Chi -square test with missing value using table1_mc

[Rodrigo Badilla]

Hi all,

I am having problems to compute the chi-square statistics based only on the valid entries (without missing values) using table1_mc from SSC.

The same problem I get with prtest, proportions are calculated with missing values.

* Example generated by -dataex-. For more info, type help dataex
clear
input float(m1 m2 age female dept offcampus) str11 comment

2 2 18 0 1 0 ""           
4 4 20 1 4 0 "Great!"     
2 2 18 1 1 0 ""           
1 1 19 1 1 1 "I'm leaving"
2 2 18 0 1 0 ""           
2 2  . . . 1 ""           
2 2 18 1 . 0 ""           
3 4 20 1 4 1 ""           
4 3 18 1 3 0 ""           
2 2 19 0 2 1 ""           
2 2 19 1 2 1 ""           
2 2 19 1 . 1 ""           
2 2 20 0 4 1 ""           
4 3 20 0 . 1 ""           
2 4 20 1 3 0 ""           
3 2 20 0 . 1 ""           
2 2 19 0 4 1 ""           
2 2 18 1 1 0 ""           
2 2 18 0 1 0 ""           
4 4 18 0 2 0 ""           
4 4 20 0 4 0 ""           
2 2 18 0 1 0 ""           
3 3 19 0 1 0 ""           
2 1 18 1 1 0 ""           
2 2  . . . 0 ""           
1 2 18 0 1 0 ""           
2 2 18 0 4 0 ""           
4 4 21 0 4 0 ""           
2 2 19 1 1 0 ""           
2 2 20 0 1 0 ""           
4 4 20 1 4 0 ""           
2 2 18 0 1 1 ""           
2 2 19 1 4 1 ""           
2 2 18 0 3 1 ""           
2 2 19 0 2 1 ""           
2 2 19 1 1 0 ""           
4 2 19 1 1 0 ""           
2 2 19 0 3 1 ""           
4 4 19 1 4 0 ""           
2 2 19 1 4 1 ""           
2 2 18 1 2 0 ""           
2 2 19 1 3 1 ""           
2 2 19 0 4 1 ""           
3 3 19 1 3 0 ""           
2 2 18 0 . 0 ""           
2 4 19 0 3 0 ""           
2 2 20 1 3 0 ""           
2 2 19 1 4 1 ""           
1 2 19 0 1 1 ""           
2 2 20 0 1 0 ""           
2 2 18 1 1 0 ""           
2 2 19 0 . 0 ""           
2 2 18 0 3 1 ""           
2 2 19 0 1 1 ""           
2 3 19 1 1 0 ""           
2 2 19 1 2 0 ""           
2 2 18 1 4 0 ""           
2 3 20 1 2 0 ""           
2 2 19 0 4 1 ""           
2 2 17 1 1 0 ""           
2 2  . . . 0 ""           
2 2 19 1 1 0 ""           
2 2 19 1 1 1 ""           
4 4 20 0 4 1 ""           
2 2 18 1 3 0 ""           
3 2 18 0 3 0 ""           
2 2 19 1 2 1 ""           
2 2 17 0 3 0 ""           
2 2 19 1 3 0 ""           
2 2 19 0 3 0 ""           
2 3 20 0 4 0 ""           
2 2 20 1 3 1 ""           
2 2 18 0 1 0 ""           
4 2 19 0 1 0 ""           
4 2 20 0 4 0 ""           
2 2 18 1 2 1 ""           
2 2 20 1 4 1 ""           
2 2 18 0 1 1 ""           
2 2 19 1 2 0 ""           
2 2 19 0 1 1 ""           
2 2 19 0 3 1 ""           
2 2 19 1 1 1 ""           
2 2 19 0 1 1 ""           
2 3 21 0 3 1 ""           
4 4 19 1 4 0 ""           
2 1 17 0 1 1 ""           
3 3 19 1 4 0 ""           
4 4 19 1 1 0 ""           
2 2 20 0 4 1 ""           
2 2 19 1 2 0 ""           
4 3 18 1 4 0 ""           
4 3 20 0 2 0 ""           
3 2 18 1 3 0 ""           
2 3 18 1 4 0 ""           
2 2 19 0 1 0 ""           
3 3 18 0 4 0 ""           
3 4 20 1 2 0 ""           
2 2 19 0 2 1 ""           
4 4 19 1 4 0 ""           
2 3 20 1 2 0 ""           
end


tab female dept,  chi2


           |                    dept
    female |         1          2          3          4 |     Total
-----------+--------------------------------------------+----------
         0 |        18          5          9         12 |        44
         1 |        14         10          9         14 |        47
-----------+--------------------------------------------+----------
     Total |        32         15         18         26 |        91

          Pearson chi2(3) =   2.2240   Pr = 0.527



tab female dept, miss chi2

          |                          dept
    female |         1          2          3          4          . |     Total
-----------+-------------------------------------------------------+----------
         0 |        18          5          9         12          4 |        48
         1 |        14         10          9         14          2 |        49
         . |         0          0          0          0          3 |         3
-----------+-------------------------------------------------------+----------
     Total |        32         15         18         26          9 |       100

          Pearson chi2(8) =  34.0972   Pr = 0.000



table1_mc, by(female) miss onecol vars(dept cat)  extraspace statistic test clear

  +--------------------------------+
  | factor   N_0   N_1   m_0   m_1 |
  |--------------------------------|
  | dept      48    49     0     0 |
  +--------------------------------+
   N_ ... #records used below,   m_ ... #records not used
 
  +------------------------------------------------------------------------------+
  |              female = 0   female = 1   Test         Statistic        p-value |
  |------------------------------------------------------------------------------|
  |              N=48         N=49                                               |
  |------------------------------------------------------------------------------|
  | dept                                   Chi-square   Chi2(4)=  2.98    0.56   |
  |    1         18 (38%)     14 (29%)                                           |
  |    2         5 (10%)      10 (20%)                                           |
  |    3         9 (19%)      9 (18%)                                            |
  |    4         12 (25%)     14 (29%)                                           |
  |    Missing   4 (  8%)     2 (  4%)                                           |
  +------------------------------------------------------------------------------+
Data are presented as n (%).

