Stats wonks: My Twitter polls seem to converge to roughly their final distribution around n=100, no matter what the final n (which ranges from 200-450). What can I conclude from that?
Conversation
Replying to
Consider a poll option with 50% probability (this is the worst case in terms of variance). With n = 100 the standard deviation for the observed proportion is 5 percentage points, so roughly 95% of the time it will be within 10 pp of the final result.
1
3
Replying to
that your followers are not quite as heterogenous
2
Replying to
Pretty good heuristic version of the central limit theorem:
In any random process, if you measure N counts of something, then you should consider that the true number (i.e., the one that perfectly represents the mean) is within the range N +/- sqrt{N}.
2
Replying to
the number of 9's for your desired confidence interval is satisfied at ~100, unless there are systematic errors or nonstandard distributions (that is my guess)






