Cecilia Hyunjung Mo and Katharine M.Conn when we consider attitudes around the education sys- on average 0.121 points (p=0.096)lower than the con- tem.We find that TFA participants are 74 percentage trol group on this measure.5 This difference is eco points (p 0.005)more likely to feel that "systemic nomically meaningful,as it represents 40 percent of injustices that perpetuate inequity throughout society" the mean value of nonadmits.However.this difference contribute to the income-based education achievement is only weakly significant at the optimal bandwidth gap,which represents an 11-percent increase relative As the IAT test was a supplement to the survey,and to the mean value of nonadmits.Participants more fre- thus subject to a smaller sample size,and the optimal quently disagree that"students from low-income back- bandwidth recommended by Imbens and Kalyanara- grounds have the same opportunities as those from man(2011)is a conservative estimate for a fuzzy RDD high-income backgrounds";there is an 11.3 percentage we extend the bandwidth measures to twice the op- point differential(p <0.001),which represents a sub- timal value and find that the impact of TFA partici- stantial 24-percent decrease relative to the mean value pation on the IAT score is a decrease of 0.087 points of nonadmits.Likewise,TFA participants are less likely (p =0.038).To place this result in context,we con- to attribute blame to the poor for class divisions in ed- sider the level of skin-color-based prejudice for white ucational achievement.For example,we find that par- Hispanic,and black Americans,as intergroup bias re- ticipants are 8.5 percentage points (p =0.012)more search suggests that skin color-based prejudice would likely to disagree that poor families"do not value ed- be lower for those of darker skin color(Fu et al.2012; ucation as much as richer families,"which represents a Billig and Tajfel 1973).We find that our treatment ef- 13-percent decrease relative to the mean value of non- fect is roughly equivalent to the 0.109 point difference admits. in skin-tone-based prejudice between white and His- Accompanying decreased blaming of poor commu- panic Americans and a third of the 0.319 point differ- nities,TFA participants are less likely to blame minor- ence in skin-tone-based prejudice between whites and ity groups.We find that TFA participation results in a African Americans (see column(3),rows(13)-(14)of decrease of 12.6 percentage points (p <0.001)in our Table E.9 in Online Appendix E).26 The fact that we racial resentment index,which represents a sizable 58- see an effect on the IAT is notable,as the IAT is a mea- percent decrease relative to the mean value of nonad- sure of automatic and unconscious attitudes,which are mits.To further put this effect in context,the reduc- difficult to shift (Rydell and McConnell 2006).27 tion in racial resentment index is 72 percent of the dif- Finally,not only does TFA participation result in a ference between how black and white Americans an- decrease in certain measures of prejudice,but partici- swer these questions in the 2008 ANES (see column pants are also more likely to report feelings of "partic- (7),row (12)of Table E.9 in Online Appendix E).Un- ular closeness"(in "ideas,interests,and feeling about packing this index,we see that participants are more things")to both African Americans and Hispanics,the likely to attribute racial inequality in this country to two most-served minority populations within the orga- systemic and historical factors than to lack of agency nization.In the 2014-2015 academic year,48 percent of or effort on the part of black Americans.Participants the student population at TFA placement schools were are 12.3 percentage points(p =0.001)more likely to African American,while 35 percent were Hispanic. disagree with the statement that"if blacks would only Specifically,compared to nonparticipants,participants try harder they would be just as well off as whites,"and report feeling 8.7 percentage points(p=0.030)closer S5.501g are 15.8 percentage points (p <0.001)more likely to to African Americans and 2.0 percentage points (p= disagree with the statement that blacks should"(over- 0.731)closer to Hispanics,though the latter effect is not come)prejudice and (work)their way up...without any statistically significant. special favors."From the perspective of societal injus- In interpreting these closeness measures,however. tice,they are also more likely to attribute any diffi- if intergroup contact incites change,we expect there culty in upward social mobility on the part of black to be differential effects on closeness depending on Americans to"generations of slavery and discrimina- the racial demographic group with which TFA partici- tion"(11.8 percentage points,p<0.001)and racial dis- pants comes into contact.We leverage the fact that the crimination in today's society (11.7 percentage points, p<0.001). Further,compared to the control group,TFA par 25 The Black-White IAT was implemented on the 2007 TFA cohort ticipants are 10.6 percentage points (p<0.001)less in 2010,and,consistent with our findings,implicit black-white preju- satisfied with the treatment of minority groups in our dice decreases after participating in TFA(Fryer and Dobbie 2015). 26 This benchmarking estimate is based upon the skin-tone IAT data society as a whole,which represents a 27-percent de- from Harvard University's Project Implicit. crease relative to the mean value of nonadmits.When 27 This finding also helps ameliorate concerns of social desirability considering the assessment of discrimination against bias,though it is highly unlikely that there would be differences in various minority groups separately,the degree of this bias levels at the cutoff,as admission is based upon predicted teacher dissatisfaction differential ranges from approximately effectiveness and not class-and race-based resentment.As an ad- ditional non-self-reported measure,we consider the ethnic fraction- 10 percentage points for Asian Americans,Hispanics. alization of respondents'zip codes.We see evidence of geographic and immigrants (p=0.000-0.007)to 173 percentage sorting within 3 years of participation,where TFA participants are points (p 0.001)for black Americans. living in more diverse communities than non-TFA participants (4 We also find evidence of prejudice reduction through percentage points,p =0.07).This is additional suggestive evidence our measure of implicit bias toward darker skin color- that there is an increase in closeness to diverse communities Source:Teach For America "School and Student Demographics the skin-tone IAT.We find that TFA participants score 2014-2015.” 