|
|
Simulation of overdispersion in LGM |
|
Message/Author |
|
|
I am MC-simulating overdispersed count data (nb) generated from a quadratic growth curve model of an outcome with 5 timepoints. The specification of the model-implied outcome variances will result in dispersion-parameter values that are equal to the specified variances. Given the restriction of the model-implied outcome intercepts to zero and the NegBin2 Variance-Function (mu_i+alpha*mu_i^2) it not obvious to me why this has to happen. Example: Specified: i BY y1-y5@1; s BY y1@0 y2@1 y3@2 y4@3 y5@4; q BY y1@0 y2@1 y3@4 y4@9 y5@16; [y1-y5*0]; !!!!!!!!!! y1*3.000; !!!!!!!!!! y2*5.000; !!!!!!!!!! y3*5.000; !!!!!!!!!! y4*3.000; !!!!!!!!!! y5*3.000; !!!!!!!!!! [i*0.5 s*0.25 q*-0.2]; i*1.0; s*0.2; q*0.01; i WITH s*-0.3; i WITH q*-0.05; s WITH q*0; Results in: Dispersion Y1 3.000 Y2 5.000 Y3 5.000 Y4 3.000 Y5 3.000 Thank you! |
|
|
With a count variable, the name of the count variable refers to the dispersion parameter, alpha, not the variance. So when you say y1*3 you are giving the alpha parameter value. |
|
Back to top |
|
|