Abstract
In regression discontinuity (RD) with a running variable S crossing a known cutoff c, an unexpectedly small break magnitude is due to S being a mis-measured version of the genuine running variable G. Has all been lost, and is RD useless when G≠S? This paper proves three things. First, when P(G=S)=0, nonparametric RD identification fails. Second, when P(G=S)>0, although the usual RD effect on the margin E(·|G=c) is not nonparametrically identified, the “effect on the truthful margin” E(·|G=S=c) is. Third, under a no-selection-problem assumption, the effect on the truthful margin becomes the effect on the margin; the no-selection-problem assumption is unnecessary, as long as the effect on the truthful margin is taken as a parameter of interest.
Acknowledgments
This research has been supported by the National Research Foundation of Korea Grant funded by the Korean Government (NRF-2014S1A5B1014360). This research has been also partially funded by a Korea University Grant.
Appendix
Example of (2.2)
Suppose, for an error term ε,
where ε|(G=g, V=v) is symmetric around 0 with Fε|G=g,V=v (e) continuous in (g, v, e). This type of D can occur if 1[c≤G] is a legal eligibility for a program and 1[0≤G+ε] is the individual participation decision (Battistin and Rettore 2008, and Battistin et al. 2009). Then we have
This is continuous in s ∀g, and discontinuous at g=c for any s with Fε|G=c,V=s–c (c)>0; if the support of ε|(G=g, V=v) is (–∞, ∞) ∀(g, v), then J(g, s) is discontinuous at g=c ∀s.
Proof of Theorem 1
Observe
With |fV (v)|≤μ for some μ, it holds that
For SRD with D=1[c≤G], it holds that
This is continuous in s because fV and fS are continuous. For FRD, replace 1[c≤g] with J(g, s) to obtain
which is also continuous in s, again invoking the dominated convergence (J(g, s) is bounded as D is bounded).
Proof of Theorem 2
First, for Theorem 2(i), observe that
since fW|Q=0,G=g (w) is uniformly bounded over (g, w), we can invoke the dominated convergence: the continuity of fW|Q=0,G=g (w) in w a.e. PG and the continuity of P(Q=1|G=g) and fG (g) render the continuity of fS (s). This proves (i).
Second, for Theorem 2(ii), observe that
For SRD D=1[c≤G], setting g=c and reversing the inequality G≤g to c≤G in the last display gives
(A.1) is in fact a special case of FRD, for which
From this, we can obtain E(D|S=c+)–E(D|S=c–), which is, however, hard to interpret. Instead, invoke the continuity of E{J(G, S)|S=s, Q=0} and P(Q=1|S=s) to see the first term of (A.2) drop out of E(D|S=c+)–E(D|S=c–) to lead to (3.2), and then (3.3) under SRD. To see when the continuity of E{J(G, S)|S=s, Q=0} in s holds, rewrite (A.2) as
If fG|W=s,Q=0(g) is continuous in s for a.e. PG and |fG|W=s,Q=0(g)|≤μ(g) for any s with
Third, for Theorem 2(iii), working analogously to what was done for E(D|S=s), we have
This, along with L(g, s) satisfying the same (dis-) continuity condition in (2.2) for J(g, s), proves Theorem 2(iii) because
where the right and left limits for G and S are along the same sequence, as the limits are taken on J(s, s) and L(s, s).
