Identifying patterns in dichotomous variables

[Miguel Costa]

Hi,

I have a dataset like the example below (except that I have 14 variables achado*_mlc and not just 3, and 101 obserb) in which 'olho' is the identifying variable .

What I am struggling to do is to find a way to discover the combination of any of these 14 'achado*_mlc' variables - all dichotomous, that better predicts 'chave' having as 'PCV' as answer (chave is also dichotomous). I would really appreciate some suggestions and how to implement them on Stata. I am using Stata 13.0 for Win 64-bit.

Thank you in advance for your help.
Best
Miguel

dataset example:

olho	achado1_mlc	achado3_mlc	achado5_mlc	chave
1	1	1	1	PCV
2	0	0	0	No PCV
3	0	0	0	No PCV
4	1	1	1	PCV
5	1	1	1	PCV
6	1	1	1	PCV
7	0	0	0	No PCV
8	0	0	0	No PCV
9	0	0	0	No PCV
10	1	1	0	PCV
11	0	0	0	No PCV
12	1	0	1	PCV
13	1	1	0	No PCV
14	1	1	1	PCV
15	0	0	0	No PCV
16	1	0	0	PCV
17	0	0	0	No PCV
18	0	0	0	No PCV
19	0	0	0	No PCV
20	0	0	1	PCV

Sorry for the very poor formatting. I tried to fix it. unsuccessfully

[Carlo Lazzaro]

Miguel:
probably the first step you may want to take is taking a look at -egen- and see how to combine all the -achado*- in a single variable (let's call it -Combine-):

egen Combine=rowtotal( achado*)

[Svend Juul]

Carlo's suggestion is a good start. Assuming that chave is a binary 0/1 variable with value labels, the groups command is useful to look patterns. Find it on SSC:

ssc install groups
groups achado* chave

You could also:

logistic chave achado*

[Miguel Costa]

Thank you for your very useful suggestions!

Kind regards
Miguel