Fits a count regression model to topic prevalence data, auto-selecting between Poisson, Negative Binomial, and Zero-Inflated Negative Binomial based on dispersion ratio and zero-inflation diagnostics.
Usage
fit_topic_prevalence_model(
topic_proportions,
metadata,
formula,
model_type = "auto",
zero_inflation_threshold = 0.5,
count_multiplier = 1000,
max_iterations = 200
)Arguments
- topic_proportions
Numeric vector of topic proportions (0-1) for one topic.
- metadata
Data frame of document-level covariates.
- formula
Model formula (character or formula object). Response variable is created internally as
topic_count.- model_type
Model selection strategy:
"auto"(default),"poisson","negbin", or"zeroinfl".- zero_inflation_threshold
Proportion of zeros above which a zero-inflated model is attempted (default: 0.5).
- count_multiplier
Multiplier to convert proportions to pseudo-counts (default: 1000).
- max_iterations
Maximum iterations for model fitting (default: 200).