Changing display format of correlations

[Nick Cox]

A colleague wanted display of correlations from correlate with fewer decimal places. The best answer I could come up with was pushing the resulting matrix through matrix list.

*  You can pick up the correlation result as a matrix and display the elements as you wish. Or you can push that into your own matrix, which is not so ephemeral. 

. sysuse auto, clear
(1978 automobile data)

. corr length-gear_ratio
(obs=74)

             |   length     turn displa~t gear_r~o
-------------+------------------------------------
      length |   1.0000
        turn |   0.8643   1.0000
displacement |   0.8351   0.7768   1.0000
  gear_ratio |  -0.6964  -0.6763  -0.8289   1.0000

. mat li r(C), format(%9.2f)

symmetric r(C)[4,4]
                    length          turn  displacement    gear_ratio
      length          1.00
        turn          0.86          1.00
displacement          0.84          0.78          1.00
  gear_ratio         -0.70         -0.68         -0.83          1.00

. mat rho = r(C)

. mat list rho, format(%3.2f)

symmetric rho[4,4]
                    length          turn  displacement    gear_ratio
      length          1.00
        turn          0.86          1.00
displacement          0.84          0.78          1.00
  gear_ratio         -0.70         -0.68         -0.83          1.00

The header text -- here symmetric rho[4.4] -- could be suppressed with an option noheader.

A similar trick should work with commands for rank correlations.

The code for pwcorr could be hacked to allow a user-specified format. But in that case, as in the others, I agree with my colleague's implication that there should be an option on the official commands to do this directly. Am I missing something simpler?

Incidentally, corrci from the Stata Journal has such an option, and several others, but there are always users unwilling or unable to install community-contributed commands.

 
SJ-21-3 pr0041_4  . . . . . . . . . . . . . . . . . Software update for corrci
        (help corrci, corrcii if installed) . . . . . . . . . . . .  N. J. Cox
        Q3/21   SJ 21(3):847
        improves explanation of the format() option and fixes a bug
        concerning saving results to a new dataset

SJ-20-4 pr0041_3  . . . . . . . . . . . . . . . . . Software update for corrci
        (help corrci, corrcii if installed) . . . . . . . . . . . .  N. J. Cox
        Q4/20   SJ 20(4):1028--1030
        corrects code for a bias correction used if (and only if) the
        fisher option is specified

SJ-17-3 pr0041_2  . . . . . . . . . . . . . . . . . Software update for corrci
        (help corrci, corrcii if installed) . . . . . . . . . . . .  N. J. Cox
        Q3/17   SJ 17(3):779
        new options added

SJ-10-4 pr0041_1  . . . . . . . . . . . . . . . . . Software update for corrci
        (help corrci, corrcii if installed) . . . . . . . . . . . .  N. J. Cox
        Q4/10   SJ 10(4):691
        update to fix corrci so that it always saves r-class results

SJ-8-3  pr0041  .  Speaking Stata: Corr. with confidence, Fisher's z revisited
        (help corrci, corrcii if installed) . . . . . . . . . . . .  N. J. Cox
        Q3/08   SJ 8(3):413--439
        reviews Fisher's z transformation and its inverse, the
        hyperbolic tangent, and reviews their use in inference
        with correlations

 . corrci length-gear_ratio

(obs=74)

                           correlations and 95% limits
length       turn              0.864    0.792    0.913
length       displacement      0.835    0.750    0.893
length       gear_ratio       -0.696   -0.798   -0.556
turn         displacement      0.777    0.667    0.854
turn         gear_ratio       -0.676   -0.784   -0.530
displacement gear_ratio       -0.829   -0.889   -0.741

. corrci length-gear_ratio, format(%3.2f)

(obs=74)

                           correlations and 95% limits
length       turn               0.86     0.79     0.91
length       displacement       0.84     0.75     0.89
length       gear_ratio        -0.70    -0.80    -0.56
turn         displacement       0.78     0.67     0.85
turn         gear_ratio        -0.68    -0.78    -0.53
displacement gear_ratio        -0.83    -0.89    -0.74

* corrci has other options, including matrix output and saving to a dataset

[Maarten Buis]

Here is an alternative way of doing this without downloading community contributed packages. It is more lines of code, but has the advantage that afterwards you can directly export the table to LaTeX, Word, Excel, and other formats using collect export.

