Twoway lfit (without constant)

[Jay Jeong]

Is there a way to generate figure (twoway lfit) with constant excluded in the regression?

Using two simple variables, I did a regression without constant term (which is identical measuring the correlation).

The regression result is as below. The estimated coefficient is negative, indicating negative correlation.

When I do twoway lfit, however, the figure shows positive relation. I believe that this is because lfit includes constant term.


To double check my guess, I did the regression again using constant term included. As expected, the estimated coefficient is positive in this case.

In short, is there a way to perform "twoway lift" with constant term excluded? Is there user written ado command?

Thanks for your help a lot.

[Justin Niakamal]

This will do what you want, but usually, you don't want to suppress the constant term.

reg cum_automation_infer routineshare , nocons
predict yhat
 
tw line yhat routineshare

[Nick Cox]

The results in #1 don't seem persuasive that omitting an intercept is a good idea. If anything fitting the mean is the simpler model to go for.

[Nick Cox]

Here's some different technique. More work is needed on the graph.

 sysuse auto , clear
(1978 automobile data)

. gen gp1000mi = 1000 / mpg

. regress gp1000mi weight, noconstant

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(1, 73)        =   4067.54
       Model |  194889.452         1  194889.452   Prob > F        =    0.0000
    Residual |  3497.67227        73  47.9133187   R-squared       =    0.9824
-------------+----------------------------------   Adj R-squared   =    0.9821
       Total |  198387.125        74  2680.90709   Root MSE        =    6.9219

------------------------------------------------------------------------------
    gp1000mi | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
      weight |   .0164665   .0002582    63.78   0.000     .0159519    .0169811
------------------------------------------------------------------------------

. scatter gp1000mi weight || function `=_b[weight]'*x, ra(weight)

. scatter gp1000mi weight || function `=_b[weight]'*x, ra(0 5000)

.