Does rural entrepreneurship pay?

The location specificity may be broader than one’s hometown. For example, Krupka (2009) suggests people who grow up near a beach may learn to surf, and while this investment is not specific to their particular hometown, it is limited to places that are similar to their childhood residence. In an examination of return migration patterns, Schlottmann and Herzog (1982) suggest that people prefer to “return” to any rural area in their home region or perhaps any rural setting at all. Yu et al. (2017) find that return migrants are more likely to start a business than the comparable migrants living in other provinces in China. Returnees are found to be less financially constrained with access to broader social networks. However, some studies find that in-migrant entrepreneurs may maintain strong relationships with key informants in the innovation process, which is outside of local territory and therefore are more likely to start a business (Kalantaridis and Bika 2006)

See McKeever et al. (2015) for a detailed literature review of embeddedness in entrepreneurship.

Alternatively, we define non-metropolitan areas (with RUCC ranging from 4 to 9) as rural areas and metropolitan areas (with RUCC ranging from 0 to 3) as urban areas. It generates qualitatively similar results, though with different magnitudes of the coefficients.

Throughout the study, sample weights are used to correct for differences in probability of response between survey years and between alumni within survey years. Weights are constructed so as to relate the number of respondents in each college-cohort cell to the number in the university. The weighted data are used to obtain consistent estimators of population.

Our survey shows that the medium salary is $67,500 for bachelor degree holders and $87,500 for advanced degree holders. To benchmark the official income statistics, we find that the medium annual salary is $51,454 for bachelor degree holders and $64,207 for advanced degree holders, according to Bureau of Labor Statistics (BLS) of USA (see https://​www.​bls.​gov/​webapps/​legacy/​cpswktab5.​htm). Furthermore, according to the ranking reported by the US News Best Colleges, Iowa State University in our study is ranked #113 out of 311 colleges (see https://​www.​usnews.​com/​best-colleges/​iowa-state-university-1869f). Considering that our surveyed university is ranked above the average colleges in the USA, the income reported in the survey is reasonable and comparable to official income statistics reported in BLS.

For example, for the subsample of rural entrepreneurs (D₁ = 1 and D₂ = 1) we have \( f\left({Y}_3|{D}_1=1,{D}_2=1\right)=\frac{1}{P_4}{\int}_{-{C}_2}^{\infty }{\int}_{-{C}_1}^{\infty}\frac{1}{P_4}h\left({\varepsilon}_1,{\varepsilon}_{2,}\frac{1}{P_4}\right)d{\varepsilon}_1,d{\varepsilon}_{2,} \) where h(⋅,  ⋅ ; ⋅) denotes the trivariate normal density for ε, and the conditioning argument X₃ is suppressed for readability.

Note that we are assuming these decisions are made simultaneously, not sequentially. Over 80% of survey respondents indicated that where they lived was an important or very important in choosing their business location and more than half of the entrepreneurs in the sample started their business within 5 years of choosing their residence.

We recognize that the types of jobs or businesses matter for location choices and entrepreneurship choices. In fact, about 11% of rural workers are in the agricultural industry, 15% in education and 2.2% in information technology industry. In contrast, employment in urban areas is more dispersely distributed with only 3.7% in agricultural industry, 7.5% in education and 6% in information technology. However, the choice of industry occupation, even the specific firm, is simultaneous with location and entrepreneurship decisions, and therefore endogenous. All of the right-hand variables we include in our model pre-date the location and entrepreneurship choice. Therefore, we choose not to include industry fixed effects in our analysis. However, we still conducted a robustness check by controlling for the industry fixed effects in the bivariate probit regression of location and entrepreneurship and the income regressions. We obtained qualitatively similar results, although we lose some of significance of some variables only in the income regressions. In the corresponding whole sample income regressions of Table 4, we get qualitatively similar results when selection is not corrected. Once double selection is corrected, the lambdas and rural location variables are not significant but entrepreneurship and the interaction term between rural location and entrepreneurship variables are still significant (0.55 with t value 2.14 and − 0.22 with t value 2.16 respectively). In corresponding subsample income regressions of Table 7, the lambdas are no longer significant after industry fixed effects are included in bivariate probit models and income regressions. Due to limited space, regression results are not reported here but detailed information on the industry categories surveyed in the questionnaire and corresponding regression results are available from the authors upon request.

The two variables (i.e., the measure of location-specific capital) Rural Origin and the preference toward diversity, Type, are not statistically significant in the income regressions, which indicate both of them are satisfactory exclusion variables.

It is worth pointing out that parents may own a farm. Because land is essentially not movable, children of farm owners are restricted to return to hometowns if they will inherit farm businesses. Therefore, our estimation might be subject to the bias from the possible endogenous selection of returning to family businesses. However, the survey shows that the majority of rural businesses are newly started but not inherited. There is not a significant difference in the rate of starting a new business between rural entrepreneurs whose parents have a farm and those whose parents do not own a farm (92.4 vs 94%). Furthermore, we perform several robustness regressions. We include a dummy variable equal to one if parents have a farm in the bivariate probit model. The coefficient of the parent farm variable is not significant, nor are the other regression results fundamentally changed. We also include both the parent business variable and the parent farm variable, and we get similar results. For parsimonious reasons, we only include the variable of parents having a business in all analyses. Regression results are available upon request.

As aforementioned, we are cautious in controlling industry fixed effect in Table 3 because choices of industries could be endogenous in location and occupational choices models. As a robustness check, we added a series of industry dummy variables, the marginal effects of rural origin on rural entrepreneurship are 2.3% (whole sample), 4.4% (return sample), and − 0.2% (non-return sample) with p values of 0.000, 0.000, and 0.055 respectively. The regression results are qualitatively unchanged.

There are possibly additional two reasons. One is that entrepreneurs may under-report their income relative to workers. Hamilton (2000) argues that the self-employed have an incentive to under-report income due to tax incentives. Our data, however, were collected via a survey rather than from administrative tax records, so it is not obvious individuals have the same incentives to underreport their income. Moreover, unless rural entrepreneurs and urban entrepreneurs underreport income systematically differently, our conclusions will not be substantially affected. We are not aware of factors that may motivate rural and urban entrepreneurs to underreport their income levels differently. The other is that urban firms enjoy a premium from a density that a rural market lacks (Faberman and Freedman 2016).

Exp(0.143) − 1 = 0.15.

This is the same argument Lazear (2009) presents in his skill-weights approach to human capital. In his model, as the market becomes thicker, the probability there is another firm that requires a worker’s particular skills increases. Workers in denser markets are therefore more certain to receive an alternative wage offer from a different firm that dominates their wage at their current firm; they have higher opportunity costs of staying in their current position. As a result, wages of those who stay in the current firm and those who leave equalize through the competitive market for wages and all skills become general skills. In thinner markets, however, the probability of receiving an alternative competitive wage offer is lower, and so skills remain more firm specific.