Missing Data, Statistical Power of Likelihood-Ratio Test and Differential Item Functioning in NECO Mathematics Examination in Nigeria

The study examined the percentage of missing data by persons and items, and the effect of missing data on statistical power of likelihood-ratio test across differential item functioning magnitude The study adopted the ex-post facto research design. The population consisted of 1,034,629 candidates that sat for the June/July 2017 NECO mathematics examination. The study sample comprised all the 194,009 students that sat for the examination in the six Southwestern states of Nigeria. Data collected was analysed using frequency count, percentage, Likelihood-ratio Test and Multiple Imputation Chained Equation and T-test respectively. Results showed that 42.2% of examinees had one or more missing responses and that all the items of the 2017 SSCE Mathematics test attracted missing responses. The result also showed that 56 of the 60 items of the NECO Mathematics functioned differentially with respect to gender and that 55 of the 56 items displaying DIF flagged non-uniform DIF. Furthermore, results showed that likelihood ratio DIF test method identifies more differential functioning items when missing responses of examinees were replaced with Multiple Imputation Chain Equation and that there is no significant difference in the power of loglikelihood ratio test in detecting DIF items under traditional method of imputing missing data and the mice method. The study concluded that missing data had no significant influence on the statistical power of likelihood-ratio test for detecting differential item functioning in mathematics examination