Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Fixed/Random Effects models on NOT panel data


From   Gerardo Infante <chitastiera@yahoo.it>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   st: Fixed/Random Effects models on NOT panel data
Date   Wed, 23 Nov 2011 11:30:54 +0000 (GMT)

Hi all,

I have data coming from an experiment on intertemporal decision making. In particular, I have "n" individuals who made decisions in "k" sequences. Each sequence is made up of "T" periods. For example, participant A played 5 sequences, each one made up of 10 periods (i.e. after period 10, a new sequence started, until the last sequence is completed).

I have created my dataset by stacking all observations together using, among my variables, "subject", "sequence" and "period". Of course, I cannot treat this dataset as Panel Data, unless I create an "index" variable, by combining "sequence" and "period" (but this latter solution does not work for my regression analysis). In regression analysis I have used this approach: regress using Pooled OLS, Random Effects and Fixed Effects (both by using the xtreg command). Also, I have used the Breusch-Pagan LM test to discriminate between RE and P.OLS, the F-Test to discriminate between FE and P.OLS and the Hausman Test to discriminate between FE and RE. Moreover, to try and take into account the variability and correlation of residuals within and between subjects I have used "cluster s.e.", in particular clustering around subjects (vce(cluster subject)).

Is this approach correct, given that I am not using Panel Data? Also, in the past I have received a comment stating that it is ALWAYS better (in these cases) to use Random Effects instead of Fixed Effect (apparently, regardless of what suggested by the Hausman test). I personally think that the assumptions underlying RE models are not so easy to meet, hence one should not rely on these model without comparing it to the FE counterpart. Do you have any inputs on this?

Thank you for any comments and suggestions.

Best wishes,

Gerardo

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index