My sup has just asked me to find the effect sizes of the correlations we've found.

I thought the r was the effect size!?

Confused.

Sneaks, this section of a book explains it all nicely: http://books.google.com/books?id=AUDoy-lSe_EC&pg=PA412&dq=effect+size+for+correlations&hl=en&ei=9En2TOfELIW6hAeRx4DNBQ&sa=X&oi=book_result&ct=result&resnum=3&sqi=2&ved=0CC4Q6AEwAg#v=onepage&q=effect%20size%20for%20correlations&f=false

Sadly, can't get my hands on this book.:-)

thanks Wal, I'm still confused.

Basically I have a correlation matrix, with significance values, so I have e.g.

chocolate eating is positively correlated with Sneaks' weight gain (r = .8, p<.001).

And she wants me to find the effect size. Surely its .8 ??

I would go back to sup and ask for clarification, just say 'is this what you asked me to do since we got x correlation as the result'. then you would seem only forgetful. Have you had a look at Howell or Andy field stat books (field has a website too)?

Hi Sneaks, has your supervisor asked you to clarify the number or degrees of freedom? Usually in stats the 'effect size' can be a combination of the p value, the test (r) value and the sample number taken into account ie n or d.f. For example, if you have r=0.8 and p=0.001 but the sample was only 4, it wouldn't be an effective stats result but if the sample size was 200 and this was your stats output it would be very 'effective'.

Of course I may be completely talking rubbish.............:$

======= Date Modified 02 Dec 2010 09:24:24 =======
The r is the degree of association between two variables (the correlation). What you need to state is the "r-squared" which is the variance explained, the strength of the relationship between two variables.

I thought r was used as effect size in correlations, but this can be converted to other efffect siz measures (such as Cohen's d) so maybe this is what your supervisor wants. Or maybe they just want you to explicitly state whether it is large, medium or small. Probably best to check with them, especially if they are not a stats guru they may not know r is the effect size.

The effect size is the beta coefficient (B) for your X variable from a regression model. So you need to do a linear regression for X on Y and report B along with its 95%CI and p-value. In the example you gave this effect size will tell you that for every, say, gram increase in chocolate consumption (X) you increase in your weight (Y) by B kgs. I hope that helps.