Sampling, from Duncan Coker
With all the sampling being done in at least two areas of recent hot debate, employment data and election polls, it brings to my mind the issue of sampling bias.
How many samples is optimal? I believe too many samples and the risk of Type 1 errors (accepting a false prediction or causation) goes up.
I use this approach towards the markets, so this is relevant to me. Is there an optimal number to avoid "water boarding" the data? 10 seems about right based on practice but statistical justification would be appreciated.