Fifteen variables are included in the model. They are grouped into three
categories: (i) resource characteristics, (ii) socioeconomic variables, and (iii)
school characteristic variables. Resource characteristics are policy variables.
They include percentage of teachers with MA or higher degree, number of
students per teacher, school expenditures per student, and school enrollment.
Impacts of these variables on student performance have received considerable
interest in the literature since they indicate where public policy initiatives might
improve student performance. For example, class size that may have little affect
on the conditional mean, can be investigated with regression quantiles for
its impact on the conditional lower tail of the achievement distribution. Socioeconomic
and school characteristics variables are included to control for
other factors influencing student performance.
Our analysis uses ACT data of Illinois public high school students in 1996.
They were matched to school characteristics data taken from the Illinois Goal
Assessment Program (IGAP). Most of the IGAP data contains school-level
information for each public school in Illinois. Several variables however are
available only at the district level. In most cases there is only one or two
schools in a district so that the district values are likely a good proxy for individual
schools in the district. An important exception is Chicago where all
62 schools are in the same district. This means that the values of district-level
variables are constant across Chicago schools. The data on school characteristics
is supplemented with 1990 census data on socioeconomic variables
matched to school zip codes. The census data includes information on the
percentage of single parent families with children, and the educational
achievement of families in the local zip code. Variables and descriptive statistics
are listed in Table 1. A dummy variable for Chicago is included in the
regressions because Chicago is a single district; there is no variation in its
district measured variables. Another reason is that it is likely that Chicago is
subject to selection e¤ects. Better students in Chicago are more likely to attend
non public schools than better students outside of Chicago.