Marginal Effect Analysis of a Conditional Logit Regression

[Vince Williams]

Marginal Effect Analysis of a Conditional Logit Regression
Dear Stata programmers,
I have firm-year data, and I need help with a conditional logit regression. I have pre-period (post_period variable=0) and post period (post period variable=1).

Here are the two equations I have:
dummy_1= a + b1* dummy_x + b2*dummy_y +e
dummy_2= a + b1* dummy_x + b2*dummy_y +e


The e_dummy is for both he pre and the post period while the a_dummy is for just the post period.

1. The code I used to estimate the conditional logit regressions is below. Is this the correct code for a conditional logit

[CODE]
clogit dummy_1 dummy_x dummy_y if post ==0, group(id) vce(cluster id)
clogit dummy_1 dummy_x dummy_y if post ==1, group(id) vce(cluster id)
clogit dummy_2 dummy_x dummy_y if post ==1, group(id) vce(cluster id)
[CODE]

1. I am looking for a marginal effects analysis:

	Post=1 for dummy 1
	dummy_x	dummy_y
Mean	-0.071	0.126
IPROB at 0	0.535	0.489
IPROB at 1	0.678	0.745
Change	0.143	0.256

I would like to get the code for the predicted probability (iprob) when dummy_x =0 or dummy_x =1 while dummy_y is at its mean value. Iprob is estimated each year using the observations in all prior years separately for post=0 and post=1. I have the numbers for the full sample above and the sample below is only 200 observations. Still, I am interested in generating this code and I am not sure how to do it.

* Example generated by -dataex-. For more info, type help dataex
clear
input long id int date float(dummy_1 dummy_2 dummy_x dummy_y post)
  4864 18262 0 0 -1 0 1
 30647 14975 0 0  0 0 0
 11499 18747 0 0  0 0 1
  4311 17531 0 0  0 0 1
 15855 19083 0 1  0 0 1
 10297 13695 0 0  0 1 0
 25124 18627 1 0  0 0 1
 10991 18992 0 0  0 0 1
 30479 13634 0 0  0 0 0
 62388 17166 1 1  1 0 1
 31578 14609 1 0  1 0 0
 31843 17897 1 0  0 0 1
 12697 13514 1 0 -1 0 0
139665 16436 0 0  0 0 1
  5496 16801 0 0  0 0 1
148950 17531 0 0  0 0 1
  8542 17166 0 0 -1 0 1
 19150 18262 0 1  0 1 1
  1050 18627 0 0  0 0 1
 66065 18992 0 0  0 0 1
 30152 17531 1 0  1 0 1
 65967 14244 0 0  0 0 0
184700 18992 0 0  0 0 1
  2697 18992 0 0 -1 0 1
 26021 17166 0 0  0 0 1
 30008 16801 0 0  0 0 1
 29339 14975 0 0  1 0 0
  2295 16982 0 0  0 0 1
102045 16436 0 0 -1 0 1
 12075 16617 0 0  0 0 1
116770 17439 0 0  0 0 1
 62745 14244 0 0  0 0 0
 10156 18627 0 0  0 0 1
 65640 16556 0 0  0 0 1
 11618 14244 0 0 -1 0 0
 16775 14244 1 0  0 0 0
  9069 14183 0 0  1 0 0
 12212 16436 0 0  0 0 1
  8293 16436 0 0  0 0 1
 17239 17166 0 0 -1 0 1
134384 16436 0 0  0 0 1
166208 18627 0 0  0 0 1
106900 14975 0 0  0 0 0
  1914 16436 0 0  0 0 1
 14538 13514 0 0  0 0 0
 14986 14244 1 0  1 0 0
  8133 13514 0 0  0 0 0
  6135 18627 0 0  0 0 1
  4737 16436 0 0  0 0 1
  6078 17897 0 0  0 0 1
 14001 13514 0 0  0 0 0
 30751 13604 0 0  0 0 0
  8240 17166 1 0  0 1 1
140044 17166 1 0  0 0 1
  3962 17897 0 0 -1 1 1
 11910 17836 0 0  0 0 1
  7228 16921 0 0  0 0 1
  4839 17166 0 0  0 0 1
178706 17897 0 0  0 0 1
 12458 16982 0 0  0 0 1
144809 17531 0 0 -1 0 1
118042 18627 0 0  0 0 1
  1274 18262 0 1  0 1 1
 65244 16617 0 0  0 0 1
 21107 13879 0 0  0 0 0
 60893 16436 0 0  0 0 1
 14140 18627 0 0  0 0 1
 14340 14609 1 0  1 1 0
  1611 18627 0 0  0 1 1
  6653 17987 0 1  0 0 1
  6223 18992 0 1  0 0 1
  2034 16436 0 0  0 0 1
 16888 13879 0 0  0 1 0
 24720 16801 0 0  0 0 1
  4124 14425 1 0  0 0 0
 11513 16436 0 0  0 0 1
 29645 14975 0 0  0 0 0
 27932 18262 0 0  0 0 1
224604 16436 0 0  0 0 1
 10991 16436 0 0  0 0 1
125474 17166 1 0  1 0 1
 13081 13665 0 0 -1 0 0
 21519 16740 0 0  0 0 1
 28485 13969 0 0  0 0 0
 12635 16801 0 0  0 0 1
 65886 18627 0 0  0 0 1
 11904 13453 1 0 -1 0 0
 12761 17531 0 1  0 0 1
  1773 16436 0 1  0 0 1
 24689 13514 0 0  0 1 0
 27783 14975 0 0  0 1 0
 15101 18992 0 0  0 0 1
 15617 18992 0 1  0 1 1
166482 17439 0 0  0 0 1
 63172 16679 0 0  0 0 1
163948 18627 0 0  0 0 1
138205 17531 0 0 -1 0 1
 64166 16436 1 0  0 0 1
  3607 14244 0 0  0 0 0
 62197 14244 1 0 -1 0 0
end
format %td date

[Vince Williams]

3. I am also looking a table output like this showing the portfolio ranking for IPROB.Given that dummy_x has three different values (−1, 0, 1) and dummy_y has two (0, 1), IRPOB can only take six
different values each year. Because IPROB can only take on six different values each year, we combine the observations with the highest two IPROB values to form group 5

	Post=1 for dummy 1
	IPROB	dummy_1
1
2
3
4
5
5-1