Hi Sarah,
On 23/03/07, Sarah Robins <[log in to unmask]> wrote:
> Hi Helen,
>
> My rusty brain is creaking into gear, and as far as I remember (and I'm sure
> someone will correct me if I'm wrong), there are three components (or
> assumptions), that if they have been met, you may (should?) use a parametric
> test:
>
> 1) That the sampling is random (or that allocation to conditions is random
> e.g. In a controlled experiment);
Random from the population of interest.
> 2) That the data are from a normal distribution;
For correlation, it's that the data are from a bivariate normal
distribution. For regression, t-tests, etc, it's that the residuals
are normally distributed.
> And 3) That any variance within groups is approximately equal between them
> (equality of variance).
>
That only matters for the independent samples t-test, and in fact, it
doesn't matter for the independent samples t-test, unless you have
very different sample sizes in each group. And there, you just used an
unpooled variance t-test, so it doesn't matter.
> I don't know about other statistical packages, but SPSS can check the
> variance and normality of your data, and some good graphs and a keen eyeball
> can really help too :o)
>
That's certainly true.
Jeremy
--
Jeremy Miles
Learning statistics blog: www.jeremymiles.co.uk/learningstats
|