Cecilia Hyunjung Mo and Katharine M.Conn Second,TFA places their participants in the lowest decision,matriculation decision,placement informa- income schools in America.Over 80 percent of the stu- tion,and demographic characteristics),and we utilize dents taught by TFA corps members qualify for free this information for all applicants who made it to the or reduced-price lunch(FRPL)and are either African final round of interviews in the application process for American or Hispanic.The socioeconomic make-up the 2007-2015 application cycles.While over 380,000 of the student population is intentional,as TFA seeks applied to TFA during this period,we restrict our fo- "partnership with communities where there is a dispar- cus to the third of applicants who were finalists for ad- ity in educational opportunity along lines of race and mission,and hence,at least close to being admitted. class,"and all partner schools have "at least 60 per- This amounts to a sample size of 120,417 Our primary cent of students eligible for FRPL,a common proxy focus is on alumni starting from the 2007 cohort be- for need."10 cause a selection process that involved the creation of Third,extended contextualized intergroup contact an admission cutoff score was instituted in 2007 Since between advantaged and disadvantaged populations at the time of data collection,the 2014 and 2015 cohorts occurs.As full-time teachers charged to help address were still participating in TFA,they have not fully been education inequality for two years,TFA corps mem- "treated,"and are excluded from the main analyses.For bers are actively in contact with low-income students the 2007 to 2013 cohorts,we have data on 91,752 appli- and their families for an extended period.Participants cants(see Online Appendix A for a detailed descrip- have the opportunity to view their students'well-being tion of the final sample size). and level of achievement in light of their familial, school,community,and societal context,which gives them a more nuanced view of the realities under Survey Data which systemic inequalities might form.Moreover. their interactions with disadvantaged communities Data Collection.On October 1.2015.we emailed ap plicants invitations to participate in an online survey. are contextualized within a social and institutional The survey stayed active for six months,closing on service context to advance the economic success of March 31,2016.12 Of the 91,752 TFA applicants from low-income students. the 2007-2013 cohorts that were targeted,272 at least Finally.TFA is nearly ideal from the standpoint of started the survey and 21.1 percent completed the sur- teasing out causality.In 2007 TFA instituted a selection vey.13 Among the 31,376 TFA alumni(2007-2013 corps process with a cutoff threshold that enables us to im- members),33.8 percent at least started the survey and plement a quasi-experimental regression discontinuity 271 percent completed the survey.Of the 60,376 ap- analysis.Our identification strategy exploits the fact plicants who did not participate in TFA,23.8 percent that TFA admission is a discontinuous function of an at least started the survey and 179 percent finished the applicant's selection score,which represents TFA's as- survey.The survey completion response rate(AAPOR sessment of how effective the applicant will be in the RR1 response rate)and partial response rate(AAPOR classroom.The ability to leverage a selection process RR2 response rate)information by application cycle that enables causal inference.coupled with the visibil- are shown in Figure A.1a and Figure A.1b in Online ity and attractiveness of TFA as a national service pro- Appendix A,respectively.There are no notable differ- gram for advantaged individuals to come into extended ences in response rates by application year. S5.501g contextualized intergroup contact with disadvantaged The average participant in our survey is 29 years individuals,makes TFA a strong case to consider in this old,has a college GPA of 3.52,and went to a selec- study tive undergraduate school (see Table B.2 in Online Ap- pendix B).A minority received a Pell Grant in college DATA AND MEASUREMENT (31.0 percent).Approximately 70 percent of the study sample are female (72.5 percent)and white(69.8 per- TFA selection data and an original national survey of cent),and 94.1 percent of study participants have par- TFA applicants are employed to test our predictions. ents with a post-secondary education.Over half of the Exact question wordings and coding schemes of each participants identify with a religion(58.1 percent).and of our measures are provided in Online Appendix F. nearly half of the study participants are upper class Unless noted otherwise,questions were recoded to be or upper middle class Americans (49.0 percent).Al- between 0 and 1 so that treatment effects can be inter- laying concerns of survey response bias,we find that preted in percentage point terms. our participant population is generally representative of the overall TFA applicant population that made it to the final stage of the application process on each of Selection Data TFA maintains detailed selection data (e.g.,contact in- 11 We used email addresses that applicants provided during the ap. formation,application year,selection score,admissions plication process.We did not update email addresses for alumn based on TFA records to ensure that we did not have fewer invalid emails for alumni. 