One sided t-test

[Francois Durant]

Hi,

I would like to record the t-stat of the ttest of the following regression: w_H => w_L. The following command gives me the p-value but not the t-stat. I would like to record it in Excel with the command "putexcel". Who can I do that? Thanks.

 sysuse auto, clear

xtile weight_HL=weight, nquantiles(2)
gen w_H=1 if weight_HL==2
replace w_H=0 if weight_HL==1
gen w_L=1 if weight_HL==1
replace w_L=0 if weight_HL==2


regress price w_H w_L weight length mpg, noconstant

test _b[w_H] = _b[w_L]
local sign_d_HL = sign(_b[w_H] - _b[w_L])
display "Ho: coef (w_H-w_L) => 0  p-value = " 1-ttail(r(df_r),`sign_d_HL'*sqrt(r(F)))

[Carlo Lazzaro]

Francois:
I do hope that what follows might be helpful:

. display "Ho: coef (w_H-w_L) => 0  t= " invttail(r(df_r), 1-ttail(r(df_r),`sign_d_HL'*sqrt(r(F))))  " p-value = " 1-ttail(r(df_r),`sign_
> d_HL'*sqrt(r(F)))
Ho: coef (w_H-w_L) => 0  t= 3.1920331 p-value = .00106403

[Francois Durant]

thanks a lot Carlo! Do you know how I could save that in excel with the "putexcel" command? Where can I see the results in memory?

[Carlo Lazzaro]

Francois:
no, sorry.
I'm not familiar with -putexcel-.

[Francois Durant]

Carlo,

Thanks again for your code. I am currently running it with my data.

Here is what I am getting:

. display "Ho: coef (w_H-w_L) => 0  t= " invttail(r(df_r), ttail(r(df_r),`sign_d_HL'*sqrt(r(F))))  " p-value = " ttail(r(df_r),
> `sign_d_HL'*sqrt(r(F)))
Ho: coef (w_H-w_L) => 0  t= 1.69504 p-value = .04552803

The p-value is <5% and the t = 1.695. The t value (corresponding to the p-value) seems low. I may be wrong on that but I just want to check that everything is right... Can you confirm? Thanks

[Carlo Lazzaro]

Francois:
the code seems right; I do not know your seemingly revised data (vs post#1), hence I cannot commen any further.

[Clyde Schechter]

Remember, you specifically asked for a one-sided t-test, so the p-values will be lower than what you are accustomed to seeing in the more usual two-sided tests. With respect to your particular example, if the degrees of freedom is 35 or higher, then that t-statistic does indeed correspond to a p-values less than 0.05.

Finally, let me also suggest you not make a fetish out of 0.05 significance. Your work will probably be more fruitful if you focus on the magnitude of the effect.

[Francois Durant]

Thanks Clyde

[Francois Durant]

And of course, thanks a lot Carlo. Your code is right and very helpful!