The Ideology of Education: The Commonwealth, the Market, and America's Schools

Methodological Appendix

All data used in the analyses for this book are publically available, or can be obtained from the author. I used three basic statistics packages to perform the analyses: Stata, SPSS, and NCSS. All OLS models were estimated using NCSS and/or SPSS; all pooled and GLS models were estimated using Stata. Technical details and data sources are presented by table, which hopefully should make it easier for the statistically inclined to judge the validity of the results without an undue departure from the main text. TABLE 2.1 Table 2.1 reports the results of a pooled time series analysis predicting state-level productivity measured as GSP per worker. Although a powerful technique that combines the advantages of both cross-sectional and time series analysis, pooled models also import and compound the problems of both approaches (Stimson 1985, Sayrs 1989). Such models often generate complicated error structures that violate the basic assumptions of ordinary least squares (OLS). A number of substitute estimators have been suggested to deal with the effects of these violations. In this particular case, a series of diagnostics indicated a problem with unit effects— in essence there were state level impacts not being accounted for in the model. One of the most common approaches to dealing with these problems are to generate estimators using generalized least squares (GLS), and the GLS estimators are reported in the table. An alternate estimation technique is a fixed effects (FE), or least squares with dummy variables (LSDV) approach, which addresses the unit effects problem without the assumption of randomness required of GLS. There were two drawbacks to employing FE in this case: (1) High degrees of multicollinearity and the unit dummies (measures of “specific ignorance”) accounting for a good 157 Methodological Appendix deal of variance. (2) FE cannot estimate time invariant variables, which are zeroed out of the model. This drops four variables (percent with high school diploma, equity in educational spending, unionization and political competition) from the model. These were theoretically important to retain. Accordingly, I opted to report the GLS coefficients, though the coefficients from both OLS and FE estimation techniques are included in the range of values indicated by the Leamer bounds in Table 2.1. The data were stacked traditionally (see Sayrs 1989), and the time period and states used in the analysis were largely dictated by data availability : Sharkansky’s (1969) culture scale does not include Alaska and Hawaii; no score is available for Louisiana in the political competition measure; Washington does not report aggregated test score results; and South Dakota was dropped because of missing data in some years. Statelevel ACT scores—necessary to calculate a test score index that avoids the selection bias problem—were not available for all years after 1990. This left data on 45 states for nine years. Sources of Data Productivity: GSP per worker in constant (1992) dollars. U.S. Department of Commerce, Bureau of Economic Analysis, Regional Economic Information System (REIS) CD-ROM. Bureau of the Census. Various Years. Statistical Abstracts of the United States. Washington, D.C., U.S. Government Printing Office. Policy Activism: Total number of tax incentives and other support packages listed in Tables 1, 2, and 3 of the Site Selection Handbook. Industrial Developments Site Selection Handbook. Annual political issue. Atlanta, GA: Conway Publications. National Productivity: GSP per worker in constant (1996) dollars. REIS CD-ROM, and Statistical Abstracts of the United States, various years. Unionization: A dummy variable where 1 = 15 percent or more of workers are labor union members, and 0 = less than 15 percent are members. Calculated from data reported in the Statistical Abstracts, various years. Political Competition: Index of state-level political competition calculated by Holbrook and Van Dunk (1993). Unified Control: A dummy variable where 1 = unified partisan control of legislature and governer’s office, 0 = divided partisan control. Calculated from data reported in the Statistical Abstracts, various years. Size of the Public Sector: Total state tax collections measured as a percentage of GSP. REIS CD-ROM. Bureau of the Census, Various Years. 158 Methodological Appendix Federal Expenditures by State. Washington, D.C.: U.S. Government Printing Office. Defense Expenditures: Total Defense Department expenditures within a state in constant (1990 Texas) dollars. Office of Management and Budget . Annual. Federal Expenditures by State. Washington, D.C.: U.S. Government Printing Office. Percent of GSP Devoted to Education: Total state and local government expenditure on education as a percent...