I have two distinct groups made of a different numbers of subjects (111 Millions and 126 Millions). My goal is to evaluate how many subjects in the two groups encounter 10 different diseases. For this purpose, I build a table (reporting here only 4 out of 10 diseases) as follows:

Disease Group A Group B
A 23 M 19 M
B 45 M 18 M
C 19 M 18 M
D 21 M 20 M

In this case there are no means involved: I'm simply counting the number of occurrences (frequency) within each group and for each disease. Is there a way to check whether the the difference is statistically significant for each disease between the two groups? I would proceed with a Chi-squared test isolating each disease, building a contingency table as follows and then run the test.

Disease A Group A Group B Sum
Infected 23 M 19 M 42 M
Not-Infected 88 M 107 M 195 M
Sum 111 M 126 M

Is, in this case the Chi-squared test, the most appropriate test?


With such large samples, do you think results of a statistical test adds much value?

Formally, you could used prop.test in R (or essentially equivalently, a chi-squared test on a 2-by-2 table).

prop.test(c(23*10^6,19*10^6), c(111*10^6,126*10^6), cor=F)

        2-sample test for equality of proportions 
        without continuity correction

data:  c(23 * 10^6, 19 * 10^6) out of c(111 * 10^6, 126 * 10^6)
X-squared = 1288000, df = 1, p-value < 2.2e-16
alternative hypothesis: two.sided
95 percent confidence interval:
 0.05631563 0.05651148
sample estimates:
   prop 1    prop 2 
0.2072072 0.1507937 

With millions of subjects is there any doubt that proportions $0.2091$ and $0.1508$ differ? Or does a P-value (predictably) extremely near $0$ somehow seem impressive--or tick someone's supposedly mandatory box?

Note: Output for chisq.test in R follows:

TBL = 10^6*matrix(c(23,19,88,107), byrow=T, nrow=2)
        [,1]     [,2]
[1,] 2.3e+07 1.90e+07
[2,] 8.8e+07 1.07e+08

chisq.test(TBL, cor=F)

         Pearson's Chi-squared test

data:  TBL
X-squared = 1288000, df = 1, p-value < 2.2e-16
  • $\begingroup$ Thanks for your answer. I was only wondering whether other comparisons were possible besides the simple Chi-squared $\endgroup$
    – SavioD
    Apr 7 at 14:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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