collect layout with regression coefficients for continuous independent variables

[Scott Adams]

Hi,
I am having issues using -collect- to make a table of regression coefficients especially when an independent variable is continuous.
Here are three options I have tried but each is unsatisfactory to me, as noted in the comment.

. clear all

. set obs 1000
Number of observations (_N) was 0, now 1,000.

. gen x=runiformint(0,1)

. label var x "Outcome"

. gen c = runiform()

. lab var c "Continuous"

. gen i = runiformint(1,3)

. label var i "Discrete"

. cap collect drop c1

. collect create c1
(current collection is c1)

. collect OR=_r_b P=_r_p _r_ci,name(c1) : /// 
>         logistic x c.c i.i

Logistic regression                                     Number of obs =  1,000
                                                        LR chi2(3)    =   1.43
                                                        Prob > chi2   = 0.6974
Log likelihood = -692.03768                             Pseudo R2     = 0.0010

------------------------------------------------------------------------------
           x | Odds ratio   Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
           c |   1.289403   .2840073     1.15   0.249     .8373382    1.985531
             |
           i |
          2  |   1.033049   .1635239     0.21   0.837      .757499    1.408834
          3  |   1.042367   .1573393     0.27   0.783     .7754193    1.401214
             |
       _cons |   .9074241   .1386773    -0.64   0.525     .6725511    1.224321
------------------------------------------------------------------------------
Note: _cons estimates baseline odds.

. est store e

.         
. collect dims //no dim for c?

Collection dimensions
Collection: c1
-----------------------------------------
                   Dimension   No. levels
-----------------------------------------
Layout, style, header, label
                      cmdset   1         
                       coleq   1         
                     colname   9         
           colname_remainder   1         
                           i   3         
               program_class   2         
                      result   46        
                 result_type   3         
                     rowname   1         

Style only
                border_block   4         
                   cell_type   4         
-----------------------------------------

. 
. collect style header, level(label) title(label)

. collect style row  stack ,spacer

. collect style cell result, nformat(%4.2f)

. collect style cell _r_ci, cidelimiter(",") sformat("(%s)")

. //option 1
. collect layout (c i)(result[OR _r_ci])            // ?? does not show c at all ??
(dimension c not found)

Collection: c1
      Rows: c i
   Columns: result[OR _r_ci]
   Table 1: 4 x 2

-----------------------------
         | Result    Result  
         |     OR    95% CI  
---------+-------------------
Discrete |                   
  1      |   1.00            
  2      |   1.03 (0.76,1.41)
  3      |   1.04 (0.78,1.40)
-----------------------------

. 
. //option 2
. collect layout (colname)(result[OR _r_ci])        // ?? does not show var label for discrete ??

Collection: c1
      Rows: colname
   Columns: result[OR _r_ci]
   Table 1: 8 x 2

-----------------------------------------------------
                                 | Result    Result  
                                 |     OR    95% CI  
---------------------------------+-------------------
Covariate names and column names |                   
  Continuous                     |   1.29 (0.84,1.99)
                                 |                   
  Discrete=1                     |   1.00            
  Discrete=2                     |   1.03 (0.76,1.41)
  Discrete=3                     |   1.04 (0.78,1.40)
                                 |                   
  Intercept                      |   0.91 (0.67,1.22)
-----------------------------------------------------

. 
. //option 3                      ?? cannot get CI next to OR instead of below??
. etable , estimates(e) cstat(_r_b, label("OR") nformat("%03.2f")) cstat(_r_ci, label("95%CI
> ") cidelimiter(",") sformat("(%s)") nformat("%03.2f")) 

----------------------------------
                            x     
----------------------------------
Continuous                    1.29
                       (0.84,1.99)
Discrete                          
  2                           1.03
                       (0.76,1.41)
  3                           1.04
                       (0.78,1.40)
Intercept                     0.91
                       (0.67,1.22)
Number of observations        1000
----------------------------------

I also could not get the desired table with the Tables Builder.
It would look like,

-----------------------------
         | Result    Result  
         |     OR    95% CI  
---------+-------------------
Continuous | 1.29 (0.84, 1.99)

Discrete |                   
  1      |   1.00            
  2      |   1.03 (0.76,1.41)
  3      |   1.04 (0.78,1.40)
-----------------------

Thanks,
Scott

[Scott Adams]

OK here's a solution I found

collect set c1
collect style header colname, level(label) title(hide)

collect layout (colname[c] i) (result[OR] result[_r_ci]) (), name(c1)

Not especially intuitive to me that continuous variables require colname but factors do not, but ... it seems to work.

[Jeff Pitblado (StataCorp)]

Factor variables found in colname are added to tags on the
collected items to help with alternative layouts, but are not necessary
for the intended/requested table.

All model coefficients exist as levels in dimension colname.

Here is a rework of the above example:

clear all
set obs 1000
gen x=runiformint(0,1)
label var x "Outcome"
gen c = runiform()
lab var c "Continuous"
gen i = runiformint(1,3)
label var i "Discrete"

collect create c1
collect : logistic x c i.i

* coefficient names are stored in dimension -colname-
collect dims
collect levelsof colname

* target cell styles to the results of interest
collect style cell result[_r_b _r_ci], nformat(%4.2f)
collect style cell result[_r_ci], cidelimiter(",") sformat("(%s)")

* turn off factor variables binder so that the row headers can stack
* their levels
collect style row stack , nobinder spacer

* show result's label in header
collect style header result, title(label)
* center repeated column headers
collect style column, dups(center)

* change the label for the collected odds ratios
collect label levels result _r_b "OR", modify

* specify the coefficients and results to report in the table
collect layout (colname[c i]) (result[_r_b _r_ci])

Here is the resulting table

. collect layout (colname[c i]) (result[_r_b _r_ci])

Collection: c1
      Rows: colname[c i]
   Columns: result[_r_b _r_ci]
   Table 1: 6 x 2

-----------------------------
           |      Result
           |   OR    95% CI
-----------+-----------------
Continuous | 1.03 (0.66,1.58)
           |
Discrete   |
  1        | 1.00
  2        | 0.96 (0.71,1.31)
  3        | 1.00 (0.73,1.35)
-----------------------------

[Scott Adams]

Thanks Jeff.
Your construction is similar to my Option 2 in using colname as the rows. But I do not see the difference that causes yours to show the variable name "Discrete" and the level labels, where mine showed Discrete=1, Discrete=2, and Discrete=3.
Which collect statement triggered that?
Thanks,
Scott

[Jeff Pitblado (StataCorp)]

* turn off factor variables binder so that the row headers can stack
* their levels
collect style row stack , nobinder spacer