Calculating p-values after oprobit using ereturn

[Susan Biancani]

I am trying to calculate p-values for coefficients after an oprobit regression, using the directions in "Stata tip 53: Where did my p-values go?" (http://www.stata-journal.com/sjpdf.h...iclenum=st0137)

Here is the example:
. local t = _b[foreign]/_se[foreign]
. di 2*ttail(e(df_r),abs(`t ́)) However, when I do ereturn list, there is no df_r, only df_m (results of ereturn list shown below). I'm not sure what the difference is, but when I try to use df_m to calculate the p-values, I do not get the correct values.
Any advice on how to calculate the p-values would be much appreciated!
Thanks,
Susan


. ereturn list

scalars:
e(rank) = 42
e(N) = 220503
e(ic) = 3
e(k) = 42
e(k_eq) = 11
e(k_dv) = 1
e(converged) = 1
e(rc) = 0
e(ll) = -401494.4658840121
e(k_eq_model) = 1
e(ll_0) = -414969.3276823036
e(df_m) = 32
e(chi2) = 26949.723596583
e(p) = 0
e(N_cd) = 0
e(k_cat) = 11
e(k_aux) = 10
e(r2_p) = .0324719464774701

macros:
e(cmdline) : "oprobit ar_emp_LTR tenuremonths _Icycle_* _Izone_* _Ib.."
e(cmd) : "oprobit"
e(predict) : "oprobi_p"
e(title) : "Ordered probit regression"
e(chi2type) : "LR"
e(opt) : "moptimize"
e(vce) : "oim"
e(user) : "mopt__oprobit_d2()"
e(ml_method) : "d2"
e(technique) : "nr"
e(which) : "max"
e(depvar) : "ar_emp_LTR"
e(properties) : "b V"

matrices:
e(b) : 1 x 42
e(V) : 42 x 42
e(cat) : 1 x 11
e(ilog) : 1 x 20
e(gradient) : 1 x 42

functions:
e(sample)

[Richard Williams]

You are using the formula for t when you should be using the formula for z. See section 4 of Maarten's paper. See if this works:

local z = _b[foreign]/_se[foreign]
di 2*normal(-abs(`z'))

[Clyde Schechter]

Or you can also find the p-values directly in the matrix r(table) that -oprobit- returns. Depending on what you plan to do, extracting them from that matrix may be easier than calculating them from _b and _se, or not.

[Richard Williams]

I didn't know about r(table). I wonder why oprobit gets special treatment, e.g. you don't get r(table) after regress. Requests like Susan's are very common, so why not make r(table) a universal feature of estimation commands?

[Clyde Schechter]

I think all the estimation commands now return r(table).

 . sysuse auto
(1978 Automobile Data)
  . regress price mpg i.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 |
    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
------------------------------------------------------------------------------
  . return list
  scalars:
              r(level) =  95
  macros:
             r(label2) : "(base)"
  matrices:
              r(table) :  9 x 4
  . matrix list r(table)
  r(table)[9,4]
                            0b.          1.           
               mpg     foreign     foreign       _cons
     b  -294.19553           0   1767.2922   11905.415
    se   55.691719           .   700.15797   1158.6345
     t  -5.2825724           .   2.5241336   10.275385
pvalue   1.333e-06           .   .01383634   1.085e-15
    ll  -405.24167           .    371.2169   9595.1638
    ul   -183.1494           .   3163.3676   14215.667
    df          71          71          71          71
  crit   1.9939434   1.9939434   1.9939434   1.9939434
 eform           0           0           0           0

.

[Richard Williams]

I stand corrected. I was doing ereturn list, not return list, and I didn't see it after running regress. Anyway, nice enhancement. Searching through -help whatsnew-, I see that this was added in Stata 12. For more info see the bottom of -help ereturn- . Programmers should probably try to add this to their user-written estimation routines.

