individual p-values- chi square test

[ching wong]

Hi

I used the simple chi square test command, tab var1 var2, chi2. It gives the overall p-value (0.000)

What command should I use in order to get the individual p-value of "No group" & the p-value of "Yes"group. (i.e. as below)



| Unknown Highest D High Disa Moderate Less Disa Least Dis | Total
-----------+------------------------------------------------------------------+----------
No | 56 658 634 564 537 553 | 3,002
Yes | 5 900 869 883 797 848 | 4,302
-----------+------------------------------------------------------------------+----------
Total | 61 1,558 1,503 1,447 1,334 1,401 | 7,304

Pearson chi2(5) = 70.9560 Pr = 0.000


Thanks

[Nick Cox]

Such P-values are, as a matter of principle, not defined. Separate P-values for each row are only defined with a test of a different hypothesis for that row alone.

[Marcos Almeida]

Perhaps I didn't get the point in #1.

However, shall it be the chi-square goodness of fit test, where say, one wishes to "test"he distribution of a variable (under a "if" clause") , I wonder whether we could consider the "no" group under a univariate frequency distribution.

If so ( and I'm not absolutely sure about that), the SJ/STB - tab_chi - (Nick Cox) might provide what Ching wishes:

. set obs 6
number of observations (_N) was 6, now 6

. input race sex freq

          race        sex       freq
  1. 0 0 12
  2. 0 1 84
  3. 1 0 11
  4. 1 1 13
  5. 2 0 14
  6. 2 1 11

. label define race 0 "White" 1 "Black" 2 "Other"

. label define sex 0 "Female" 1 "Male"

. label values sex sex

. label values race race

. expand freq
(139 observations created)

. tab sex race, chi2

           |               race
       sex |     White      Black      Other |     Total
-----------+---------------------------------+----------
    Female |        12         11         14 |        37 
      Male |        84         13         11 |       108 
-----------+---------------------------------+----------
     Total |        96         24         25 |       145 

          Pearson chi2(2) =  25.9934   Pr = 0.000

. chitest race if sex ==1

Chi-square test:
    observed frequencies from race
    expected frequencies equal

         Pearson chi2(107) = 140.8857   Pr =  0.016
likelihood-ratio chi2(107) = 109.3733   Pr =  0.418

                                           residuals
     +-------------------------------------------------+
     | observed   expected   notes   classic   Pearson |
     |-------------------------------------------------|
  1. |        0      0.324      **    -0.324    -0.569 |
  2. |        1      0.324      **     0.676     1.187 |
  3. |        2      0.324      **     1.676     2.944 |
  4. |        0      0.324      **    -0.324    -0.569 |
  5. |        0      0.324      **    -0.324    -0.569 |
     |-------------------------------------------------|
  6. |        0      0.324      **    -0.324    -0.569 |
  7. |        0      0.324      **    -0.324    -0.569 |
  8. |        0      0.324      **    -0.324    -0.569 |
  9. |        0      0.324      **    -0.324    -0.569 |
 10. |        0      0.324      **    -0.324    -0.569 |
     |-------------------------------------------------|
  (output omitted)
     |-------------------------------------------------|
101. |        2      0.324      **     1.676     2.944 |
102. |        2      0.324      **     1.676     2.944 |
103. |        2      0.324      **     1.676     2.944 |
104. |        2      0.324      **     1.676     2.944 |
105. |        2      0.324      **     1.676     2.944 |
     |-------------------------------------------------|
106. |        2      0.324      **     1.676     2.944 |
107. |        2      0.324      **     1.676     2.944 |
108. |        2      0.324      **     1.676     2.944 |
     +-------------------------------------------------+

** 0 <  expected < 1

I hope to have helped. Shall this strategy be incorrect, I also hope to learn from that.

Best,

Marcos

[Marcos Almeida]

Editing the second part of the previous messsage:

. set obs 3
number of observations (_N) was 0, now 3

. input male

          male
  1. 84
  2. 13
  3. 11

. chitest male

Chi-square test:
    observed frequencies from male
    expected frequencies equal

         Pearson chi2(2) =  96.0556   Pr =  0.000
likelihood-ratio chi2(2) =  89.7795   Pr =  0.000

                                   residuals
     +-----------------------------------------+
     | observed   expected   classic   Pearson |
     |-----------------------------------------|
  1. |       84     36.000    48.000     8.000 |
  2. |       13     36.000   -23.000    -3.833 |
  3. |       11     36.000   -25.000    -4.167 |

[Nick Cox]

Marcos: Your posts and mine (#2) are consistent. You could test each row of the table against a hypothesis of equal frequencies, or any other hypothesis about those frequencies alone. Somehow I doubt this is the question, but the OP has not returned to the thread.

For the record tab_chi is available from SSC. It has never been published through the STB or Stata Journal.

[Marcos Almeida]

Thanks for the feedback Nick! Also, thanks for the correction on the source (SSC) of tab_chi.

Below, the "reason" I thought it was SJ/STB. Due to discraction, I just read the information on the first line:

2 packages found (Stata Journal and STB listed first)
-----------------------------------------------------

tab_chi from http://www.stata.com/users/njc
tab_chi Tabulation and chi-square test programs. / Programs by Nicholas J.
Cox, Durham University <[email protected]>. / Statalist distribution,
22 March 1999. / See help tab_chi

tab_chi from http://fmwww.bc.edu/RePEc/bocode/t
'TAB_CHI': modules for tabulation of multiple variables in Stata 8.2 or
better / chitest and chitesti carry out chi-square tests for univariate /
frequency distributions. chitesti is the immediate command / version of
chitest. Both require Stata 8.2; earlier versions are / available as

[Nick Cox]

Indeed. I didn't mention the prehistoric version on the Stata website itself. But (e.g.) results like that below (truncated) imply that anyone watching with Stata 5 should download from there.

. type http://www.stata.com/users/njc/tab_chi/chitest.ado
*! 1.4.4 NJC 26 January 1999
* 1.4.3  2 Sept 1997
* 1.4.2  13 August 1997
* 1.4.2, 1.4.3 indicate keyboard entry (undocumented option)
* 1.4.1  30 April 1997     save emean in S_7
* 1.4.0  20 December 1996
program def chitest
    version 5.0