Location decisions of charter schools: an examination of Michigan

ABSTRACT

Using school level data we examine which factors influence charter school location decisions. We augment previous research by employing a panel dataset, recently developed geographic techniques to measure distances and define areas, and employing a hurdle model to deal with the excess zero problem. The main results of our research indicate that, after controlling for other factors, charters are more likely to locate in areas with higher median income, fewer students who receive a free or reduced lunch, lower reading scores, a larger percentage of black students, and more racial diversity.

KEYWORDS:

• Charter school location
• charter school decisions
• school choice

JEL CLASSIFICATION:

• I21
• I28

Acknowledgements

The authors would like to thank Matthew Winden for his helpful suggestions. We would also like to thank the participants of the 2012 Midwest Economics Association Meetings who commented on an earlier version of this paper.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Stoddard and Corcoran (2007) do something similar, but it is not the emphasis of their paper.

2 Since these are non-nested models the Vuong test is the most appropriate to examine which is superior from a statistical point of view (Winkelmann 2008, 188).

3 An example of where zero-inflated model would be more appropriate is in a buying decision where in stage one you decide whether or not to buy and stage two you decide how much to buy. If once you ‘go out to buy’ you still could buy zero then a zero-inflated model would be more appropriate.

4 Approximately $1000 less per student on average; this equates to approximately 80% of what traditional schools receive.

5 Intermediate School Districts (ISD’s) assist local school districts in providing programs and services.

6 With aggregate data it is impossible to separate school performance from student ability and effort.

7 In Michigan, to be proficient means to score in one of the top two categories out of four. This paper uses test score results for grades 3–9, as these data are the most complete.

8 Special education and English language learner data had issues with unreported counts due to student privacy laws. To address this, the maximum percent of enrollment that could possibly be enrolled in special education is assigned to those missing cases. A regression was then run to confirm that these imputed cases do not significantly affect the dependent variable.

9 Area level statistics had missing data issues. The American Community Survey has data for years 2009–2012. We distribute the 2009 information to the years that come previous (2001–2008).

10 Michigan also changed the exam in the 2011/2012 school year, the last year of our panel. We do not see much of a change in our data so we did not include an additional dummy for this year in our main estimations since we did not see a substantial drop in test scores in this year and including it does not qualitatively change our main results. These results are available from the authors.