A comparative look at private and public schools' class size determinants

Abstract

This paper tests three theories of class size determination: that schools assign better‐behaved students, higher quality teachers, or higher‐achieving students into larger classes. Furthermore, we estimate how these methodologies differ between public and private schools. Using a nationally representative sample from the USA, we show that, within public schools, third‐grade class size is correlated with first‐grade ability and, to a lesser extent, first‐grade behavior. Private schools, however, appear to assign teachers reporting greater control over school policy to larger classes and teachers with more experience to smaller classes. Class size determination is due to uniquely different processes within public and private schools.

Keywords:

• class size
• private schools

JEL codes:

• L33
• H40
• I21

Acknowledgements

We thank seminar participants at the University of Georgia, University of South Florida, American Educational Finance Association, Southern Economic Association Meetings, and the Clemson University brown‐bag lunch for comments on an earlier version of this paper. We also thank Jeff DeSimone, Jahn Hakes, Rey Hernández, Curtis Simon, John Warner, and Lei Zhang for helpful conversations. Errors or deficiencies that have survived this counsel are most assuredly ours alone.

Notes

1. Several of these, for example, focus on the Tennessee STAR experiment such as Krueger (1999).

2. In Texas, for example, class size is explicitly compensatory. Texas allows state funds to be used to reduce class size for compensatory education if school can show a certain percentage of students meet the state eligibility criteria for students at risk of dropping out of school. http://www.tea.state.tx.us/school.finance/audit/resguide12/comped/comped-09.html (accessed June 7, 2009).

3. In the elementary school data we use, the correlation for first‐grade test scores and behavior is 0.2413. This is significant at the 1% level.

4. Authors' calculations from NCES (2007).

5. Authors' calculations from the Schools and Staffing Survey, 2003–2004 (NCES, 2004). About 78% of public school teachers are ‘members of a teachers' union or an employee association similar to a union’; only 7% of private school teachers are.

6. The US Department of Education (2000) provides an overview of state regulations of teacher certification in private schools. Goldhaber and Brewer (2000) discuss state differences in teacher certification for public schools. For initial certification, most states require a public school teacher to pass a standardized test. Almost all prospective teachers pass these tests on the first try.

7. Authors' calculations using the ECLS.

8. For instance, parents may pressure schools to assign their child to particular classes or particular teachers may be more adept at receiving plum assignments (Sieber 1982; Oakes 1995). Although we cannot observe who influences the process, we test for non‐random assignment of teachers and students and differential processes of non‐random assignment within public and private schools.

9. Class sizes range from 10 to 35. Although the choice is discrete, the dependent variable takes on a large number of values. As such, limited depended variable estimation techniques are unwieldy.

10. We estimate similar regressions without fixed effects using a broad set of school‐level controls. These estimates are qualitatively similar for both samples, although point estimates are frequently larger.

11. The reported differences in within‐school standard deviations are all statistically significant at the 5% level.

12. The summary statistics report the raw exam scores; however, we standardize the exam scores in the regression analysis.

13. Parental measures of behavior may be more subjective than teachers.’ As a robustness check, we construct the behavior index incorporating the teachers' responses to four behavior questions and omitting the parental response. The results are similar. We thank an anonymous referee for the suggestion.

14. In results not presented here, we recode the middle group of ‘some influence’ as a one so that these teachers are counted as having control over school policy. This change does not affect the public school results substantially. However, when coded this way, private school teachers with control over school policy appear in smaller, instead of larger classes as in Table 2. The smaller class size result is not statistically significant in the fuller specifications analogous to Columns (5) through (7) in Table 2.

15. Most elementary school teachers, both public and private, graduate from schools of education with a bachelor's degree in education. In our dataset, about 78% of public and private school third‐grade teachers list their undergraduate major as elementary education.

16. Results are not sensitive to coding the middle group of ‘neither agree nor disagree’ as a 1 instead of a 0.

17. Hedges, Laine, and Greenwald (1994a and 1994b) criticize Hanushek's (1989) meta‐analyses and review the same evidence to find that inputs do matter.

18. An exception to the natural experiments is Hoxby (2000a); using natural population variation she does not find significant evidence of class size effects.

19. Seven of the eight goals listed as options were the same for public and private principals. The eighth choice for private school principals was ‘fostering religious or spiritual development.’ The eighth choice for public school teachers was ‘promoting multi‐cultural awareness or understanding.’

20. This calculation is determined using sample weights.