We applied a computationally practical form of probit analysis for multiple response variables to data on early childhood development of four psychiatric disorders: disruptive disorders (DD - attention deficit disorders, oppositional defiant disorder, conduct disorder); adjustment disorders (ADJ); emotional disorders (ED - all anxiety disorders, depression); and other DSM-III-R Axis I disorders (OTHER). In addition to estimating the intercept slope and higher order polynomial terms for each age versus diagnosis regression, we estimated simultaneously the correlation among the four diagnostic categories. We then took into account the correlation found among these four diagnostic categories when testing the hypothesis of no age effect, which would have been ignored in a piecemeal univariate approach. Regression lines for diagnostic prevalence indicate a linear increase for OTHER disorders, and a curvilinear increase for ED. We then used expected frequencies of individual response patterns (that is, the 24 = 16 possible diagnostic combinations) in obtaining more precise estimates of diagnostic comorbidity and its relation to age. We further generalize the Beck and Gibbons model to alternative specification of the random-effects distribution (that is, they assumed multivariate normality), illustrate how one can estimate the random-effects distribution empirically, and study the robustness of parameter estimates to specification of the random-effects distribution.
|Original language||English (US)|
|Number of pages||13|
|Journal||Statistics in Medicine|
|State||Published - Nov 15 1998|
ASJC Scopus subject areas
- Statistics and Probability