[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

From |
Michael Ingre <Michael.Ingre@ipm.ki.se> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
st: gllamm model with random slopes in time varying covariates |

Date |
Fri, 18 Jun 2004 17:25:49 +0200 |

Dear Statalist

I would like to regress individual differences in change over time in time varying covariates on individual differences in change over time in of my dependent variable. This calls for a random coefficient model with -gllamm- where random effects (slope in covariates) are regressed on another random effect (slope in the dependent variable).

The reason for trying this model is that I want to show that individual differences in the change over time (slope) in the dependent variable is "explained" by individual differences in the change over time (slope) in my covariates.

I have no problem with defining a model with a random intercept and slope for the dependent variable. And regressing a random effect on another random effect is also quite straightforward.

My problem is the slope for the covariates. I first thought that defining a random effect that multiply the interaction covariate*time would do the trick. But now I'm not sure any more. The model (I'm not sure of any more) is described below:

. gen const = 1

. gen cov1_time = cov1*time

. gen cov2_time = cov2*time

. eq int: const

. eq slope0: time

. eq slope1: cov1_time

. eq slope2: cov2_time

. matrix b = (0,1,1,1\ 0,0,0,0\0,0,0,0\0,0,0,0)

. gllamm y time cov1 cov2 , i(id) link(ident) fam(gauss) nrf(4) ///

. eq(slope0 intercept slope1 slope2) nocorrel bmatrix(b) trace dots adapt

This is complicated stuff for me. I'm now questioning the meaning of my slope1 and slope2. They multiply the random effect with the interaction cov*time. But does this mean that they describe individual differences in change in covariates over time? Or what do they mean?

In the case I am on the right track: Should I include the interaction cov*time as a fixed effect also?

The answer I'm afraid of is that I need to model three parallel growth curve models to answer my question (6 random effects 3 intercepts and 3 slopes).

Please, help me out. I cant think straight anymore (!!!)

Thanks for your time!

Michael

WARNING: Don't try this at home, unless you have a supercomputer or lots of time!

------------------------------------------------

Michael Ingre , PhD student & Research Associate

Department of Psychology, Stockholm University &

National Institute for Psychosocial Medicine IPM

Box 230, 171 77 Stockholm, Sweden

*

* For searches and help try:

* http://www.stata.com/support/faqs/res/findit.html

* http://www.stata.com/support/statalist/faq

* http://www.ats.ucla.edu/stat/stata/

- Prev by Date:
**st: table command** - Next by Date:
**st: gllamm: geqs() and bmatrix() together** - Previous by thread:
**st: table command** - Next by thread:
**st: gllamm: geqs() and bmatrix() together** - Index(es):

© Copyright 1996–2016 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |