Lower-Class Mobilization and Policy Linkage in the U.S. States: A Correction

Abstract

Hill, Leighley, and Hinton-Andersson (1995) report a series of empirical analyses intended to test four hypotheses about the association between lower-class mobilization and public policies in the American states. Those hypotheses are: (1) the higher the turnout of the lower class, the more liberal is state welfare policy; (2) the higher economic and fiscal stress, the lower the association between lower-class turnout and welfare policy; (3) liberal party control of state government is the linkage mechanism that relates lower-class mobilization to policy; and (4) party competition is the linkage mechanism that relates lower-class mobilization to policy. In this note we explain how the incorrect handling of missing data in those original analyses led to an erroneous conclusion about the third hypothesis test, an incorrect sign reported for one control variable in the first two hypothesis tests, and small changes in the significance of a few other control variables. We explain the cause of the errors, report corrected analyses, and close with brief observations about the theoretical and practical implications of the correct empirical results. The error in the original analyses arose because missing data codes for three variables in a cross-sectional data set originally created in SPSSX were interpreted as valid values when the data set was expanded into a pooled, cross-sectional time-series design and analyzed with the MICROCRUNCH software package. Those three variables were for state party competition, mass political liberalism, and state party elite liberalism, and the missing data affect four states. Thus the missing data values biased the coefficients reported for these variables in the original analyses. The optimal procedures for handling such missing data problems are unsettled (Bartels and Brady 1991, 136-7). For the purposes of this note we report analyses that are conservative by some, but not all, criteriaanalysis of only those cases with complete data on the variables under study. This procedure appears especially appropriate in light of the absence of a consensus on an alternative one and the small percentage of total cases eliminated from consideration by the procedure in the present instance.