UPCOMING WEBINAR: DIY Data Cleaning & Prep: Super Fast, Super Good This article describes how to go from a choice model output created in Displayr from an alternative-specific experimental design:
To a Choice Modeling Simulator that shows fixed alternative scenarios:
Requirements
- A Hierarchical Bayes, Latent Class Analysis or Multinomial Logit choice model analysis output created in Displayr (see How to Do the Statistical Analysis of Choice-Based Conjoint Data) which uses an alternative-specific experimental design such as Alternative specific - Random or Alternative specific - Federov (see How to Create an Experimental Design for Choice-Based Conjoint).
Method
In this example, we are using an alternative-specific design that includes 2 brands, one with 3 features and one with the same 3 features, plus an additional one. To create a simulator from a design like this, you will need to create the simulator like normal and configure the controls to hardcode the specific alternative and hide its non-relevant dropdowns, see Technical details below for more detail on how this all works within the modeling.
Create the standard simulator
- Select the choice model output in your Displayr document.
- In the object inspector, click the Simulator button under Data > Simulation.
- Enter the number that corresponds to the number of alternatives in your design. In this example, it is 2 for the 2 alternative brands.
- At the next prompt, select Yes to include the Alternative attribute. This ensures that any utilities associated with the alternative by name are included in the model (i.e. in this case brand power/recognition).
- If your choice model includes a 'None of these' alternative, select Yes to include it in your simulator.
The default simulator setup is below where dropdowns for all attributes are created for each alternative. Now we need to hardcode the specific alternatives and non-relevant dropdowns, see Technical details for an explanation behind this.
Fix the simulator alternatives to be specific alternatives from the design
6. Because only some of the dropdowns are relevant for alternatives in the design, we will have to hardcode alternatives in the simulator (the columns of dropdowns) to be specific alternatives from the design. To do so, remove the top row of the simulator: 1) the Alternative text box on the left and 2) the top row of combo boxes (in this case the ones showing BrandA). You can right-click those outputs select Delete. This will cause the preference shares to go into error, but that will be fixed later.
7. Relabel the Alternative 1 and Alternative 2 text boxes at the very top of the page to the design alternatives. Note, the labels need to be exactly the same as in your design. In this example, this is BrandA for the first column and BrandB for the second column.
8. Select the 2 modified text boxes and copy their names from General > General > Name. In this example, they are text.10 and text.21.
9. Expand the hidden preference share calculation at the bottom left of the page (it will still be in error).
10. Under Data > R Code > Edit R Code, on the first line you will see code to create the simulator scenario which is a list of all alternatives and a list of the attributes selected for each in the simulator. In this example, "Alternative 1", "Alternative 2" and "None of these" are listed and each is equal to a list of attributes referencing the names of the Alternative combo boxes you deleted.
To hardcode these alternatives in the simulator to be your design alternatives in the text boxes, replace those Alternative = combo box names with their respective text box names from step 8 (text.10 and text.21).
The Calculation output will recalculate and the errors should be fixed.
Remove the attributes that don't apply to each alternative
There are two ways you can hardcode non-relevant attributes for each alternative depending on the amount of attributes and how fool-proof you'd like to set things up. You can either:
- Move the non-relevant combo boxes to a hidden separate page and note not to change them. This also requires no R code editing.
- Delete the non-relevant combo boxes and in the R Code delete the references to the non-relevant attributes from each alternative list. This ensures that nobody will accidentally change these settings.
To hide the combo boxes (the quickest method):
11. In the Report pane, hover and click + > Page > (select layout) to create a new page and right-click > Hide to hide it and its contents. You can also edit the title to say something about not using or changing the page.
12. Select all the combo boxes that don't apply to each alternative column.
13. Drag them on top of the new hidden page in the Pages pane to move them there to hide them.
To hardcode the R CODE (the most fool-proof method):
11. Select all the combo boxes that don't apply to each alternative column, right-click, and select Delete. This will make the preference share calculation error, but this will be fixed in the next step.
12. Go to Data > R Code > Edit R Code on your preference share calculation off the bottom left of the page and remove, the non-relevant parameters from each alternative's list. In this case, all the "BrandB" attribute items from the Alternative 1 (BrandA) and all the "BrandA" attribute items from the Alternative 2 (BrandB). In the example code below, there are only "BrandA" attributes for Alternative 1 and only "BrandB" attributes for Alternative 2.
scenario = list('Alternative 1' = list('Alternative' = text.10,'BrandA.Feature1' = cBrandA_Feature1.5,'BrandA.Feature2' = cBrandA_Feature2.5,'BrandA.Feature3' = cBrandA_Feature3.5),
'Alternative 2' = list('Alternative' = text.21,'BrandB.Feature1' = cBrandB_Feature1.6,'BrandB.Feature2' = cBrandB_Feature2.6,'BrandB.Feature3' = cBrandB_Feature3.6,'BrandB.Feature4' = cBrandB_Feature4.6),
'None of these' = list('Alternative' = 'None of these'))
13. Once all references to the deleted combo boxes and attributes have been removed, this calculation will become valid again.
Tidy up your simulator
14. Rename the text boxes in the list of attributes on the left to their generic feature, and drag and drop each alternative's corresponding feature's dropbox in alignment.
15. Any attributes only shown to one alternative can also be better placed.
16. [OPTIONAL]: You can additionally customize all the various components of the simulator by clicking on them to modify. Options for the combo box controls are in the object inspector under Control > Item list, while the appearance of all the outputs can be customized in the Appearance > Appearance section. Click here for an example.
Technical details
By default, a choice model simulator will include a combo box for each attribute by the number of alternatives selected in the script, whereby each column is not fixed to a specific alternative. When working with alternative-specific simulators, however, not all attributes will apply to all alternatives. We can effectively ignore non-applicable attributes for an alternative by setting those attributes to the first level of that attribute. This is because the preference share is calculated as a whole across all attributes in relative terms (not absolute). The utilities for the first level of each attribute set to 0 to be the baseline for the other levels to measure their relative utilities. Thus in hardcoding the first level for non-relevant attributes, those will have 0 as their utility and not affect the preference share calculation. Note, this approach only works with alternative-specific designs, and is not a way to model different designs after fielding the study.
Next
How to Do the Statistical Analysis of Choice-Based Conjoint Data
How to Create an Experimental Design for Choice-Based Conjoint