Some Limits Using Random Slope Models to Measure Academic Growth

Alder Graduate School of Education Senior Research Scientist Dan Wright published an article in the open access journal Frontiers in Education, a publication focused on assessment, testing, and applied measurement. Wright’s article, “Some Limits Using Random Slope Models to Measure Academic Growth,” explores random slope multilevel (RSM) models and ways in which some of the aberrant results of RSM can be avoided through the use of a combination of methodologies.

It is common to estimate student and school growth for many statistics (e.g., test scores, truancy, and graduation rates). One statistical technique that has been used for this is called an RSM model. This paper shows an example where this method went wrong: Schools with graduation rates that increased by about 10% each year were given lower growth scores than schools that dropped by about this amount. Several computer simulations were conducted and showed that the RSM model should not be used to measure growth if there are fewer than six time points of data. An alternative, using ordinary least squares estimates, does not produce these highly discrepant results and is preferred in situations with few time points. Computer code for both of these methods is included in the paper.

 

Full Text Article