Does Extension Work? Impacts of a Program to Assist Limited-Resource Farmers in Virginia

Historically Black Land-Grant Colleges and Universities (HBCUs) (1890 institutions and Tuskegee University, among others) have made significant advances in carrying out their tripartite mission of teaching, research, and extension. Despite limited resources compared to their 1862 counterparts, and virtually no research funding until the 1970s, they have played an important role in conducting and extending the results of research on problems confronting limited-resource farmers, households, and rural communities.¹

In recent years, the U.S. Department of Agriculture (USDA) loan and technical assistance program called the Small Farm Outreach, Training, and Technical Assistance Program (2501) has strengthened efforts by the 1890 institutions to assist limited-resource farmers. In Virginia, this federally funded program is administered by Virginia State University (VSU), an 1890 institution. The goal of the 2501 program is to provide technical, management, and other income-enhancing information services to limited-resource farmers. In Virginia, the program currently reaches approximately 400 limited-resource farmers, many of them minority farmers. There are approximately 1,200 minority farmers in Virginia; many of them with limited resources.

The federal government has provided the bulk of the funding for this small-farm outreach program with supplemental funds provided by the Commonwealth of Virginia. However, in a period of scarce funds and increased demand for accountability from funding agencies and other stakeholders, institutions charged with implementing extension and outreach programs are challenged to “develop meaningful outcome measures that allow for adequate determination of their effectiveness and returns to public investment in these programs” (Essel, Clarke, and Tegene). In addition to helping justify levels of public investments, quantitative information about the benefits derived from extension programs can guide program design and help allocate resources.

Despite these challenges, there has been little systematic effort to quantitatively link public investments in 1890 colleges such as VSU to program outcomes. The main objective of this article is to empirically evaluate the impact on participants in the Virginia Small Farm Outreach, Training, and Technical Assistance Program as administered by VSU (and funded in part by the 2501 program). This evaluation provides guidance on how to structure subsequent programs targeted at limited resource and minority farmers. Such guidance is especially timely as the USDA revises its programs to assist minority farmers in response to the consent decree signed in January 1999. This decree effectively settled a class action lawsuit against the USDA by thousands of African-American farmers.

The economic impact of the program is measured in terms of its effect on the incomes of limited-resource participant farmers in Virginia compared with their net farm incomes had they not participated. The measurement method employed corrects for potential biases resulting from individual and household-level heterogeneity. Such heterogeneity leads to endogenous correlations between program participation decisions and farm incomes. Without correcting for this endogeneity, estimates of benefits of participation may be biased. We use measures of access to extension services to identify the impact of the (endogenous) participation and intensity of participation decisions on net farm income. The logic of these instruments is that access to such services has no impact on farm income except as felt through program participation and intensity of program participation.

Background on 2501 Program

U.S. Department of Agriculture initiated the Small Farm Outreach, Training, and Technical Assistance Program during fiscal year 1983. Its dual goal was to limit historical trends toward reductions in ownership of farmland by African Americans and to support HBCUs. The Farm Services Agency (FSA) entered into cooperative agreements with Land Grant institutions; each award recipient hires farm management specialists to provide one-on-one and group training.² The objective is to enable participants to operate their enterprises independently and produce adequate income (http://www.USDA.gov/agency/outreach/2501his.htm). The Virginia program is designed to improve the financial viability and standard of living of small-scale and socially disadvantaged farm operators through outreach and technical assistance. It provides educational programming in 35 counties, attempting to help participant farmers acquire, manage, and dispose of financial and production resources. It assists farmers with filling out loan applications for FSA loans, record keeping and budgeting, preparing farm and home plans, developing marketing plans, and developing high-value alternative enterprises and activities. Home visits are made by outreach agents associated with the program, and workshop and production meetings are held. The program helps participating farmers enhance their skills in making timely business decisions.

Over its 10-year life, the program has reached roughly 500 farmers. The 2501 program is also available in other states, and nationally more than 8,000 farmers have participated. Between FY 1994 and 2001, some $37 million was allocated to the national program by USDA, and the program grew to include other institutions as well, such as the 1862 Land Grant colleges. Qualitative assessments of the program indicate that it has been successful in ensuring the long-term sustainability of African-American farms (Hargrove). As a follow-up to this qualitative work, we investigate the impacts on farm income of program participation in Virginia.

