Cecilia Hyunjung Mo and Katharine M.Conn Another threat to causal interpretation is if appli- Broadly,our results indicate that TFA participants cants and interviewers can manipulate the admission are much more likely to lose faith in political insti- score.This is theoretically impossible because neither tutions than the nonadmit "control"group,indicat- the applicants nor the interviewers are aware of the ing a sense that the political status quo is not fair- cutoff score.For further verification of nonmanipula- a decrease in"denial of discrimination."On our in- tion at the cutoff,we test the null hypothesis of continu- dex of systemic injustice measures,we find that par- ity of the density of the forcing variable-the admission ticipating in TFA decreases an individual's support for score-at the cutoff.Reassuringly,we find that there is the current political system by 10.4 percentage points no discontinuity at the cutoff in the density function of (p=0.005).Specifically,participants are 9.1 percentage the admissions score(p =0.27). points (p=0.032)less likely to respect "the political institutions of the United States"and are 10.2 percent- RESULTS age points(p=0.003)less likely to feel that"citizens basic rights are well protected."These drops are quite First we verify that being above the cutoff is an appro- large.Let us consider our results against Haiti,a coun- priate instrument for admission into and participating try that has consistently had among the lowest levels in TFA.This assumption is indeed robust;at the cutoff, of political system support in the Americas over the there is a 28.7 percentage point(p <0.001)bump in the last decade.We see that the decrease in system support admission rate and a 24.9 percentage point(p<0.001) due to TFA participation,as measured by our index, increase in TFA participation(see Figures A.3(a)and is nearly equivalent to the difference in political sys- A.3(b),respectively,and Table A.1 in Online Appendix tem support between the United States and Haiti(86 A). percent;see column(7),row (3)of Table E.9 in Online Paipaopepecv talimg Appendix E).23 Participation in TFA is also linked to a greater per- We detect (1)an increase in perceptions of systemic ception of class-based injustice,and participants are 4号元 injustice against the disadvantaged per prediction 1;(2) more likely to attribute poverty to underlying sys- a decrease in both class-based and racial resentment- temic issues and other external factors than to a lack increased beliefs that situational or environmental fac- of individual effort.We detect a 9.3 percentage point tors are the root cause of outcomes for those who are (p =0.004)increase in participants'support of pro- disadvantaged rather than the disposition of disadvan- poor policy perspectives (class-based injustice index), taged individuals per prediction 2:and(3)a decrease in which represents a 20-percent increase relative to the prejudice and an increase in identification with disad- mean value of this measure for nonadmits (see Table vantaged minorities per prediction 3.20 B.4 in Online Appendix B for the mean value of each We implement the quasi-experimental estimation outcome measure by admission and participation sta- strategy described above.21 The causal effect estimates tus:(1)nonadmit,(2)nonmatriculants,and (3)matricu- 是 from a fuzzy RDD analyses are reported in column(3) lants).Specifically,TFA participants are more likely to of Table 1 and visualized in Figure 2.22 All our find- argue for greater income redistribution(5.8 percentage ings reported below are based on optimum bandwidth points,p=0.049)and greater government responsi- calculations according to Imbens and Kalyanaraman bility to ensure everyone is provided for(75 percent- (2011)unless stated otherwise:however,the signifi- age points,p=0.011).To understand the magnitude cance of the RDD results are generally not sensitive to of these effects,we benchmark our effect sizes against 吧 alternative bandwidths (see Table E.7 in Online Ap- the German population,as Americans tend to priori- pendix E). tize individualism over the role of the state,whereas Germans tend to prioritize state interference over in- dividualism (Pew Research Center 2011).These two effects are equivalent to 30 percent and 37 percent, said,there is no meaningful differences in household income (p 0.654). respectively,of the difference between how the aver- With 26 outcomes,the Bonferroni correction a and the Sidak cor- age American answers these questions compared to rection a are both 0.002.Ten comparisons reported in Table 1 meet the average German (see column(7),rows (4)-(5)of the 0.002 threshold.However,it is unnecessary to employ the ad- Table E.9 in Online Appendix E).24 Further,TFA par- justed a because we find that all 26 measures are statistically mean- ticipation is linked to an increase in the belief that ingful at standard levels,and the probability of seeing this by chance is essentially zero. having a "better life"is more closely linked to "luck 30 When the average response of participants is compared to those and connections"than to hard work alone (9.3 per who declined their admission,as well as to nonadmits,we see that the centage points,p=0.026)and that "poor people are direction of differences are largely consistent with each of our three poor due to an unfair society"as opposed to "lazi- predictions(see Table B.4 in Online Appendix B).More specifically matriculants,on average,display higher perceptions of class-based ness and lack of willpower"(72 percentage points, injustice and lower racial resentment and prejudice levels than both p=0.001). This general dissatisfaction with the broader politi- An inspection of response averages by score near the cutoff for cal system and external blame attribution is detectable each outcome of interest are provided in the Figures D.6-D.7 in On- line Appendix D,and provide visual evidence that there are shifts at First-stage and reduced-form results are reported in column (1) 23 Source:2010 AmericasBarometer. and column(2),respectively,in Table 1. 24 Source:World Values Survey (Wave 6). 