Calibration Models in Statistic

Abstract

This thesis considered statistical calibration models. It focused on Eiesinhart s calibration model for two tests X & Y where X is an exact but expensive and slow test and Y is less expensive but quick and cheap test. The objective was to investigate through a simulation experiment the effect of the degree of linear dependency between X & Y as well as sample size on confidence interval estimation of the forecasted value of X .It is shown that changes on these factors have no significant effect on the degree of confidence. Also this thesis hosts a comparison between logistic regression model and calibration linear model . Both models were applied on random sample of 120 people , 100 are infected with blood cancer and 20 are fit . And we have 3 independent variables , age , pcv , mch . When applying both models we discovered that the values of standard errors in calibration regression model are less than the value of standard errors in logistic regression model , meaning that calibration regression method was better . Some other results were reached , like when applying logistic all variables mentioned above have significant influence on cancer infection , we also found that pcv variable is the most influential in cancer infection , followed by the rest age and msh.