Abstract
The present study examined indirect effects of principal leadership on the mathematics achievement of 254,475 15-year-old students from 10,313 schools in 32 OECD economies. Results showed that the students could be divided into three categories (Disadvantaged, Average, and Privileged) differing in levels of student SES and prior achievement, parental academic expectations, and access to school resources. Results also showed that principal leadership effects accounted for a greater proportion of between-school achievement variance for Disadvantaged vis-à-vis Privileged or Average students. In particular, instructional leadership had the largest positive effect on Disadvantaged vis-à-vis other students’ achievement via the mediating variables of teacher autonomy and morale. Distributed leadership negatively affected the achievement of Disadvantaged but not other students. The negative effects of principal goal-setting were the largest while those of principal problem-solving were the smallest for Disadvantaged students. The study contributes to the literature by examining contextual influences on the leadership–achievement relationship.
Notes
1. According to Cheema (Citation2014), different methods of handling cases with missing values (eg listwise deletion or multiple imputation) yield comparable unbiased parameter estimates if the percentage of missing values is low. Therefore, listwise deletion of cases with missing values was used in the present study in view of the relatively low percentage of missing values.
2. While HLM is perhaps the most commonly used multilevel methodology in the analysis of nested data in school effectiveness studies (eg Liu, van Damme, Gielen, & van den Noortgate, Citation2015), some researchers use another technique: multilevel structural equation modelling (SEM). However, HLM was used in the present study because the primary purpose was to compare the pattern of mediation of leadership effects among different latent classes of students (which HLM is capable of) and not the testing of measurement models or comparison of competing structural models (which SEM is designed for). Furthermore, the nested HLM models enabled the change in principal leadership regression coefficients to be compared before and after the mediating school variables were added. This comparison is essential in the test for mediation (Baron & Kenny, Citation1986).
3. The pattern of results for MathPV2 to MathPV5 was similar to that for MathPV1 (see Tables respectively in appendix). Therefore, only the detailed results for MathPV1 are presented in the article.