The results of this study build upon the existing literature. The first notable study is Soffer and Korenich (1961) in which the authors conclude that right-to-work laws do not contribute to the growth of nonagricultural employment and industrial development within a state. However, in reaching this conclusion, the authors employ a rather simple analysis of variance methodology that ignores the influence of widely accepted confounding factors on state employment. Newman (1983) employed multiple regression to control for many of these other factors commonly thought to influence employment and found that right-to-work laws exert a statistically significant and positive impact on changes in employment, a result that is particularly strong in labor-intensive industries, the focus of this study. Newman (1984) extends the earlier work and offers evidence that the impact of right-to-work laws diminish over time and eventually disappear. Despite improvements in the employed methodology, both of Newman’s (1983 and 1984) analyses continue to suffer from omitted variable bias due to the exclusion of unmeasurable geographic characteristics and neither study addresses spatial dependence in any way.[*]
Holmes (1998) made what many have argued to be the first significant attempt to address the omitted variable bias due to unmeasurable geographic characteristics. Holmes relied on county-level analysis at the state borders between states with differing right-to-work policies. Geographic characteristics are likely to be similar across contiguous border counties. Thus, Holmes explains, the issue involving distinguishing the effects of state policy from the unmeasurable geographic characteristics is mitigated.
Holmes’ expectation is that we should observe distinct changes in manufacturing activity at state borders where the policy differs. He uses two measures for manufacturing activity at the county level: 1) manufacturing employment as a percentage of total private nonagricultural employment in 1992, and 2) the growth rate in manufacturing employment from 1947 through 1992.
In his primary model explaining the manufacturing share of total private nonfarm employment, Holmes included a right-to-work binary variable and two distance functions to control for a county’s proximity to a right-to-work policy border and on which side of that border the county fell. Holmes determined that controlling for geography impacts the estimated coefficient on the right-to-work binary variable. The average cumulative increase in the manufacturing share of private nonfarm employment on the right-to-work side of the border is roughly 6.6%. This is about a third larger than the estimated increase when geography is not controlled for.
While Holmes’ attempt to control for unmeasured geographic factors was an improvement upon the status quo at that time in the right-to-work literature, the use of the ordinary least squares estimator is likely not appropriate. The dependent variable — the manufacturing share of nonfarm employment — may be spatially correlated due to agglomeration economies or measurement error. In such a cases, OLS produces coefficient estimates that are biased and inconsistent. Further, it is possible that the OLS residuals are spatially correlated; this is, in part, because the ad hoc distance functions are likely to result in correlation in the error terms across counties. In such cases, OLS estimates are unbiased but inefficient.
While limited in scope due to focusing on only one right-to-work state, namely Idaho, Wilbanks and Reed (2001) and Dinlersoz and Hernandez-Murillo (2002) are noteworthy in that they attempt to address the omitted variables bias and correlated errors in ways that differ from Holmes (1998). Using a treatment and control group comparison approach and controlling for a host of demographic and geographic characteristics such as educational attainment, race, population growth prior to right-to-work adoption, industry composition, and urban-rural status, Wilbanks and Reed (2001) find that manufacturing employment growth in Idaho was greater than that of the control groups. Dinlersoz and Hernandez-Murillo (2002) employ a similar methodology and find that Idaho counties experienced higher annual growth in manufacturing employment than the nearly stagnant growth in the neighboring states. Further, the improvement between the pre- and post-right-to-work law growth rates in manufacturing employment was larger in Idaho than in neighboring states.
None of the aforementioned studies adequately addresses the potential for spatial correlation in the dependent variable or in the residuals. Arguably, the first paper to do so is Kalenkoski and Lacombe (2006). They separately address spatial correlation in the dependent variable and residuals by estimating the spatial autoregressive model, or SAR, and the spatial error model, or SEM. Kalenkoski and Lacombe (2006) also argue that they better control for the demographic characteristics that influence demand for and supply of labor or to explain public attitudes toward state policy. Specifically, they include measures of age distribution, race and gender composition, educational attainment, and degree of urbanization. The authors also examine a larger number of industries rather than just focusing on manufacturing.
The approach employed by Kalenkoski and Lacombe (2006) serves as a primary guide to the empirical approach employed in this study. The primary findings of their study indicate that right-to-work legislation is associated with a 2.12% increase in the manufacturing share of private employment. This finding is nearly 30% lower than their estimate from their specification that does not control for the spatial correlation. Right-to-work is found to reduce the share of private employment in the agriculture, forestry, fishing and hunting, mining and some service industries. However, right-to-work is positively associated with employment share in the information and professional, scientific, management, administration and waste management industries.
[*] More detailed discussions of these early studies — Soffer and Korenich (1961) and Newman (1983 and 1984) — can be found in the literature reviews in Moore and Newman (1985) and Moore (1998).