I see.
Moving from "Can you do it?" to "Should you do it?" my impression is that clustering will just throw away information that you probably should be using in the modelling.
Nick
[email protected]
Claudio Cruz Cázares
Thank you Nick!
In fact I want to try a partition cluster analysis (kmeans). I agree that the time series in the panel will be collapsed as well but after the cluster results I will paste the resulting cluster as a new variable in the original panel and split the sample base in this results. I hope this will work!
De: Nick Cox <[email protected]>
> What kind of clustering would you like? That depends on your
> definition of the individuals to be clustered. You could summarize
> panels in whatever way you like and cluster those summaries also
> in whatever way you like.
> That's one pathway. But it would collapse the time series
> information in the panels.
>
> But I don't think there's a definition of cluster analysis that is
> automatically obvious for panels and I don't think there's Stata
> software specifically designed for any such decision.
claudio cruz
> I have a panel dataset and I want to create different scenarios to
> evaluate the impact of X in Y using a dynamic model. So, for
> creating the different scenarios I want to create clusters based
> on x1, x2 and x3.
>
> So far, I can't found any information if there is a specific
> cluster analysis for panel data. Does anyone know what could I
> do?
>
> I was thinking in one solution. I'm not sure of doing a pooled
> cluster since some firms may be in t-1 at Cluster_a and in t may
> be at Cluster_b, right? I think it doesn't have any logic. Thus,
> I was thinking in collapsing the data by the mean of x1, x2 and x3
> for each firm and then create the clusters. Is that correct?
>
*
* 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/
