Testing a non-zero null hypothesis in a regression

[Sarah Goldberg]

Hello,

In my current project, I need to test whether a coefficient in my regression is significantly different from 1 (not from 0). I realize that I can obtain this information by simply running the regression and running a test on the coefficient afterwards, but I was hoping there would be a different method that actually incorporates this test into the regression table. This would be extrodinarily useful because I have been using esttab to create Latex tables from my regressions.

My code is as follows:

eststo: xi: xtreg var1 var2 var3 i.year, fe 
test var2 = 1
.
.
.
esttab using table.tex, drop(_Iyear*) stats(N) cells(b(star fmt(2)) se(par)) replace

As is, this code produces a table in which coefficients are starred if they are significantly different from 0, and I can separately report whether the var2 coefficient is significantly different from 1. Is there any way that I can combine the regression and the test to automatically report the results of that test in the table itself?

Any help would be very much appreciated.

Thanks,
Sarah

[Brendan Halpin]

gen y2 = var1 - var2
reg y2 var2 var3

This works for simple cases: does it work for yours?

[Richard Williams]

Brendan's code sounds right to me, except I wonder if it will zap other statistics of interest, e.g. in a regular regression things like R^2 and the model F get zapped when you do this.

[Felix Leung]

I believe you could use an -estadd- after your -test- command.

Here's an example:

sysuse auto, clear

eststo m1: regress price weight length displacement

test weight = 1
estadd scalar pvalue = r(p)

esttab m1 using table.tex, replace ///
    cells(b(star fmt(2)) se(par)) ///
    stats(pvalue N, fmt(3 0)) ///
    subs("pvalue" "p-value ($ H_0: \beta_{weight} = 1$)") label noabbrev

And here's the table.tex returned:

 
 {  
 \def\sym#1{\ifmmode^{#1}\else\(^{#1}\)\fi}  
 \begin{tabular}{l*{1}{c}}  
 \hline\hline  
                     &\multicolumn{1}{c}{(1)}\\  
                     &\multicolumn{1}{c}{Price}\\  
                     &        b/se\\  
 \hline  
 Weight (lbs.)       &        4.61\sym{**} \\  
                     &      (1.40)         \\  
 Length (in.)        &      -97.63\sym{*}  \\  
                     &     (39.57)         \\  
 Displacement (cu. in.)&        0.73         \\  
                     &      (6.97)         \\  
 Constant            &    10440.63\sym{*}  \\  
                     &   (4369.32)         \\  
 \hline  
 p-value ($ H_0: \beta_{weight} = 1$)              &       0.012         \\  
 N                   &          74         \\  
 \hline\hline  
 \end{tabular}  
 }