Can’t get a p-value with SVY

[Jen Brooks]

Hello,

I’m running some Chi2 test with weighted ACS data but when I run the pearson test I don’t get a p-value. Anyone else having this problem? Am I doing something wrong?

Output

. svy: tab employed female if disability==1, col pearson
(running tabulate on estimation sample)

Number of obs = 4,027
Population size = 421,744
Replications = 80

-------------------------------
RECODE of | RECODE of sex (SEX)
esr (ESR) | 0 1 Total
----------+--------------------
0 | .6829 .6695 .6761
1 | .3171 .3305 .3239
|
Total | 1 1 1
-------------------------------
Key: column proportion

Pearson:
Uncorrected chi2(1) = 0.8295

.
end of do-file

[Carlo Lazzaro]

Jen:
welcome to this forum.
Actually, Stata seems to give back the p-value for the Design-based Pearson test only:

use https://www.stata-press.com/data/r17/nhanes2.dta
. svy: tab houssiz sex if race ==1, col pearson
(running tabulate on estimation sample)

Number of strata = 31                            Number of obs   =       9,065
Number of PSUs   = 62                            Population size = 102,999,549
                                                 Design df       =          31

-------------------------------------
Number of |
people in |            Sex           
household |    Male   Female    Total
----------+--------------------------
        1 |   .1285    .1584    .1441
        2 |   .2961    .2982    .2972
        3 |   .1893    .1862    .1877
        4 |   .2032    .1804    .1914
        5 |   .1029    .1017    .1023
        6 |   .0472    .0432    .0452
        7 |   .0227     .019    .0208
        8 |    .005    .0069     .006
        9 |   .0039    .0029    .0034
       10 | 4.4e-04    .0013  8.8e-04
       11 | 4.4e-04  1.9e-04  3.1e-04
       12 |       0    .0011  5.5e-04
       13 | 5.8e-05  3.6e-04  2.2e-04
       14 | 2.0e-04        0  9.8e-05
          | 
    Total |       1        1        1
-------------------------------------
Key: Column proportion

  Pearson:
    Uncorrected   chi2(13)        =   33.6073
    Design-based  F(8.34, 258.44) =    2.2233     P = 0.0243

. ereturn list

scalars:
               e(df_r) =  31
      e(N_strata_omit) =  0
          e(singleton) =  0
             e(census) =  0
             e(stages) =  1
           e(cun_Penl) =  33.60733785429137
             e(F_Penl) =  2.148822234417541
           e(df1_Penl) =  8.775157075049131
           e(df2_Penl) =  272.0298693265231
             e(p_Penl) =  .0270691984810268
           e(cun_Pear) =  33.60733785429137
             e(F_Pear) =  2.223277741548805
           e(df1_Pear) =  8.336802397765183
           e(df2_Pear) =  258.4408743307207
                  e(N) =  9065
           e(N_strata) =  31
              e(N_psu) =  62
              e(N_pop) =  102999549
               e(rank) =  25
            e(cvgdeff) =  .6938695576311746
             e(mgdeff) =  1.203068263879086
              e(total) =  102999549
                  e(c) =  2
                  e(r) =  14
            e(pun_LLW) =  .
            e(Fun_LLW) =  .
              e(p_LLW) =  .
              e(F_LLW) =  .
            e(cun_LLW) =  .
               e(p_LR) =  .
             e(df2_LR) =  258.4408743307207
             e(df1_LR) =  8.336802397765183
               e(F_LR) =  .
             e(cun_LR) =  .
             e(p_LRnl) =  .
           e(df2_LRnl) =  272.0298693265231
           e(df1_LRnl) =  8.775157075049131
             e(F_LRnl) =  .
           e(cun_LRnl) =  .
             e(p_Pear) =  .0243278991288622

That said:
1) -ereturn- is an useful command to retrieve Stata results that are not displayed in the outcome tables;
2) whenever Stata does not allow researchers to do a given statistical procedure/return a given result, there are sound methodological reasons for that.

[Jen Brooks]

Hi Carlo,

Thanks for your help. I followed your suggestion but Stata will not produce a Design-based Pearson test. Any thoughts on why? I’ll do some more research myself.


.
. svy: tab lfp disability, col pearson
(running tabulate on estimation sample)

Number of obs = 47,182
Population size = 5,527,953
Replications = 80

-------------------------------
RECODE of | RECODE of disall
esr (ESR) | 0 1 Total
----------+--------------------
0 | .2178 .6307 .2493
1 | .7822 .3693 .7507
|
Total | 1 1 1
-------------------------------
Key: column proportion

Pearson:
Uncorrected chi2(1) = 3028.8469

. ereturn list

scalars:
e(N_strata_omit) = 0
e(singleton) = 0
e(census) = 0
e(stages) = 1
e(N_reps) = 80
e(N_misreps) = 0
e(k_exp) = 0
e(k_eexp) = 4
e(k_extra) = 0
e(cun_Penl) = 3028.846891525951
e(cun_Pear) = 3028.846891525951
e(cun_LRnl) = 2566.433340678475
e(cun_LR) = 2566.433340678475
e(cun_LLW) = 1793.301345060757
e(N) = 47182
e(N_strata) = .
e(N_pop) = 5527953
e(rank) = 4
e(cvgdeff) = 0
e(mgdeff) = 2.567150308285274
e(total) = 5527953
e(c) = 2
e(r) = 2
e(N_psu) = .

[Carlo Lazzaro]

Jen:
in your case -ereturn- does not report degrees of freedom, hence p-value cannot be calculated (why this happens I do not know).
In addition, I was wrong, because -ereturn- (in my example, though) does report the p-value for the Uncorrected Pearson's test:
:

 e(p_Penl) =  .0270691984810268

Sorry that I cannot be more helpful.

[David Radwin]

Can you use method #3 of https://stats.oarc.ucla.edu/stata/fa...th-survey-data instead of a chi-square test?