Etable from large number of models

[Janine Stubbs]

I have fourty predictors (x1 ... 40), each to anlayse in univariate regressions and would like to report the results in a compact table like this (ultimately I would also like to report margins for binary variables):

	y	CI
x1	-0.1	[-0.1 -0.0]
x2	0.1	[-0.7 1.0]
x3	0.4	[-0.4 1.2]
x4	-0.9	[-2.1 0.3]
Number of observations	147	147

The progress so far yields an enormous table with a lot of white space (here shown only for predictors x1 to x4).

I've toyed with importing from Excel and reshaping in Stata (with limited success), but wonder if there is a better way?

Thankyou!

* Example generated by -dataex-. For more info, type help dataex
clear
input float(x1 x2 x3 x4 y)
32 0 0 0 22
16 0 1 0 20
18 0 0 0 23
25 0 1 0 21
31 0 1 1 20
29 0 1 0 23
30 0 0 0 20
29 0 0 0 22
22 0 1 0 23
18 0 1 0 23
29 0 0 0 20
40 0 1 1 20
19 0 1 0 21
27 1 1 1 23
34 0 1 1 23
28 0 1 0 22
18 0 1 0 20
40 0 1 0 20
30 1 0 0 20
16 1 1 0 22
29 1 0 0 23
27 0 0 0 21
21 0 0 1 20
25 0 1 1 21
32 0 1 0 20
31 1 0 0 20
41 1 0 0 20
22 0 1 0 22
27 0 0 0 22
34 0 0 0 23
29 0 0 0 24
21 0 0 1 23
24 0 0 0 21
27 0 1 0 21
29 0 0 0 20
33 1 0 0 23
38 0 1 0 23
22 1 1 0 21
34 1 1 0 22
24 0 1 0 20
22 0 0 0 23
25 0 0 1 23
27 0 1 1 21
36 0 0 0 21
39 0 0 0 21
40 0 0 0 21
25 1 0 0 22
24 0 0 0 23
35 0 0 0 20
34 0 0 0 20
24 0 0 0 20
25 0 0 0 22
38 1 0 0 20
36 0 1 1 21
26 0 0 0 20
31 1 0 0 20
28 1 0 0 22
42 0 0 1 23
37 0 0 1 21
22 1 0 0 21
24 0 0 0 23
21 0 1 0 21
20 0 1 0 21
38 0 1 0 20
24 0 0 0 23
24 0 1 0 27
38 1 1 0 27
17 0 0 0 27
24 0 1 0 28
22 0 0 0 27
28 0 1 0 27
27 0 0 0 26
29 1 0 0 25
21 1 1 0 26
25 1 0 0 26
21 0 1 0 29
34 1 1 0 23
16 1 0 0 20
33 0 0 0 21
29 0 1 0 21
39 0 0 0 21
39 0 0 0 22
34 0 1 0 21
29 0 1 0 22
25 0 0 0 23
22 1 0 0 21
22 1 0 0 21
27 1 0 0 20
21 0 0 0 23
22 0 1 0 23
23 0 0 1 21
27 1 1 0 21
27 0 0 0 23
32 0 0 1 21
37 1 1 0 29
21 1 0 0 21
17 1 0 0 27
21 0 0 0 20
40 0 0 0 20
21 1 0 0 23
end

foreach var of varlist x1 x2 x3 x4 {
qui reg y `var'
estimates store m`var'
}

etable, replace estimates(m*) cstat(_r_b, nformat(%6.1f)) cstat(_r_ci, cidelimiter(" to") nformat(%6.1f))

[Jeff Pitblado (StataCorp)]

You are almost there. You just need to use some collect commands to change header styles for the results of interest and change the layout.

* show header labels for coefficients and CI
collect style header result[_r_b _r_ci], level(label)
* change layout
collect layout (colname) (result[_r_b _r_ci]#stars[value])

Here is the resulting table.

------------------------------
   Coefficient      95% CI    
------------------------------
x1        -0.1 [-0.1 to   0.0]
x2         0.5 [-0.5 to   1.4]
x3         0.6 [-0.3 to   1.5]
x4        -0.7 [-2.0 to   0.5]
------------------------------

In your example there are no missing values in any of the x variables, so all the estimation results use the same observations. In this case you could arrange it so the common sample size shows up at the end with the following code.

collect addtag sample[N] , fortags(cmdset[1]#result[N])
collect composite define bN = _r_b N
collect label levels result bN "y"
collect layout (colname sample) (result[bN _r_ci]#stars[value])

Here is the resulting table.

-----------------------
     y       95% CI    
-----------------------
x1 -0.1 [-0.1 to   0.0]
x2  0.5 [-0.5 to   1.4]
x3  0.6 [-0.3 to   1.5]
x4 -0.7 [-2.0 to   0.5]
N   100                
-----------------------

[Janine Stubbs]

So elegant! Thankyou Jeff for showing me that.

Janine