Conceptual Framework and Empirical Specification

The model we employ examines decisions to participate in the program, intensity of participation, and impacts of these variables on net farm income. Let Y_P represent net farm income if the limited-resource farmer participates in the Virginia Small Farm Outreach, Training and Technical Assistance Program (hereafter referred to as the Small Farm Program) and Y_NP represent income following nonparticipation. Earnings in each case may be specified as:

(1)

(2)

where X_i is a vector of determinants of farm income and the subscript i denotes the individual (farmer) in question. The expected gain from participation is denoted by E(Δ) =E(Y_P−Y_NP). The individual is assumed to decide to participate in the program if expected gains from participation, net of participation costs, exceed zero. Define:

(3)

where Z_i is a vector of factors representing the determinants of participation, γ is a vector of parameters, and ɛ_i is an error term. I^*_i is unobserved. A person is assumed to participate if I^*_i > 0 and we observe the binary outcome I_i= 1. Otherwise, I_i= 0. If the program only affects earnings through the first element of β (an intercept shifter), the model of program participation's effect on income may be rewritten as:

(4)

where I_i is a dummy variable representing program participation. The effect of the program on the dependent variable is measured through α. The dummy variable cannot, however, be treated as exogenous if the decision to participate is based on individual self-selection (i.e., as shown in equation (3)). Such conditions might arise because the program purposefully targets more disadvantaged farmers, or because farmers with certain unobserved characteristics might be more eager to participate. Endogeneity of participation in (4) means Cov(U_i, ɛ_i) ≠ 0, and if it is ignored and equation (4) is estimated using OLS, the resulting parameters will be biased and inconsistent. One such solution to this problem is the use of instrumental variables techniques (Robinson, Heckman).

In addition to simple participation in the 2501 program, increased intensity of participation may also raise farm incomes. We measure intensity of participation through the number of visits by program outreach agents to the farm in question. This count of events may be modeled using Poisson regressions. Denote the incidence rate for the ith farmer as R_i and let this rate be:

(5)

where φ_i is the regression error. Now, examine the impact of R_i on net farm income by rewriting equation (4) as:

(6)

Intensity of participation can affect income in the same way as participation in the 2501 program. As intensity is also likely to be endogenous to the income-generating process (Cov (η_i, φ_i) ≠ 0), instrumental variables can also be used to estimate equation (6). This procedure is similar to those used in other studies of participation in programs such as the Food Stamp Program or other social welfare programs (Levedahl, Capps and Kramer). More recently, participation models have been used to analyze agricultural assistance programs such as the Conservation Reserve Program and forestry assistance programs (McLean-Meyinsse, Hui, and Joseph). In all these studies, the authors find that income plays a significant explanatory role in the participation decision. Indeed, a recent study of farmer participation in the U.S. farm programs found that “income is the best predictor of farmer's attitudes and behaviors” (Thomas and Thigpen).

As a consequence of this joint endogeneity of program participation and the resulting income conditioned on participation, the participation decision variables cannot reasonably be thought to be exogenous to the stochastic mechanism that determines the farmers' income. In other words, the mechanism that determines the income levels of limited-resource farmers also determines who participates and with what intensity.

Data Collection

Data were collected using a mail survey followed by a phone call to nonrespondents. Virginia State University provided a list of 400 farmers to form part of the data base for the analysis. The Virginia Agricultural Statistics Service (VASS) was instructed to select a group of 400 farmers whose racial composition, asset positions, and farm income situations mirror those of the VSU data. Like the VSU group, the VASS group is composed of farmers whose gross annual sales do not exceed $150,000. The control group is constructed using nonparticipants from both data sets.

The questionnaire was pilot-tested during fall 2000 by a VSU extension agent and one of the co-authors of this article. Virginia Agricultural Statistics Service then mailed the revised questionnaire to the 800 farmers during winter 2001. Overall, nonresponses were a major problem, resulting in only 205 usable questionnaires. The major subcategories of nonresponses included nonfarmers who responded to the survey (135), and deceased, absent, or otherwise unavailable farmers. Once these were omitted, the remaining major subcategories of nonresponses were income related, asset related, or related to the distance to extension and FSA offices.

