How to Select the Regression Type for Driver Analysis

Introduction

There are lots of potential outcome variables that can be used when conducting a driver analysis,
from five-point rating scales through to utilities from conjoint studies. To conduct a valid driver analysis
we need to select an appropriate regression type or generalized linear model (GLM) which is consistent with our data.

This article describes how to select the Regression Type suitable for your Driver Analysis. 

Requirements

A Driver Analysis output (See: How to Do Driver Analysis).

Method

1. The type of Regression type required will depend on the outcome variable used in your Regression model. To find the Outcome variable select the output and go Inputs > Linear Regression > Outcome. 
2. OPTIONAL: To view all potential outcomes of the dependent variable go to the Data Sets tree, select the Outcome variable and go to the object inspector > Properties > DATA VALUES > Values. Alternatively, you can also drag and drop the variable onto the Page to create a table containing all outputs. 
3. Go to Inputs > Regression Type and select the appropriate model depending on the number of categories of the Outcome variable.  
   

   Outcome variable	Example	Regression Type
   Two categories	1. Yes / Selected 2. No / Not selected	Binary logit (also known as logistic regression)
   Three to 11 ordered categories	1. Hate 2. Dislike 3. Acceptable 4. Like 5. Love	Ordered logit
   12 or more ordered categories	How would you rate your happiness on a scale of 0 to 100	Linear regression
   Net Promoter Score (NPS)	-100: Detractor 0: Passive/Neutral 100: Promoter	Linear regression
   Purchase or usage quantities	Number of cans of coke consumed per week	NBD (or, if you get a weird message, quasi-Poisson regression)
   Utilities from a conjoint or MaxDiff study	That is, a variable that contains the estimated utilities for each respondent	Linear regression
   Probabilities or shares from conjoint and MaxDiff studies	That is, a variable that contains the probabilities for each respondent	Use the utilities instead of the preference shares, and then use Linear regression.

See Also

How to Do Driver Analysis

Webinar 

eBook