You can check that lamentably table1_mc use missing values to Chi2 test.

[Clyde Schechter]

You can check that lamentably table1_mc use missing values to Chi2 test.

Yes, but it's doing that because you specified the -missing- option in your -table1_mc- command. Remove that and you will get what you want.

[Rodrigo Badilla]

Thanks Clyde Schechter for you reply. Removing miss option

table1_mc, by(female)  onecol vars(dept cat)  extraspace statistic test clear

  +--------------------------------+
  | factor   N_0   N_1   m_0   m_1 |
  |--------------------------------|
  | dept      44    47     4     2 |
  +--------------------------------+
   N_ ... #records used below,   m_ ... #records not used
 
  +--------------------------------------------------------------------------+
  |          female = 0   female = 1   Test         Statistic        p-value |
  |--------------------------------------------------------------------------|
  |          N=48         N=49                                               |
  |--------------------------------------------------------------------------|
  | dept                               Chi-square   Chi2(3)=  2.22    0.53   |
  |    1     18 (41%)     14 (30%)                                           |
  |    2     5 (11%)      10 (21%)                                           |
  |    3     9 (20%)      9 (19%)                                            |
  |    4     12 (27%)     14 (30%)                                           |
  +--------------------------------------------------------------------------+



Only remove missing frequencies and yes change a little the chi-square p-value not was my expected result.

Because results of



	
	
 tab female dept, miss chi2
give a p-value=0.0000, I can not get this result with table1_mc

           |                          dept
    female |         1          2          3          4          . |     Total
-----------+-------------------------------------------------------+----------
         0 |        18          5          9         12          4 |        48
         1 |        14         10          9         14          2 |        49
         . |         0          0          0          0          3 |         3
-----------+-------------------------------------------------------+----------
     Total |        32         15         18         26          9 |       100

          Pearson chi2(8) =  34.0972   Pr = 0.000

[Clyde Schechter]

I'm not following you. When you remove the missing option from the -table1_mc- command you are getting:

  +--------------------------------------------------------------------------+
  |          female = 0   female = 1   Test         Statistic        p-value |
  |--------------------------------------------------------------------------|
  |          N=48         N=49                                               |
  |--------------------------------------------------------------------------|
  | dept                               Chi-square   Chi2(3)=  2.22    0.53   |
  |    1     18 (41%)     14 (30%)                                           |
  |    2     5 (11%)      10 (21%)                                           |
  |    3     9 (20%)      9 (19%)                                            |
  |    4     12 (27%)     14 (30%)                                           |
  +--------------------------------------------------------------------------+

and -tab female dept, chi2- gives you

           |                    dept
    female |         1          2          3          4 |     Total
-----------+--------------------------------------------+----------
         0 |        18          5          9         12 |        44
         1 |        14         10          9         14 |        47
-----------+--------------------------------------------+----------
     Total |        32         15         18         26 |        91

          Pearson chi2(3) =   2.2240   Pr = 0.527

and these results are the same except that -tab- gives you 3 decimal places in the p-value whereas table1_mc gives you 2.

So, what is the problem?

[Rodrigo Badilla]

I expect get this chi- square result whith table1_mc:

 tab female dept, miss chi2        

           |                          dept
    female |         1          2          3          4          . |     Total
-----------+-------------------------------------------------------+----------
         0 |        18          5          9         12          4 |        48
         1 |        14         10          9         14          2 |        49
         . |         0          0          0          0          3 |         3
-----------+-------------------------------------------------------+----------
     Total |        32         15         18         26          9 |       100

          Pearson chi2(8) =  34.0972   Pr = 0.000

Pr=0.0000, Not Accept H0

and whith table1_mc chi-square test:

Pr=0.53 or 0.56, I Accept H0

[Clyde Schechter]

Oh, I see what's going on. The problem is that your data is not compliant with the requirements of table1_mc. From its help file:

      by(varname)           group observations by varname, which must be either (i) string, or (ii) numeric and contain only non-negative integers, whether or not a
                              value label is attached

[emphasis added]
You can't have missing values in the -by()- variable of table1_mc. And, in fact, looking carefully at the output it's giving you, you can see that it is taking the missing values of variable dept into account, but ignoring those of variable female in doing its calculations.

You can work around this by something like this:

clonevar female2 = female
replace female2 = 9 if missing(female)
label define female2 0 "Male"    1    "Female"    9    "Missing"
label values female2 female2
tab female2 dept, miss chi2
table1_mc, by(female2) miss onecol vars(dept cat)  extraspace statistic test

By the way, -table1_mc- is not an official Stata command. When posting about user-written commands, it is best practice to mention that fact and indicate where it comes from. In this case, it is by Mark Chatfield and available from SSC.

[Rodrigo Badilla]

Thanks Clyde Schechter for your reply.

I thought that miss option in table1_mc was like miss in tabulate x y, miss chi2.

I will use your command.

Regards
Rodrigo