getting the pvalue from t stat

[Pietro Raviola]

Hi everybody,

I managed to store the coefficients, the standard errors and the t_stats of a particular regressor for a set of regressions and I would be interested to know the corresponding p_value. Do you know how to get them?
I have tried with (2 * ttail(e(df_r), abs(_b[jobfather]/_se[jobfather]))) but it returns me missing values (.)

Thanks

Pietro

[Maarten Buis]

That works for me:

. sysuse auto, clear
(1978 Automobile Data)

. regress price mpg foreign

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(2, 71)        =     14.07
       Model |   180261702         2  90130850.8   Prob > F        =    0.0000
    Residual |   454803695        71  6405685.84   R-squared       =    0.2838
-------------+----------------------------------   Adj R-squared   =    0.2637
       Total |   635065396        73  8699525.97   Root MSE        =    2530.9

------------------------------------------------------------------------------
       price |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         mpg |  -294.1955   55.69172    -5.28   0.000    -405.2417   -183.1494
     foreign |   1767.292    700.158     2.52   0.014     371.2169    3163.368
       _cons |   11905.42   1158.634    10.28   0.000     9595.164    14215.67
------------------------------------------------------------------------------

. di (2 * ttail(e(df_r), abs(_b[foreign]/_se[foreign])))
.01383634

So the error occurs somewhere else. One possible guess would be that the standard error is missing in your model. However, there are many other possible reasons, so without additional information there is little more we can say.

[Pietro Raviola]

Maybe the problem is that I'm trying to run it in a loop?

foreach y of varlist `y'{
fracglm `y' `controlVars1', vce(robust) link(p)
post `myfile' (_b[jobfather]) (_se[jobfather]) (2 * ttail(e(df_r), abs(_b[jobfather]/_se[jobfather])))
}

it provides me correct coefficient and standard errors, it seems wired it does not work for p_values

[Maarten Buis]

What is the command fracglm? Does that return e(df_r)? My suspicion is that that is not the case, a t-test is inappropriate for that model, and that you need to compare the Wald statistic with the standard normal distribution. See the last part of: http://www.stata-journal.com/article...article=st0137

[Emad Shehata]

clear all
 input t y x1 x2
 1  89.1  96.7   101
 2  99.2  98.1 100.1
 3    99   100   100
 4   100 104.9  90.6
 5 111.6 104.9  86.5
 6 122.2 109.5  89.7
 7 117.6 110.8  90.6
 8 121.1 112.3  82.8
 9   136 109.3  70.1
10 154.2 105.3  65.4
11 153.6 101.7  61.3
12 158.5  95.4  62.5
13 140.6  96.4  63.6
14 136.2  97.6  52.6
15   168 102.4  59.7
16 154.3 101.6  59.5
17   149 103.8  61.3
 end

 tsset t
qui reg y x1 x2 t
 matrix b=e(b)'
 matrix v=e(V)
 scalar df=e(df_r)
 mata: v= st_matrix("v")
 mata: se= sqrt(diagonal(v))
 mata: b= st_matrix("b")
 mata: t=  b:/se
 mata: df= st_numscalar("df")
 mata: p= 2*ttail(df,abs(t))
 mata: se= st_matrix("se",se)
 mata: t= st_matrix("t",t)
 mata: p= st_matrix("p",p)
 matrix stat = b , se , t , p
 matrix colnames stat = "B"  "SE"  "T"  "P>|t|"
 matrix rownames stat = x1 x2 t _cons
 reg y x1 x2 t
 matlist stat

.  reg y x1 x2 t

      Source |       SS       df       MS              Number of obs =      17
-------------+------------------------------           F(  3,    13) =   27.44
       Model |  8335.47821     3  2778.49274           Prob > F      =  0.0000
    Residual |   1316.4803    13  101.267715           R-squared     =  0.8636
-------------+------------------------------           Adj R-squared =  0.8321
       Total |  9651.95851    16  603.247407           Root MSE      =  10.063

------------------------------------------------------------------------------
           y |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
          x1 |   .4213162   .4933296     0.85   0.409    -.6444576     1.48709
          x2 |  -.9885772   .4572538    -2.16   0.050    -1.976414   -.0007405
           t |   1.322487   1.509992     0.88   0.397    -1.939652    4.584627
       _cons |   150.1613   60.72759     2.47   0.028     18.96734    281.3553
------------------------------------------------------------------------------

. matlist stat

             |         B         SE          T      P>|t|
-------------+--------------------------------------------
          x1 |  .4213162   .4933296   .8540258   .4085527
          x2 | -.9885772   .4572538  -2.161988   .0498522
           t |  1.322487   1.509992   .8758242   .3970202
       _cons |  150.1613   60.72759   2.472703   .0279919

[Emad Shehata]

Notes:
The above example valid when
scalar df=e(df_r)
Elsewhere choose a valid expression or number

[Maarten Buis]

The above example valid when
scalar df=e(df_r)
Elsewhere choose a valid expression or numbe

If e(df_r) is not defined by your estimation command, then you should not choose a number. The t-distribution is nice, but only works in specific models. The most common example is linear regression. It is convention in Stata to return e(df_r) when the t-distribution is appropriate, otherwise leave it empty. So if e(df_r) is empty, then that is a clear sign, that the t-distribution is inappropriate. An example of a large set of models where this is the case is any model estimated using maximum (quasi-)likelihood, where you will need to compare the Wald statistic with the standard normal distribution rather than a t-distribution.

[Emad Shehata]

I mean about the number is degrees of freedom
Not any number
And some models not all df scalar are the same formula =e(df_r)

[Richard Williams]

fracglm is a user-written routine (by me) that is very similar to Stata 14's fracreg (but it only requires Stata 11.2). Like fracrec it returns z values, not t-values. The correct formula should be something like

. display 2 * (1 - normal(1.96))
.04999579

Note too that, after many/most estimation commands, if you do

mat list r(table)

you get a lot of information, including the p values. This information could be extracted via matrix commands if a user wanted to.

I've never officially released fracglm, partly because fracreg came along and greatly reduced the need for it. But it is useful for users condemned to using older versions of Stata. Also, at least for the moment, there is a bug in fracreg that causes it to report incorrect degrees of freedom but that is supposed to be fixed soon. If anyone wants fracglm more information is available at

http://www3.nd.edu/~rwilliam/xsoc739...onseModels.pdf