2. Literature Review

In the 1960s, economists and sociologists (Burchinal, Reference Burchinal1961; Ford and Box, Reference Ford and Box1967; Haller and Sewell, Reference Haller and Sewell1967; Portes, Haller, and Sewell, Reference Portes, Haller and Sewell1968) extensively analyzed farming as an occupation. Much of the focus was on what factors determined the choice of farming as an occupation for the sons and daughters of farm operators. Gasson (Reference Gasson1969) surveyed eastern England farmers to assess their sons’ levels of aspiration to become farmers and whether they had realistic economic outlooks for agriculture. Gasson (Reference Gasson1968, p. 317) had earlier written, “Sons whose fathers have farms which are too small to provide a living should be made aware of their own prospects in farming, the range of opportunities open to them and the satisfactions that could be gained in other occupations before they commit to an unrewarding struggle in farming.” Gasson's (Reference Gasson1969) survey was conducted at a time when, in America, younger, more educated people were leaving farms and heading to cities and towns in search of a better future. Overall, the U.S. agricultural sector has experienced a decline in both the absolute level and relative importance of farm employment during the 20th century. Most rural areas of the United States have been reshaped by continued farm loss and an outmigration of residents (Mishra et al., Reference Mishra, El-Osta, Morehart, Johnson and Hopkins2002).

Although the literature investigating factors affecting individuals’ entry into farming is relatively scarce, a significant literature has amassed investigating the factors affecting farm exit (Breustedt and Glauben, Reference Breustedt and Glauben2007; Kimhi and Bollman, Reference Kimhi and Bollman1999; Mishra, Fannin, and Joo, Reference Mishra, Fannin and Joo2014; Pietola, Väre, and Lansink, Reference Pietola, Väre and Lansink2003; Weiss, Reference Weiss1999). Understanding the dynamics of farm exit behavior provides insight into the entry decision, as factors leading to exit would likely be those that discourage entry. For example, Rahelizatovo and Gillespie (Reference Rahelizatovo and Gillespie1999) showed that regional differences in milk production tended to simultaneously encourage exit and reduce entry or vice versa. Zepeda (Reference Zepeda1995) examined factors influencing net new entry, the difference between entry and exits, inherently recognizing that many of the same factors would influence both. Thus, we review much of the literature on farm exit, with implications for entry.

Previous research has shown mixed effects of government policy on farm exit. D’Antoni, Mishra, and Barkley (Reference D’Antoni, Mishra and Barkley2012) concluded that agricultural subsidies were a factor in pushing labor off the farm. Specifically, increased direct government payments resulted in greater migration of labor from agriculture—subsidies acted as a pure income transfer. There is also evidence to suggest that long-standing conservation programs and commodity buyouts may be attributable to migration from agriculture (Edwards and DeHaven, Reference Edwards and DeHaven2001; USDA, 2011). On the other hand, a strand of literature focused on the effects of government payments on the labor allocation decisions of farm operators and spouses (Ahearn, El-Osta, and Dewbre, Reference Ahearn, El-Osta and Dewbre2006; El-Osta, Mishra, and Ahearn, Reference El-Osta, Mishra and Ahearn2004; Goodwin and Mishra, Reference Goodwin and Mishra2004; Mishra, Fannin, and Joo, Reference Mishra, Fannin and Joo2014; Serra, Goodwin, and Featherstone, Reference Serra, Goodwin and Featherstone2005) has found that government payments, particularly decoupled payments (direct payments), decreased off-farm work by operators, essentially slowing migration out of agriculture. A study by Key and Roberts (Reference Key and Roberts2009) suggested that a reduction in coupled support payments would reduce the marginal value of on-farm work, thereby increasing off-farm labor hours, unlike the increase in the wealth effect due to coupled payments, which would have a decreasing effect. Furthermore, investigating structural change on Connecticut dairy farms, Foltz (Reference Foltz2004) showed that U.S. dairy price policies such as the Milk Income Loss Contract, which establishes a price floor for dairy products, reduced the rate of dairy farm exit in Connecticut during 1997–2001.