EDIT: Or, maybe Stata just does this for programmers. My oglm program returns r(table), even though I never explicitly told it to do so.

[Susan Biancani]

Thanks so much, this is great!
If I want to use r(table), how do I access the p-value of a specific variable, e.g. the value for mpg in your example?

[Richard Williams]

P values are in the 4th row of the matrix. Getting the column numbers straight might get confusing because factor variables take up multiple columns, Try something like

. sysuse auto, clear
(1978 Automobile Data)

. regress price mpg i.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 |
    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
------------------------------------------------------------------------------

. mat list r(table)

r(table)[9,4]
                            0b.          1.            
               mpg     foreign     foreign       _cons
     b  -294.19553           0   1767.2922   11905.415
    se   55.691719           .   700.15797   1158.6345
     t  -5.2825724           .   2.5241336   10.275385
pvalue   1.333e-06           .   .01383634   1.085e-15
    ll  -405.24167           .    371.2169   9595.1638
    ul   -183.1494           .   3163.3676   14215.667
    df          71          71          71          71
  crit   1.9939434   1.9939434   1.9939434   1.9939434
 eform           0           0           0           0

. mat mytable = r(table)

. scalar pval = mytable[4,1]

. di pval
1.333e-06

You probably have something more ambitiious in mind though. What would you ultimately like?

[Rich Goldstein]

one caveat; as I have previously said re: r(table), I recommend that one first make the matrix into one of your own; e.g., mat def junk=r(table); saved results can disappear and will unless you make them "permanent"

[Richard Williams]

Maybe it is just me, but I have a hard time working with returned matrices unless I make a copy of them. I can never get the syntax right. Could I do the following working directly with r(table)?

mat mytable = r(table)
scalar pval = mytable[4,1]

[Jeff Pitblado (StataCorp)]

The el() function works with all matrices except e(b), e(V), and e(Cns).

Here is Richard's example without making a copy of r(table).

. sysuse auto
(1978 Automobile Data)

. regress price mpg i.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 |
    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
------------------------------------------------------------------------------

. matrix list r(table)

r(table)[9,4]
                            0b.          1.            
               mpg     foreign     foreign       _cons
     b  -294.19553           0   1767.2922   11905.415
    se   55.691719           .   700.15797   1158.6345
     t  -5.2825724           .   2.5241336   10.275385
pvalue   1.333e-06           .   .01383634   1.085e-15
    ll  -405.24167           .    371.2169   9595.1638
    ul   -183.1494           .   3163.3676   14215.667
    df          71          71          71          71
  crit   1.9939434   1.9939434   1.9939434   1.9939434
 eform           0           0           0           0

. scalar pval = el(r(table),4,1)

. di pval
1.333e-06

[Richard Williams]

Thanks Jeff. I suspected it was something like that. I think it is still probably better to make a copy of the matrix, partly so you don't lose it, and partly because the syntax seems a bit simpler.

[Susan Biancani]

Thanks, Richard. The example you gave is great.
Thanks all for the help!

[Richard Williams]

Sure. I like threads where I learn something (or in this case finally learn something, since Rich Goldstein pointed out r(table) in an earlier thread and I totally missed its usefulness.) You can either learn some matrix manipulation commands now; or extract the values one by one (instead of calculating them) and do whatever it is you were originally planning,

[Jae Li]

You are using the formula for t when you should be using the formula for z. See section 4 of Maarten's paper. See if this works:

local z = _b[foreign]/_se[foreign]
di 2*normal(-abs(`z'))

@Richard Williams Hi Richard, may I ask you a question about the p-values with respect to z statistics and t statistics ?

When reporting z statistics, p-value = 2*normal(-abs(z)) where z = _b[foreign]/_se[foreign]

However, when reporting t statistics, p-value becomes 2*ttail(e(df_r),abs(t)) where t = _b[foreign]/_se[foreign]

Thus, are these above correct? Many thanks in advance!!