Optimizing school choice: conjoint analysis of parent preferences
Date
2005
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
The market critique of public education maintains that conventional school assignment plans give schools local monopolies over students, so that schools have little incentive to be efficient. A common premise is that a school choice program, in which parents get to select the public schools their children attend, will improve the overall quality of public education by forcing schools to compete for students. A counter-premise is that a choice program really lets students (or their parents) compete for schools, so that students with competitive advantages (e.g., high income, high athletic or academic achievement, highly involved parents) will be concentrated in a few elite schools while the majority of public schools are doomed to mediocrity. ☐ This thesis presents a simulated school choice experiment in which parents of elementary school students evaluated various hypothetical middle schools described by multiple attributes. The thesis explores two hypotheses. The first is that schools are positional goods, valued as much for their perceived relative quality as for any absolute quality. The positional nature of schools is reflected in a community consensus as to which schools are good or bad. School reputations apparently evolve slowly, and do not necessarily reflect objective measures of school quality such as average student performance on standardized tests. Parental sensitivity to school reputations suggests that a school’s reputation may implicitly signal the quality of the child attending it. This thesis gauges the degree of positionality (importance of reputation per se) reflected in parent preferences among alternative schools. ☐ The second hypothesis is that conjoint methods can be extended to analyze respondent preferences among highly complex goods with a large number of variable attributes. A common criticism of large conjoint designs is that respondents cannot be expected to keep track of more than a few attributes at once. But I argue that consumers deal with information overload all the time when choosing between multi-attribute goods. There is no reason to assume that subjects in a conjoint choice experiment cannot use the same information-filtering heuristics they use in actual market situations involving complex choices. The real problem with large conjoint designs is that different respondents apparently key on different attributes, and some attributes may be desirable for some respondents but undesirable for others. This calls into question the validity of conventional random utility models in which individual respondent preferences are assumed to vary randomly around a core preference structure estimated from aggregated choice data. In fact, it may be more meaningful to analyze the distributions of individual preferences rather than simply characterize group mean preferences. ☐ A survey team interviewed 95 parents. The survey included a conjoint choice exercise in which each respondent evaluated (rated) eight hypothetical schools described by 18 attributes. A follow-up question asked respondents to estimate the weight of each attribute in determining their overall evaluations of the schools. ☐ I test these hypotheses via logit and tobit regressions of respondents’ ratings of schools against the school attributes. The empirical preference models all demonstrated a high degree of parent sensitivity to school reputations per se, independent of average standardized test scores or other school characteristics. Surprisingly, when I incorporated the attribute weights reported by the respondents into the regression models, they did not improve model goodness of fit. I test a method for determining empirical attribute weights for individual respondents by regressing each respondent’s set of school ratings against the school attributes one at a time.