We all love stats

[MarkyP05]

Not really sure what topic this question fits under but here goes...

for the purpose of the question im a investigating the influence of adverts upon the UK population and seeing if there is a difference between males and females

right, there are stats tests looking at relationships...ie correlations (e.g. spearmans, pearsons) , in this example say i was looking at whether the amount of hours of adverts viewed increased the purchases of retail goods.

Say there was a score of 0.76 for men and 0.61 for women (both siginificant)

Is there a stats test that examines whether there is a significant difference between these two results...

so in other words my question is... is there a stats test that would tell me whether there is a significant difference between two correlations. Is one significantly stronger than the other?

[EuanC]

I don't know if there is a test that examines if there is a sig diff between the two, but....... a correlation of 0.76 is a stronger correlation than one of 0.61.

I seem to remember from my uni days that the nearer the correlation coefficient is to + 1 then the stronger the correlation [that is a positive correlation i.e as one variable increases so does the other one].

A strong negative correlation [i.e. the nearer to -1 the stronger] suggests that as one variable increases so the other variable decreases.

I hope this makes sense. I

t is also probably worth keeping in mind that just because there is a correlation between 2 variables it does not mean that there is a direct causal relationship. Taking the "New York" example : there is a strong positive correlation between an increase in the sale of ice cream and an increase in the murder rate. However, this does not mean that eating ice cream causes people to murder. Rather it is the case that the increase in both is down to an independent variable - the increase in temperature.

[JudgeJ]

There are lots of other things to consider apart from the simple correlation displayed within the two sets of data (which I presume is what you mean by your �scores� of 0.76 and 0.61). Among these things, you will need to consider the residuals (the differences between your observed values and those provided by your�line of best fit� � the lines which produced your two values). The residual values should be small in value and random in direction. If they are not, your correlation may not be valid. There is a test called the Durbin-Watson test that will establish this. However this will only test the two individual sets of data you have analysed. I cannot think of anything which tests whether the difference between them is significant. As EuanC says, 0.76 is higher than 0.61, and that is all you can say. You would need to ask more questions of your sample to establish if and how the advertising effects men and woman differently.

This is all part of establishing the �causal relationship� between the things you are observing. This is different to the statistical relationship, but is equally, if not more important.

[whowhatwhy]

chi-squared test should give you a statistical significance between these figures? Google it.

[JudgeJ]

chi-squared examines the variation between observed and expected. Have we started with an expectation in this case?

[rojash]

"Have we started with an expectation in this case?"
Maybe I'm wrong but I would have thought that you could treat either as the expected. e.g "Having produced a result for males, I expect the result for females to be the same". That should give an indication as to whether the difference is significant.