Teaching, Learning, Policy & Leadership Theses and Dissertations

This study examines the influence of institutional selectivity on alumni giving among public research universities, using a conceptual framework based on the economics of nonprofit organizations. This study introduces a dynamic panel modeling technique, which addresses many limitations that more traditional statistical models have when applied to panel data with lagged or "dynamic" variables. Using panel data from 147 public universities over 11 years, the analysis for this study compares the results from ordinary least squares (OLS), fixed effects, and instrumental variable (2SLS) regression with a dynamic fixed effects panel model using a procedure proposed by Blundell and Bond (1998). This dynamic panel modeling technique allows researchers to simultaneously discern the relationship between variables and take into account the possible endogeneity and omitted variable biases, as well as determine conditional convergence or divergence of the values of key observed variables over time.

The results indicate that ordinary least squares, fixed effects, and instrumental variable regression models yield different coefficients, standard errors, and probability values for hypothesis tests. Results from the most robust technique, a dynamic panel fixed effects model using system generalized method of moments, did not indicate that a statistically significant relationship exists between student selectivity and alumni giving. However, the presence of a law or medical program and institutional wealth were statistically significant. Additionally, there is no evidence of convergence or divergence of alumni giving rates.

The results from this analysis have a number of implications. First, the statistically insignificant relationship between selectivity and alumni giving challenges a major paradigm in the literature regarding the influence of this measure of prestige on alumni giving. Future studies should test the influence of other conceptions of prestige and donative support, using dynamic panel modeling, to see if the results are similar. Second, this analysis shows that statistical models prominent in the literature can yield misleading results when applied to panel data. Researchers, therefore, must take great care in using the most appropriate technique when examining dynamic panel data. Finally, this analysis indicates that more complex modeling techniques are required to study alumni giving over time.