test of proportions using prtest

[Lara ingram]

Hi statalist,

I am trying to run a two sample test of proportion using two groups (women who report physical abuse versus women who report emotional abuse):

Both variables are binary and take the value of 1 for those who report abuse.

 prtest emo2, by (physical)

Two-sample test of proportions                     0: Number of obs =     8172
                                                   1: Number of obs =     2839
------------------------------------------------------------------------------
       Group |       Mean   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
           0 |   .0527411   .0024725                       .047895    .0575872
           1 |   .4938359   .0093833                       .475445    .5122267
-------------+----------------------------------------------------------------
        diff |  -.4410948   .0097036                     -.4601134   -.4220761
             |  under Ho:   .0081151   -54.35   0.000
------------------------------------------------------------------------------
        diff = prop(0) - prop(1)                                  z = -54.3547
    Ho: diff = 0

    Ha: diff < 0                 Ha: diff != 0                 Ha: diff > 0
 Pr(Z < z) = 0.0000         Pr(|Z| > |z|) = 0.0000          Pr(Z > z) = 1.0000

I am trying to evaluate whether those who experience physical abuse also report emotional abuse.

From what I understand, prop(0) - prop (1) is the proportion of women who did not report physical violence but reported psychological violence minus the proportion of women who did report physical violence and also reported psychological violence. The hypothesis that women who experience physical violence are also more likely to report emotional violence is ha: diff<0 and the very low p value associated with it suggests we should reject the null and accept that alternative hypothesis.

I am a bit confused and I am unsure if this is correct?

[Carlo Lazzaro]

Lara:
I think you would be better off with:
-label-ling your group;
-then going -logistic- or -logit-.

[Lara ingram]

Thank you Carlo. I went with bi-variate analysis using logit and then used the pearson chi2 to check if there is a statistically significant relationship between the two types of violence. However, I get quite a large chi2 value, is this a cause for concern?

 svy: ta emotional physical, pearson
(running tabulate on estimation sample)

Number of strata   =         1                  Number of obs     =     11,011
Number of PSUs     =    11,011                  Population size   =  1,604,696
                                                Design df         =     11,010

-------------------------------
          |      physical      
emotional |     0      1  Total
----------+--------------------
        0 | .6911   .153  .8441
        1 | .0346  .1213  .1559
          | 
    Total | .7258  .2742      1
-------------------------------
  Key:  cell proportion

  Pearson:
    Uncorrected   chi2(1)         = 2592.6503
    Design-based  F(1, 11010)     =  259.5581     P = 0.0000

[Carlo Lazzaro]

Lara:
no, I do not think that this is a cause of concern.
I we go back to chi2 formula, we see that the numerator of the running sum is squared; hence, it is easy to reach sky-rocketing values.

[Lara ingram]

Thank you for the help Carlo.

[Blue Sky]

Carlo Lazzaro after the results Lara posted, how would she calculate an effect size for the Design-based and uncorrected chi2?

[Blue Sky]

Carlo Lazzaro after the results Lara posted, how would she calculate an effect size for the Design-based and uncorrected chi2?

Thank you Carlo. I went with bi-variate analysis using logit and then used the pearson chi2 to check if there is a statistically significant relationship between the two types of violence. However, I get quite a large chi2 value, is this a cause for concern?

 svy: ta emotional physical, pearson
(running tabulate on estimation sample)

Number of strata = 1 Number of obs = 11,011
Number of PSUs = 11,011 Population size = 1,604,696
Design df = 11,010

-------------------------------
| physical
emotional | 0 1 Total
----------+--------------------
0 | .6911 .153 .8441
1 | .0346 .1213 .1559
|
Total | .7258 .2742 1
-------------------------------
Key: cell proportion

Pearson:
Uncorrected chi2(1) = 2592.6503
Design-based F(1, 11010) = 259.5581 P = 0.0000