730Cecilia Hyunjung Mo and Katharine M. Conn Another threat to causal interpretation is if appli￾cants and interviewers can manipulate the admission score. This is theoretically impossible because neither the applicants nor the interviewers are aware of the cutoff score. For further verification of nonmanipula￾tion at the cutoff, we test the null hypothesis of continu￾ity of the density of the forcing variable—the admission score—at the cutoff. Reassuringly, we find that there is no discontinuity at the cutoff in the density function of the admissions score (p = 0.27). RESULTS First we verify that being above the cutoff is an appro￾priate instrument for admission into and participating in TFA. This assumption is indeed robust; at the cutoff, there is a 28.7 percentage point (p < 0.001) bump in the admission rate and a 24.9 percentage point (p < 0.001) increase in TFA participation (see Figures A.3(a) and A.3(b), respectively, and Table A.1 in Online Appendix A). Overall, we find strong evidence that, ceteris paribus, participation in TFA increases perspective-taking.19 We detect (1) an increase in perceptions of systemic injustice against the disadvantaged per prediction 1; (2) a decrease in both class-based and racial resentment— increased beliefs that situational or environmental fac￾tors are the root cause of outcomes for those who are disadvantaged rather than the disposition of disadvan￾taged individuals per prediction 2; and (3) a decrease in prejudice and an increase in identification with disad￾vantaged minorities per prediction 3. 20 We implement the quasi-experimental estimation strategy described above.21 The causal effect estimates from a fuzzy RDD analyses are reported in column (3) of Table 1 and visualized in Figure 2. 22 All our find￾ings reported below are based on optimum bandwidth calculations according to Imbens and Kalyanaraman (2011) unless stated otherwise; however, the signifi￾cance of the RDD results are generally not sensitive to alternative bandwidths (see Table E.7 in Online Ap￾pendix E). said, there is no meaningful differences in household income (p = 0.654). 19 With 26 outcomes, the Bonferroni correction α and the Sidak cor￾rection α are both 0.002. Ten comparisons reported in Table 1 meet the 0.002 threshold. However, it is unnecessary to employ the ad￾justed α because we find that all 26 measures are statistically mean￾ingful at standard levels, and the probability of seeing this by chance is essentially zero. 20 When the average response of participants is compared to those who declined their admission, as well as to nonadmits, we see that the direction of differences are largely consistent with each of our three predictions (see Table B.4 in Online Appendix B). More specifically, matriculants, on average, display higher perceptions of class-based injustice and lower racial resentment and prejudice levels than both nonadmits and nonmatriculants. 21 An inspection of response averages by score near the cutoff for each outcome of interest are provided in the Figures D.6– D.7 in On￾line Appendix D, and provide visual evidence that there are shifts at the cutoff. 22 First-stage and reduced-form results are reported in column (1) and column (2), respectively, in Table 1. Broadly, our results indicate that TFA participants are much more likely to lose faith in political insti￾tutions than the nonadmit “control” group, indicat￾ing a sense that the political status quo is not fair— a decrease in “denial of discrimination.” On our in￾dex of systemic injustice measures, we find that par￾ticipating in TFA decreases an individual’s support for the current political system by 10.4 percentage points (p = 0.005). Specifically, participants are 9.1 percentage points (p = 0.032) less likely to respect “the political institutions of the United States” and are 10.2 percent￾age points (p = 0.003) less likely to feel that “citizens’ basic rights are well protected.” These drops are quite large. Let us consider our results against Haiti, a coun￾try that has consistently had among the lowest levels of political system support in the Americas over the last decade.We see that the decrease in system support due to TFA participation, as measured by our index, is nearly equivalent to the difference in political sys￾tem support between the United States and Haiti (86 percent; see column (7), row (3) of Table E.9 in Online Appendix E).23 Participation in TFA is also linked to a greater per￾ception of class-based injustice, and participants are more likely to attribute poverty to underlying sys￾temic issues and other external factors than to a lack of individual effort. We detect a 9.3 percentage point (p = 0.004) increase in participants’ support of pro￾poor policy perspectives (class-based injustice index), which represents a 20-percent increase relative to the mean value of this measure for nonadmits (see Table B.4 in Online Appendix B for the mean value of each outcome measure by admission and participation sta￾tus: (1) nonadmit, (2) nonmatriculants, and (3) matricu￾lants). Specifically, TFA participants are more likely to argue for greater income redistribution (5.8 percentage points, p = 0.049) and greater government responsi￾bility to ensure everyone is provided for (7.5 percent￾age points, p = 0.011). To understand the magnitude of these effects, we benchmark our effect sizes against the German population, as Americans tend to priori￾tize individualism over the role of the state, whereas Germans tend to prioritize state interference over in￾dividualism (Pew Research Center 2011). These two effects are equivalent to 30 percent and 37 percent, respectively, of the difference between how the aver￾age American answers these questions compared to the average German (see column (7), rows (4)–(5) of Table E.9 in Online Appendix E).24 Further, TFA par￾ticipation is linked to an increase in the belief that having a “better life” is more closely linked to “luck and connections” than to hard work alone (9.3 per￾centage points, p = 0.026) and that “poor people are poor due to an unfair society” as opposed to “lazi￾ness and lack of willpower” (7.2 percentage points, p = 0.001). This general dissatisfaction with the broader politi￾cal system and external blame attribution is detectable 23 Source: 2010 AmericasBarometer. 24 Source: World Values Survey (Wave 6). 730 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:05, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0003055418000412