Performs the Hausman-McFadden specification test to check the Independence of Irrelevant Alternatives (IIA) assumption for logit models.
Usage
fishset_iia_test(
project,
model_name,
fit_name = NULL,
omitted_zones = NULL,
robust = FALSE,
...
)Arguments
- project
Character string. Name of the project.
- model_name
Character string. Name of the specific model design used.
- fit_name
Character string. Name of the full model fit object previously saved in the project database (created by
fishset_fit).- omitted_zones
Character vector (Optional). The names of the zones (alternatives) to exclude from the restricted model. These must match the zone labels found in the original data. If NULL (default), a random zone is selected.
- robust
Logical. Default FALSE. If TRUE, uses numerically stable utility values, mirroring the fit in
fishset_fit.- ...
Additional arguments passed to the optimization of the restricted model (e.g.,
control,start_values).
Value
A list object of class "fishset_iia" containing:
- statistic
The Hausman chi-squared test statistic.
- p_value
The p-value of the test.
- df
Degrees of freedom (number of coefficients compared).
- full_coefs
Coefficients from the full model (subsetted for comparison).
- restricted_coefs
Coefficients from the restricted model.
- description
Text interpretation of the result.
Details
The test compares the estimates from a full model (fitted on all alternatives) against a restricted model (fitted on a subset of alternatives). If the IIA assumption holds, the coefficients common to both models should not be systematically different.
A significant p-value (typically < 0.05) indicates that the IIA assumption has been violated, suggesting that the logit model may be misspecified (e.g., unobserved correlation between alternatives). Note: Alternative-Specific Constants (ASCs) are excluded from the statistical comparison to prevent reference-level shift bias.