Impact of missing data on cluster analysis in education

In the research conducted, when at least one of the data forming the sample cannot be obtained, the problem of missing data arises. The issue of missing data may have occurred due to reasons such as the way data was obtained, insufficient time, or intentionally not answering the question. In almost all of the studies conducted under these conditions, missing data can occur. The resulting missing (incomplete) data can largely affect the outcome of the research. Different statistical methods are used to reduce the effects of these missing data. Among the methods of dealing with missing data, data deletion and imputation methods are the most commonly used methods. In the conducted study, data imputation methods were used to perform cluster analysis using data obtained from data published by the Ministry of National Education (2023) and the Turkish Statistical Institute (2023) classified by provinces During data selection, living indices and education indices were preferred. 8% of the data was randomly selected to obtain the missing data set. Initially, analysis methods without missing data were used during the clustering analysis. Among the cluster analysis methods, within-group mean correlation method, ward method, k-means method, and two-step clustering method were used. Euclidean distance was preferred as the distance measure. The SPSS 22.0 program was used during the analysis process. It was observed that the clusters changed and outliers occurred in the cluster analysis methods used. It was understood that the median imputation method is the most preferred method because it is not affected by outliers in case of missing data. Since other methods were affected by outliers, the clustering results showed variations. The results of the analysis are included in the study.