12 Participants received uptoeight email reminders regarding survey 9 Source:Teach For America's "School and Student Demographics participation,and incentives were offered to increase response rates 2014-2015." see Online Appendix G for additional details) 10 Source:www.teachforamerica.org/tfa-on-the-record/responses/ Among applicants for the 2014 and 2015 cohorts,26.8 percent at april-22-2014-nation (accessed March 18,2016). least started the survey,and 19.4 percent completed the survey. 726
Cecilia Hyunjung Mo and Katharine M. Conn Second, TFA places their participants in the lowest income schools in America. Over 80 percent of the students taught by TFA corps members qualify for free or reduced-price lunch (FRPL) and are either African American or Hispanic.9 The socioeconomic make-up of the student population is intentional, as TFA seeks “partnership with communities where there is a disparity in educational opportunity along lines of race and class,” and all partner schools have “at least 60 percent of students eligible for FRPL, a common proxy for need.”10 Third, extended contextualized intergroup contact between advantaged and disadvantaged populations occurs. As full-time teachers charged to help address education inequality for two years, TFA corps members are actively in contact with low-income students and their families for an extended period. Participants have the opportunity to view their students’ well-being and level of achievement in light of their familial, school, community, and societal context, which gives them a more nuanced view of the realities under which systemic inequalities might form. Moreover, their interactions with disadvantaged communities are contextualized within a social and institutional service context to advance the economic success of low-income students. Finally, TFA is nearly ideal from the standpoint of teasing out causality. In 2007, TFA instituted a selection process with a cutoff threshold that enables us to implement a quasi-experimental regression discontinuity analysis. Our identification strategy exploits the fact that TFA admission is a discontinuous function of an applicant’s selection score, which represents TFA’s assessment of how effective the applicant will be in the classroom. The ability to leverage a selection process that enables causal inference, coupled with the visibility and attractiveness of TFA as a national service program for advantaged individuals to come into extended contextualized intergroup contact with disadvantaged individuals,makes TFA a strong case to consider in this study. DATA AND MEASUREMENT TFA selection data and an original national survey of TFA applicants are employed to test our predictions. Exact question wordings and coding schemes of each of our measures are provided in Online Appendix F. Unless noted otherwise, questions were recoded to be between 0 and 1 so that treatment effects can be interpreted in percentage point terms. Selection Data TFA maintains detailed selection data (e.g., contact information, application year, selection score, admissions 9 Source: Teach For America’s “School and Student Demographics 2014–2015.” 10 Source: www.teachforamerica.org/tfa-on-the-record/responses/ april-22-2014-nation (accessed March 18, 2016). decision, matriculation decision, placement information, and demographic characteristics), and we utilize this information for all applicants who made it to the final round of interviews in the application process for the 2007–2015 application cycles. While over 380,000 applied to TFA during this period, we restrict our focus to the third of applicants who were finalists for admission, and hence, at least close to being admitted. This amounts to a sample size of 120,417. Our primary focus is on alumni starting from the 2007 cohort because a selection process that involved the creation of an admission cutoff score was instituted in 2007. Since at the time of data collection, the 2014 and 2015 cohorts were still participating in TFA, they have not fully been “treated,” and are excluded from the main analyses.For the 2007 to 2013 cohorts, we have data on 91,752 applicants (see Online Appendix A for a detailed description of the final sample size). Survey Data Data Collection. On October 1, 2015, we emailed applicants invitations to participate in an online survey.11 The survey stayed active for six months, closing on March 31, 2016.12 Of the 91,752 TFA applicants from the 2007–2013 cohorts that were targeted, 27.2 at least started the survey and 21.1 percent