Explaining the substantial inter-domain and over-time correlations in student achievement: the importance of stable student attributes

ABSTRACT

Multi-domain and longitudinal studies of student achievement routinely find moderate to strong correlations across achievement domains and even stronger within-domain correlations over time. The purpose of this study is to examine the sources of these patterns analysing student achievement in 5 domains across Years 3, 5 and 7. The analysis is of longitudinal population data of over 22,000 students and utilizes fixed-effects models to incorporate stable general and domain-specific latent factors. These latent factors correspond to a general cognitive-ability-like factor and specific aptitudes in particular, or types of, subject areas. The preferred model incorporates both general and domain-specific latent factors with stronger effects for the general factor, although the domain-specific factors are particularly strong for spelling and numeracy. When taking into account general and domain-specific latent factors, the effects of student’s socioeconomic status (SES) and school SES are trivial.

KEYWORDS:

• Achievement
• longitudinal studies
• educational career
• structural equation modelling
• fixed-effects models

Acknowledgement

This research uses unit record data from the NAPLAN administrative data collection, provided by the Department of Education and Training (DET) in Victoria. The findings and interpretations reported in this paper are those of the author and should not be attributed to DET or any other branch of the Victorian government.

Notes on contributor

Gary N. Marks is currently a Public Policy Fellow, Office of Government, Policy & Strategy, Australian Catholic University. He has published in a range of research areas including school leaving, cross-national differences in student performance, the school-to-work transition, the youth and adult labour markets, and over-time changes in social stratification.

Notes

1. Technical and other documentation is available online: http://www.nap.edu.au/results-and-reports/national-reports.html

2. The computer code for these analyses is available from the author.

3. If the covariances were not specified to be estimated, then the model assumes the covariances are zero, that is, the latent factor is uncorrelated with the predictor variables. This is equivalent to a random-effects model.

4. Details of the measures can be found at https://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_calis_sect077.htm

5. The covariances between the latent variables are estimated in this model, but there is no term specified to provide estimates.