======= Date Modified 25 03 2010 22:03:17 =======
I've done a wee independent t-test looking at whether there was a difference in the level of trust for politicians between those who had and hadn't heard (whoever they are) of the MP's expenses scandal. I've done the test, which is significant, but I have also calculated the effect size (r) which is small (.09).

We never covered effect sizes in quant methods, and so I'm unsure about what I have found. Am I right in saying that even though there is significance between the 2 groups that being in either the yes or no camp has no substantial effect on your level of trust?

If someone could help me out I would be most appreciative.

:-)

I think that's right, although you've found a significant different it only accounts for 0.9% of the total variance, so its a very small effect.

see page 4 onwards of this: http://www.statisticshell.com/effectsizes.pdf

How many participants do you have in each group?

A lot, its taken from the BES 2010 and the sample is about 15,000. Although the Yes outweigh the No's, the no group still has over 1000 people in it.

maybe you need the sample sizes more equal in order to get a bigger effect size (just completely guessing here!)

Aye, I did consider that. But it's ok because having a small effect size is not actually a bad thing for the analysis at all.

Thanks for your help

So, basically, can I say that whether you are in the yes camp or no camp it has little effect on how you think of politicians even though there is a significant difference between the two groups?

Yes, you can say there is a very small, but statistically significant difference in the level of trust between the two groups. BTW, if you are doing a 2-sample t-test the effect size is just the difference of the two means. What scale did you use to quantify level of trust? To be used in a t-test it should have at least 10 values, e.g. a trust score from 0 to 10.

0 to 10 yes