Statistical Test Question

[Adalbertus]

I am looking for the correct statistical test and didn't find it yet.

My data consist of two groups of people for which I have yes/no answers. My question is whether they differ significantly.

As a more real-world example, imagine two football clubs. Of each member, you know whether he smokes or not. The question would be if one of the clubs attracts significantly more smokers than the other.

Good ideas are welcome.

[Proteus_MST]

Maybe a t-test?
But no, wait this would require a normal distribution

[Eli]

Re: Statistical Test Question

Originally posted by Adalbertus
As a more real-world example, imagine two football clubs. Of each member, you know whether he smokes or not. The question would be if one of the clubs attracts significantly more smokers than the other.

Good ideas are welcome.

Umm... You look what percentage of people from each group answered yes?

[Adalbertus]

@Eli, That's what I know. What I want to know is if this has a statistical significance.

[Kuciwalker]

Depends on the sample size and the difference in response rate...

[Meticulous Man]

I don't know, is a test of significance needed in this case? Maybe you could just say something like "members of club B are 15% more likely to be smokers so there are strong grounds for the assumption that..." IIUC, when testing statistical significance you still always have to arbitrarily choose a level of significance (5%, 1%...)

Disclaimer: I'm no expert on Statistics

[Main_Brain]

Simple Binominal test with true propablility of 29% for smoking.

[Lul Thyme]

This should give you an idea to start :

http://en.wikipedia.org/wiki/Statistical_hypothesis_testing#Common_test_statistics

It mostly depends on sample size and what other assumptions you can make.

[Meticulous Man]

I think Adalbertus has a scenario where the whole population has been examined...

Of each member, you know whether he smokes or not. The question would be if one of the clubs attracts significantly more smokers than the other.

[CrONoS]

Which software do you use?

[CrONoS]

Wait; I'll open my stat book and I'll come back

[CrONoS]

The best test that you can do is: Chi-square test

A t test is inappropriate for this one.

http://en.wikipedia.org/wiki/Chi-square_test

[DrSpike]

Thinking about it in terms of tests is not really the most helpful. As has been given earlier, the distributions in question are binomial, which is the key.

Once you have constructed the two binomial distributions, using the sample probabilities of 'success' (here smoking) there are several sensible approaches to testing whether those probabilities differ. A simple chi-square test like that linked may be best if you have little data - other tests may have more power if you have more.

[Nikolai]

[snoopy369]

Chi square test is definitely my suggestion as well given the available data.