Connectedness and Expectations: How minority teachers can improve educational outcomes for minority students

Abstract

Research in the field of representative bureaucracy provides evidence that the presence of minority teachers can improve educational outcomes for minority students. We test two possible mechanisms by examining if the presence of minority teachers increases how ‘connected’ minority students feel to their school and the student's educational aspirations. Previous research has established a strong link between both of these factors and educational and non-educational outcomes. We find that increasing representation of African American and Latino/a teachers increases educational expectations for African American students, while increasing representation of Latino/a teachers increases school connectedness and educational expectations for Latino/a students.

Key words:

• Representation
• street-level bureaucrat
• school connectedness
• expectations

Notes

1. Admittedly, we still rely on aggregate measures of teacher representation and cannot directly test the relationship between a specific student and a specific teacher. However, the mechanisms that we are examining (role modeling and school connectedness) are not necessarily tied to the direct interaction between a student and teacher. In fact, several studies find a change in the self-perceptions of female students after very brief and even indirect exposure to a female role model (Gardiner et al. 2007; Greene et al. 1982; Stout et al. 2011). Although, purely individual level data would be preferable for some analyses, the data we use are appropriate to capture how the expectations and feelings of connectedness for minority students may change as they see more teachers that look like them.

2. Most of the extant research has almost exclusively (Pitts and Lewis 2009; Thielemann and Stewart 1996; Van Gool 2008 are notable exceptions) focused on three identities – race, ethnicity, and sex – all three of which are tied to immutable and visible demographic characteristics. However, recently researchers have found links for identities linked to occupation (Close et al. 2011), sexual orientation (Pitts 2011), language (Turgeon and Gagnon 2013), and veteran status (Gade and Wilkins 2013).

3. Regrettably we were unable to interview any African American male teachers. In the state of Georgia, 82 per cent of the African American school teachers are female (Nweke et al. 2004). Given this, our race variable is really a measure of African American female teachers.

4. The first wave of the survey was completed during the 1994–1995 school year. The Add Health data set is one of two nationally representative data sets that would allow us to examine our questions and the other data set (NLSY) is also from the late 1990s. We understand that the age of these data is problematic, however our qualitative data gathered in 2012–2013 offers a robustness check on the quantitative findings and convincingly supports them. While we agree that the educational environment has dramatically changed and that the demographics of schools have shifted significantly, our discussions with teachers demonstrate that teachers still act as role models for students and influence how connected a student feels to their school.

5. Due to the complex survey design of Add Health, sampling weights are available; however, we choose not to use weights in this analysis for several reasons. First, our analyses do not fit the three criteria (need to correct for heteroskedasticity, need to correct for endogenous sampling, or need to identify partial effects in the presence of heterogeneous effects) for requiring sample weighting offered by Solon et al. (2013). Second, we are interested in the behavioural responses of the study participants to teacher representation and are not seeking to make population-level projections based on our estimates. Finally, in our analyses, we control for mother's education and other individual characteristics, and we adjust the standard errors for intra-cluster correlations at the school-level, which partially account for selection into the sample due to the survey's design. Several studies in the behavioural economics literature do not use the sample weights in their analyses; this is a particularly common practice in studies utilizing the Add Health Data (Rees and Sabia 2010a, 2010b, 2012; Sabia and Rees 2012).

6. The higher rate of college expectations among African Americans is to be expected in this sample because the Add Health over-samples African American students with highly educated parents.

7. Note that the coefficients are very similar to the ordered probit coefficients (Appendix Table 1) (the marginal effects which are not shown for space considerations are also similar), and the HLM coefficients (Appendix Table 2).

8. Just as with the college expectations results, the coefficients from the connectedness regressions are similar to the ordered probit coefficients (and marginal effects not shown) (Appendix Table 1) and the HLM coefficients (Appendix Table 2).