Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

st: Re: Making Cohorts

From   ajjee <[email protected]>
To   [email protected]
Subject   st: Re: Making Cohorts
Date   Thu, 28 Oct 2010 16:31:35 -0700 (PDT)

Dear Maarten
Thanks again.

I think I was unable to explain my problem. My data is on 3 year surveys (6
waves) for 15 countries covering near 1000 from each country in each wave.
As in each wave of survey, an independent sample is drawn from each country,
it is not true panel data (respondents are not same in each wave) and it is
repeated cross sections over time or pseudo panel data. So its not possible
to apply the normal panel data techniques on this data. My dependent
variable is binary so I'll use logistic regression. For  this I 'm following
the method described in  

Gassner, Katharina, 1998. "An estimation of UK telephone access demand using
Pseudo-Panel data," Utilities Policy, Elsevier, vol. 7(3), pages 143-154,
He described the method as below:

"Suppose that yit is a 0–1 variable indicating whether a household has
access to a telephone or not, and that this indicator variable is a linear
function of explanatory variables. We acknowledge this being an unusual
assumption as in reality it is not possible for a 0–1 indicator to be a
linear function of variables, since the linear function can take any value
whereas the indicator is binary. However, the estimation procedure which we
have adopted involves subsequent aggregation of the individual data. We
removed yit to be linear in the explanatory variables in order for the
aggregation to give a relation between the share of 1’s and the average of
the explanatory variables."

Thats why I wrote a linear function and asked for how to make a cohort data
according to that. 

Thanks and Regards,


Maarten buis wrote:
> --- On Thu, 28/10/10, ajjee <wrote:
>> I want to implement the following strategy on my data.
>> Consider now the basic linear individual effect model
>> yit = ai + bXit + uit    i = 1,..N; t = 1,..T       eq(1)
>> where Xit is a (K x 1) vector of explanatory variables
>> which we assume exogenous to the model, index t and
>> i refer to time and individuals respectively.
>> Assuming, for simplicity, that there is a unique regressor
>> (K = 1), if we aggregate all observations to cohort level,
>> the resulting model can be written as
>> (y-bar)_ct = (a-bar)_ct + b(x-bar)_ct + (u-bar)_ct  c = 1,..C
>> where (x-bar)_ct is the average value of all observed
>> xit’s in cohort c at time t, and analogously for the other
>> variables in the model. The resulting data set is a pseudo
>> panel with repeated observations over T periods and C
>> cohorts.
>> now I want to make cohorts (say for) birth year and then
>> want to aggregate the data as above.  But I have 15 countries
>> also and I want to analyse particular attitudes in countries
>> based on three year surveys(6 waves).
> You earlier stated that you wanted to do a -logit- analysis. 
> This type of averaging does not work in non-linear models, 
> like -logit-. You can do the computations, but the results 
> just don't mean what you think they mean. Instead you should 
> look at -xtlogit- and -xtmelogit-.
> Hope this helps,
> Maarten
> --------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
> --------------------------
> *
> *   For searches and help try:
> *
> *
> *

View this message in context:
Sent from the Statalist mailing list archive at

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index