Keeping Land in Grass: Re-Enrollment Motivations with the Environmental Quality Incentive Program after the Conservation Reserve Program

Conservation programs from the United States Department of Agriculture (USDA) have been ongoing for decades - the Farm Services Agency's (FSA) Conservation Reserve Program (CRP) celebrated its 35-year anniversary in 2020 (FSA, USDA, a) and the Natural Resource Conservation Service's (NRCS) Environmental Quality Incentives Program (EQIP) has been in place since 1996 (NRCS, USDA, a). Many farmers and millions of acres of have been enrolled in these programs since their inception. The CRP program retires land to achieve conservation goals such as soil health, water quality and wildlife habitat. The EQIP program is a working lands program, assisting producers to apply conservation practices while continuing to use the land for production. For an excellent description of such programs, please see Maher et al. (2003).

Social science research concerning the adoption of federal conservation programs is extensive. Ranjan et al. (2019) found 171 studies conducted from 1982 and 2017 examining the adoption of conservation practices and programs. These studies produced numerous predictors of producer attitudes and behavior toward conservation programs, with “few consistent determinants of conservation adoption” (Ranjan et al., 2019). Factors found to influence adoption of conservation practices include farm-level social characteristics, farm-level biophysical characteristics, farm-level management and financial characteristics, and plus exogenous factors (Prokopy et al., 2019). The list of individual, social and ecological factors is long, and consistent support for any one factor, or group of factors, remains elusive.

Liu et al. (2021) argued that the long-term ecological benefits of EQIP, as well as CRP, can be attributed to those programs only if producers would have implemented practices without payment, thus making them “additional” conservation practices. This notion is important because the persistence of environmental benefits after CRP should be a valuable policy outcome. Barnes et al. (2019) investigated re-enrollment and persistence of participation in conservation programs. They found that producers with expiring CRP contracts were interested in continuing in the program, though they had mixed success in doing so. In addition, they found that management plans in a post-CRP situation were generally amenable to retaining the conservation gains from CRP.

Pathak et al. (2021) found that participants in programs like EQIP displayed a higher “intensity of adoption” of conservation practices. They found that participating in cost-share programs often led to an increase in the number of conservation practices, whether part of the cost-share contract or not. The salient question is whether these behaviors will persist through time. Skaggs et al. (1994) concluded that younger producers concerned about soil erosion were far more likely to either re-enroll into CRP or simply continue conservation practices. Dayer et al. (2018) put it this way: “…what remains largely unknown is what happens when payments stop: persistence or reversion? And why?”

This research seeks to evaluate attitudes and plans for producers in four Upper Midwest US states as they decided what to do with their acres after they did not re-enroll in CRP. We sampled one group of potential respondents who did not re-enroll in CRP and made other decisions regarding their land. A second group of potential respondents subsequently placed those former CRP acres under EQIP. Our primary question is whether the two groups leaving CRP, those who enrolled in EQIP and those who did not, have different attitudes or see different options for the former CRP land.

The literature cited above led us to reject an evaluation of factors leading to the adoption of conservation programs. Farmers with at least one expired CRP contract have already made an adoption decision regarding that program on that land. Instead, we are examining the attitudinal differences between two groups who have already adopted conservation programs and need to make decisions for their future conservation plans.

Our goal is to try to assess whether these acres will be placed on to a trajectory likely to retain grassland conservation gains over time. We are interested in two questions. First, what are the options and motivations of producers who cannot re-enroll in the CRP conservation program? Second, what are the options and motivations of producers that enroll their former CRP land into a second conservation program, EQIP?

We built a sample frame from a list of producers who had at least one CRP contract expire between 2015 and 2020 in the states of Colorado, Montana, North Dakota or South Dakota. We received samples from two sources, the FSA and NRCS. First, from the FSA, we obtained a list of landowners that had at least one expired CRP contract from 2015 to 2020 in the target states. That group was split into two populations. The first were producers who returned former CRP land to some form of production. This list contained 6,514 names and addresses from which we drew a random sample of 2,200. The second were producers who enrolled former CRP land in the NRCS's EQIP program and are implementing EQIP contracts focused on keeping their former CRP land in grasslands. This list contained contact information for 668 producers. We surveyed the entire group.

