Psychometric analysis using Stata
In this one-day Workshop (held over 2 consecutive half-days May 27th -28th 2008), the instructor (DrJon Heron) presented a guided trek through the sharpest and most effective means of preparing your data for psychometric analysis, whether it be in MPlus, Stata, R or any other statistical package.
Part 2 - ALSPAC data and the data-pipeline
Using ALSPAC data from (a) the moods and feelings questionnaire, and (b) the EAS temperament survey, Jon Heron described how one would implement a number of the steps of what we refer to as the data-pipeline, in Stata.
Firstly the importance of adequately preparing one's dataset was stressed, by renaming variables for greater clarity, creating additional data by deriving binaries from categorical data for greater modelling flexibility, and using summarize/codebook to check for incorrect values.
Polychoric correlations (for non-continuous data) were introduced, along with a comparison with traditional Pearson's product moment correlations followed by a discussion of the inadequacy of the latter when assumptions do not hold. This lead to an examination of polychoric-PCA, again with it's more familiar counterpart PCA.
Finally, non-parametric scaling models were covered, with Loevinger's H and the Mokken Scale Procedure which can both be implemented within Stata.