734Cecilia Hyunjung Mo and Katharine M. Conn when we consider attitudes around the education sys￾tem. We find that TFA participants are 7.4 percentage points (p = 0.005) more likely to feel that “systemic injustices that perpetuate inequity throughout society” contribute to the income-based education achievement gap, which represents an 11-percent increase relative to the mean value of nonadmits. Participants more fre￾quently disagree that “students from low-income back￾grounds have the same opportunities as those from high-income backgrounds”; there is an 11.3 percentage point differential (p < 0.001), which represents a sub￾stantial 24-percent decrease relative to the mean value of nonadmits. Likewise, TFA participants are less likely to attribute blame to the poor for class divisions in ed￾ucational achievement. For example, we find that par￾ticipants are 8.5 percentage points (p = 0.012) more likely to disagree that poor families “do not value ed￾ucation as much as richer families,” which represents a 13-percent decrease relative to the mean value of non￾admits. Accompanying decreased blaming of poor commu￾nities, TFA participants are less likely to blame minor￾ity groups. We find that TFA participation results in a decrease of 12.6 percentage points (p < 0.001) in our racial resentment index, which represents a sizable 58- percent decrease relative to the mean value of nonad￾mits. To further put this effect in context, the reduc￾tion in racial resentment index is 72 percent of the dif￾ference between how black and white Americans an￾swer these questions in the 2008 ANES (see column (7), row (12) of Table E.9 in Online Appendix E). Un￾packing this index, we see that participants are more likely to attribute racial inequality in this country to systemic and historical factors than to lack of agency or effort on the part of black Americans. Participants are 12.3 percentage points (p = 0.001) more likely to disagree with the statement that “if blacks would only try harder they would be just as well off as whites,” and are 15.8 percentage points (p < 0.001) more likely to disagree with the statement that blacks should “(over￾come) prejudice and (work) their way up...without any special favors.” From the perspective of societal injus￾tice, they are also more likely to attribute any diffi￾culty in upward social mobility on the part of black Americans to “generations of slavery and discrimina￾tion” (11.8 percentage points, p < 0.001) and racial dis￾crimination in today’s society (11.7 percentage points, p < 0.001). Further, compared to the control group, TFA par￾ticipants are 10.6 percentage points (p < 0.001) less satisfied with the treatment of minority groups in our society as a whole, which represents a 27-percent de￾crease relative to the mean value of nonadmits. When considering the assessment of discrimination against various minority groups separately, the degree of this dissatisfaction differential ranges from approximately 10 percentage points for Asian Americans, Hispanics, and immigrants (p = 0.000–0.007) to 17.3 percentage points (p < 0.001) for black Americans. We also find evidence of prejudice reduction through our measure of implicit bias toward darker skin color— the skin-tone IAT. We find that TFA participants score on average 0.121 points (p = 0.096) lower than the con￾trol group on this measure.25 This difference is eco￾nomically meaningful, as it represents 40 percent of the mean value of nonadmits. However, this difference is only weakly significant at the optimal bandwidth. As the IAT test was a supplement to the survey, and thus subject to a smaller sample size, and the optimal bandwidth recommended by Imbens and Kalyanara￾man (2011) is a conservative estimate for a fuzzy RDD, we extend the bandwidth measures to twice the op￾timal value and find that the impact of TFA partici￾pation on the IAT score is a decrease of 0.087 points (p = 0.038). To place this result in context, we con￾sider the level of skin-color-based prejudice for white, Hispanic, and black Americans, as intergroup bias re￾search suggests that skin color-based prejudice would be lower for those of darker skin color (Fu et al. 2012; Billig and Tajfel 1973). We find that our treatment ef￾fect is roughly equivalent to the 0.109 point difference in skin-tone-based prejudice between white and His￾panic Americans and a third of the 0.319 point differ￾ence in skin-tone-based prejudice between whites and African Americans (see column (3), rows (13)–(14) of Table E.9 in Online Appendix E).26 The fact that we see an effect on the IAT is notable, as the IAT is a mea￾sure of automatic and unconscious attitudes, which are difficult to shift (Rydell and McConnell 2006).27 Finally, not only does TFA participation result in a decrease in certain measures of prejudice, but partici￾pants are also more likely to report feelings of “partic￾ular closeness” (in “ideas, interests, and feeling about things”) to both African Americans and Hispanics, the two most-served minority populations within the orga￾nization. In the 2014–2015 academic year, 48 percent of the student population at TFA placement schools were African American, while 35 percent were Hispanic.28 Specifically, compared to nonparticipants, participants report feeling 8.7 percentage points (p = 0.030) closer to African Americans and 2.0 percentage points (p = 0.731) closer to Hispanics, though the latter effect is not statistically significant. In interpreting these closeness measures, however, if intergroup contact incites change, we expect there to be differential effects on closeness depending on the racial demographic group with which TFA partici￾pants comes into contact. We leverage the fact that the 25 The Black-White IAT was implemented on the 2007 TFA cohort in 2010, and, consistent with our findings, implicit black-white preju￾dice decreases after participating in TFA (Fryer and Dobbie 2015). 26 This benchmarking estimate is based upon the skin-tone IAT data from Harvard University’s Project Implicit. 27 This finding also helps ameliorate concerns of social desirability bias, though it is highly unlikely that there would be differences in bias levels at the cutoff, as admission is based upon predicted teacher effectiveness and not class- and race-based resentment. As an ad￾ditional non-self-reported measure, we consider the ethnic fraction￾alization of respondents’ zip codes. We see evidence of geographic sorting within 3 years of participation, where TFA participants are living in more diverse communities than non-TFA participants (4 percentage points, p = 0.07). This is additional suggestive evidence that there is an increase in closeness to diverse communities. 28 Source: Teach For America “School and Student Demographics 2014–2015.” 734 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