-xtdpdml-, -xtabond- or -cmp-?

[Connie Gao]

Hi All:

I want to investigate the causality relationship between work and health. My model is like this:

health (i=work(it-1)+xb+e (1)
work(it)=work(it-1)+xb+u (2)

Health is continuous variable, work is categorical variable. It is a typical cross-lagged panel model. I know -xtdpdml-, -xtabond- and -cmp- can help me with this model. But I don't know which is more suitable.

If use -xtabond-, then I will use one lagged or twice lagged work value with first difference (FD) method, at the cost of losing two waves due to the lagged values.
If use-xtdpdml-, with fixed effect, it seems lagged work variable is still related with error term as lagged work is related with average error.
If use-cmp-, it seems fixed effect or first difference is not allowed. I am not sure whether Mundlak random effect is possible in -cmp-.

I would appreciate if you can give some suggestions on this.

Thank you,

Connie

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions.

I suspect you could even estimate this with xtivreg or xtivreg2. I suspect xtdpdml is correctly specified to give you consistent estimates but I'm not sure. One nice thing about xtdpdml is that it will give you the SEM code for the estimation so you can see exactly what it is doing and can manipulate the SEM code yourself.

To confuse you even more, I would suggest you look at xthybrid and Mundalk estimators. If cmp does random effects, I don't see why you couldn't code it to do a Mundlak or hybrid estimator.

[Richard Williams]

I don't really understand your concern with using xtdpdml. To read more about it, you can see the support page at

https://www3.nd.edu/~rwilliam/dynamic/index.html

Also this article does a nice comparison of different approaches:

https://osf.io/4h8nu/