This preview shows pages 1–5. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Categorical Data Analysis  Lei Sun 1 CHL 5210  Statistical Analysis of Qualitative Data Topic: Logistic Regression Outline Single predictor (2 levels, > 2 levels, quantitative). Parameter interpretation. Inference of the parameter. Multiple predictors (next lecture). Categorical Data Analysis  Lei Sun 2 Introduction to logistic regression (logit model). So far: twoway contingency tables. Most studies: several explanatory variables (categorical and continuous). Goal: describe their effects on response variables, including relevant in teractions, smoothed estimates of response probabilities. Most important case: logistic regression (logit model), a linear model for the logit transformation of a binomial parameter. Also: loglinear model (Poisson regression model), a linear model for the log of a Poisson mean. We will study the following. Single predictor (2 levels, > 2 levels, quantitative). Parameter interpretation. Inference of the parameter. Multiple predictors. Interaction. Model building (model fit and selection). Categorical Data Analysis  Lei Sun 3 Logistic regression with one covariate with two levels. Background. Y : binary response, outcomes denoted by 0 and 1. e.g. birth weight, normal (0) or low weight (1). X : binary predictor, two levels denoted by 0 and 1. e.g. smoking status, no (0) or yes (1). We are interested in defining E [ Y ] = P ( Y = 1) = ( X ): reflects its dependence on value of the predictor, X . A linear probability model: P ( Y = 1) = ( X ) = + X Major structural defect: probabilities fall between 0 and 1. But linear functions take values over entire real line. Interpretation of : is the change in for a oneunit increase in X . We want to construct our model so that: Predicted value of P ( Y = 1) = ( X ) is bounded between 0 and 1. The regression coefficient is measured on a meaningful scale. We can construct statistical inferences which approximately model the Y i , i = 1 , ..., n , as Bernoulli random variables. Categorical Data Analysis  Lei Sun 4 Consider logit of as a linear function of X . logit ( ( X )) = log ( X ) 1 ( X ) = + X = ( X ) = exp ( + X ) 1 + exp ( + X ) . If = 1 and = 2, ( X ) = + X . ( X ) = exp ( + X ) 1+ exp ( + X ) . ( X ) = exp ( X ) 1+ exp ( X ) ....
View
Full
Document
 Fall '11
 LeiSun

Click to edit the document details