Skip to main content
Fig. 4 | BMC Medical Research Methodology

Fig. 4

From: The effect of missing data and imputation on the detection of bias in cognitive testing using differential item functioning methods

Fig. 4

DIF estimates for hearing impairment in the reference scenario (No Additional Missingness) compared to low missingness scenarios with and without hot deck single imputation (single imputation) and Multiple Imputation by Chained Equations (MICE) (multiple imputation) for select cognitive test items in the Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS). Each dot represents a difference between the two groups (impaired and unimpaired) for a single threshold of the ordinal test score. When the difference in thresholds is significantly different from zero (indicated by an unfilled circle), this indicates DIF. Differences between the reference scenario and each of the missingness scenarios illustrates the magnitude of error due to missingness. The scenarios with no imputation are shifted to the left as compared to the reference estimates with no additional missingness, indicating systematic DIF estimation error. However, estimates with both single and multiple imputation more closely align with the reference estimates, illustrating the reduction of DIF estimation error with both imputation strategies

Back to article page