anyone any good at statistical tests.

[HauntedLittleLunatic]

If I have a survey as follows:

The answer to q1 can be male or female.
The answer to question 2 can be yes or no.

My hypothesis would be that if you answer female to q1 you are more likely to answer yes to q2 (i know that's not a null hypothesis but hey ho!).

How can test this statistically with 2 sets of categorical data?

Hypothetical sample size approx 1000.
Hypothetical 50/50 male female response to q1
Hypothetical response rate of 10% yes overall to q2.

Hope that isn't too cryptic (not intended to be...but it is a hypothetical research proposal I'm working on and I have to say how I will analyse the data....)

Hope you can see why its in geeky stuff...

Can you not just find the percentage of males that say yes and the percentage of females that say yes, the compare the percentages?

Is that oversimplifying it? That's the sort of answer I would expect my GCSE Statistics pupils to come up with so probably quite basic.

I think you say for Q2, "let the null hypothesis = yes", and then you try to prove or disprove this.

I could do that jodidi, but I would need to test to see if the differenxes in percentages were statistically significant. This is for a masters project so I think they will expect more than from gcse students, but I could hope :o

I am used to analysing numerical data which I can stick into a normal distribution etc. so categorical data is alien to me.

Graph pad has a good online calculator for 'fishers exact test' - I think this should give you what you need.

Graph pad has a good online calculator for 'fishers exact test' - I think this should give you what you need.

I really knew I was oversimplifying but I just like to join in. Glad some people with more recent experience with proper statistics have turned up.

www.graphpad.com/welcome.htm

www.graphpad.com/welcome.htm

Erm, Google 'Chi Squared test' - decade old memories of studying stats makes me think that's what you want.

Just had a look at wiki and fishers looks like it could do the job.

My only concern was that it states that it replaces chi2 for smaller samples (and other criteria) but I had ruled out chi2 as not being right for this so I'm not 100% sure...

Two tailed student T-test?

Im not sure where you get your 10% hypothesis from though?

Is question one - what sex are you or something else? Surely if you are 50% likely to say F to question one, and this makes you more likely to answer yes to question two? Then surely it would be nearer 50% still? How more likely? Need more info

Have a look here - bardeen.physics.csbsju/stats/ i used this for some stats i had to do, i THINK it told me what test was more suitable for my data set and believe me, i am NOT a statistician.

You need a Chi Squared test. It starts with a crosstabulation then tests the expected vs the observed distribution. If you need a reference I would recommend Siegel and Castellan "nonparametric statistics for the behavioural sciences" because it's very readable!

What spring forward said. Also if you are going to use SPSS, get a copy of Pallant's survival guide.

Ok this is social sciences (so don't believe in a null hypothesis!).

My 10% hypothesis is not what I'm testing. I'm looking at a level uptake rates for a.given subject (which I know are around 10% typically). I was trying to give you a handle for the fact that the number of reapondants likely to say yes would be very low compared to sample size....thought it might help.

Thought I couldn't do chi2 with only 1 degree of freedom :S...

I have a PhD in stats, it's the chi square test you need. why did you discount it?

I normally associate chi2 with observed/expected comparisons.

Would o treat my answers to q2 as my observed/expected (ie yes/no responses)?

Yes. Chance would predict one distribution (50/50), your hypothesis another. Chance is your null hypothesis, which your survey results may reject.

Of m/f, y/n responses, that is!

Could it not also be a correlation? I.e. a significant relationship between the answers to Q1 and Q2?

But I don't predict 50/50 in q2.

The number of people that answer yes in q2 won't total 100%

So if I was to tabulate (this may go horribly wrong) I might get:

..........M........F
Yes....80......30
No....720....170

So 10% of males say yes to q2 and 15% of females say yes. I want to know if the difference between the 10% and 15% is significantly different.

Can I do this?

(and I know I've skewed the answer to q1 - this could represent any ratio of male to female).

And sorry its all a bit hypothetical...it is literally a hypothetical research proposal I am trying to formulate...

And yes I am looking for a correlation between female and yes.....but I know that 90% of all pupils (male and female) will say no...so correlation co-efficients are.going to be low...I think...

Yes, that's fine. On qu2 chance (the expected distribution) would predict 50/50 y/n, your observed distribution may be different, and it's that difference you'd be testing IYSWIM.

A correlation wouldn't work because this is nominal data, btw, without an inherent order.

Ok...I get what you are saying.

I can see how chi2 will tell me that 10% is Sig diff from 50% and the same for the 15%. How do I get it to tell me that 10% is diff to 15%....

I really should go and find my stats book...

Ok...have found my stats test. Think I had only ever done univariate which is why I was confused...I can see that a bivariate (probably with Yates correction thingumy...) Should work...

Ty for sorting me out...