RE: st: CIs around Qreg trend line

["Nick Cox" <n.j.cox@durham.ac.uk>]

(100 - 95) / 200 is .025 as I presume you know or can work out for
yourself. 

The result is put in a local macro called -level- which is then fed as
one argument to the -invttail()- function. In this case, you get a 95%
confidence interval. The other argument is the number of residual
degrees of freedom to play with. The element "df" is a clue here. 

To get this example, I modified some code that you can see for a similar
problem in 

. ssc type rcspline.ado 

I use this form because it is then easy to see that for "different
values of 95", you would just plug in something else instead of 95, e.g.


local level = (100 - 90) / 200 

is the different line needed for a 90% confidence interval. 

Nick 
n.j.cox@durham.ac.uk 

Jeffrey B. Wolpin

Thank you.  The example syntax worked great.
Question:  
What is being computed by the line "local level = (100 - 95) / 200"?
Is this line in any way dependent upon the number of observations?
Alternatively, what does 200 signify.

Nick Cox

sysuse auto, clear 
gen gpm = 1/mpg
qreg gpm weight 
predict pred 
predict se, stdp  
local level = (100 - 95) / 200  
gen ul = pred + invttail(e(df_r), `level') * se 
gen ll = pred - invttail(e(df_r), `level') * se 
twoway rarea ll ul weight, sort astyle(ci) || ///
line pred weight, sort || ///
scatter gpm weight, legend(off) ytitle(Gallons per mile)

Jeffrey B. Wolpin

In Stata 10, assuming a qreg regression with one predictor, is there a
way to develop confidence intervals around the qreg trend line (or any
quantile for that matter)?

Also, once developed, what would be the syntax to then graph the scatter
plot, trend line and confidence intervals?

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/