completed the survey.13 Among the 31,376 TFA alumni (2007–2013 corps members), 33.8 percent at least started the survey and 27.1 percent completed the survey. Of the 60,376 applicants who did not participate in TFA, 23.8 percent at least started the survey and 17.9 percent finished the survey. The survey completion response rate (AAPOR RR1 response rate) and partial response rate (AAPOR RR2 response rate) information by application cycle are shown in Figure A.1a and Figure A.1b in Online Appendix A, respectively. There are no notable differences in response rates by application year. The average participant in our survey is 29 years old, has a college GPA of 3.52, and went to a selective undergraduate school (see Table B.2 in Online Appendix B). A minority received a Pell Grant in college (31.0 percent). Approximately 70 percent of the study sample are female (72.5 percent) and white (69.8 percent), and 94.1 percent of study participants have parents with a post-secondary education. Over half of the participants identify with a religion (58.1 percent), and nearly half of the study participants are upper class or upper middle class Americans (49.0 percent). Allaying concerns of survey response bias, we find that our participant population is generally representative of the overall TFA applicant population that made it to the final stage of the application process on each of 11 We used email addresses that applicants provided during the application process. We did not update email addresses for alumni based on TFA records to ensure that we did not have fewer invalid emails for alumni. 12 Participants received up to eight email reminders regarding survey participation, and incentives were offered to increase response rates (see Online Appendix G for additional details). 13 Among applicants for the 2014 and 2015 cohorts, 26.8 percent at least started the survey, and 19.4 percent completed the survey. 726 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
When Do the Advantaged See the Disadvantages of Others? the demographic characteristic in the application file, effort or the system,we assessed three questions.We apart from race,alongside selection score and appli- measured beliefs on whether poor families do not value cation year.Our participant sample skews somewhat education as much as richer families,and whether sys- more white:however,the skew is similar for both our temic injustices perpetuating inequity throughout so- admitted and nonadmitted survey sample (see Online ciety"are contributors to the inequality in educational Appendix B for additional details). achievement in the US"(response options:0 not a contributor/does not occur->1=main contributor). Outcome Measurement.There are four batteries that were asked to capture whether there is enhanced Additionally,we assess the extent to which a respon- dent believes that "students from low income back perspective-taking for disadvantaged populations with grounds have the same educational opportunities as respect to class and race:(1)systemic injustice,(2) students from high income backgrounds"(response class-based injustice,(3)the relationship between class and education inequality,and(4)racial injustice.These options:0=strongly disagree1 strongly agree). Racial Injustice:The racial injustice battery included questions map onto our three predictions:(1)reduction four questions from the standard racial resentment or in“denial of discrimination,”(2)reduction in“actor symbolic racism measures forwarded by Kinder and observer bias,"and (3)reduction in prejudice levels and increased identification with disadvantaged popu- Sanders (1996)and Henry and Sears (2002).15 Addi- tionally,we asked"How much racial discrimination do lations.Table B.3 in Online Appendix B provides sum- mary statistics of each of our outcome measures.14 you feel there is in the US today,limiting the chances of Systemic Injustice:We measured attitudes around individuals from particular racial groups to get ahead?" systemic injustice with two measures from a political (response options:0 none at all 1 a great deal).We also considered an index of this discrimina support index(Booth and Seligson 2009)that assess tion measure and the four racial resentment measures, the level of respect an individual has for U.S.political institutions (response options:0=not at all1 =a which we refer to as the racial resentment index,given the Cronbach's alpha score is 0.86. lot)and the extent to which citizens'basic rights are We also asked a series of questions about the respon- protected by the U.S.political system (response op- dent's level of satisfaction with the treatment of each of tions:0=not at all-1 =a lot).We also consider an the of the following minority groups(response options: & index of these two measures(system support index);the Cronbach's alpha score is 0.71,which is acceptably high. 0=very dissatisfied-1 very satisfied):Asians,His- Class-Based Injustice:We considered four questions panics,blacks,Muslims