3 ms·

>n=50 (closer to 100 with the control group) is massive for a brain-imaging study. The problem I see with it being a small group is not necessarily that the re

by nv-vn 11y ago

>n=50 (closer to 100 with the control group) is massive for a brain-imaging study. The problem I see with it being a small group is not necessarily that the results are inaccurate because of size/lack of statistical power, but more so because of the fact that a lot of variables will not show up in a small sample. For example, it is possible that certain diseases that they found to be more common in one sex than the other are only more common in caucasians and the same relationship does not occur with other races (since this happened in Hungary, it is very unlikely that non-white people were a part of this study). The numbers determined here probably apply very well to people in Hungary but they might be very different from what we'd find in the United States or in India, for example.

omginternets 11y ago
What you're describing is an epiphenomenon, and increasing the sample size won't fix this. In fact, increasing the sample-size (i.e. over-powering a study) increases the propensity for spurious correlation. This may well be an epiphenomenon, but that's why reproductibility (not statistical significance) is the gold-standard of science.
- tomp 11y ago
  > In fact, increasing the sample-size (i.e. over-powering a study) increases the propensity for spurious correlation. Huh? If there aren't just 2 kinds of brain, but instead 30 different kinds, then a study with 300 brains is much more likely to catch all 30 kinds of brains than a study with just 50 brains.
  - omginternets 11y ago
    So you're worried about a representative sample? This is -- again -- only indirectly related to sample size. If the sample was randomly selected and the statistical power is great enough, then any correlation it's missing is marginal; this is a feature, not a bug! You're not entirely wrong insofar as brain-imagery studies rarely conduct random samples, but increasing your sample size won't fix that.

by nv-vn 11y ago

omginternets 11y ago

What you're describing is an epiphenomenon, and increasing the sample size won't fix this. In fact, increasing the sample-size (i.e. over-powering a study) increases the propensity for spurious correlation. This may well be an epiphenomenon, but that's why reproductibility (not statistical significance) is the gold-standard of science.

tomp 11y ago
> In fact, increasing the sample-size (i.e. over-powering a study) increases the propensity for spurious correlation. Huh? If there aren't just 2 kinds of brain, but instead 30 different kinds, then a study with 300 brains is much more likely to catch all 30 kinds of brains than a study with just 50 brains.
- omginternets 11y ago
  So you're worried about a representative sample? This is -- again -- only indirectly related to sample size. If the sample was randomly selected and the statistical power is great enough, then any correlation it's missing is marginal; this is a feature, not a bug! You're not entirely wrong insofar as brain-imagery studies rarely conduct random samples, but increasing your sample size won't fix that.