Mathematically, 10 is generally considered enough to make broad conclusions, more is always better though. OTOH, that assumes a fairly uniform random sampling. The reason more is better is because it helps leverage against the non-uniform random sampling.
I honestly don't understand how 10 people is a large enough sample to make broad conclusions about health and medicine for 7 billion people. Can you please explain the statistics behind that?
If you hypothesize that the odds of a coin flipping heads is 50% and you get heads 9 out of 10 times, the odds of getting that result if the odds were really 50% is like 1/500. You should be pretty confident that if you were the flip the coin billions of times you wouldn't get close to 50% heads based on looking at 10 samples.
Do you not think selection bias could be a major factor?
Hypothetically, if this study was undertaken in small-town Wisconsin, and genetic factors could even remotely be involved, the samples wouldn't be representative in the slightest for most genes in the world.
It absolutely is a factor, which is why I cautioned about a uniform, random sampling. Selection bias implies non-uniformity, and 10, 100, or 1000 samples won't fix selection bias since it is, by definition, non-uniform across the intended subjects.
It's good that you question it, it should always be one of the first questions asked when looking at these sorts of statistics.
I was just pointing out that the number 35, by itself, is plenty to justify a conclusion, and because perfect uniformity is not completely possible, more tends to be better. If the conclusion is faulty, it isn't typically due to the sample size of 35 being too low.
In psychopathology specifically, though, you're not making inferences about 7 billion people; you're just making inferences about the population that actually has the disease. If only three people in the world have ever had Rudolff's Aphasia, then you'll want to study them to help the fourth guy down the line.