It is not surprising that farmers filling out the survey would omit questions about their financial situation. More telling is that a substantial number of farmers do not know the distance either to their FSA or their Extension Service office. As the estimation results in the next section bear out, if the nonresponse is due to a lack of awareness about the proximity of the office, this could have an adverse effect on the farmers' income level.

Table 1 presents variable definitions and summary statistics. The dependent variable in equations (4) and (6), is computed as the dollar value of the previous three-year average of net farm income. The other income variable (OFINC) is computed similarly as an average of three years' prior income. Participants in the 2501 program have higher incomes, larger farm acreages, and slightly higher asset values than nonparticipants (table 1). Participants, however, have lower education levels, higher debt loads, and are more likely to be black. Twenty-seven percent of participants did not finish high school while 80% of nonparticipants finished high school or attended college. Participants are also more likely to have been turned down for a loan (one of the targeting criteria for the 2501 program) and are more likely than nonparticipants to have been visited by an extension agent.

Among all respondents to the questionnaire, 69% received no visits from Small Farm Program outreach agents; 9% received one visit; 10%, two to four visits; 8%, five to eight visits; 3%, nine to 15 visits; and 1%, 16 to 20 visits.

Estimation Results

Equations (4) and (6) were first estimated without treating participation and intensity of participation as endogenous to the farm income-generating process. The purpose of this exercise is to understand how estimates of program impacts are affected by self-selection. Neither participation variable is found to have a significant impact on farm income (see Appendix for details). The regression results are similar regardless of whether the participation variable (PART in column 1) or the number of farm visits by agents (NOVISIT in column 2) is included as a regressor. However, if these variables (as we suspect) are endogenous to farm income generation, then all the parameters will be biased and inconsistent (see Robinson for a discussion of this bias). In both cases, participation in the program was not found to have a statistically significant impact on net farm income. However, as noted above, because this program is targeted specifically toward farmers who are financially stressed and because the most stressed farmers might feel a need to join, self-selection and endogenous participation are likely to contribute to a downward bias on these estimated parameters. To examine this bias, we use a two-stage procedure to estimate the structural parameters in equations (4) and (6).

The first-stage estimates express the probability of participation and number of visits by agents as a function of all exogenous variables in the system. These equations are estimated using a probit and a Poisson regression, respectively. The structural estimates of the impact of probability of participation and number of visits on farm income are identified in the second stage using the variables PRVISIT, NOLOAN, DISTF, and DEBT (see table 1 for a description) as identifying instruments. Each of these variables is expected to only affect farm income through their impact on participation and intensity. They are logical candidates for use as identifying instruments.

Table 2 presents the first-stage estimates of participation and number of visits. The probit model for program participation shows that only race and prior visit by an extension agent are significant determinants of current participation in the Small Farm Program. Black farmers and those that received assistance from extension prior to 1990 were both more likely to participate in the program. Farm size, level of earned income off the farm, farm financial position, and household characteristics all had no significant impact on program participation. Oddly, the fact that a person was turned down for a loan (NOLOAN) is insignificant in the participation probit. This factor is supposed to be one of the targeting criteria for the Small Farm Program.

Efforts to understand the determinants of number of visits by the agents were slightly more satisfactory (column 2 in table 2). The pseudo R² for this regression is above 0.2 and coefficients on an additional two variables (the dummy variable for beef production and the farm debt variable) were statistically significant at the 5% level. Farmers with higher debt loads were likely to receive more visits from the agents, and beef producers, though they were no more or less likely to participate, received fewer visits, conditional on program participation. The regression results from table 2 were used to generate predicted values of the propensity to participate and numbers of visits. These predicted values were used in the second-stage regressions.

Table 3 presents the second-stage estimates of the structural determinants of farm income. The parameter estimates for the determinants of net farm income are similar regardless of whether participation or intensity of participation was modeled. Higher values of off-farm income are associated with lower net farm income, and more financial and human (in the form of the household size variable) assets are associated with higher farm incomes. Land assets were not significant contributors to farm income in either regression, although the parameter estimates are positive, as expected. Higher off-farm incomes mean less specialization in agriculture in Virginia, where more than 95% of farm operators report working off the farm and 40% work more than 200 days off the farm (USDA). Less specialization, in turn, is associated with lower net-farm incomes, and limited-resource farmers often use farm expenses to offset off-farm earnings for tax purposes. The negative coefficient³ for beef producers is further evidence that off-farm income is likely to substitute for on-farm income for part-time farmers; beef producers tend to be part-time farmers in Virginia.

