Publication: Büyüme Eğrisi Modellerinin Tahmininde Genelleştirilmiş Tahmin Denklemlerinin Kullanılabilirliğinin İrdelenmesi
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Abstract
Büyüme, genetik ve çevre arasındaki etkileşim sonucunda organizmanın hacminde ve kütlesinde meydana gelen artışı ifade eder. Büyüme eğrisi modelleri büyüme sürecinde zamana bağlı olarak meydana gelen değişimlerin açıklanması için oldukça kullanışlı yöntemlerdir. Geleneksel büyüme eğrisi modelleri tekrar eden ölçümler arasındaki ilişkiyi dikkate almamaktadır. Bu durum hatalı tahminlere ve bilgi kaybına neden olabilmektedir. Genelleştirilmiş Tahmin Denklemleri (GTD) yöntemi tekrar eden ölçümlerden kaynaklanan oto-korelasyonları dikkate alarak analizi gerçekleştirdiğinden daha güvenilir sonuçlar üretebilmektedir. Bu çalışmada, tekrar eden ölçümlerden kaynaklanan etkilerin dikkate alınabilmesi için GTD yöntemi kullanılarak elde edilen farklı korelasyon (Autoregressive, Exchangeable, Independent, M-Dependent ve Unstructured) matrisleri kullanılarak oluşturulan sabit etkili büyüme eğrisi modelleri Akaike Bilgi Kriter (AIC) değerleri kullanılarak iki farklı örnek büyüklüğünde karşılaştırılmıştır. Çalışma sonucunda, büyük örnek durumunda bağımsız (Independent) ve değiştirilebilir (Exchangeable) korelasyon matrislerinin kullanıldığı modeller en iyi model, M-bağımlı (M-dependent) ve Pearson korelasyon matrisinin kullanıldığı model ise en kötü modeller olarak belirlenmiştir. Küçük örnek durumunda ise yine bağımsız ve değiştirilebilir korelasyon matrislerinin kullanıldığı modeller en iyi model olarak belirlenirken en kötü model ise M-bağımlı korelasyon matrisinin kullanıldığı model olarak belirlenmiştir. Bu sonuçlar zaman noktaları arasındaki ilişki derecesinin sabit olduğu matrislerden elde edilmesi açısından ilgi çekicidir. Anahtar Kelimeler: Geleneksel Büyüme Eğrisi Modeli, Genelleştirilmiş Tahmin Denklemleri (GTD), Tekrarlanan Ölçümlü Veriler, Kovaryans Matrisi, Akaike Bilgi Kriteri (AIC).
Growth expresses the increase of organism?s volume and mass resultant of interaction between genotype and environment. Growth curve models are useful tools for interpreting the temporal conversions during growth. Conventional growth curve models ignore the relationship among repeated measurements; this case could lead to erroneous predictions and information loss. Generalized Estimating Equations (GEE) is a method used for analyzing the data by taking the correlations among the observations into account, which yields more reliable results. In this study, fixed effect growth curve models was created using different correlation matrices (Autoregressive, Exchangeable, Independent, M-Dependent and Unstructured) acquired by using GEE to take the effects derived from repeated measurements into account were compared on two size of sample by using Akaike Information Criterion (AIC). In present study, while the models which use independent and exchangeable correlation matrices were determined as the best models, the models which use M-dependent and Pearson correlation matrices were worst models on large sample size. Whereas models which use independent and exchangeable correlation matrices were determined as the best models, the models which use M-dependent correlation matrix were worst model on small sample size. These results are noteworthy in terms of obtained from the matrices that relation rates among time points were constant. Key Words: Traditional Growth Curve Model, Generalized Estimating Equations (GEE), Repeated Measurement Data, Covariance Matrix, Akaike Information Criterion (AIC).
Growth expresses the increase of organism?s volume and mass resultant of interaction between genotype and environment. Growth curve models are useful tools for interpreting the temporal conversions during growth. Conventional growth curve models ignore the relationship among repeated measurements; this case could lead to erroneous predictions and information loss. Generalized Estimating Equations (GEE) is a method used for analyzing the data by taking the correlations among the observations into account, which yields more reliable results. In this study, fixed effect growth curve models was created using different correlation matrices (Autoregressive, Exchangeable, Independent, M-Dependent and Unstructured) acquired by using GEE to take the effects derived from repeated measurements into account were compared on two size of sample by using Akaike Information Criterion (AIC). In present study, while the models which use independent and exchangeable correlation matrices were determined as the best models, the models which use M-dependent and Pearson correlation matrices were worst models on large sample size. Whereas models which use independent and exchangeable correlation matrices were determined as the best models, the models which use M-dependent correlation matrix were worst model on small sample size. These results are noteworthy in terms of obtained from the matrices that relation rates among time points were constant. Key Words: Traditional Growth Curve Model, Generalized Estimating Equations (GEE), Repeated Measurement Data, Covariance Matrix, Akaike Information Criterion (AIC).
Description
Tez (yüksek lisans) -- Ondokuz Mayıs Üniversitesi, 2013
Libra Kayıt No: 66142
Libra Kayıt No: 66142
Citation
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
82
