I have a question concernig the work on pooled data. I want to estimate a MIMIC-Model with pooled data over different time periods. How is it possible to take the pooling bias into account? Do there excist any option in MPlus which allows a multistep residual analysis?
I assume that you have several data sets that you want to analyze together that contain different observations but the same variables. You could use multiple group analysis where you test whether these data sets come from the same population.
There is no multistep residual analysis in Mplus.
C. Lechner posted on Wednesday, April 23, 2014 - 6:23 am
Suppose I am interested in lagged effects of a variable X on a categorical variable Y, moderated by a cluster-level variable Z. I have pooled longitudinal data from participants who took part in 2-4 waves. I want to look at lagged effects between pairs of waves (i.e., 1-2, 2-3, 3-4) in a single analysis to maximize power.
Can I use the DATA WIDETOLONG command in conjunction with TYPE=COMPLEX TWOLEVEL to take into account both the clustering due to repeated measurement and the clustering due to cluster sampling? My idea was to use WIDETOLONG to create lagged effects by pairing x's measured at time 1-3 with y's and z's measured at time 2-4; and to then analyze these data as a multilevel model (three-level analysis) without the repetition-variable:
DATA WIDETOLONG: WIDE = x1-x3 | y2-y4 | z2-z4; LONG = x | y | z ; IDVARIABLE = person; REPETITION = time;
ANALYSIS: TYPE = TWOLEVEL COMPLEX RANDOM; ESTIMATOR = MLR;
MODEL: %WITHIN% S | Y ON X; %BETWEEN % Y S ON Z; Y WITH S;
This does not seem to account for the fact that the cluster-variable Z was measured repeatedly, or does it? Are there any alternatives in Mplus to analyse such data?
I would have time as wide format and TYPE=TWOLEVEL. This is a good way to take into account that z is a repeated measure. See Example 9.12.
C. Lechner posted on Wednesday, April 23, 2014 - 1:53 pm
Dear Linda, thank you. Yes, I thought about such a model. I would, however, like to pool data from four waves and look at lagged effects between any combination of two subsequent time points (1-2, 2-3, 3-4); that is, my analytic model has only two time points that stem from four waves of data collection. For some people, I have three such "slices" consisting of two waves, for most I have two, for others I have only one "slice" of two waves. At any rate, these "slices" are repeated measures nested in persons. I was unsure how to modify the model in example 9.12 accordingly. Any suggestions?