Incremental R2

[Jakub Cirkowski]

Hello,

I am conducting an analysis of how much variance in a share price can be explained by changes in 16 accounting variables. One of my outputs is presented below. I am considering presenting an incremental R2 for each explanatory variable.

The incremental R2 is simply the difference between R2 from a regression using all 16 variables and R2 regression using 15 variables (excluding the variable that we are calculating the incremental R2 for)

I have tried to calculate it manually but it is a little bit time-consuming as I need to do it for 16 variables across industries.

I was wondering if somebody conducted the study using incremental R2 and know how to generate a nice list of variables and corresponding incremental R2s?

      Source |       SS           df       MS      Number of obs   =    91,455
-------------+----------------------------------   F(16, 91438)    =   6237.52
       Model |  3371191.93        16  210699.496   Prob > F        =    0.0000
    Residual |  3088719.95    91,438   33.779391   R-squared       =    0.5219
-------------+----------------------------------   Adj R-squared   =    0.5218
       Total |  6459911.88    91,454  70.6356407   Root MSE        =     5.812

------------------------------------------------------------------------------
         prc |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        atpr |   -.030514   .0014333   -21.29   0.000    -.0333232   -.0277048
      capxpr |   .0325993   .0124087     2.63   0.009     .0082784    .0569201
       ceqpr |   .6497327   .0041183   157.77   0.000     .6416609    .6578044
        chpr |   .5618771   .0115592    48.61   0.000     .5392213     .584533
      cogspr |  -.3351217   .0136862   -24.49   0.000    -.3619466   -.3082968
       dvcpr |   1.698806   .0453667    37.45   0.000     1.609887    1.787724
     intanpr |   .1766914   .0059984    29.46   0.000     .1649346    .1884482
        ibpr |   .6382094   .0178205    35.81   0.000     .6032814    .6731373
     oancfpr |   .0359895    .002705    13.30   0.000     .0306877    .0412913
      revtpr |    .325594   .0136197    23.91   0.000     .2988996    .3522885
       spipr |  -.5048722   .0242737   -20.80   0.000    -.5524484   -.4572961
       xadpr |  -.1427368   .0308852    -4.62   0.000    -.2032714   -.0822022
       xrdpr |   1.516386   .0413511    36.67   0.000     1.435338    1.597434
      xsgapr |   -.332726   .0146801   -22.67   0.000    -.3614989   -.3039532
   revgrowpr |    .014171   .0023716     5.98   0.000     .0095227    .0188193
       ocipr |   .0423746   .0122841     3.45   0.001     .0182979    .0664513
       _cons |   2.273077   .0260565    87.24   0.000     2.222006    2.324147
------------------------------------------------------------------------------

Thank you for taking your time to help me,

Jakub

[Carole J. Wilson]

Here's a little loop that estimates a full regression, then compares that r-squared value to a regression without one of the predictors:

sysuse auto, clear
local varlist16 "foreign turn weight"
reg price foreign turn weight
local r2_16 = `e(r2)'
foreach var of local varlist16 {
    local varlist15: list varlist16 - var
    qui reg price `varlist15'
    local r2_`var' = `e(r2)'
    noi di "regression without `var' :  r-squared is `r2_`var''"
    noi di "difference between full regression & regression without `var' is " `r2_16'-`r2_`var''
    }

Note that this is not a general solution. If you have factor variable notation (you do not in the example you provide), this won't work correctly.