Estimating policy effects using spatial regression discontinuity

| No Comments

I created a poster to present at GIS Day, an event organized by the University of Minnesota Geographic Information Science Student Organization. I hadn't heard of GIS Day until last week, so I had to spend all weekend on the analysis, write-up, and formatting. Thankfully I already had the data and the research topic ready to go. I'm grateful to David Card and Alan Krueger, authors of Myth and Measurement: The New Economics of the Minimum Wage, for sharing data from their study. Here are some excerpts from my poster:

Estimating policy effects using spatial regression discontinuity: The case of New Jersey's minimum wage increase

Background
Estimating causal effects from policy-level interventions is an important aim in the field of program evaluation, but policies are typically implemented in geographically defined jurisdictions, such as school districts or states, and not by randomly assigning participants to a treatment or control group. Consistent with the Education Sciences Reform Act of 2002, the U.S. Department of Education gives preferential treatment to causal research based on "random assignment experiments or designs ... [that] eliminate plausible competing explanations." Geographic information systems (GIS) are not widely used in education research but may help isolate competing explanations when estimating the effect of a policy on educational outcomes.

Purpose
Can GIS help evaluators and policy analysts comply with federal priorities for causal research in education? The purpose of this study is to apply GIS and cross-disciplinary inquiry (i.e., across education, geography, economics, and statistics) to the case of treatment assignment based on geographic borders. Geographic information from a well-known study of minimum wage effects by Card and Krueger (1994) was harnessed with GIS software. Data were then re-analyzed using regression discontinuity (RD). The strengths and limitations of GIS and spatial RD are discussed in the context of statistical results.

Results
The final model explains about 3 percent of total variation in the dependent variable (i.e., post-pre change in FTE employees). Of the models examined, the final is the most parsimonious with significant higher order terms. The confidence interval for the mean difference in Y at the PA-NJ border (CI0.95 = -6.3, 3.64) suggests that raising the minimum wage by 19 percent had an insignificant effect on employment in the food service industry. This finding corresponds with Card and Krueger's conclusion.

             Estimate Std. Error t value Pr(>|t|)
(Intercept) 2.309174 2.595784 0.89 0.374
NJ -1.330037 2.526658 -0.53 0.599
Distance 0.224291 0.081157 2.76 0.006
NJ*Distance -0.340753 0.132150 -2.58 0.010
Distance^2 0.001632 0.000653 2.50 0.013
Note: Heteroscedasticity-consistent standard errors
Residual standard error: 8.82 on 374 degrees of freedom
(25 observations deleted due to missingness)
Multiple R-squared: 0.0411, Adjusted R-squared: 0.0308
F-statistic: 4.84 on 4 and 374 DF, p-value: 0.00082

Conclusions
- GIS can help educational researchers harness geographic information to evaluate programs and policies.
- Spatial RD holds promise as a quasi-experimental evaluation tool because educational and other policies are frequently implemented along geographic boundaries, rather than by randomly assigning students or citizens, which requires stringent modeling of the assignment process to minimize competing explanations.
- More methodological inquiry is needed to judge how well and under what conditions spatial RD yields unbiased estimates.

PA_NJ_Map.png

Fitted_line.png

Click on the image to access the full, readable poster:
Poster_Handout_111908.png

Leave a comment