How to Remove Random Choosers from a Choice-Based Conjoint Model

This article describes how to use the root likelihood (RLH) to identify respondents who appear to have randomly chosen options when answering a choice-based conjoint questionnaire. It then describes how to remove these respondents from further analyses.

Requirements

A document containing a choice model.

Method

Fit a Hierarchical Bayes model. See How to do the Statistical Analysis of Choice-Based Conjoint Data. Preferably, leave out the Alternative attribute by unchecking Model > Alternative-specific constants.
Save the RLH as a variable by pressing Data > Save Variable(s) > RLH (Root Likelihood) in the object inspector . Note that if the Hierarchical Bayes model has Questions left out for cross-validation, this step will create multiple RLH variables. Both of these variables may be used in this exercise, but it is recommended that you turn off cross-validation in this process.
Examine the distribution of the RLH statistic
- Select the variable created in the previous step.
- Create a histogram (Visualization > Distributions > Histogram) and click and drag to draw the visualization.
Duplicate the Hierarchical Bayes model in step 1. It is usually best to select the entire page in the Reports tree, right-click, and press Duplicate.
In the object inspector , set Respondent Data > Data Source to Simulate choices from priors and Simulated sample size to the sample size. Provided that no priors have been set (as may have occurred if using an efficient or partial profiles design), this will simulate data under the assumption that people are randomly choosing alternatives.
Plot the distribution of the RLH for the simulated data:
- Insert a calculation
- Enter the code of
```
hist(choice.model.2$rlh)
```
- Replace choice.model.2 with the name of the choice model that contains the simulated choices.
Create a calculation with the following code:
```
random.rlh.cutoff = quantile(choice.model.2$rlh, .95)
```
replacing choice.model.2 with the name of the choice model and .95 with some other percentile if you desire (0.95 will set the cutoff to the value that 95% of simulated respondents are below).
Create a filter variable as follows:
- Hover over a variable in Data Sources, and press '+' > Custom Code > R Numeric
- Activate the cursor in the Code Editor , click on the RLH variable, and type in > random.rlh.cutoff (e.g., if your model is called choice.model, then the code will appear as:
```
`RLH from choice.model` > random.rlh.cutoff
```
- Label: Non-Random Choosers
- Name: nonRandom
- Check Usable as a filter in the object inspector .
Duplicate the initial model, and apply the filter.