices. These include farmer training, farm environmental planning and monitoring, environmentally sound use of pesticides, maintenance of biodiversity, and landscape management.
↵3. While general support is granted on the basis of an annual application, eligibility for special support is determined on the basis of a detailed environmental plan and farm visits, and approved farms sign a 5- to 10-year contract. Participation in the general subprogram is a prerequisite for registering in the special subprogram.
↵4. The climatic conditions in northern Finland are not suitable for grain production.
↵5. The asymmetries arise from renegotiations with the EU on which parts of Finland are eligible for less-favored-area support, from transitional support that was only payable during the first years in the EU and was gradually phased out, and from national payments in support regions A and B that have been renegotiated with the commission every few years.
↵6. Nitrogen fertilization rates for each year have been computed by dividing fertilizer expenditure by fertilizer price, under the assumption that farms use the compound fertilizer typically used in grain production in Finland (20% nitrogen content).
↵7. The studies differ in terms of how they derive land allocation equations. Lacroix and Thomas (2011) obtain land use equations by differentiating the profit function with respect to crop area subsidies, Arnade and Kelch (2007) by setting the shadow price of land equal to the observed land prices. Fezzi and Bateman (2011) follow the original Chambers and Just approach and derive land use share equations using the first-order conditions of farms’ profit maximization problem.
↵8. Based on the results of Koundouri et al. (2009), we assume that farmers are risk-neutral. Using the same profitability bookkeeping data as the present study over the years 1992-2003, Koundouri et al. found evidence that farmers were risk-averse prior to Finland’s accession to the EU in 1995 and risk-loving thereafter, due to the increase in the nonrandom part of farm income brought by the introduction of the CAP. For the period 1995-2003, Koundouri et al. estimated the risk premium to be between -2% and 2% of farm profit. Given the small magnitude of the risk premium, we consider the assumption of risk-neutrality over the 1996–2005 period a reasonable approximation.
↵9. See Eurostat: http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=apro_cpp_crop&lang=en.
↵10. Under the agricultural conditions in Finland, grains have similar nutrient loading potential, and they are typically grouped together in land use analyses (see, e.g., Helin, Laukkanen, and Koikkalainen 2006; Ekholm et al. 2007).
↵11. In agricultural statistics this land use category includes land under annual crops and land temporarily fallow.
↵12. Set-aside was compulsory for farms receiving CAP arable area payments in 1992-2007. The set-aside subsidy was only paid to set-aside area exceeding the mandatory area. Small farms were exempt from the requirement. To deal with the presence of voluntary and mandatory set-aside, we proceed as follows: if a farm’s observed set-aside area exceeds the CAP requirement, we treat the difference as voluntary set-aside and include it in the land set-aside equation. If the observed set-aside is less than the CAP requirement, we assume that the farm is exempt and that its entire set-aside area is voluntary. Finally, if the reported set-aside area equals the CAP requirement, we assume that there is no voluntary set-aside, and assign the value zero to set-aside in the land allocation equation.
↵13. The independent variables are the farmer’s age, the price of pesticides, the price of labor, the share of the total land that is rented, the price of grain, the number of animals on the farm, total farm size, and province dummies (our sample covers 17 provinces).
↵14. The bookkeeping data do not have entries for agri-environmental support for 1995, Finland’s year of accession to the EU. The EU single-farm payment was introduced in Finland in 2006. The single-farm payment is based on the land that farms manage or own, not on the crops that they produce.
↵15. Because data on grain revenue were missing for some farms and we sought to avoid a misreporting error, grain revenue for each farm and each year was computed as the median (per-hectare) grain revenue in the support region in that year multiplied by the grain area of the farm.
↵16. The CAP arable-crop subsidy for each support region was calculated as the weighted average of the subsidies for barley, wheat, oats, and rye, where the weights were the average shares of land allocated to each crop in the support region on the farms in our sample.
↵17. Monetary values have been converted into 2005 euros using the consumer price index (see Statistics Finland: URL: www.stat.fi/til/khi/index_en.html).
↵18. The full set of estimated coefficients, including results of the first-stage tobit model (with the amount of the FAEP special subprogram payments received by the farm as the dependent variable), is available from the authors upon request.
↵19. The profit function is not strictly convex, which indicates that farms do not adjust their land allocation and input use in response to changes in grain price, for price changes of the magnitude observed in our data. This is not surprising given that farm subsidies are an important source of income for Finnish grain farms: on average, the total subsidies for land in grain production amount to 1.4 times the total grain revenue on the sample. Furthermore, the land use choice is between grains and set-aside, where the latter only earns an area subsidy.
