We validated the power-procedure and have good results with 10 subjects... With less than 10 subjects, it will be hard to estimate effect size and variance, as you'll have less than 9 df.
Although a power analysis should be part of any fMRI experiment, unfortunately not everyone does it. The drawback then is that you won't have statistical control over the statistical power/false negative rate (how much of the true activation you won't find).
But <at least> you should fix the number of subjects in advance (eg. 20) and only do analyses once the whole sample is collected. If you just increase sample size until you get significance, you don't control for the false positive rate, and in that case it doesn't even make sense to do statistical inference... This harms the validity and reproducibility of your results.
Read the links that Chris send, it gives an intuitive explanation why you cannot do your analysis like that...
|