Using Literacy and Numeracy Assessment Tool (LNAT) data-extraction file variables in a binary logistic regression to predict module completion

Greyling, Willfred (2015) Using Literacy and Numeracy Assessment Tool (LNAT) data-extraction file variables in a binary logistic regression to predict module completion. Wintec/TEC report, 30 September 2015. (Unpublished)

[img] PDF - Accepted Version
Restricted to Repository staff only

861Kb

Abstract or Summary

In this, the last of four reports, we provide a very brief account of two analyses. First, we report the outcome of a chi-square analysis of the association between two variables: initial step score and module-completion category with the latter variable analysed as a multi-levelled module-completion variable (i.e. six levels of completion) and then as a binary variable (categorised as Pass/Fail). We also report on a second analysis: a binary logistic regression. In Appendix A, we report the findings for the chi-square analyses. Our aim was in both cases (and for both reading and numeracy) to test the null hypothesis that there was no association between initial reading/numeracy step levels and module completion. In all cases we rejected the null hypotheses, concluding that there was an association between the levels of the two variables for both the reading and the numeracy data sets. We offer a brief interpretation of these findings, using adjusted residuals as a guide to identify cells in the contingency table which showed meaningful associations. In Appendix B, we report the results of a binary logistic regression which included two biographical variables (Gender and Ethnicity), as well as a continuous variable, Initial Scale Score, as predictors of the dichotomous Module Completion variable. We opted for a binary logistic regression to ensure that the sample size would be adequate – a multinomial logistic regression would have required a sample of at least 2880 observations, distributed evenly across 436 cells (with no fewer than 5 observations per cell) (Lachenicht, 2002). This statistical procedure is used to identify independent variables (both continuous and/or categorical) (in our case, gender, ethnicity, English as a first language and initial scale score) that predict membership of the levels of a dichotomous variable (in this case, the Pass/Fail levels of Module completion). We found that the predictors explained a very modest 7% to 9.4% of the classification model for reading, and an equally modest 7.7% to 10.3% of the classification model for numeracy. We agree with the second reviewer of this research that a much richer data set (with more predictors) is needed to predict membership of the pass/fail categories of module completion. By implication, the LNAT variables explain a limited proportion of the variance on the dichotomous pass/fail of module completion.

Item Type:Report
Keywords that describe the item:binary logistic regression, literacy, numeracy, module completions
Subjects:L Education > LC Special aspects of education
Divisions:Schools > Centre for Foundation Studies
ID Code:3913
Deposited By:
Deposited On:12 Oct 2015 02:57
Last Modified:10 Apr 2017 02:34

Repository Staff Only: item control page