Boylamsal Verilerde Çok Düzeyli Doğrusal Regresyon ve Kantil Regresyon Yöntemlerinin Karşılaştırılması
![](/xmlui/themes/Mirage2//images/mime2.png)
View/ Open
Date
2023Author
Kafkas, Begüm
xmlui.dri2xhtml.METS-1.0.item-emb
Acik erisimxmlui.mirage2.itemSummaryView.MetaData
Show full item recordAbstract
Classical simple linear regression analysis assumes that the observations are independent and identical. This assumption is not met in situations such as hierarchical data. The fact that the principle of independence of observations in hierarchical data is violated causes biased results in regression analysis. Multilevel modeling techniques are used to solve such problems. However, in cases where the distribution is skewed, quantile regression models, which are used as an alternative method, have been developed due to the low efficiency of the methods used in multilevel modeling. In cases where classical regression conditions are not met, information about each point of the distribution can be obtained with quantile regression models, which are advantageous. In this study, the performances of multilevel regression (random intercept and random slope model) and multilevel quantile regression analysis methods were compared in data structures with three longitudinal observation values belonging to individuals with different sample sizes and different distributions. For this purpose, model-data fit values, absolute error and bias values, and changes in intercept and slope coefficients at different quantile levels (0.10; 0.25; 0.50; 0.75; 0.90) and different sample sizes (50; 500; 1000) in multilevel regression and multilevel quantile regression methods in longitudinal data sets were analysed. The variation of absolute error and bias values was examined. In the analysis, multilevel quantile regression analysis were performed with qrLMM and nlme, which are sub-packages of the R program. As a result of the analyses, it was seen that multilevel quantile regression made more unbiased predictions than multilevel regression analysis in all data structures. In addition, although there was no significant difference between the intercept value obtained from multilevel regression and the intercept value obtained from the median in any case, it was observed that a significant difference occurred between the slope coefficients in many cases.