Is there a rule of thumb about extrapolating from a small sample? When might it be reasonable (valid) to do so?
The short answer is that, scientifically speaking, it’s not reasonable to do so.
The result will be invariably inconclusive and null when the sample size is too small. It compromises the conclusions.
Often, it is indicative of a huge bias/skew. It needs to take into account factors as to which subsets are agreeing to do it and how they may be different from other subsets.
So what is a reasonable sample size for a survey? Well, it partially depends on the type of survey, complexity, how random, type of analysis & how fancy the statistics are (ie: multivariate analysis, etc), and other things. But, you asked about a general rule of thumb… So in general, most agree that the minimum sample size should be no smaller than 100… If your entire population is smaller than 100 then you need EVERYONE to take the survey, not a sample.
Maximum should generally be up to 10% of the applicable population, but probably no more than 1,000 --again, there are exceptions, but generally speaking, doing more than 1,000 usually won’t add much to the accuracy and will just be a waste of time and resources for little to no added benefit.
Again, assuming everything is well organized/structured and minimizes other bias and influence. And again, in general. There are plenty of things that can make results inconclusive, null, compromised, and not statistically significant.
Strong response, Carver, thank you.
“Most polls are flawed.” Is that a fair assumption?
And data people don’t like flaws.
So data mining instead. 24/7 tracking of behavior. And physiological responses. With full demographic profiles underlying.
Passé. Frosting on the cake. A mere diversion.
It’s why “apps” are the worst thing to happen to the Internet. Where functionality can be delivered over a web browser that can be closed, every piddling little site shoves its “app” at you to eat up your phone resources and do constant metrics and more.
Edit: Yes I’m conflating the Internet with the web. Sue me.
and the app you rode in on?
Having said that with confidence…
…good ol’ Nielsen’s got its digital engine revving:
but still sends out snail mail (with landline follow-ups) requesting your household’s participation in a telephone survey.
So maybe polls are still among the topmost in a multilayer cake?
What do I know…
This is still one of my favorites:
A tiny sample, yes, but I find its findings definitive!