Fixed Effects Cohort analysis

[Guest]

(shouldnt have called it fixed effects in the title. Sorry. My mistake.)

I am using 30 waves of panel data to conduct a fixed-effects analysis.

I have a linear dep. variable (satisfaction) and several independent var. (economic mobility 0/1, health, age.). I want to test whether the influence of economic mobility on satisfaction is different depending on the cohort group.

I am aware of the APC problem. So far I have used period dummies (years of highest satisfaction as base) and age (cohort implicit).

Could anyone guide me how to procede further? As a FE model wont calculate the effect of cohort, should I use a RE model? Or could I possibly illustrate it as a graph? How can I see the cohort effect?

I have seen papers using ML or FGLS. But maybe there is a way around it.

[Marko Ledic]

This might be of interest to you

Thomas Plümper and Vera E. Troeger (2007). Efficient Estimation of Time-Invariant and Rarely Changing Variables in Finite Sample Panel Analyses with Unit Fixed Effects, Political Analysis, 15: 124-139.

Boyce, C. J. (2010). Understanding fixed effects in human well-being. Journal of Economic Psychology, 31, 1-16.

HTH
Marko

[Stephen Jenkins]

Regarding the first article mentioned by Marko and its methods, any user would be wise to also read the critiques by Greene as well as Breusch, Ward, Ngyen and Kompass (also published in Political Analysis 19 (2), Spring 2011, Symposium on Fixed-Effects Vector Decomposition ). There is also the authors' robust rejoinder to look at too in the same issue ... followed by further comments ...

[Guest]

Reading the papers (thank you!) and in particular the critique thereof I am coming to a point where I have to say that the fevd both exceeds my current statistical skills (though with time and effort I would surely be able to overcome that) as well as seems to be quiet problematic (as in regard to Greene et al. and Breusch et al.).

May I ask if there is any other way to test my assumption?

As I couldn't interpret coefficients of time-constant variables in a hybrid model, would it be alright to interpret an interaction effect between cohort and my dichotome economic mobility variable ? Then plotting this with a conditional effect plot.

I have been trying the following (I am mostly interested in the effect of downward):

xtreg satis upward health c.age##c.age##c.age  i.syear c.downward##i.cohgr, re vce (cluster pid)
margins, at (downward = (0/1) cohgr = (1/8))
marginsplot

I am unsure whether this approach is correct, though I do get a good looking plot.(And I am unsure whether I should demean health and probably age?)

[Guest]

Can someone help me with this? I have just changed the code a little bit (removed the c before "downward" in the interaction term), but otherwise I am still stuck whether this is a valid approach.

[Carlos Becerra]

I'm by no means an expert but I am working with the APC package with command apc_cglim for a repeated survey analysis and it seems to work very well. It is described by Yang and Land:

Yang, Y., Fu, W., and Land, K. 2004. A Methodological Comparison of
Age-Period-Cohort Models: The Intrinsic Estimator and Conventional
Generalized Linear Models. Sociological Methodology 34(1), 75-110.

Hope this helps.

[Steve Samuels]

Please that long-standing Stata etiquette is to register with full real names, as all the other posters to this thread have done. You can re-register via the CONTACT US button at the bottom of the page.