chi2 explanation and cchi2 contribution

[Tommaso Salvitti]

Hello everyone, I find myself in a situation of great difficulty in commenting on the data in this table in which I applied the chi2 test I have three type of school (A,B e C) and 4 tipe of school subjects (1,2,3,4) independent of each other
I used the instruction written below and I added the "cchi2" option to understand which box gave rise to the significance of chi2

 tab school subject, row column cell chi2 exact cchi2

My output is

Type of school	1	2	3	4	TOTAL
A	1 0.00 33.33 2.56 0.74	0 0.1 0.00 0.00 0.00	2 3.5 66.67 7.69 1.47	0 1.5 0.00 0.00 0.00	3 5.1 100.00 2.21 2.21
B	17 2.3 41.46 43.59 12.50	1 0.00 2.44 25.00 0.74	14 4.8 34.15 53.85 10.29	9 6.2 21.95 13.43 6.62	41 13.4 100.00 30.15 30.15
C	21 1.1 22.83 53.85 15.44	3 0.0 3.26 75.00 2.21	10 3.3 10.87 38.46 7.35	58 3.5 63.04 86.57 42.65	92 7.9 100.00 67.65 67.65
TOTAL	39 3.5 28.68 100.00 28.68	4 0.2 2.94 100.00 2.94	26 11.7 19.12 100.00 19.12	67 11.2 49.26 100.00 49.26	136 26.5 100.00 100.00 100.00

Pearson chi2(6) = 26.5116 Pr=0.000
Fisher's exact= 0.000

the threshold value corresponding to six degrees of freedom is 12.59. i don't have this value in the table..How to explain the significance of the test? can everybody help me?
Many thanks in advance

[Nick Cox]

I find so-called Pearson residuals much easier to work with. These are (observed - expected) / root of expected. A rule of thumb is that values more than 2 in absolute value indicate notable departures from null expectation. See tabchi from tab_chi on SSC for one implementation.

[Tommaso Salvitti]

you are right!!!
but I rephrased the question better...despite this no one answers me...Is it possible that I'm always wrong?

[Tommaso Salvitti]

i don't understand and how to use the instruction?

[Tommaso Salvitti]

I use this instruction

tabchi school subject, adj noo noe

my output is:

Type of school	1	2	3	4
A	2.842	-0.728	2.930	-4.608
B	-0.132	-0.358	2.845	-2.00
C	-2.808	0.799	-3.499	5.002

4 cells with expected frequency < 5
2 cells with expected frequency < 1

Pearson chi2(6) = 34.9316 Pr = 0.000
likelihood-ratio chi2(6) = 34.6548 Pr = 0.000


HOW TO INTERPRET THE OUTPUT? POST-HOC COMPARISONS? THANK YOU VERY MUCH FOR HELPING

[Nick Cox]

Clear association between your variables. Positive residuals = more than expected under null, and so on.

[Tommaso Salvitti]

and if I wanted to make comparisons with a couple what could I say?

[Tommaso Salvitti]

I mean 2vs1; 3vs1; 4vs1; 3vs2; 4vs2; 4vs3...What could i say?thanks a lot

[Tommaso Salvitti]

But I want to know which column proportions are significantly different from the others. IF SOMEBODY CAN HELP ME

[Tommaso Salvitti]

Thanks to everybody for the help

[Tommaso Salvitti]

Thank you..@NickCox#6...and according to you and to everybody...is it correct use chisquared even if some cells contain values like 0?
Thanks million in advanced
Tommaso

[Tommaso Salvitti]

I ask this question because i think that if we use the Fisher exact test...we cannot use tab_chi...Is it true?

[Nick Cox]

#11 I see no problem in observed frequencies of zero. Expected frequencies of zero are a different story. I don't know if that answers your question.

#12 I don't understand your question here. tab_chi is a package. Some of its commands have nothing to do with the Fisher test and others allow it in the standard way by allowing tabulate options. The package does not add to, subtract from, or inhibit or prohibit using what is already available in official Stata for Fisher's test.

[Tommaso Salvitti]

Ok i ask this question because for my knowledge i thought that csquared test mus be used when the cell cointain non zero..probably i was wrong... thanks a lot...if if i use fisher...for post hoc test i 'll do comparison for each group i think it is correct...thanks a lot...

[Nick Cox]

You're touching on questions of statistical style and taste. I don't myself ever follow chi-square testing with other tests. Either there's a pattern which is worth talking about and a chi-square test is made public to disarm sceptics. Or there is a need to move to proper modelling, usually a loglinear model. Lots and lots of little tests are, in my experience, not usually productive scientifically or statistically and raise doubts about, as you say, what might be called post hockery.