| Abstract: | Development criteria observed as time series on nominal variables pose a problem for research design. Markov chains are attractive in the first instance because they can handle such multivariate multinomial data sets, with results that are described and demonstrated on Career Development Study data. The power of the formal probability model is purchased at the expense of assumptions that may be unrealistic. It is argued and demonstrated that where the Markov chain theory for the data breaks down is precisely where the most interesting part of the message contained in the data resides. The Markov chain is shown to be a better null hypothesis against which to array the psychometric or other prediction model than is the usually employed random walk. A new prediction system that combines information from the Markov chain with the psychometric predictors is described and operated. |
