EstimateEffect

EstimateEffect is a function in the R package stm used to quantify how document metadata covariates influence topic distributions in a structural topic model. It enables researchers to assess how factors such as author, time, or demographics relate to the prevalence of topics across documents, and, optionally, how covariates relate to the content of topics.

Usage and input: After fitting a stm model (for example with the stm function), estimateEffect takes a formula of the form 1:K ~ covariates, the fitted model object, and the document metadata (or data frame) containing the covariates. The 1:K notation specifies the topics for which effects are estimated. The function uses the model’s posterior distribution to estimate covariate effects for each topic and provides standard errors and uncertainty estimates. It can produce outputs suitable for plotting and reporting, including point estimates, confidence intervals, and test statistics. The uncertainty argument controls how uncertainty is quantified, typically via the model’s posterior distribution or simulated draws.

Output and interpretation: The result is an object with topic-specific effect estimates across covariate values. Users can summarize these effects and visualize them with built-in plotting methods, such as plots that display the relationship between a covariate and topic prevalence. Interpretations focus on associations between covariates and topic proportions, conditional on the model specification; they are not causal without stronger assumptions.

Notes: EstimateEffect is central for exploring how metadata shapes topic structure in stm analyses, complementing the model fit and content analyses. It is commonly used in conjunction with plotting and summary tools to communicate how topics vary with covariates across the dataset.