↵20. The elasticities of voluntary set-aside with respect to subsidies have been calculated using the estimated coefficients for the set-aside land in equation [9], which account for the fact that only a proportion of farms has voluntary set-aside.
↵21. Results from a comprehensive analysis of land use changes in the US, based on microlevel data, also suggest that federal farm payments have boosted crop acreage, partially offsetting cropland retirement induced by the CRP and falling net returns on crops (Lubowski, Plantinga, and Stavins 2008). Findings from a farm-level analysis of the production effects of US farm programs also suggest that government programs, even largely decoupled payments, increase growth in farm size (Key, Lubowski, and Roberts 2005).
↵22. The FAEP general subprogram requires a 1 m wide field margin along main drains and 3 m wide filter strips along streams and other waterways. Farms participating in the FAEP special subprogram may have constructed wider riparian zones. In what follows, we refer to all of these buffers as filter strips. The CAP payments require field margins 0.6 m in width along main drains and waterways. By cover crops we refer to any wintertime crop cover, including crop stubble left behind after harvest.
↵23. Helin, Laukkanen, and Koikkalainen (2006) have calibrated the coefficients for an area that covers approximately support regions A and B in our study. The calibration draws on physical models predicting nitrogen and phosphorus loads (SOIL-N and IceCream, respectively) and monitoring data on nitrogen and phosphorus loads from agricultural land for year 2003.
↵24. We will account for the distance to the receiving body of water in our damage specification.
↵25. This mix was the most commonly used mix for grains in the study period in Finland.
↵26. The function predicting nitrogen load was developed by Simmelsgaard (1991) and Simmelsgaard and Djurhus (1998), and those predicting dissolved and particulate phosphorus losses by Uusitalo and Jansson (2002). Similar environmental production functions have been applied by Lankoski, Ollikainen, and Uusitalo (2006) and Laukkanen and Nauges (2011).
↵27. These organisms are able to bind nitrogen from the atmosphere to phosphorus in the water. The ratio of nitrogen and phosphorus in algae averages 7.2 (Redfield, Ketchum, and Richards 1963), and due to nitrogen fixation, phosphorus entering the water can result in the conversion of an average of 7.2 times its weight of atmospheric nitrogen into a form available to aquatic plants, thereby potentially causing 7.2 times more eutrophication than nitrogen.
↵28. Kuismanen (2000), using a labor supply model, estimated the dead-weight loss of Finnish taxation to be 15%.
↵29. The average nitrogen loads for the factual and counterfactual simulations were 19.2 kg/ha and 21.6 kg/ha, and the average phosphorus loads 1.2 kg/ha and 1.4 kg/ha. The simulations reproduce magnitudes of nutrient loads that are in line with ecological assessments (Rekolainen et al. 1995; Vuorenmaa et al. 2002).
↵30. Approximately 78% of the nitrogen load and 54% of the phosphorus load from the study area finds its way into the Baltic Sea. Lepistöo et al. (2006) provide an estimate for the proportion of the nitrogen load that reaches the sea. We thank Petri Ekholm of the Finnish Environment Institute for providing an estimate for the proportion of the phosphorus load. The estimate has been calculated from a regression model documented by Rankinen et al. (2010).
↵31. As alternative model specifications we considered the simple probit model, as well as the fixed effect and random effect conditional logit models. The estimation results of the random effect panel probit model and corresponding specification tests indicate that random effects should be included. Thus, the random effect probit model is preferred to the simple probit model. The overall fit of the fixed effect conditional logit model was low, and the likelihood ratio test of global model significance was rejected at the 5% level of significance. Finally, the estimation of a random effect conditional logit model produced results that are very close to those obtained with the random effect probit model.
References
In this issue
Jump to section
- Article
- Abstract
- I. Introduction
- II. Background
- III. Behavioral Model: Farms’ Decisions On Land Allocation and Input Use
- IV. Model Specification and Estimation
- V. Data
- VI. Estimation Results and Simulations
- VII. Benefits and Costs of The Agri-Environmental Payments
- VIII. Conclusion
- Appendix A Description Of Shonkwiler and Yen’S Approach To Control For Censoring of Land Set-Aside
- Appendix B Summary Statistics For The Sample And The Population of Finnish Grain Farms
- Appendix C Price Statistics
- Appendix D Estimation Results (Four Equations Out of Six)
- Footnotes
- References
- Figures & Data
- Info & Metrics
- References