0
$\begingroup$

I want to compare the frequency of a categorical value among 4 groups, what statistical test should I use? (I am using SPSS). I am looking for a statistical test that would allow me to say: the frequency of value "V" depends on the group and the groups' frequencies are statistically different for that value.

I write here an example: Group 1 shows value "V" 10 times, Group 2 shows value "V" 15 times... Group 4 shows value "V" 40 times.

Thanks a lot in advance.

$\endgroup$
1
  • $\begingroup$ It is hard to understand your question. If you mean you want to test if "V" appears in 4 groups equally, then use multinomial test to test null hypothesis $H_0: \pi_1=0.25, \pi_2=0.25, \pi_3=0.25, \pi_4=0.25$ $\endgroup$
    – user158565
    Commented Aug 9, 2019 at 15:37

1 Answer 1

0
$\begingroup$

Results will depend on the numbers of non-v events in each group.

If data are as below, with counts of v in a and non-v in b, so that the total number of events in each group is 100, then the null hypothesis that probabilities of v are homogeneous among groups is overwhelming rejected with P-value nearly 0. The test is a chi-squared test of homogeneity. (Output from R.)

a = c(10,15,50,40)
b = c(90,85,50,60)
DTA = rbind(a,b)
DTA
  [,1] [,2] [,3] [,4]
a   10   15   50   40
b   90   85   50   60

chisq.test(DTA)

        Pearson's Chi-squared test

data:  DTA
X-squared = 54.615, df = 3, p-value = 8.296e-12

By contrast, if the counts are as shown below, so that the proportion of v events is roughly 1/3 in all groups, then a chi-squared test for homogeneity comes nowhere near rejection:

a = c(10,15,50,40)
b = c(30,50,160,125)
DTA = rbind(a,b)
DTA
  [,1] [,2] [,3] [,4]
a   10   15   50   40
b   30   50  160  125
chisq.test(DTA)

        Pearson's Chi-squared test

data:  DTA
X-squared = 0.061404, df = 3, p-value = 0.996

Such chi-squared tests find expected counts for each cell (in your case $2*4=8$ cells). Then the expected counts are compared with observed counts.

For the second test above, here are the observed and expected counts, For a valid test, all expected counts must exceed 3, and almost all should exceed 5. (No difficulty here, because the smallest expected count is almost 10.)

v.out = chisq.test(DTA)
v.out$obs
  [,1] [,2] [,3] [,4]
a   10   15   50   40
b   30   50  160  125
v.out$exp
       [,1]     [,2]     [,3]      [,4]
a  9.583333 15.57292  50.3125  39.53125
b 30.416667 49.42708 159.6875 125.46875

Notes: (a) This test of homogeneity of variances is mathematically identical to a test of indepencence of v/non-v and your categories--even though the phrasing of the interpretation of results may be different. (b) In such a chi-squared test, it is important to compare counts, not proportions. (c) Also, you must compare counts of v against counts of non-v; not counts of v against total group counts. (d) Other methods of analyzing such data are possible, but I think you will find these chi-squared tests in SPSS.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.