The neighborhood context of school openings: Charter school expansion and socioeconomic ascent in the United States

ABSTRACT

In this study, we examine the association between neighborhood socioeconomic ascent from 1990 to 2010 and charter school openings from 2010 to 2016 using a national sample of school attendance boundaries in the U.S. We first index school attendance boundaries into typologies according to their demographic profiles in 2010, and changes along multiple components of socioeconomic status from 1990 to 2010. We then examine whether charter schools were increasingly opening in neighborhoods that experienced SES ascent and find notable variation between different types of neighborhoods where charters expanded. When charters opened in ascendant neighborhoods, they located proportionally more often in racially and socioeconomically diverse urban areas compared to prior years. Results provide a rich descriptive portrait of the neighborhood context of recent charter expansion.

Acknowledgments

We thank the anonymous reviewers and editor for very helpful comments on previous versions.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Data on charter school orientation (e.g., religious, market-based, mission-driven aims; etc.) are limited, thus restricting explicit examinations at a broader geographic scale of how charter school decisions about where to locate may vary based on differing organizational aims.

2. Our analyses explore charter school openings from 2010 to 2016, so we need data from 2008–2009 to confirm whether a charter school in 2009–2010 existed in the prior academic year.

3. About 95% of census blocks nest entirely within SABs (Saporito et al., 2007).

4. Data are at the school/SAB-tract level, and tracts appear multiple times if located in multiple SABs. Tracts are assigned a unique reapportionment weight for each associated SAB. We normalized census boundaries to 2010 definitions using the Longitudinal Tract Database (LTDB) (see J. Logan et al. (2014)).

5. While information on demographic and population counts are derived from the decennial census (a full-count survey), SES measures are derived from the 1990 long form (Summary File 3 (SF3)) decennial census (a 1-in-6 sample) and the 2008–2012 ACS 5-year estimates (a roughly 1-in-8 sample when pooled across 5 years). Researchers concerned with margin of error in the SF3 and ACS data have proposed two techniques to reduce uncertainty that essentially require either combining geographies or combining measures (J. R. Logan et al., 2020). One approach is to aggregate census tracts to larger areal units such that the larger sample sizes would produce more reliable estimates. Another suggestion is to use dimension reduction techniques, such as factor analysis, to classify geodemographic groups using a composite of multiple measures rather than focusing on single indicator (Spielman & Singleton, 2015). The logic underlying this approach, as Spielman and Singleton (2015) demonstrate, is that “estimation errors in each indicator are likely to be independent of one another, so that random errors from multiple indicators will average to zero” (J. R. Logan et al., 2020, p. 2). In this study, we employ both techniques to address potential noise in the SES estimates by aggregating tracts (which are smaller geographic units) up to the SAB level and collapsing SES data on multiple indicators to construct composite SES factor scores.

6. SES data for 2010 are derived from 2008–2012 ACS (herein, 2010 for brevity).

7. Because we want the factors to be correlated, we used promax (oblique) rotation. While economic measures explain the most, all five indicators loaded highly onto a single factor, as expected, in both years (1990 and 2010), with high correlation and communality. We also constructed factor scores using logged values for median household income, median rent, and median home value, but doing so did not substantively alter eigenvalues or factor scores.

8. We began with tracts for two reasons. First, because we were interested in SES trajectories of neighborhoods in our sample relative to all neighborhoods located within the same metropolitan region, we calculated SES factor scores using the entire universe of tracts for each MSA represented in our sample. SABINS does not provide complete coverage for all SABs within a given MSA. We thus reapportioned tracts to SABs to take advantage of complete information for each MSA’s ecosystem of neighborhoods when ranking SABs by SES. Second, because our sample includes fewer SABs in each MSA, a ten-percentage point change in a SAB’s rank (relative to other SABs in the same MSA) may not capture meaningful SES ascent.

9. Our threshold of 10 points should capture substantive increases in the relative position of neighborhoods compared to others within the same MSA, even with the larger margin of error induced by the ACS’s sampling design. Robustness checks using alternate thresholds (8, 12, and 15%) yielded similar patterns of ascent.

10. We employed a two-decade time frame for methodological and conceptual reasons since the timing of ascent does not always neatly align with decennial censuses.

11. To ensure cluster analyses produced consistent typologies, we replicated our two-step PCA and cluster analysis ten times, which produced strongly correlated typologies between sets (R =.93). We also repeated our empirical analyses using these ten sets of replicated clusters, and substantive patterns held.

12. We measure baseline neighborhood charter presence as the number of continuing charter schools. These are charter schools that were operational during the 2008–2009 academic year and remained operational at the start of the 2009–2010 academic year. Continuing charter schools are also be referenced in this study as “existing charter” schools, since these schools were already operational during the baseline academic year.

13. Ascendant neighborhood types have similar baseline values and mean trajectories of SES ascent, on average.

14. Note that multiple charter schools can open in the same ascendant neighborhood. For analyses accompanying Figure 2, we take the perspective of the school, examining whether or not each charter opening occurs in an ascendant neighborhood. That is, we observe all charter openings (and all continuing charters) regardless of whether they occur in the same neighborhood.

15. SABs are represented multiple times in our dataset if (a) they received a new charter or TPS after 2010; or (b) continuing charter and TPSs were located in the same SAB in 2010.

16. In sensitivity analyses, we performed models using an alternative measure defining ascent from 2000 to 2010. While the predicted probabilities of charter openings were slightly higher than our preferred measure of ascent, patterns held across all neighborhood types.

Additional information

Notes on contributors

Jennifer Candipan

Jennifer Candipan is an Assistant Professor of Sociology at Brown University. Her research agenda broadly investigates how social and spatial contexts produce racial/ethnic, health, and economic disparities. Recent work examines neighborhood change, segregated urban mobility, links between housing and education markets, and how neighborhood and school processes shape inequality.

Noli Brazil

Noli Brazil is an Assistant Professor of Community and Regional Development in the Department of Human Ecology at the University of California, Davis. His research agenda spans multiple areas of inquiry connected by an interest in understanding the influence of place and space in generating social, economic, and health inequalities.