split sample and run logit models and capture numbers to make a new dataset

[Nishan Lamichhane]

* Example generated by -dataex-. For more info, type help dataex
clear
input double(trans_bef_pregnancy maternalage) byte parity_cat float event
0 37 2 1
0 24 2 0
0 27 2 1
0 35 1 0
0 23 1 0
0 33 1 0
1 35 2 1
0 20 1 0
0 35 3 0
0 28 2 0
0 42 3 0
0 33 1 0
0 28 2 0
0 21 1 0
0 31 1 1
0 27 2 0
0 23 2 0
0 31 1 0
0 40 2 0
0 28 1 0
0 34 2 0
0 28 1 .
0 30 3 0
1 34 1 0
0 24 2 0
0 39 4 0
0 30 4 0
0 28 1 0
0 18 1 0
0 29 2 0
0 30 1 0
1 30 1 1
0 40 2 0
0 43 2 0
0 37 2 0
1 20 1 0
0 29 1 0
0 35 4 0
0 34 2 0
0 25 1 0
0 31 3 0
0 39 2 0
0 37 2 0
0 22 1 1
1 25 1 1
0 23 1 0
0 26 1 0
0 43 1 0
0 29 2 0
1 31 2 0
0 31 2 0
0 27 1 0
0 33 1 0
0 32 1 1
0 32 2 0
0 28 2 0
0 22 1 0
1 35 1 1
0 36 3 0
0 29 2 0
0 27 1 0
1 39 3 .
0 32 1 0
0 33 1 0
0 33 2 0
0 32 1 0
1 40 2 0
0 41 4 0
0 36 3 0
0 42 2 1
0 25 1 0
1 27 1 1
0 26 1 0
0 35 2 0
0 23 1 1
0 29 3 0
0 27 1 0
0 33 2 0
0 27 1 0
0 32 2 0
1 23 1 0
0 29 1 1
0 37 2 0
0 32 4 0
0 31 4 0
0 37 3 0
1 36 2 0
0 24 2 0
0 30 2 0
0 33 1 0
0 38 4 0
0 28 1 1
0 30 3 0
0 25 1 0
1 28 3 1
0 36 2 0
0 29 2 0
1 32 3 0
0 41 1 0
0 28 3 0
end
label values parity_cat parity_cat
label def parity_cat 1 "1", modify
label def parity_cat 2 "2", modify
label def parity_cat 3 "3", modify
label def parity_cat 4 "4 or above", modify

For the data above: i have the following commands.

set seed 2200

// splitting randomly the data into training 50%, validation 25% and test set 25%
splitsample, generate(sample) split(0.5 0.25 0.25)
lab var sample "training, validation or test set"
lab define sample 1 "training set" 2 "validation set" 3 "test set"
lab val sample sample

// let's run a logistic model on the training set
logit event b1.momorigin_cat i.parity_cat trans_bef_pregnancy maternalage b4.abo if n_test>=2 &sample==1 , or

//predicted values for validation and test samples

predict pred_val if sample==2 & (event==0 | event==1)
sum pred_val, de
local sum_pred_val=r(sum)

predict pred_test if sample==3 & (event==0 | event==1)
sum pred_test, de
local sum_pred_test=r(sum)


// to see how the sum of predicted probabilities look like
tabstat pred_val pred_test, stat(sum mean N)
tab event sample


//Now I want to run simulations with different sumsamples like 50, 25 ,25 above but for diffferent seeds fro example seed : 1 to 1000.
// How and would like to see how sum of pred_val and pred_test differ from the actual observed events given by tab event sample,
//What could be the best approach to use

[Nishan Lamichhane]

I figured out this the following way:

gen sum_p2=.
gen sum_p3=.
qui forv i=1/1000 {
local k=`i'
set seed `k'

// splitting randomly the data into training 50%, validation 25% and test set 25%
splitsample, generate(sample) split(0.5 0.25 0.25)

//doing logit
logit event  var2 var2......
pred p2 if sample==2
sum p2
replace sum_p2=r(sum) in `i' 

pred p3 if sample==3
sum p3
replace sum_p3=r(sum) in `i'

drop sample
}