The EQIP practices in the sample were focused on the grazing and rangeland conservation project practices are listed in Table 1. NRCS provided us with a complete list of EQIP participants using at least one of the practices in Table 1.

Both groups were surveyed by mail, with a web-based option. Overall response rates were low. The EQIP group response rate was 16.9% and the CRP group response rate was 13.7%. The low response rates and corresponding smaller sample sizes discouraged us from performing many types of analysis. For example, evaluating responses by state proves difficult due to low sample sizes for some states producing unreliable test statistics. Total returns by state are presented in Table 2.

The state numbers presented in Table 2 show 25 respondents listed as “Other”. A significant number of possible respondents in both groups had only business names in the database. Numerous types were represented including Limited Liability Corporations/Partnerships (LLC/LLP), corporations and various types of family trusts. Among our respondents, this represented almost 15% of the total returned surveys. Numerous potential respondents contacted us to say that they were legally operating the farm for the business entity in question but did not have enough information to answer the survey. Much of the acreage on such farms is leased to other operations.

An expected criticism of low response rates is that the data will suffer from non-response bias. Most research on the issue finds that nonresponse bias is more likely to affect univariate relationships and that it is reduced dramatically in multivariate settings (Amaya and Presser, 2017; Coon et al., 2020; Groves, 2006; Hellevik, 2016; Hendra and Hill, 2019). Still, to be prudent in this situation, we have chosen to limit our analysis and interpretation of results to the relationships we find in the sample we have. We cannot generalize any of these results to larger populations.

Finally, we take an alternative approach to analyzing response differences between the CRP and CRP/EQIP groups. Almost all research studies in this area report statistically significant differences between producers who adopt conservation practices and those who do not by applying p values to parameter estimates. While important, they do not offer much insight into how important a result might be relative to the stated hypotheses. “Effect size indicators are descriptive statistics that estimate the magnitudes of relationships or effects; they are meant to encourage substantive, rather than statistical, interpretations of significance (Brooks et al., 2013). Effect size is useful in this study because attitudinal differences between the two groups are hypothesized to be small since they both have experience with conservation programs. We have chosen a nonparametric approach to examining group differences and will report three effect sizes to evaluate those differences.

Both samples were presented with three sets of attitudinal items that allow us to compare responses across groups. Our questions are Likert-type scales of various lengths. Each scale is scored as Strongly Agree = 1, Agree = 2, Neither = 3, Disagree = 4, Strongly Disagree = 5. In discussing our analysis, the term “lower” is associated with agreement, while “higher” is associated with disagreement with the statement in the survey item.

The approach most often used to compare responses between two groups in this situation is the two-sample t-test. This ubiquitous test assumes sample independence, equal variance and normal distributions for the populations (Kloke and McKean, 2015). Attitudinal items such as those presented here are often in ordinal, Likert-type scales. At first glance, this type of categorical data appears to violate one or more of the assumptions of a two-sample t-test. The central limit theorem is sometimes invoked to ignore these violations, simply treat the ordinal data as if it were continuous and apply the t-test (de Winter and Dodou, 2012). Ordinal data often violates the normal distribution assumption about the population means. After examining box plots, histograms and tests to the data, we concluded that these data are non-normally distributed.

Non-parametric tests for two-sample situations can be used to evaluate Likert-scales for two independent samples and we agree with Cliff (1993) that ordinal questions deserve ordinal answers. We used the two-sample Wilcoxon rank-sum test. The Wilcoxon procedure tests that hypothesis that two groups are from the same population, with the alternative being that one group is “stochastically larger” than the other (Kloke and McKean, 2015). We performed Wilcoxon tests using the “wilcox_test” function in the “rstatix” package of the R software (Kassambara, 2022; R Core Team, 2022). We set up the data to determine if the CRP group has higher scores on our items than does the EQIP group.

In addition, we are reporting effect sizes along with the non-parametric statistics. Reporting effect sizes is a standard practice in psychology and medicine because they offer readers an easily understood way to distinguish between statistical and practical significance (Kirk, 1996). They allow us to estimate not just the statistically significant differences, but to estimate how large, and important, the difference might be. Effect sizes for nonparametric data are not as numerous as for normally distributed data, but we are reporting two. One effect size estimate used commonly for nonparametric situations is Cohen's r. Estimates of r are used in two-sample situations because the measure is independent of sample size (Fritz et al., 2012). Cohen advises that r effects be interpreted as follows: 0.10 – 0.30 are small, 0.30 – 0.50 are medium and >.50 are large (Cohen, 1988).