Finally, Mishra and El-Osta (Reference Mishra and El-Osta2008) argue that by reducing market risk, government farm programs create a disincentive for farmers to leave the industry. While studying farm succession patterns in U.S. agriculture, they found that farm program payments tended to increase the likelihood of having a succession plan and, in turn, increase the likelihood that the successor was a family member—a case where the farm was in existence. If this is the case, then two issues arise. First, is there a possibility that a person working off the farm at one stage would want to return and choose farming as his/her occupation? Second, if one chooses to become a farmer, what is the goal of the farmer? Is it to take over operation of the farm from a family member, develop a farm business as an additional source of income, or develop a farm business as a primary source of income?

3. Conceptual Framework

Previous research on occupational choice has focused on the neoclassical human capital theory (Becker, Reference Becker1964). Boskin's (Reference Boskin1974) study of U.S. men and women confirms three goals in occupational choice: workers choose occupations that (1) maximize the discounted present value of potential lifetime earnings, (2) entail the lowest training costs, and (3) offer the lowest discounted present value of expected earnings forgone due to unemployment. In the simplest form, human capital theory of occupational choice embodies two postulates. First, the variety of occupational choices is limited by the amount of training required. Training is purchased as an investment in human capital. Second, it is hypothesized that individuals select occupations in a manner that maximizes the present value of their lifetime earnings. Implicitly, selection is equivalent to maximizing the rate of return on one's human capital investment. The concepts of present value and internal rate of return are best suited for analyzing the effects of expected income on occupational choice. Lucas (Reference Lucas1977) pointed out that human capital theory is preoccupied with wage and personal characteristics and tends to overlook nonpecuniary characteristics of jobs. Lucas (Reference Lucas1977) stipulates that utility gained from one's job is a function of wage, nonwage job features, and personal characteristics. A study by Schmidt and Strauss (Reference Schmidt and Strauss1975) develops a theoretical model of occupational choice.

For the purposes of this study, two occupational categories are relevant, off-farm employment and farming. We first use a probit model to estimate the likelihood that an individual i chose farming as an occupation after working off the farm. In other words, we calculate the probability that the current farm operator worked off the farm before taking on farming as his or her main occupation. We then estimate, while benefiting from the theoretical underpinning of occupational decision making of individuals as highlighted by Schmidt and Strauss (Reference Schmidt and Strauss1975), an ordered probit model to examine reasons for entering farming, conditional on the farmer having held an off-farm job prior to entering.

4. Data

Our data are from the ARMS, which is an annual survey conducted by the USDA Economic Research Service and the National Agricultural Statistics Service. The ARMS collects data to measure the financial conditions (farm income, expenses, assets, and debts) and operating characteristics of farm businesses, the cost of producing agricultural commodities, and the well-being of farm operator households. Each observation in the ARMS represents a number of similar farms, the particular number being the survey expansion factor (or the inverse of the probability of the surveyed farm being selected for surveying), which is referred to henceforth as the survey weight, or w_i (i = 1, …, n, where n denotes sample size).

A farm is defined by the USDA as an establishment that sold or normally would have sold at least $1,000 of agricultural products during the year (see http://www.ers.usda.gov/topics/farm-economy/farm-household-well-being/glossary.aspx). We maintain the null hypothesis of homoscedastic errors. Farms are organized as sole proprietorships, partnerships, family corporations, nonfamily corporations, or cooperatives. Data are collected from the senior operator (i.e., the person who makes most of the management decisions) on each farm. We excluded approximately 40,000 farms—namely, those organized as nonfamily corporations or cooperatives and those run by hired managers.Footnote ⁴

The ARMS provides data necessary for estimating operator household income. The Current Population Survey (CPS), conducted by the U.S. Census Bureau, is the source of official U.S. household income statistics (U.S. Department of Commerce, U.S. Census Bureau, 2011). Thus, calculating an estimate of farm household income from ARMS that is consistent with CPS methodology allows comparison between farm operator householdFootnote ⁵ income and all U.S. households. The CPS definition of self-employment income is net revenue from the operation of a business by a person's own account. CPS self-employment income includes income received as cash but excludes in-kind or nonmoney receipts; its definition departs from a strictly cash concept by deducting depreciation, a noncash business expense, from income. Total income for the farm operator household includes that from the farm business, other farming activities, and earned and unearned off-farm sources.

