md.quicklogit: Estimate MaxDiff utilities quickly with an aggregate logit...

In cnchapman/choicetools: Tools for Choice Modeling, Conjoint Analysis, and MaxDiff analysis of Best-Worst Surveys

Estimate MaxDiff utilities quickly with an aggregate logit model

Description

md.quicklogit is intended to give a quick check to see whether your data are structure correctly and have reasonable estimates, before investing time to run a hierarchical Bayes model (md.hb()). A common error is to have the values reversed for the "best" and "worst" observations. That will appear in the results with obviously preferred items showing low preference, while poor items show high preference.

The fix in that case is to relabel those data points such that the "best" items' value is greater than the value for the worst items (the exact values don't matter). For example, if your data accidentally code best as 1, and worst as 2, you could replace all of the best observations with a value of 3. Then run md.quicklogit again.

Note that md.quicklogit drops 1 item for model identification, and thus reports K-1 estimates. For example, if you have 15 items you will see 14 estimated parameters in a summary of the return object.

Usage

md.quicklogit(md.define, preadapt.only = TRUE)

Arguments

md.define	A study object as documented for parse.md.qualtrics()
preadapt.only	An experimental parameter for adaptive models, to be documented in the future. For now, leave this as TRUE.

Value

A model object from mlogit::mlogit() with parameter estimates for K-1 levels of your MaxDiff items. Use md.plot.logit() to plot the results.

See Also

[md.hb()] for the recommended hierarchical Bayes estimation.

cnchapman/choicetools documentation built on May 28, 2023, 9:14 a.m.