Second, we estimate the effect size Cliff 's Delta, δ. This effect size measure determines the degree of overlap between two ordinal distributions (Wan et al., 2020). It ranges from −1 to 1, with 0 indicating complete overlap between the distributions. The sign of δ indicates which groups is dominant: when the data from the second group (EQIP) has a higher score (more likely to disagree) than the first group (CRP), then Cliff 's δ is negative. A suggested interpretation for the absolute value of δ are |δ| < 0.147 – Negligible, 0.147 ≤ |δ| < 0.330 – Small, 0.330 ≤ |δ| < 0.474 – Medium and |δ| ≥ 0.474 – Large (Wan et al., 2020).

The third effect size statistic we report is a nonparametric extension of what is commonly referred to as the “common language effect” (CL) statistic (Vargha and Delaney 2000; McGraw and Wong 1992). CL estimates “the probability that a randomly chosen member of Group 1 scores higher than a randomly chosen member of Group 2 (Ruscio, 2008).” Delaney and Vargha (2002) estimated a CL effect for non-parametric data, labeled A. Simulation studies showed that estimates of A “yielded unbiased estimates of parameters and stable standard errors across normal and nonnormal population distributions as well as homogenous and heterogenous population variances” (Ruscio, 2008). A is sometimes referred to as a measure of “stochastic superiority” (Delaney and Vargha, 2002). Values for A range from 0 to 1, with a value of 0.5 indicating that there is no difference between the groups. The estimates of Delaney & Vargha's A can be interpreted as the probability that the CRP group will score higher (more likely to disagree) than the EQIP group on an item. A suggested interpretation for values of A < 0.56 are small, from 0.56 – 0.71 are medium, and ≥ 0.71 should be considered large (Vargha and Delaney, 2000).

We presented data from a survey of producers in the upper Great Plains who had at least one CRP contract expire in the last seven years. The survey covered two groups, one with a retired CRP contract and a second with a retired CRP contract who then transitioned that land into the EQIP program. Re-enrollment results were compelling: the majority of the CRP group considered re-enrolling while the majority of the EQIP group did not. We found that 50% of our EQIP group had previously enrolled in EQIP. The CRP group primarily moved to crops on the former CRP land, while EQIP respondents focused on cattle and grazing. Awareness of the EQIP program on the part of the CRP group was 58%.

This mitigates against adoption of EQIP by these producers. Almost 18% of the CRP group had participated in EQIP in the past and over 6% currently are so this level or awareness indicates that if NRCS wanted to recruit more participants for EQIP, there appears to be room for success.

We compared groups regarding what they considered doing with the land at the end of their respective programs, CRP and EQIP. The CRP group reported being more likely to convert land back to crops, the EQIP group reported being more likely to keep their land in grass, and both supported the idea of placing it back into another conservation program. The respondents were already participants in federal conservation programs. Regardless of their motivations, they had already adopted a conservation-based program in return for payment. It seems reasonable that they would be more similar than different.

Attitude items about the farm and stewardship showed that the EQIP group was indeed composed of ranchers as questions concerning livestock prices and ranching produced answers significantly different from the CRP group. But again, the effect sizes for all questions were small. Finally, the attitude items concerning agricultural practices and the producer's role in agriculture saw a similar pattern. Statistically significant differences for some items were apparent, but with very small effects. In this case, our conclusions may vary, but all have a few underlying themes.

Most of the CRP group would be re-enrolled if they could be. As reported above, over 80% of the CRP group considered going back to CRP. The fact that they did not may point to the program not accepting their applications even though they have experience with CRP.

But most interestingly, their attitudes about stewardship, the land and conservation ideas were very similar. Though years of research may have identified certain attitudes or motivations that sometimes underpin the adoption of conservation practices, in this case, this particular set of attitudes did not differentiate these two sample groups. Over time, both groups participated in USDA conservation programs, were still willing to do so and displayed a convergence of motivations and attitudes about their participation. This leads us to conclude that making additional programs such as EQIP or the USDA Conservation Stewardship Program (CSP) more widely known to farmers not re-enrolling in CRP might provide a foundation for the conservation persistence policy makers are seeking.