Marketing Research的课件- marketing_research_q_a_assignment.ppt

Marketing Research Mark Vroegrijk MSc. QA: Assignment Some common issues… 1. Factor Analysis ● What to do when reliability of summated scales is low? 2. Logistic Regression ● How to interpret model parameters (variable effects)? ● How to use both an estimation and validation sample? 3. Cluster Analysis ● How to justify the choice for a particular clustering method? ● How to combine hierarchical and non-hierarchical methods? 4+5. ANOVA Post-Hoc ● Why do I need both ANOVA and post-hoc analyses? 1. Factor Analysis When choosing to represent factors by summated scales… …reliability (i.e. Cronbachs alphas) of some factor scales may fall below acceptable levels (i.e. 0.6)! Not uncommon, often happens when… …correlations between items in the same factor are quite weak (but still high relative to those between variables in different factors) …factor is composed of only a limited number of variables Can be a strong motivation to switch representation method! both surrogate variables and factor scores do not require reliability checks, as they either dont lump multiple variables together (summated scales), or represent the factors exactly as they are (factor scores) 2. Logistic Regression What is the direction of independent variables effects? ● Interpret Exp(β)s → when independent variable increases by 1 unit, the odds ratio will be multiplied by Exp(β): Odds ratio: (= ) (= ) ● So, if Exp(β) 1, odds ratio becomes larger… → positive effect! (probability on event [value 1 on DV] increases relative to probability on no event [value 0 on DV]) ● Similarly, if Exp(β) 1, odds ratio becomes smaller… → negative effect! (probability on event [value 1 on DV] decreases relative to probability on no event [value 0 on DV]) 2. Logistic Regression (contd) What is the size of the independent variables effects? 1. Absolu