Hacker News Reader

Hi Nathan, Thanks for your comment. And no offense taken. :)

Most people are surprised how "few" respondents it takes to get to 5% margin of error at 95% confidence levels: https://en.wikipedia.org/wiki/Margin_of_error#Different_conf...

In this case, .98/sqrt(365) = 5.1%

So, perhaps counterintuitively, 300-400 respondents gives good read on large populations (like Snapchat users...or the US population!).

View on HN

Assuming the respondents are uniformly randomly chosen, it takes surprisingly few people to get a pretty high confidence interval. I'm not convinced that the respondents are uniform, but I expect that there is a sufficient distribution that the results are still interesting. 3x vs. 1.5x seems like a small difference to me, but I am not an investor or user, so....

Yes, this exactly. The assumption of uniform distribution is not one to be taken lightly.

Reply on news.ycombinator.com