. clear all

. sysuse auto, clear
(1978 automobile data)

. corr length-gear_ratio
(obs=74)

             |   length     turn displa~t gear_r~o
-------------+------------------------------------
      length |   1.0000
        turn |   0.8643   1.0000
displacement |   0.8351   0.7768   1.0000
  gear_ratio |  -0.6964  -0.6763  -0.8289   1.0000


. collect get corr=vech(r(C))

. collect style header result, level(hide)

. collect style cell, nformat(%9.2f)

. collect layout (rowname#result) (roweq)

Collection: default
      Rows: rowname#result
   Columns: roweq
   Table 1: 4 x 4

-----------------------------------------------------------------------------------------
                       | Length (in.) Turn circle (ft.) Displacement (cu. in.) Gear ratio
-----------------------+-----------------------------------------------------------------
Length (in.)           |         1.00                                                    
Turn circle (ft.)      |         0.86              1.00                                  
Displacement (cu. in.) |         0.84              0.78                   1.00           
Gear ratio             |        -0.70             -0.68                  -0.83       1.00
-----------------------------------------------------------------------------------------

[John Mullahy]

I recall suggesting on Statalist at some earlier date that the cformat, sformat, pformat suite of settings be expanded to accommodate other results' displays (e.g. correlation matrixes). This seems to me an obvious and easy solution, but maybe one that would be challenging to implement under the hood.

. sysuse auto
(1978 automobile data)

. reg price mpg

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(1, 72)        =     20.26
       Model |   139449474         1   139449474   Prob > F        =    0.0000
    Residual |   495615923        72  6883554.48   R-squared       =    0.2196
-------------+----------------------------------   Adj R-squared   =    0.2087
       Total |   635065396        73  8699525.97   Root MSE        =    2623.7

------------------------------------------------------------------------------
       price | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |  -238.8943   53.07669    -4.50   0.000    -344.7008   -133.0879
       _cons |   11253.06   1170.813     9.61   0.000     8919.088    13587.03
------------------------------------------------------------------------------

. corr price mpg
(obs=74)

             |    price      mpg
-------------+------------------
       price |   1.0000
         mpg |  -0.4686   1.0000


. set cformat %9.2f

. reg price mpg

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(1, 72)        =     20.26
       Model |   139449474         1   139449474   Prob > F        =    0.0000
    Residual |   495615923        72  6883554.48   R-squared       =    0.2196
-------------+----------------------------------   Adj R-squared   =    0.2087
       Total |   635065396        73  8699525.97   Root MSE        =    2623.7

------------------------------------------------------------------------------
       price | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |    -238.89      53.08    -4.50   0.000      -344.70     -133.09
       _cons |   11253.06    1170.81     9.61   0.000      8919.09    13587.03
------------------------------------------------------------------------------

. corr price mpg
(obs=74)

             |    price      mpg
-------------+------------------
       price |   1.0000
         mpg |  -0.4686   1.0000


.

[Nick Cox]

#2 Maarten Buis helpfully explained how to use collect. An advantage of this route is that you get to see variable labels. There's presumably a route that uses variable names only when you prefer them or there is not enough space for variable labels.

#3 John Mullahy Indeed in practice those options seem quite independent of correlate commands.

[Evgeny Saburov]

It's actually possible to wrap what Maarten Buis proposed in a -table- one-liner:

sysuse auto, clear

table (rowname) (roweq), command(vech(r(C)): corr length-gear_ratio) nformat ("%9.2f")

And to drop variable labels one can run this afterwards:

collect label drop rowname
collect label drop roweq
collect preview

[Nick Cox]

Evgeny Saburov Thanks to you too. #5 certainly shows how the collect code can be compressed to one line.

I want to flag, just in case any readers missed it, that the first work-around in #1 doesn't hinge on installing anything extra.

I still favour something like an explicit format() option for these commands -- or a little more subtly rformat(), pformat(), etc.

As John Mullahy hinted a generic setting would also be welcome. so that rformat defaults to some format with 4 d.p. but is honoured by correlation commands if you set it to something different When I learned about correlations in the 1960s they were almost always presented with 2 d.p. Giving more d.p. seems on a par with grade inflation and is defensible insofar as a researcher can always round but knife-edge cases can always exist. 0.2345 might mean 0.23449 or 0.23451, and so on. Sure, you can always look at saved results, but the issue is only wanting small changes from StataCorp to ease a researcher's work.