and immigrants.We consider each measure separately,and as a simple index,which from the World Values Survey that center on blaming we refer to as the discrimination index given high in- those who are poor for being poor as opposed to an ternal consistency of these measures:the Cronbach's external entity (e.g.,government)or force (e.g.,misfor- alpha score is 0.85. tune or lack of fairness),which have been found to be Racial Prejudice:We employed two measures to cap- strongly predictive of support for government welfare policies(Alesina,Glaeser,and Sacerdote 2001).For in- ture prejudice.First,we implemented a skin-tone Im- plicit Association Test(IAT),a method for gauging un- stance,if people perceive the poor as lazy,then indi- viduals are less likely to support redistributive policies. conscious antipathy toward various groups.The IAT has commonly been used in psychology(Greenwald, Namely,we provided the respondent with four pairs of McGhee.and Schwartz 1998:Greenwald.Nosek.and statements and assess which statement in each pair in dividuals agree with more:(1)"We need larger income Banaji 2003),and increasingly in political science to predict political behavior (Arcuri et al.2008:Mo 2015) differences as incentives for individual effort"(coded and policy judgments (Malhotra,Margalit,and Mo as 0)versus "Incomes should be made more equal" 2013;Perez 2010).The IAT is a method designed to (coded as 1);(2)"People should take more responsi- bility to provide for themselves"(coded as 0)versus capture the strength of associations linking social cate- gories (dark skin color versus light skin color)to eval- 'Government should take more responsibility to en- sure that everyone is provided for"(coded as 1);(3) uative anchors (good versus bad). "In the long run,hard work usually brings a better life' The difference in categorization performance is ar- (coded as 0)versus "Hard work doesn't generally bring gued to capture"implicit"(system 1)attitudes that are success-it's more a matter of luck and connections automatic,as opposed to "explicit"(system 2)attitudes that are effortful and conscious(Kahneman 2003).The (coded as 1):and (4)"People are poor because of lazi- IAT effect is a D score,which ranges from-2 to 2, ness and lack of willpower"(coded as 0)versus"People where negative (positive)numbers indicate an implicit are poor because of an unfair society"(coded as 1).We bias favoring darker(lighter)skin tones over lighter also consider an index of these four measures,which (darker)skin-tones and 0 indicates neutrality (see On- we call the class-based injustice index:the Cronbach's alpha score is 0.77 line Appendix C for additional details on the IAT). Class-Based Education Inequality:To capture beliefs on whether education inequality is due to individual i5 We included a question on the extent to which respondents agree that black Americans have gotten less than they deserve;agree that black Americans should overcome prejudice without special favors; agree that it is really just a matter of black Americans working harder 14 Direct questions of whether a respondent feels increased to be just as well off as whites;and agree that slavery and discrimina- perspective-taking were not asked given social desirability bias con- tion has made it difficult for black Americans to work their way up siderations. (response options:0=strongly disagree1 strongly agree). 727
When Do the Advantaged See the Disadvantages of Others? the demographic characteristic in the application file, apart from race, alongside selection score and application year. Our participant sample skews somewhat more white; however, the skew is similar for both our admitted and nonadmitted survey sample (see Online Appendix B for additional details). Outcome Measurement. There are four batteries that were asked to capture whether there is enhanced perspective-taking for disadvantaged populations with respect to class and race: (1) systemic injustice, (2) class-based injustice, (3) the relationship between class and education inequality, and (4) racial injustice. These questions map onto our three predictions: (1) reduction in “denial of discrimination,” (2) reduction in “actorobserver bias,” and (3) reduction in prejudice levels and increased identification with disadvantaged populations. Table B.3 in Online Appendix B provides summary statistics of each of our outcome measures.14 Systemic Injustice: We measured attitudes around systemic injustice with two measures from a political support index (Booth and Seligson 2009) that assess the level of respect an individual has for U.S. political institutions (response options: 0 = not at all → 1 = a lot) and the extent to which citizens’ basic rights are protected by the U.S. political system (response options: 0 = not at all → 1 = a lot). We also consider an index of these two measures (system support index); the Cronbach’s alpha score is 0.71,which is acceptably high. Class-Based Injustice: We considered four questions