marginscontplot problem.

[River Huang]

Dear All, I have a dataset with two variables (simplified). The basic statistics

. sum indep ir2d1

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
       indep |     35,432    .3323226    .1119013          0   .5555556
       ir2d1 |     34,059    .5667652    .1640421   .2095384   .9267846

When I use, say the first 1,000 observations, the command marginscontplot works and gives me the figure. However, when I increase to say, 10,000, the estimation result is

. keep in 1/10000
(47,372 observations deleted)

. reg indep ir2d1 c.ir2d1#c.ir2d1, 

      Source |       SS           df       MS      Number of obs   =     8,888
-------------+----------------------------------   F(2, 8885)      =      5.93
       Model |  .218049191         2  .109024596   Prob > F        =    0.0027
    Residual |  163.360103     8,885  .018386056   R-squared       =    0.0013
-------------+----------------------------------   Adj R-squared   =    0.0011
       Total |  163.578153     8,887  .018406454   Root MSE        =     .1356

---------------------------------------------------------------------------------
          indep |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
----------------+----------------------------------------------------------------
          ir2d1 |   .1730636   .0502702     3.44   0.001     .0745223    .2716048
                |
c.ir2d1#c.ir2d1 |  -.1495413   .0442941    -3.38   0.001    -.2363681   -.0627146
                |
          _cons |   .2565559   .0135907    18.88   0.000      .229915    .2831968
---------------------------------------------------------------------------------

and error message is

. marginscontplot ir2d1, ci
invalid numlist has too many elements
r(123);

It seems to me that this mcp/marginscontplot can not be applied to a large dataset? Any suggestions? Thanks.

[River Huang]

By the way, please search marginscontplot and install.

[Jeff Pitblado (StataCorp)]

marginscontplot is written by Patrick Royston, the Stata Journal article that goes with this command is https://www.stata-journal.com/articl...article=gr0056.

Consider using the at1() option to specify at which values to compute and plot the marginal predictions.

Here is a simple example using simulated data:

set seed 12345
set obs 10000
gen x = rnormal()
gen y = x - x*x + 1 + rnormal()
regress y c.x##c.x

* following throws and error because of too many values
cap noi marginscontplot x

* specify at which values of 'x' to compute the marginal predictions
marginscontplot x, at1(.2(.01).9)

Here is a log of the output

. set seed 12345

. set obs 10000
number of observations (_N) was 0, now 10,000

. gen x = rnormal()

. gen y = x - x*x + 1 + rnormal()

. regress y c.x##c.x

      Source |       SS           df       MS      Number of obs   =    10,000
-------------+----------------------------------   F(2, 9997)      =  14625.92
       Model |  29811.7737         2  14905.8868   Prob > F        =    0.0000
    Residual |  10188.3632     9,997  1.01914206   R-squared       =    0.7453
-------------+----------------------------------   Adj R-squared   =    0.7452
       Total |  40000.1369     9,999  4.00041373   Root MSE        =    1.0095

------------------------------------------------------------------------------
           y |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
           x |   1.010405   .0101316    99.73   0.000     .9905454    1.030265
             |
     c.x#c.x |  -1.006517   .0071026  -141.71   0.000    -1.020439   -.9925941
             |
       _cons |   1.022298   .0123168    83.00   0.000     .9981551    1.046442
------------------------------------------------------------------------------

. cap noi marginscontplot x
invalid numlist has too many elements

. marginscontplot x, at1(.2(.01).9)

Here is the resulting graph.

[River Huang]

Dear Jeff, Thank you so much. The code works quite well.