References
Battistin, E., and E. Rettore. 2008. “Ineligibles and Eligible Non-Participants as a Double Comparison Group in Regression-Discontinuity Designs.” Journal of Econometrics 142: 715–730.10.1016/j.jeconom.2007.05.006Search in Google Scholar
Battistin, E., A. Brugiavini, E. Rettore, and G. Weber. 2009. “The Retirement Consumption Puzzle: Evidence from a Regression Discontinuity Approach.” American Economic Review 99: 2209–2226.10.1257/aer.99.5.2209Search in Google Scholar
Calonico, S., M. D. Cattaneo, and R. Titiunik. 2014. “Robust Nonparametric Confidence Intervals for Regression-Discontinuity Designs.” Econometrica 82: 2295–2326.10.3982/ECTA11757Search in Google Scholar
Cappelleri, J. C., W. M. K. Trochim, T. D. Stanley, and C. S. Reichardt. 1991. “Random Measurement Error does not Bias the Treatment Effect Estimate in the Regression-Discontinuity Design: I. The Case of no Interaction.” Evaluation Review 15: 395–419.10.1177/0193841X9101500401Search in Google Scholar
Card, D., D. Lee, Z. Pei, and A. Weber. 2015. “Inference on Causal Effects in a Generalized Regression Kink Design.” Econometrica, accepted for publication.10.17848/wp15-218Search in Google Scholar
Dong, Y. 2015. “Regression Discontinuity Applications with Rounding Errors in the Running Variable.” Journal of Applied Econometrics 30: 422–446.10.1002/jae.2369Search in Google Scholar
Dong, Y., and A. Lewbel. 2015. “Identifying the Effect of Changing the Policy Threshold in Regression Discontinuity Models.” Review of Economics and Statistics, forthcoming.10.1162/REST_a_00510Search in Google Scholar
Feir, D., T. Lemieux, and V. Marmer. 2015. “Weak Identification in Fuzzy Regression Discontinuity Designs.” Journal of Business and Economic Statistics, accepted for publication.10.1080/07350015.2015.1024836Search in Google Scholar
Frandsen, B. R., Markus Frölich, and B. Melly. 2012. “Quantile Treatment Effects in the Regression Discontinuity Design.” Journal of Econometrics 168: 382–395.10.1016/j.jeconom.2012.02.004Search in Google Scholar
Hahn, J., P. Todd, and W. van der Klaauw. 2001. “Identification and Estimation of Treatment Effects with a Regression-Discontinuity Design.” Econometrica 69: 201–209.10.1111/1468-0262.00183Search in Google Scholar
Hullegie, P., and T. J. Klein. 2010. “The Effect of Private Health Insurance On Medical Care Utilization and Self-Assessed Health in Germany.” Health Economics 19: 1048–1062.10.1002/hec.1642Search in Google Scholar
Imbens, G. W., and J. D. Angrist. 1994. “Identification and Estimation of Local Average Treatment Effects.” Econometrica 62: 467–475.10.2307/2951620Search in Google Scholar
Imbens, G. W., and K. Kalyanaraman. 2012. Optimal Bandwidth Choice for the Regression Discontinuity Estimator.” Review of Economic Studies 79: 933–959.10.1093/restud/rdr043Search in Google Scholar
Imbens, G. W., and T. Lemieux. 2008. “Regression Discontinuity Designs: A Guide to Practice.” Journal of Econometrics 142: 615–635.10.1016/j.jeconom.2007.05.001Search in Google Scholar
Kim, Y. S., and M. J. Lee. 2015. “Regression-Kink Approach for Wage Effect on Male Work Hours,” Oxford Bulletin of Economics and Statistics, forthcoming.10.1111/obes.12112Search in Google Scholar
Lee, D. S., and T. Lemieux. 2010. “Regression Discontinuity Designs in Economics.” Journal of Economic Literature 48: 281–355.10.3386/w14723Search in Google Scholar
McCrary, J. 2008. “Manipulation of the Running Variable in the Regression Discontinuity Design: A Density Test.” Journal of Econometrics 142: 698–714.10.3386/t0334Search in Google Scholar
Nielsen, H. S., T. Sorensen, and C. R. Taber. 2010. “Estimating the Effect of Student Aid on College Enrollment: Evidence from a Government Grant Policy Reform.” American Economic Journal: Economic Policy 2 (2): 185–215.10.3386/w14535Search in Google Scholar
Otsu, T., K. L. Xu, and Y. Matsushita. 2013. “Estimation and Inference of Discontinuity in Density.” Journal of Business and Economic Statistics 31: 507–524.10.1080/07350015.2013.818007Search in Google Scholar
Otsu, T., K. L. Xu, and Y. Matsushita. 2015. “Empirical Likelihood for Regression Discontinuity Design.” Journal of Econometrics 186: 94–112.10.1016/j.jeconom.2014.04.023Search in Google Scholar
Schanzenbach, D. W. 2009. “Do School Lunches Contribute to Childhood Obesity?” Journal of Human Resources 44: 684–709.10.1353/jhr.2009.0021Search in Google Scholar
Thistlethwaite, D., and D. Campbell. 1960. “Regression-Discontinuity Analysis: An Alternative to the Ex Post Facto Experiment.” Journal of Educational Psychology 51: 309–317.10.1037/h0044319Search in Google Scholar
Yu, P. 2015. “Understanding Estimators of Treatment Effects in Regression Discontinuity Designs.” Econometric Reviews accepted for publication.10.1080/07474938.2013.833831Search in Google Scholar
©2017 Walter de Gruyter GmbH, Berlin/Boston