The unique feature of the 2004 ARMS is that it queried the farm operator on occupational choice. Specifically, the survey asked: “Did you (the operator) work at your current or any other off-farm job prior to becoming a farm operator?” The survey then asked the farm operator to rate the importanceFootnote ⁶ of, among others, each of the following reasons for becoming a farm operator: (a) “take over operations of the farm from a family member,” (b) “develop a farm business as an additional source of income,” and (c) “develop a farm business as a primary source of income.” This study also utilizes some auxiliary data. Specifically, local area characteristics in commuting zones are based on county-level data from the Bureau of Economic Analysis income files for 2003, the Bureau of Economic Analysis employment files for 2003, the Bureau of Labor Statistics, and the 2000 Census of Population, STF-3 file (U.S. Department of Commerce, Bureau of the Census, 2003; U.S. Department of Commerce, Bureau of Economic Analysis, 2003; U.S. Department of Labor, Bureau of Labor Statistics, 2003) (for more detail, see USDA, Economic Research Service, 2011b). Summary statistics and description of the variables used in the analysis are presented in Table 1.

Table 1. Summary Statistics for the “Sample-Selection” Equation Depicting Off-Farm Work Participation Decision by the Farm Operator, 2004

^aSource: 2004 Agricultural Resource Management Survey (version 1, phase III). Standard errors of all continuous variables are estimated using bootstrapping with 1,000 resamples. Differences in the means by off-farm work status of continuous variables are examined, with asterisk (*) indicating that the respective means within each row are statistically different (at 5% level confidence interval).

^bSource: Computed by Economic Research Service economists based on information from the U.S. Census Bureau and U.S. Bureau of Labor Statistics (for more detail, see USDA, Economic Research Service, 2011b).

Table 1 presents summary statistics for households having off-farm work while farming or prior to becoming a farmer. The survey covers 6,706 households, which expands to 2 million households when allowing for expansion factors for similar farm households. This allows us to examine previous farm entry decisions of the population of U.S. farmers. Sixty percent of the households had worked off the farm before entering the farming sector. Means comparisons suggest that the operators of farm households with no previous off-farm job were relatively older and had lower educational attainment, perhaps because of their on-farm employment alternatives and the potential for improved production performance from farming experience (see Ahearn, Perry, and El-Osta, Reference Ahearn, Perry and El-Osta1993; Goodwin, Featherstone, and Zeuli, Reference Goodwin, Featherstone and Zeuli2002). The sample covers farming-dependent counties (11%), manufacturing-dependent counties (33%), financial service and government–dependent counties (10%), population loss counties (14%), and other county types (see Figure 1 for delineation of farm location by county type). In 2003, households where the farm operator had previously held an off-farm job were present in significantly lower proportions in farming-dependent and population loss counties, as well as in counties with lower percentages of employment in the finance/insurance and construction sectors.

Figure 1. Delineation of Farm Location by County Type

5. Empirical Estimation

The ordered response model used in this study follows the ordered probit model with self-selection as delineated by De Luca and Perotti (2011):

(1)$$\begin{equation} Y_j^{*} = \beta _j^{\prime} X_j + \varepsilon _j ,\quad j = {\rm 1,}{\rm 2,}\end{equation}$$

(2)$$\begin{equation} Y_{\rm 1} {\kern 1pt} = I{\kern 1pt} {\rm (}U_j^* \ge {\rm 0)} = I(A),\end{equation}$$

(3)$$\begin{equation} Y_2 = \sum\limits_{{h} = {\rm 0}}^H {hI} {\kern 1pt} {\rm (}\alpha _{h} < Y_j^* \le \alpha _{h + 1} {\rm )}\quad{\rm if }Y_1 = {\rm 1,}\end{equation}$$

where Y*₁ and Y*₂ are continuous latent variables for the process representing sample selection and the ordered outcomes. β_j in equation (1) is a vector of k_j parameters to be estimated, X is a vector of exogenous covariates, and ε_j represents random errors with ε₁ and ε₂ following a bivariate normal distribution with zero means, unit variance, and correlation coefficient ρ. I(A) in the observational rule (equation 2) is an indicator function of event A that links the latent variable Y*₁ to the binary variable Y ₁. The observational rule (equation 3) links the latent variable Y*₂ to outcome Y ₂ with α = (α₁, ⋅⋅⋅, α_H; α₀ = −∞, and α_{H + 1} = +∞) denoting a vector of monotonically increasing thresholds, which divide Y*₂, while using h to represent the ordered outcomes of event A, into H + 1 exhaustive and mutually exclusive segments (see Greene, Reference Greene2008, p. 833). In this study, the binary variable Y ₁ represents the off-farm labor participation decision of the operator during and/or prior to entering farming (or event A), and Y ₂ represents the ordered rating (or outcome) of the importance of the reason for farm entry itself conditional on the occurrence of a prior off-farm labor participation decision.

