Non-probability sampling stuff


New Member
Hi there,
Sorry probably asking a very basic question but just want to make sure...
Am I right in saying that if I have surveyed people using a convenience sampling method (ie non-probability based) then I cannot perform any inferential statistics, only descriptives. Does this rule out Mann Whitney, Wilcoxon and other non-paramethric tests?
Thanks very much :)
Hi Amy,

You want a sample to be representative of the population, and generalize the results obtained from the statistical analysis. If you have a "convenient sample", your results may apply to your sample only, not applicable to the target population. I think you can use a nonparametric method to analyze the data, but the results may or may not be generalizable.

In some cases, it is hard to get a random sample, you can do the analysis and describe sampling method in detail, so people can make decision whether the results are good.

Hope this helps.


New Member
Another question...

Thanks so much for your prompt reply
I wanted to verify though, isn't one of the assumptions in using non-parametrics that samples need to be random? Wouldn't convenience sampling mean that I would not meet this condition?
You are right Amy. Mann Whitney test requires the samples to be independent and random. hmm... I think it cannot be used. Still, convenience samples can provide you with useful information, especially in a pilot study. To interpret the findings from a convenience sample properly, you need to characterize how your sample would differ from a random sample. In particular, pay attention to who might be left out of your convenience sample or who might be underrepresented in your sample. Also note whether the people who were left out might behave differently than the people in your convenience sample. If your outcome measures are not strongly related to this factor, you might be okay.