Determining the Appropriate Model by Using Fit Criteria According To Different Zero Value Ratios in the Dependent Variable for Nonlinear Regression
No Thumbnail Available
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Bu tez çalışmasında, sayıma dayalı olarak elde edilen bağımlı değişkendeki sıfır değerlerinin %5, %25, %50, %75 ve %100 oranlarında silinmesi sonucunda, model uyum ölçütlerine bağlı olarak, hangi doğrusal olmayan regresyon modelinin uygun olacağının belirlenmesi amaçlanmıştır. Ayrıca her bir farklı oranlarda kesilmiş sıfır gözlemler için oluşturulan regresyon modellerin birbirleriyle olan etkinlikleri ele alınmıştır. Sıfır gözlemlerin tamamı modeldeyken ve sıfır gözlemlerin %50'sine kadar silindiğinde, sıfır yayılımlı hurdle negatif binomial regresyon en iyi model olarak seçilmiştir. Sıfır gözlemlerin tamamı silindiğinde ise sıfır değer kesilmiş negatif binomial regresyon en iyi model seçilmiştir. Genellikle negatif binomial esaslı regresyonlar en iyi model olarak seçilmiştir. Çünkü Poisson regresyon ve türevleri aşırı yayılımdan olumsuz anlamda daha fazla etkilenmektedirler. Sonuç olarak, bağımlı değişkende var olan sıfır gözlemlerin, sayısal oranlarına göre uygun model belirlenmiştir.
In this study, it was aimed to determine which nonlinear regression model would be appropriate, depending on the model fit criteria, as a result of the deletion of the zero values in the dependent variable obtained based on counting at the rates of 5%, 25%, 50%, 75% and 100%. Zero-inflated hurdle negative binomial regression was chosen as the best model when all of the zero observations were in the model and up to 50% of the zero observations were deleted. When all zero observations were deleted, the truncated negative binomial regression was chosen as the best model. Generally, negative binomial-based regressions were chosen as the best model. Because Poisson regression and its derivatives are more negatively affected by overdispersion. As a result, the appropriate model was determined according to the numerical ratios of the zero observations in the dependent variable.
In this study, it was aimed to determine which nonlinear regression model would be appropriate, depending on the model fit criteria, as a result of the deletion of the zero values in the dependent variable obtained based on counting at the rates of 5%, 25%, 50%, 75% and 100%. Zero-inflated hurdle negative binomial regression was chosen as the best model when all of the zero observations were in the model and up to 50% of the zero observations were deleted. When all zero observations were deleted, the truncated negative binomial regression was chosen as the best model. Generally, negative binomial-based regressions were chosen as the best model. Because Poisson regression and its derivatives are more negatively affected by overdispersion. As a result, the appropriate model was determined according to the numerical ratios of the zero observations in the dependent variable.
Description
Keywords
İstatistik, Doğrusal olmayan regresyon, Gözlem, Model seçimi, Yayılım politikası, Statistics, Nonlinear regression, Observation, Model selection, Expansion policy
Turkish CoHE Thesis Center URL
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
70