Comparison of influence of the missing data handling methods on measurement invariance

The aim of this thesis is to find a solution to the missing data problem in measurement invariance studies. For this purpose, the effects of regression assignment, multiple assignment and expectation maximization methods used to deal with missing data on the student questionnaires of the PISA application in 2018 were compared in the measurement invariance study based on the language variable. America, UK, Canada, Ireland, Singapore, New Zealand, Australia for English; Belgium, France, Switzerland, Canada and Luxembourg for French; Turkey, Ukraine, Russia, Greece, and Israel were chosen to target the comparison of different languages. Data were generated at 5%, 10% and 20% missing rates in MCAR and MAR mechanisms. Missing data were completed with the determined methods and measurement invariance analyzes were performed. The obtained values were compared with the values obtained from the full data sets. In the results of the research, it is stated that in the measurement invariance studies performed on MCAR, when the missing rate is 5%, all three methods of coping with the missing data used in the research can be used; expectation maximization and multiple assignment methods can be used when the missing rate is 10%; it was found that the multiple assignment method can be used when the missing rate is 20%. In measurement invariance studies performed on MAR, all three methods used in the research can be preferred to deal with missing data at all missing rates.