Content area
Full Text
School-effects research in sociology cannot be separated from concerns about causality. Purely descriptive modeling justifications are untenable. Focusing on the Catholic school effect on learning, this article demonstrates an approach that places regression modeling strategies within a specific and well-developed framework for thinking about causality. While regression models should properly remain the workhorse methodology for school-effects research, regression estimates should more often be subject to exacting interpretations and presented alongside alternative estimates of more specific parameters of interest. In this demonstration, propensity-score matching estimates of the Catholic school effect for the Catholic schooled are provided to supplement the estimates obtained by regression models. Although subject to their own set of weaknesses, the matching estimates suggest that the Catholic school effect is the strongest among those Catholic school students who, according to their observed characteristics, are least likely to attend Catholic schools. Four alternative explanations are offered for this finding, each of which should be pursued in further research.
In a series of widely read research reports, Coleman and his colleagues (Coleman and Hoffer 1987; Coleman, Hoffer, and Kilgore 1982; Hoffer, Greeley, and Coleman 1985) presented evidence that Catholic schools confer learning advantages on their students. Although vigorously contested (see Alexander and Pallas 1983, 1985; Goldberger and Cain 1982; W. R. Morgan 1983; Noell 1982; Willms 1985), their provocative findings inspired three strands of subsequent survey research in the sociology and economics of education: the evaluation of market competition models of school improvement (e.g., Chubb and Moe 1990; Figlio and Stone 1997; Hoxby 1996; Neal 1997), effective schools research (e.g., Lee and Smith 1993, 1995; Lee, Smith, and Croninger 1997), and social capital research (e.g., Carbonaro 1998; S. L. Morgan and Sorensen 1999).
In most of this school-effects research (and, to be fair, in my research as well), the limitations of observational survey data are acknowledged but rarely discussed in any depth. Thus, although the possible existence of omitted variable bias is recognized, the care with which the weaknesses of survey data are discussed declines dramatically when the specter of hidden self-- selection bias arises. An important negative consequence of suppressing forthright discussion of the specific weaknesses of available data and the limited range of conclusions that they can effectively sustain is that the need...