Regress ≠ Twoway qfit !?

[aurelien mulliez]

Hi,

I'am a bit puzzled as i got 2 different predictions from a simple quadratic regression (Y = aX +bX² + c), by 2 differents stata commands that i expected to be equivalent/exangeable.

Here is a reproductible example :

sysuse auto, clear

regress price mpg c.mpg#c.mpg /*** we fit a basic quadratic model with regress ****/
predict fitted  /*** we get the predictions ****/
sort mpg   /*** sort on mpg in order to use twoway line ***/

graph tw (scatter  price mpg) /// /**** to get the scatter plot from crude values ****/
 (line fitted mpg) /// /*** the fitted values from regress ***/
(qfit  price mpg) /*** the quadratic fitted values from qfit ***/

As you can see, there is a small difference between the 2 curves (i got another set of data were the difference is not neglectable)
I'am not sure were i'am wrong (Regress ≠ Twoway ?) or if i mess something ?

[Nick Cox]

This pivots entirely on line using straight-line segments to join data points. Change line to mspline and you should feel happier.

[aurelien mulliez]

Thank you,
I didn't think this was a so obvious little thing
Of course i feel happier ... but for a short while, as i still have a gap with my own data.
I'am looking for an explanation before to bother you again (or not )

[aurelien mulliez]

All fixed, i got a mistake on the regress line (that were aX² + c, bX was missing)
Sorry for the inconvenience and thanks again for the mspline plot tips that is better suited than line in quadratic situations.