| The objective of this study is to find appropriate statistical analysis methods for a dataset that collected from 125 families in Nanjing that suffered from acute respiratory infection (ARI).The most significant characteristic is that correlation in each family. Three different models, GEE1, multilevel model and zero-inflated model, are chose according to data characteristics and requirements of analysis, to assess risk factors and estimate the correlation of ARI intrafamily. Comparison among 3 models shows the pros and cons of 3 methods with respect to twofold of analysis.In GEE1 model the response variable is fitted with Poisson and negative binomial distributions. At the same time, the models for different structure of correlation are compared. For multilevel model, Poisson, extra Poisson and negative binomial distribution are used to fit the response variable. Zero-inflated model, with 2 parts, zero and Poisson distribution,, are transformed to conditional zero-inflated model. Logit and Log linkage function are used to promote the explanation of the result. The results of significant risk factor show that the same conclusion is concluded with the three different models. Those who are of aged less than14 have history of chronic respiratory system disease and own ill body conditions are the higher risk population of ARI.Our study shows the three models have advantages in analyzing ARI risk factors respectively.GEE1 is superior in estimation of regression coefficient. Multilevel model could be used to estimate random effect and correlation coefficient intrafamily. The risk factors are divided 2 parts with ZIP model, so that the explanation of regression coefficient is more practical and detailed. Furthermore, the application of GEE1 and Multilevel model could be extended to quantitative data and categorical data, while ZIP model is only suitable to binary data. This study provides useful direction to choose the powerful statistical models in case that the same type data are met. |
PDF Full Text Request