One of the parts of the output of a choice model is the distribution of the Respondent Coefficients (also know as Respondent Utilities). You may want to save this information into your data set for further analysis. This article describes how to take a Latent Class Analysis, Hierarchical Bayes or Ensemble Choice Model output:
And use it to create numeric variables containing respondent-level utilities (also known as individual-level coefficients) for each attribute level:
If the choice model contains a "None of these" alternative, the utility of the "None of these" parameter will need to be adjusted in order to take into account the changes made to the utilities of the other attributes. For example, if there are 3 attributes in the model (not including the Alternative attribute), and their utilities need to be shifted by +0.3, -0.5, +0.8 respectively, then the "None of these" utility will need to be shifted by +0.6 (= 0.3 - 0.5 + 0.8), in addition to any shifting done as part of the Alternative attribute.
Requirements
- A document containing a Latent Class Analysis or Hierarchical Bayes choice model or Ensemble Choice Model output created in Displayr.
Method
1. Select the Choice Model output in your document.
2. From the object inspector, under Inputs > SAVE VARIABLE(S), click the option for the type of utilities with your preferred scaling. (If using an older model without this option, you can access these under Anything > Advanced Analysis > Choice Modeling > Save Variable(s)).
- Individual-level coefficients - the respondent-level utilities unchanged.
- Utilities (Mean 0) - individual-level coefficients shifted so that for each individual and attribute the mean utility across the levels is zero.
- Utilities (Mean 0, Max Range 100) - individual-level coefficients scaled and shifted so that, for each individual and attribute, the lowest utility of any level is zero and the greatest utility of any level is 100.
- Utilities (Mean 0, Mean Range 100) - individual-level coefficients shifted so that, for each individual and attribute, the mean utility across the levels is zero. Utilities are then all multiplied by a scaling factor per individual, so that the average range of utilities per individual across the levels of each attribute is 100.
- Utilities (Min 0) - individual-level coefficients shifted so that, for each individual and attribute, the lowest utility of any level is zero.
- Utilities (Min 0, Max Range 100) - individual-level coefficients shifted so that, for each individual and attribute, the mean utility across the levels is zero. Utilities are then all multiplied by a scaling factor per individual, so that the maximum range of utilities per individual across the levels of any attribute is 100.
- Utilities (Min 0, Mean Range 100) - individual-level coefficients shifted so that, for each individual and attribute, the lowest utility of any level is zero. Utilities are then all multiplied by a scaling factor per individual, so that the average range of utilities per individual across the levels of each attribute is 100.
Individual-level coefficients are not able to be saved to variables for models with simulated data and models created using a CHO data file where respondent IDs were not specified.
A new variable set is added to the Data Sets tree containing the utilities for each attribute level as shown in the Raw Data table below:
References
McLean, M. W. (2018, July 24). How to Use Hierarchical Bayes for Choice Modeling in Displayr [Blog post]. Accessed from https://www.displayr.com/how-to-hierarchical-bayes-choice-model-displayr/.
Next
How to Create a Choice Model Utilities Plot
How to Create a Choice Model Simulator
How to Create a Choice Model Optimizer