Re: st: panel data analysis advice

[David Greenberg <dg4@nyu.edu>]

You could do structural equation modeling, treating observations for
the missing years as latent variables, with coefficients for equations
predicting those latent outcomes fixed at the values of the closest
years for which you do have observations,. David Greenberg, Sociology
Department, New York University

On Mon, Mar 10, 2014 at 5:21 PM, Theophilus Dapel <dapelzg82@gmail.com> wrote:
> Please permit me to also seek for assistance.
>
> I have a balanced but unequally spaced panel dataset:
>
> 1980
>
> 1985
>
> 1992
>
> 1996
>
> 2004 and
>
> 2010.
>
> How do I get around this?
>
> Thanks
> On 10 Mar 2014, at 21:09, Robert Paul <pt162591@yahoo.com> wrote:
>
>> Dear Statalist,
>>
>>
>> I have demographic and treatment information for patients chronic disease (N=60,000). I got permission to link a subset of my data to income data (18.5%). For this subset I have 20 years panel data.
>> The data in long format looks
>> Id     year  income age …
>> 1      1990  100   45
>> 1      1991  110   45
>> 1      1992   125   45
>> 1      1993  132   45
>> .
>> .
>> .
>>
>> My aim is
>> a-  to estimate the effect of demographic, treatment, and being chronic disease patient, on patient’s income; and
>> b-  to evaluate differences in income between patients and the general population (when linked to control population)
>>
>> to address these issues I plan
>>
>> a-  to run a Fixed and Random effects model , to start with  then run  Hausman test …
>>
>> b-  I will also get a control group for my data - (from general population without chronic disease -matched by demographic vars)  --- for this I plan to use Hausman-Taylor that utilizes the vars as instruments and provide  parameter estimate for time-invariant variable (major variable of interest – chronic disease patient or not)
>>
>>
>>
>> Dependent variable – log equivalized income
>> RHS vars – age at end of follow-up,  age^2, age at diagnosis,  treatment type
>>     1. Run  xtreg logincome age age_square age at diagnosis treatment type dummies . .  , fe
>>     2. xtreg logincome age age_square age at diagnosis treatment type dummies . . . .  , re
>>     3. xtreg logincome age age_square age at diagnosis treatment type dummies . . . , re vce(robust)  or
>>     4. xtreg logincome age age_square age at diagnosis treatment type dummies . . . , re vce(cluster id)
>>
>> The aim of using vce or cluster is to produce consistent VCE estimator when the disturbances are not identically distributed over the panels.
>>
>>
>> 5.  ** Hausman Taylor estimation
>>
>> . xthtaylor  logincome age age_square age at diagnosis treatment_type dummies, endog(age treatment type dummies)
>>
>> My question, as I am new to panel data analysis, is if I am doing the right way to address my question.
>> 1.  Do I need to calculate weights because I am using a subset of the population? If yes, how do I do that?
>> 2.  I am not sure – probably using dynamic models would be more appropriate
>> 3.  I need advice on my analysis procedure. This is of critical importance for my project.  I appreciate your valuable comments.
>> Thanks
>>
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>> *   http://www.ats.ucla.edu/stat/stata/
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

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