semanticfa
An R package that implements semantic factor analysis without respondent data. It assesses the structure of psychometric scales by performing factor analysis on the similarity matrix of scale-item embedding vectors.
Timeline (last 90 days)
semanticfa is an R package that implements semantic factor analysis: it converts scale items into semantic embeddings and performs factor analysis directly on the semantic similarity matrix among items, without requiring respondent response data.
In the demonstration, the factor retention criterion wavered between five and six factors, and under a preset tie-breaking rule five factors were ultimately retained; three factors mainly correspond to a single dimension, and two factors mix most of the extraversion items and five agreeableness items.
Under the five-factor setting, the 25-item short form generated by semanticfa (five items per dimension) outperforms all 200 random five-item combinations.
The limitations of semanticfa include: the choice of embedding model affects the results, the handling of reverse-scored items is not yet ideal, and factor retention decisions may depend on additional rules.
semanticfa is suitable for quickly screening items in the early stages of scale development, constructing candidate short forms, or assessing whether the semantic structure of translated versions is preserved, but it cannot replace confirmatory analysis based on real responses.
All Facts (5)
semanticfa is an R package that implements semantic factor analysis: it converts scale items into semantic embeddings and performs factor analysis directly on the semantic similarity matrix among items, without requiring respondent response data.
50%UnverifiedIn the demonstration, the factor retention criterion wavered between five and six factors, and under a preset tie-breaking rule five factors were ultimately retained; three factors mainly correspond to a single dimension, and two factors mix most of the extraversion items and five agreeableness items.
50%UnverifiedUnder the five-factor setting, the 25-item short form generated by semanticfa (five items per dimension) outperforms all 200 random five-item combinations.
50%UnverifiedThe limitations of semanticfa include: the choice of embedding model affects the results, the handling of reverse-scored items is not yet ideal, and factor retention decisions may depend on additional rules.
50%Unverifiedsemanticfa is suitable for quickly screening items in the early stages of scale development, constructing candidate short forms, or assessing whether the semantic structure of translated versions is preserved, but it cannot replace confirmatory analysis based on real responses.
50%