# Statistical Significance Dependent Populations

I am hoping to understand best way to test statistical significance between 2 dependent population groups.

For example, consider a usability test. When 100 subjects were tested, 50 of them clicked (=50% click rate). However, 50 of the subjects were male, 40 of whom clicked for an 80% click rate for males.

The question is that 80% statistically significant? In other words, do men click more than the population as a whole? I think I need to use a paired $t$-test, however unsure as what I would use as the mean, since these are all population proportions.

• Do you mean 50 of the 100 subjects? And significantly different from what? Do you want to know if the part is different from the whole?...the remaining part?
– John
Commented Sep 5, 2013 at 4:17
• "The question is that 80% statistically significant?" against what null? It seems like you want a two sample test, but you must clarify. Commented Sep 5, 2013 at 5:14
• The statement I want to make is that males click rate is 700% better then those who are not male. (40/50)-(((50-40)(100-50)))/(40/50)). However if the males subject population was only 1, and that 1 person clicked, i would have a much larger value, but it wouldn't be significant because 1 of 1 is not large enough to be deemed statistically significant. Commented Sep 5, 2013 at 5:43
• Do you mean that 40/50-(50-40)*(100-50)/(40/50)=-624.2? The male click rate is 40/50=80%, and the female click rate is (50-40)/(100-50)=20%? If 1 of 1 male clicked, you mean that you will have a larger male click rate (from 80% to 100%), but what becomes insignificant? male click rate = 0 or male click rate = female click rate, or others? Please edit your question to clarify. Commented Sep 5, 2013 at 7:07

You seem to have gone on a convoluted route to asking for how to assess independent count data. You've got 100 independent items. There are 40 males that clicked, 10 males that didn't click, 10 non-males that clicked, and 40 non-males that didn't click. You can easily construct what is called a contingency table (below) from those data and do a $\chi^2$ (chi-square) test for independence.

         male non-male
click    40   10
no click 10   40


Searching for the chi-square test on the internet will show you the formulas, logic, and even online calculators that can solve the problem.

• Sorry I don't have a lot of stats knowledge but from my general understanding, the populations are dependent since the male population is a subset of the total population of subjects. If the populations were independent could I use a binomial approx and check for the result being within 3 std deviations? Sorry I am totally out of my element here, but I am trying to determine statistical significance for a bunch of different usability tests where I want to know if the subset of users proportion is statistically significant to the larger population of subjects. Commented Sep 5, 2013 at 17:57
• Yes, the males are a subset of the total population but given that kind of a rule for dependence any values are dependent. Imagine I rolled a die 20 times and got 4 6's. The number 4 is dependent on the fact that I rolled 20 times. But none of those rolls is dependent on any other roll. They're independent rolls. And yes, you could use a binomial approximation and reporting a CI would be a good idea. However, that runs into some judgments on what's conservative and not. I believe it's a better way to report results but not a better test than $\chi^2$.
– John
Commented Sep 5, 2013 at 18:18