| Message/Author |
|
|
| Paul Spin posted on Tuesday, May 01, 2018 - 11:46 am
|
|
|
I have a question about interpreting the MODEL CONSTRAINT output associated with the code
VARIABLE:
NAMES ARE
id y1 y2 x1 x2 x3 w1 w2 p1 p2 n ;
USEVARIABLES ARE
y1 y2 x1 x2 x3 w1 w2 p1 p2 n ;
COUNT is y1 y2 ;
CLASSES = C(2);
Training = w1 w2 (bch);
ANALYSIS: TYPE = MIXTURE; STARTS = 0; ESTIMATOR = MLR;
MODEL:
%overall%
y1 y2 on x1 x2 x3
%c#1%
[y1] (y11);
[y2] (y12);
%c#2%
[y1] (y21);
[y2] (y22);
MODEL CONSTRAINT:
NEW(d_22_21);
d_22_21=y22-y21;
d_22_21 gives the difference between y22 and y21 and the p-value of the test that the difference is zero. My question is: does the test take account for non-independence (or clustering) of outcomes (y1, y2) at the individual-level? If not, how might I incorporate this? |
|
|
|
|
| Yes, it takes into account the non-independence of y1 and y2. This is because the var-covs of the parameter estimates are obtained from the bivariate model for both variables. |
|
|
| Paul Spin posted on Wednesday, May 02, 2018 - 11:01 am
|
|
|
| Thank you. I am thinking of the "arbitrary secondary model" as similar to a Poisson random effects regression (random intercepts at the individual level), but I am not sure if this is correct. |
|
|
|
|
I don't know what "the "arbitrary secondary model" is.
You can also add
f BY y1 y2; f@1; [f@0];
to get a residual covariance between the 2 count outcomes. |
|
|
| Paul Spin posted on Thursday, May 03, 2018 - 7:50 am
|
|
|
| Is there a way to stack y1 and y2 into y and identify a grouping variable g for the y2 observations? I am just wondering about a more direct method of estimating a random effects Poisson regression outside of the Multilevel add-on. |
|
|
|
|
| A 2-group analysis assumes that the observations in the 2 groups are independent so that wouldn't work. |
|
|
| Back to top |