More financial assets are clearly associated with higher farm incomes, and households with more members also show higher net farm incomes. This latter finding is attributable to a labor effect. Oddly, experience and higher levels of schooling do not significantly contribute to net farm incomes. Experience has the expected positive sign, but both high school (SCHOOL1) and college (SCHOOL2) schooling have negative signs. Although none of the coefficients on these variables are statistically significant, one would expect the signs to be positive, since the deleted educational category is less than high school education.

The regression results indicate that participation in the Virginia Small Farm Program alone is not enough to increase farm income. As shown in the results for equation (4), simple participation⁴ in the program does not raise farm incomes significantly (although the sign of the coefficient is positive). As intensity of participation increases, however, the effect on farm income increases and becomes significant. The coefficient on number of visits (second column) indicates that each additional visit of an agent is associated, holding everything else constant, with about a $3,300 increase in annual net farm income. We ran a similar model (model 6a) with the log of predicted number of visits on the right-hand side to allow for a nonlinear response. This variable is highly significant and indicates that at the modal number of visits (4) by 2501 agents, the marginal return to an additional visit is around $1,700.⁵ The significance of the visit variable is robust to alternative specifications of the model and provides evidence of a substantial positive impact of the program on farm incomes.

Intensity of participation's impact on net farm income is mediated through two paths. First, farmers who participate intensely in the 2501 program are assisted in receiving loans from a variety of sources. The low rate of rejection for loans for program participants may reflect 2501 program efforts to enhance loan applications. While debts on participant farms exceed those of nonparticipants, debt-to-asset ratios are lower, indicating evidence of efficient use of borrowed funds. Second, more intense participation leads to fuller interactions between agents and the farmer. These interactions allow the farmer to troubleshoot problems early. Evaluations of 2501 operations in Virginia suggest that increased interactions lead to better timing of farm operations, more active efforts to solve on-farm problems, improved planning, and quick referrals for problem solutions. More intense participation exposes the farmer to a wide means of improving farm operations, which helps raise incomes.

Evidence of bias created by the endogeneity of program participation is provided by comparing the results from the annex table and table 3. A formal test of endogeneity of participation and intensity of participation was performed (Davidson and MacKinnon).⁶ This test produced a p-value of 0.114 and 0.065, respectively, for equations (4) and (6). In the former case, there is only weak evidence that endogeneity of program participation is a problem. Part of this result may be due to the weak fit of the probit-reduced form used to generate the predictions. The number of visits by Small Farm Program outreach agents in equation (6) is clearly endogenous to the income generation process. Once endogeneity of participation is properly instrumented, the parameter estimates of the participation and intensity of participation variables grow in magnitude and significance. These results indicate that lower resource farmers are self-selecting into the program; without accounting for this selection, the econometric models indicate no significant income impact of the program. Once this endogeneity is controlled for, the program's impacts become stronger and more significant.

Conclusion

The Small Farm Program for limited-resource farmers in Virginia appears to significantly increase participants' net farm income, provided there is sufficient intensity of participation. Limited contact between agents and farmers, as measured by a single visit by an agent, has no significant effect on income. This finding is consistent with those of more qualitative analyses, which show that the most successful 2501 programs are those that recruit participants, assist farmers in obtaining loans, establish cooperatives, and introduce farmers to alternative enterprises (Hargrove). The results demonstrate that a successful program will involve multiple contacts with individual farmers.

Aggregate benefits of the program may approach $5 million per year (for a cost of a few hundred thousand dollars). However, extrapolation based on the benefits per visit and the number of visits is questionable, as impacts are shown to be nonlinear as the number of visits increases. The importance of accounting for endogeneity of participation is clear in the analysis. Lower resource farmers self-select into the program, which would bias the results unless it is accounted for in the model.

The findings clearly suggest a rationale for deepening the intensity of participation before broadening the program to include nonparticipants. Without such deepening, in the form of multiple farm visits by program agents, the program's impact will be lessened—active and intense participation makes the program work. The quantitative results in this article complement and strongly support the earlier qualitative evaluations of the 2501 program, providing additional evidence that the program yields significant benefits for farmers who have intensive interaction with the program.