Estimation of the sample-selection ordered probit model (De Luca and Perotti, Reference De Luca and Perotti2011) is based on the maximization of the log-likelihood function ${\rm log }L(\tilde \theta )$, which has an underlying random sample of n observations {(Y _1i, Y _2i,X _1i,X _2i); i = 1, …, n} over the parameter space Φ = ℜ^{k ₁ + k ₂ + Hx( − 1, 1)}:

(4)$$\begin{eqnarray} \mathop {{\rm Max\ log }L{\rm (}\tilde \theta {\rm )}}\limits_{\tilde \theta \in {\kern 1pt} \Phi } = \sum\limits_{{i} = {\rm 1}}^{\rm n} {\left[ {{\rm (1}{\rm - }Y_{{\rm 1}{i}} {\rm )}\log \pi _{0i} {\rm (}\theta {\rm )} + \sum\limits_{{h} = 0}^{H} {Y_{{\rm 1}{i}} I{\rm (}Y_{{\rm 2}{i}} = h{\rm )}{\kern 1pt} \log \pi _{1hi} {\rm (}\theta {\rm )}} } \right]} {\rm ,}\nonumber\\ \end{eqnarray}$$

where θ = (β₁, β₂, α, ρ) is a vector of parameters of the “event” and the “outcome” models, and (π₀, π₁₀, ⋅⋅⋅, π_1H) represents the conditional probabilities reflecting the H + 2 possible realizations of Y ₁ and Y ₂, which are estimated as follows:

(5)$$\begin{eqnarray} \pi _0 {\kern 1pt} {\rm (}\tilde \theta {\rm )} &=& {\rm Pr}{\rm \{ }Y_1 = {\rm 0\} } = {\rm 1} - \Phi {\rm (}\beta _{\rm 1}^{\rm '} X_1 ),\nonumber\\ \pi _{1h} {\kern 1pt} {\rm (}\tilde \theta {\rm )} &=& {\rm Pr}{\rm \{ }Y_1 = 1,Y_2 = h{\rm \} } \nonumber\\ &=& \Phi _2 {\rm (}\beta _{\rm 1}^{\rm '} X_1 ,\alpha _{h{\kern 1pt} + 1} - \beta _{\rm 2}^{\rm '} X_2 ; - \rho ) - \Phi _2 {\rm (}\beta _{\rm 1}^{\rm '} X_1 ,\alpha _{h{\kern 1pt} } - \beta _{\rm 2}^{\rm '} X_2 ; - \rho {\kern 1pt} ),\quad \end{eqnarray}$$

with Φ denoting the standardized normal distribution, Φ₂ denoting the bivariate normal distribution with means and variances of 0 and 1, respectively, and correlation coefficient ρ.

The marginal effects of the explanatory variables of the sample-selection ordered probit model across the h _1,2,3 ordered outcomes as considered in this study are computed as in the following (see Duncan, Khattak, and Council, Reference Duncan, Khattak and Council1998):

(6)$$\begin{eqnarray} \partial \pi _{1h} {\kern 1pt} {\rm (}Y_1 = 1,Y_2 = h)/{\kern 1pt} \partial {\kern 1pt} X_2 &=& - [\Phi _2 {\rm (}\beta _{\rm 1}^{\rm '} X_1 ,\alpha _{h{\kern 1pt} + 1} - \beta _{\rm 2}^{\rm '} X_2 ; - \rho )\nonumber\\ && - \Phi _2 {\rm (}\beta _{\rm 1}^{\rm '} X_1 ,\alpha _{h{\kern 1pt} } - \beta _{\rm 2}^{\rm '} X_2 ; - \rho {\kern 1pt} )]\beta _2 . \end{eqnarray}$$