I have bad sense for low sample sizes; is 50 people from a subpopulation meaningful to get averages from, like how hard should I try to scrape together more data points? I know how confident to be in it but not like, what usefulness other ppl take from it
Conversation
Replying to
My completely intuitive sense based on just exposure to data collection over time is that like... 150-175 starts to be a good number that makes me feel relaxed
4
32
Replying to
I would think it depends on how the samples were selected. If it's totally random than I think CLT says 50 is plenty
Replying to
It depends a lot on how heavy tailed the distribution is, for gaussian it is enough
1
Replying to
n=75 can be an acceptable number IF sampling/weighting is done properly & common sense used (be modest, beware unintuitive results). But that is a BIG if; most surveys not conducted well enough.
2
Replying to
roughly: 25 people is good for ±10%. 100 people is good for like ±5%. 500 is like ±2%. 2500 is ±1%.
5
18
Replying to
The averages of said sub population probably will have little change between 50 and 150 responses. You can have some fun measuring the changes, but, don’t think you’ll see much.
Replying to
doesn’t it depend on the standard deviation of the data as well as your desired level of confidence?
Replying to
Sample size determines your resolving power.
If you're looking at big effect sizes, you don't need much resolving power.
This is why the vaccine trials were 50k+: looking for small but important side effects.
1
2
Replying to
Here's the thing, with 50 people, and standard p-hackable p-values, it only takes 3 people to skew your data. With 100 it takes 5.
This means you either need REALLY STRONG correlations or LOTS of people to avoid mostly getting "noise" from the results.
1
Replying to
If measuring something rare among the population sampled, then you need larger numbers.
If it's something like GOP v Dem, 50 works. If you include Libertarian, Communist, & Socialist, you probably need > 1000 to get representative numbers of those.
2