Class size and learner outcomes in South African schools: The role of school socioeconomic status

ABSTRACT

Class size reduction is frequently argued to be a cost-effective way to improve learner outcomes. In the South African context, most studies conclude that greater class sizes are associated with poorer educational outcomes on average. However, given the country’s bimodal education system, it is plausible to believe that such a relationship may depend on where learners find themselves in the system. This paper merges newly available data from the 2017/18 School Monitoring Survey with external administrative data to investigate whether the relationship between class size and learner outcomes varies by school socioeconomic status. Although extreme class sizes are concentrated in poorer schools, class size is only negatively associated with learner outcomes in wealthier schools. This finding is robust to several robustness tests. This does not imply that class size does not matter. Rather, reductions may only be effective in the South African context once other school quality-related factors are addressed.

KEYWORDS:

• South Africa
• class size
• socioeconomic status
• education
• human capital
• school quality

JEL CLASSIFICATION:

• I20
• I21
• I24
• I28

Acknowledgements

The author is grateful to helpful comments from two anonymous reviewers, as well as feedback on an earlier version of the article from Professor Servaas van der Berg and the Research on Socio-Economic Policy (ReSEP) research unit based in the Department of Economics at Stellenbosch University, as well as the Development Policy Research Unit (DPRU) based in the School of Economics at the University of Cape Town.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Science, Technology, Engineering, and Mathematics.

2 Special Needs Education Schools, Specialisation Schools, and private schools are excluded.

3 In particular, the survey focused on 13 of the 15 Action Plan indicators which were measured in the 2011 School Monitoring Survey.

4 The interested reader is referred to the 2017/18 SMS report for more information on the survey’s sampling design, available here: https://www.education.gov.za/Resources/Reports.aspx.

5 The latter weights are used for teachers as observations, as suggested in the 2017/18 SMS report.

6 This is because the correlation between self-reported class size and the aforementioned error is small and insignificant, while the correlation between observed class size and the error is small and very significant (at the 1% level). As such, the properties of the ordinary least squares (OLS) regression estimates to follow need not be biased by this discrepancy in class sizes.

7 Only schools with a maximum of 100 reported learners are included, although the relationship holds without this restriction.

8 In the South African education system, public schools are categorised into five quintiles based on the relative socioeconomic status of their surrounding communities.

9 The definition of teachers here includes both School Governing Body and state-employed teachers, principals, and School Management Team members, teachers, or practitioners (including Grade R teachers). It does not include administrative staff, cleaners, caretakers, security, or student teachers on practical.

10 Specifically, the mean class sizes of Quintile 1, 2, and 3 schools are 56.34, 57.15, and 54.56 – all of which are not statistically significantly different from one another. The mean class size of Quintile 5 schools is 37.

11 A Dinaledi school is one which receives financial and other assistance from government subject to certain criteria being met. The intension of the policy is to increase participation and performance in mathematics and physical sciences in historically disadvantaged schools.

12 It should be noted that, despite their inclusion, the estimates do not differ significantly if these FE are excluded from any of the models.

13 Similar results hold when individual dummy variables for each school Quintile are included in the model as opposed to the current specification (not shown here).

14 The presence of an interaction term implies that the association between class size and the dependent variable depends on two coefficients: the coefficient for class size alone, and the coefficient for the relevant interaction term. The association in Model (3) here is computed as follows: 0.014 + (−0.786) = −0.772. Therefore, an increase in the class size of 20 learners implies a 20 × (−0.772) = 15.44 reduction in a school’s NSC pass rate, on average.

15 The SNAP survey contains data recorded on the 10th school day of each year.

16 The dependent variable in every model in Table 6 is the school NSC pass rate.

17 Given that the relevant question in the survey is phrased as ‘What is the largest class that you teach this year?’, in actuality the independent variable of interest is the minimum/maximum/median/mean maximum class size in a given school, as discussed earlier in the paper.