binary choice model after collapse

[Øyvind Snilsberg]

is it possible to use binary choice models to predict pr(y|x) after collapsing the data by group?

* Example generated by -dataex-. For more info, type help dataex
clear
input float(group individual y x)
1  1 1 8
1  2 1 9
1  3 0 7
1  4 1 7
1  5 1 8
1  6 1 5
1  7 0 4
1  8 0 5
1  9 1 8
1 10 1 9
2 11 0 7
2 12 0 5
2 13 1 8
2 14 0 3
2 15 0 4
2 16 0 5
2 17 1 4
2 18 1 9
2 19 0 4
2 20 0 5
end

[Rich Goldstein]

I don't fully understand, but I think you want a grouped version of logit (or probit); Stata used to have specific programs for this (e.g., -blogit) but now recommends glm - here is some info from one of the help files (help blogit): "Logistic and probit models for grouped data can be fit via maximum likelihood estimation using glm if pos_var is specified as the dependent variable and the family(binomial pop_var) option is specified, where pos_var and pop_var are defined as below. For a logistic model, specify the link(logit) option. For a probit model, specify the link(probit) option."

[Leonardo Guizzetti]

If you collapse by group, what then happens to x?

If you don't collapse, and you really just have one observed choice per person (given in y), and a is some continuous covariate, this reduces to simple logistic regression.

[Øyvind Snilsberg]

thank you Rich Goldstein for pointing me in the right direction. thank you for your input Leonardo Guizzetti, I did a poor job describing the data - there are multiple choices per person.