Comparison proportion of missingness among groups

[Tu Kim]

Dear Statalists,

Is there anyway to check for the difference of missingness among groups?

For e.g in the photo below, the missingness of the outcome "sexual victimization" are different among three highschool (20%, 15%, and 12%). How can I check whether these proportions of missingness of three highschool are significantly different?

Many thanks for your ideas.

Kind regards,
Kim Tu

[Carlo Lazzaro]

Tu:
you may want to coinsider something like:

logit missing i.highschool

What above requires that you create a variable -missing- beforehand.
As I do not know the format of your dataset (long or wide) I cannot help you on that.

[Nick Cox]

You could just do a chi-square test on missing versus not missing as an outcome. Here I use tabchii from tab_chi on SSC, which has some more bells and whistles than tabi.

As would @Carlo Lazzaro's similar suggestion to use logit, there is an underlying assumption that the observations are independent (and so for example, not panel data with individuals followed through time).

. clear 

. tabchii 65 251 \ 48 264 \ 38 268, replace

          observed frequency
          expected frequency

----------------------------
          |       col       
      row |       1        2
----------+-----------------
        1 |      65      251
          |  51.088  264.912
          | 
        2 |      48      264
          |  50.441  261.559
          | 
        3 |      38      268
          |  49.471  256.529
----------------------------

         Pearson chi2(2) =   7.8329   Pr = 0.020
likelihood-ratio chi2(2) =   7.7702   Pr = 0.021

. d

Contains data
 Observations:             6                  
    Variables:             8                  
---------------------------------------------------------------------------------
Variable      Storage   Display    Value
    name         type    format    label      Variable label
---------------------------------------------------------------------------------
row             byte    %8.0g                 
col             byte    %8.0g                 
observed        long    %12.0g                observed frequency
expected        double  %9.3f                 expected frequency
raw             double  %9.3f                 raw residual
Pearson         double  %9.3f                 Pearson residual
cont            double  %9.3f                 contribution to chi-square
adjusted        double  %9.3f                 adjusted residual
---------------------------------------------------------------------------------
Sorted by: row  col
     Note: Dataset has changed since last saved.

[Tu Kim]

Dear Carlo and Nick,

Thanks a lot for your comments. I appreciate it greatly!

Kind regards,
Kim Tu