6SIGMA工具6_Logistic Regression.ppt

Breakthrough Management Group Breakthrough Management Group Version 1.0 Logistic Regression Module Objectives Understand the three different types of logistic regression available for the modeling of Attribute Responses: Binary Response Variables Ordinal Response Variables Nominal Response Variables Be able to apply Logistic Regression using Minitab for both categorical and continuous input variables Logistic Regression Logistic or “Logit” Regression investigates the relationship between response variables (Y’s) and one or more predictor variables (X’s) where: Y’s are categorical, not continuous X’s can be either continuous or categorical Logistic Regression in Minitab Binary Logistic: Regresses the independent variables on a binary response (two categories) Ordinal Logistic: Regresses the independent variables on an ordinal response with 3 or more categories Nominal Logistic: Regresses the independent variables on a nominal response with 3 or more categories. The logistic procedures in Minitab can fit up to 9 categorical inputs (factors), and 50 continuous inputs (covariates) Examples of Different Types of Models Why Logistic Regression ? Binomial data violates normality, equal variance assumption ?k = nxp = nxpx(1-p) = ?k x(1-p) Variance changes as the mean changes The relationship between p, the likelihood of “success” and the predictor variables might not be linear Introducing Odds and Odds Ratios Odds = p/(1-p) = pr(event occurs)/pr(event does not occur) = O p = O/(O+1) For example, define an event as : an account is paid in 30 days: If the odds are 6 to 1, this implies: Pr(account paid) = 6/(6+1) = 6/7 Pr (account not paid) = (1/6)/(1/6 + 1) = 1/(1 + 6) = 1/7. Odds of account being paid = 6/1 = 6 Odds of account not being paid = 1/6 If the odds of the 1st shift completing their schedule are 5 to 2, and the odds of the 2nd shift completing their schedule are 10 to 1, Shift (2)’s odds of completion (10) is 4 times larger than Shift(1)’s o