Table 3.

Risk of bias assessment of included quantitative studies (ROBINS‐I).

Author	Title	Overall comment	Overall judgement	Confounding bias	Judgement	Selection bias	Judgement	Classification bias	Judgement
Metzner (1926)	Size of class for mentally retarded children	This is a well‐performed study, but unfortunately, SD's are missing and are not possible to retrieve or calculate.	Moderate risk of bias	The authors term it an experiment, but only report that groups of mentally retarded students (four treated and one control) were ‘formed’. Treated are three classes with 15 students, three classes with 20 students, three classes with 25 students, and three classes with 30 students; controls are 12 classes with 22 students. Gender is highly imbalanced, mental age and age in years are reasonably balanced, IQ is reasonably balanced, and SES is not reported (Table 3). Nothing is controlled for.	Moderate risk of bias	8%–10% were replaced due to drop‐out (p. 242), otherwise all children from the initial sample are followed from start to finish. Replacements took the same pre‐tests as students in the initial sample, but the size of the class they attended before they were included as replacements and the timing of the replacement is not reported.	Moderate risk of bias	Nothing of concern	Low risk of bias
Forness (1985)	Effects of class size on attention, communication, and disruption of mildly mentally retarded children	The study is given a rating of critical risk of bias in the confounding domain and the rest is therefore not assessed.	Critical risk of bias	Only age and gender considered (Table 3). There is some imbalance on gender (between small classes vs. medium and large) and a relatively large imbalance on age (between small classes vs. medium and large). All students are characterised as mildly/educable mentally retarded (IQ range of 50–70). Nothing is controlled for.	Critical risk of bias
MAGI Educational Services, Inc. (1995)	Results of a Statewide Research Study on the Effects of Class Size in Special Education	The study is given a rating of critical risk of bias in the confounding domain and the rest is therefore not assessed.	Critical risk of bias	No confounders considered.	Critical risk of bias

Author	Deviation bias	Judgement	Missing data	Judgement	Measurement bias	Judgement	Reporting bias	Judgement
Metzner (1926)	Eight percent of the control group and 10% of the treated (not reported separately by the four treatment groups, but reported that the range was 8%–13%) dropped out and were replaced. The size of the class they attended before they were included as replacements and the timing of the replacement is not reported.	Moderate risk of bias	Eight percent of the control group and 10% of the treated (not reported separately by the four treatment groups, but reported that the range was 8%–13%) dropped out and were replaced. Data only shown for those who were in the study by the end of the experiment.	Moderate risk of bias	No mentioning of blinding, otherwise tests are standardised tests (Stanford‐Binet and Pressey reading test)	Moderate risk of bias	No pre‐specified plan of analysis, but otherwise nothing to indicate selective reporting biases	Moderate risk of bias
Forness (1985)
MAGI Educational Services, Inc. (1995)