Date of Award

2009

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Psychology

Abstract

The question of whether social interventions are effective has become an ever more important concern as programs are asked to justify their funding and theories of action. Although the randomized control trial (RCT) is a rigorous choice of design with regards to establishing causality, there are many reasons a researcher might want to consider an alternative design. One design which has been shown to provide unbiased estimates of the program effect is the regression discontinuity (RD). However, a major problem encountered when using RD is that participants must be assigned to groups strictly based on a cutoff score - a situation which is likely to be problematic in applied settings. To date, the literature has not included quantifications of the bias caused when assignment to groups is fuzzy (i.e., the cutoff is not strictly adhered to), making it difficult for researchers to determine when the use of RD would still be appropriate.

This study used a simulation methodology to assess the extent to which fuzziness in RD created bias in the program effect estimate. The three major parameters investigated were: 1) region in which fuzzy data occurred, 2) percent of fuzzy data, and 3) the extent to which the variable driving misassignment was correlated with the outcome variable. The results showed that if the amount of fuzzy data was very small or the region in which fuzziness occurred was very narrow then the estimate of the program effect was highly unlikely to be biased. Outside of these limited conditions estimates were likely to be biased to differing degrees, with many estimates exceeding an absolute effect size of 0.10. Since applied researchers are often expecting to see small effects, it is quite possible that fairly minor amounts of fuzziness may lead to erroneous conclusions about the efficacy of the program. However, in many cases where there is fuzzy data, estimates from the RD may be fairly robust and may provide better estimates than other research designs.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.

Share

COinS