@article{article_377421, title={Use of Generalized Estimating Equations with Multiple Imputations for Missing Longitudinal Data}, journal={Yüzüncü Yıl Üniversitesi Fen Bilimleri Enstitüsü Dergisi}, volume={23}, pages={96–103}, year={2018}, url={https://izlik.org/JA76BB54ZS}, author={Ser, Gazel and Okut, Hayrettin}, keywords={Marjinal modeler,Çalışma korelasyon,Çoklu atama}, abstract={<p class="MsoNormal" style="margin:0cm;margin-bottom:.0001pt;text-indent:0cm;line-height:normal;"> <span lang="en-us" style="font-size:10pt;font-family:’Times New Roman’, serif;" xml:lang="en-us">This study aimed to assess the performance of multiple
imputation for the Generalized Estimating Equation (GEE) method, one of the
marginal model approaches. Observations with longitudinal data structure
obtained from 1044 individuals during five years were used. Smoking frequency,
response variable with Poisson distribution and the independent variables
thought likely to affect these were taken into consideration. These variables
are the individual’s alcohol use frequency, the score for friend influence on
the individual’s smoking, the score for individual’s listening to his/her
family, individual-family relationship score, marriage status of parents,
gender and age. Four different working correlation structures were examined to
determine the study correlation structure in GEE. Quasi information criterion was
used to determine the most appropriate working correlation structure to fit the
data. In estimating the missing observation, the missing observations were assumed
to be missing at random, and missing observations were estimated using multiple
imputation (MI). Thus, the GEE method was applied again to the complete data
set obtained and MI-GEE results were obtained. As a result, the appropriate working
correlation structure for GEE and MI-GEE was determined as the independent
structure, and parameter estimations were obtained using this structure. In
both cases, empirical standard error results were evaluated. Accordingly, in
the data set with missing observations, effect of alcohol use and family
relationship status (p<0.001) and of age (p<0.01) on smoking was found to
be significant in GEE results. In the MI-GEE results, effect of alcohol use,
family relationship score, gender, age (P<0.001) and the score for the
individual’s relationship with his/her family (p<0.01) was found to be
significant. Standard error estimations obtained for MI-GEE were much smaller
compared to the data with missing observations. </span> </p> <p> </p> <p> </p> <p class="MsoNormal" style="margin:0cm;margin-bottom:.0001pt;text-indent:0cm;line-height:normal;"> <b> <span lang="en-us" style="font-size:10pt;font-family:’Times New Roman’, serif;" xml:lang="en-us"> </span> </b> </p> <p> <b>  </b> </p> <b> </b>}, number={1}