dr ali akbar khadem maboudi

Quantile regression can be applied to model skewed variables، especially، when the objective is to model the tails of a response variable with highly skewed distribution. The aim of this study is to apply quantile regression to analyze urine iodine data and related factors in a Tehranian population. Data was collected in a cross-sectional study، in which 639 subjects، aged 19 years and over، were enrolled through randomized cluster sampling in Tehran between 2008-9. Due to the high skewness of 24hr urinary iodine concentrations (UIC24) and to evaluate its extreme points، two linear quantile regression models were fitted. In model I، UIC24 was regressed on iodine content of salt and daily salt intake. These variables were replaced by iodine intake in model II، both models were adjusted by age. Model coefficients were estimated using the linear programming method and simplex algorithm. Significancy of the variables were evaluated by the bootstrap method. The Akaike information criterion (AIC) was used to assess the fitting of the models. All analyses were performed using R software version 2. 12. 2. Model I showed an increase in coefficients of iodine content of salt، daily salt intake، but a decrease in age coefficient in the length of the urinary iodine concentration percentiles. Model II showed similar results، but better fit (smaller AIC) in percentiles lower than median. Compared to ordinary regression، quantile regression models showed better fit، and a more complete picture and are recommended for modeling all parts of UIC24.

Regression is one of the most important statistical tools in data analysis and study of the relationship between predictive variables and the response variable. in most issues, regression models and decision tress only can show the main effects of predictor variables on the response and considering interactions between variables does not exceed of two way and ultimately three-way, due to complexity of such interactions. To consider such interactions in the regression models, instead of individual variables in the model, we can construct a combination of them and use this combination as a new independent variable into the model Logic regression is a generalized regression and classification method that in this model, predictive variables are Boolean combinations that are made of the original binary variables. Annealing algorithm is used to find such combinations and their coefficients. randomization test or “null model test” is an overall test for signal in the data.also, cross-validation test can be used to determine the size of the logic tree model with the best predictive capability. As an example, we applied Logic Regression to predict diabetes in TLGS study.