from the World Values Survey that center on blaming those who are poor for being poor as opposed to an external entity (e.g., government) or force (e.g., misfortune or lack of fairness), which have been found to be strongly predictive of support for government welfare policies (Alesina,Glaeser, and Sacerdote 2001). For instance, if people perceive the poor as lazy, then individuals are less likely to support redistributive policies. Namely, we provided the respondent with four pairs of statements and assess which statement in each pair individuals agree with more: (1) “We need larger income differences as incentives for individual effort” (coded as 0) versus “Incomes should be made more equal” (coded as 1); (2) “People should take more responsibility to provide for themselves” (coded as 0) versus “Government should take more responsibility to ensure that everyone is provided for” (coded as 1); (3) “In the long run, hard work usually brings a better life” (coded as 0) versus “Hard work doesn’t generally bring success-it’s more a matter of luck and connections” (coded as 1); and (4) “People are poor because of laziness and lack of willpower” (coded as 0) versus “People are poor because of an unfair society” (coded as 1).We also consider an index of these four measures, which we call the class-based injustice index; the Cronbach’s alpha score is 0.77. Class-Based Education Inequality:To capture beliefs on whether education inequality is due to individual 14 Direct questions of whether a respondent feels increased perspective-taking were not asked given social desirability bias considerations. effort or the system, we assessed three questions. We measured beliefs on whether poor families do not value education as much as richer families, and whether systemic injustices perpetuating inequity throughout society “are contributors to the inequality in educational achievement in the US” (response options: 0 = not a contributor/does not occur → 1 = main contributor). Additionally, we assess the extent to which a respondent believes that “students from low income backgrounds have the same educational opportunities as students from high income backgrounds” (response options: 0 = strongly disagree → 1 = strongly agree). Racial Injustice: The racial injustice battery included four questions from the standard racial resentment or symbolic racism measures forwarded by Kinder and Sanders (1996) and Henry and Sears (2002).15 Additionally, we asked “How much racial discrimination do you feel there is in the US today,limiting the chances of individuals from particular racial groups to get ahead?” (response options: 0 = none at all → 1 = a great deal). We also considered an index of this discrimination measure and the four racial resentment measures, which we refer to as the racial resentment index, given the Cronbach’s alpha score is 0.86. We also asked a series of questions about the respondent’s level of satisfaction with the treatment of each of the of the following minority groups (response options: 0 = very dissatisfied → 1 = very satisfied): Asians, Hispanics, blacks, Muslims and immigrants. We consider each measure separately, and as a simple index, which we refer to as the discrimination index given high internal consistency of these measures; the Cronbach’s alpha score is 0.85. Racial Prejudice:We employed two measures to capture prejudice. First, we implemented a skin-tone Implicit Association Test (IAT), a method for gauging unconscious antipathy toward various groups. The IAT has commonly been used in psychology (Greenwald, McGhee, and Schwartz 1998; Greenwald, Nosek, and Banaji 2003), and increasingly in political science to predict political behavior (Arcuri et al. 2008; Mo 2015) and policy judgments (Malhotra, Margalit, and Mo 2013; Pérez 2010). The IAT is a method designed to capture the strength of associations linking social categories (dark skin color versus light skin color) to evaluative anchors (good versus bad). The difference in categorization performance is argued to capture “implicit” (system 1) attitudes that are automatic, as opposed to “explicit” (system 2) attitudes that are effortful and conscious (Kahneman 2003). The IAT effect is a D score, which ranges from −2 to 2, where negative (positive) numbers indicate an implicit bias favoring darker (lighter) skin tones over lighter (darker) skin-tones and 0 indicates neutrality (see Online Appendix C for additional details on the IAT). 15 We included a question on the extent to which respondents agree that black Americans have gotten less than they deserve; agree that black Americans should overcome prejudice without special favors; agree that it is really just a matter of black Americans working harder to be just as well off as whites; and agree that slavery and discrimination has made it difficult for black Americans to work their way up (response options: 0 = strongly disagree → 1 = strongly agree). 727 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