Category switching

[Rohit Sonika]

Dear Stata-listers,

I have a panel-data setting in which a category variable has the following dummy groups: 00, 01, 10, and 11.

I am trying to create a table that is able to show the frequency of switches between categories, i.e. from 00 to 01, 10 and 11; from 01 to 00, 10 and 11, etc. So, row and column headers are the same category groups and the table is populated by frequency stats.

Any help on how I can do this in Stata would be great. I am using Stata 13.1.

Rohit

[Joseph Chrestien]

Hi Rohit,

nice problem. I would advise you to set the table to long first (see -help reshape- for that).

When/if your data are long, perform the following :

sort id time
rename category current_category
gen next_category = current_category[_n+1]

This should get you this configuration :

ID	current_category	next_category	time
ab	00	01	1
ab	01	02	2
cd	11	11	1
cd	11	10	2

Then you can do a tabulate current_category next_category, missing
Cheers, I hope this helps.
Joseph


EDIT : I'm afraid I've mistaken : beware that when you do gen next_category = current_category[_n+1] the last observation of each ID will be defined from the next category of the next ID. So perhaps you should try something like egen next_category = current_category[_n+1], by(ID)

[Joseph Chrestien]

PS. By the way, has anybody heard of Sankey diagrams in Stata ? Nick Cox : do you think they will be featured someday ?

[Rohit Sonika]

Thanks, Joseph! Will give it a shot and see how this performs.

[Rich Goldstein]

for a more general answer, see the help for xttab (which will also give you the help for xttrans)

[Rohit Sonika]

Thanks, Rich! xttrans seems be exactly what I wanted. Joseph's solution is intuitive too. Thanks, both!

[Nick Cox]

Austin Nichols gave a tremendous talk on transition probabilities etc. at the last Boston meeting. See http://www.stata.com/meeting/boston1...14_nichols.pdf

#3 Sankey diagrams: All I can say for sure is that I have no plans to write such a routine. I find such diagrams impressive rather than directly informative.

I can't (fore)tell what anyone else might write. It doesn't seem like the kind of thing that Stata developers much want to implement, but that's just my wild guess.