Marketing Research的课件- Logistic_s.ppt

Logistic Regression Session: March 8 and 12; 2010 1. Objectives/Purpose Predict likelihood that an event will occur (yes/no) Assess variables that affect event occurrence Direction of influence Magnitude/importance 1. Objectives/Purpose Examples: What is the probability that a person will respond to a Neckerman direct mailing, and how can Neckerman adjust its mail content to increase this probability? Does improved waiting time at the checkout increase C1000’s store patronage? What distinguishes heavy from light ‘donators’ to Parents plan charity? 2. Steps Design Assumptions Model Estimation and Fit Interpreting Results Validation LR: Design Dependent variable: Dichotomous (0/1): mutually exclusive/exhaustive Reconstructed from metric variable (polar extremes approach) Independent variables: Metric or dummy variables Based on theory, intuition? Example: Advertising through in-store screens Do consumers notice ad messages on TV screens in store? How does this depend on message characteristics? Which consumers primarily notice the message? Example: In-store TV ads (ctd) Dependent: seen/not seen (recall?) Independent: Consumer: store visit frequency (klantVOD), education, Home Tv-ad seen, competitive store visited recently? Message: sound, length Research Design (ctd) Sample: # observations/#independent variables Group sizes Analysis versus Holdout sample (proportionally stratified subsamples?) Assumptions of LR Two groups for outcome variable (Extension: MNL) Robust to deviations from multivariate normality and equal VC matrices Model form: S-shaped Logistic Regression Model LR estimation: Maximum Likelihood Given: ‘Observed’ events (e.g. questionnaire: seen=1/not seen=0) yi Measures for independent variables Xki Find parameters b0, b1, … that provide best ‘match’: maximize Model Fit Likelihood value: low enough –2LL (close to 0? ) Pseudo R2 Nagelkerke R2 (close to 1?) Classification Table Hit rate ( chance?) Hosmer-Lemeshow test (no significant differenc