how can I say that my sample is representative? I needed a sample size of 605 (R, linear model, f2, small effect; Cohen). Advise? deleted
It's not at all representative for the employees of municipalities. It is representative for the subpopulation of employees willing to respond (for whatever reason - and that may be confounded with your questionnaire!).
So you tried and you failed. Shit happens. Try another way of getting responses. Or forget about that questionnaire and do something different.
It is always desirable to have the most representative sample possible, and that may be exactly what you have. In other words, this might well be the best data that it would be possible to obtain from this population on this topic. If so, then it is important to find out what you can learn from this sample, as a way to guide future researchers in this area.
You do need to be modest about the generalizability of your data, but if it is the best data available, then you can definitely make a contribution to knowledge.
Hello Fancesco,
What you may be able to do is to compare attributes of your participants with those of the target population as a whole (assuming that such descriptions of the target population exist). Is your sample comparable with respect to demographic characteristics, such as sex, ethnicity, age, level of education, job position, and so on?
While this in no way proves equivalence (should you find no difference on such variables), it can offer some credence to the suppositions that: (a) the sample was representative with respect to these characteristics; and (b) at least these characteristics won't be confounding variables biasing your results.
The other method is to chase down non-respondents, entice/elicit responses from them, and then compare their results with those of the initial responders.
Good luck with your work.
What do you think about this?
https://surveyinsights.org/?p=10981
Enlightning, from that paper: "The homogeneity assumptions researchers are willing to accept are thus key for the usability of NPSg" (NPSg = non-probability sampling). IMHO it's extremely hard to justify such assumptions, because it is usually very plausible that the topic of the questionnaire is related to the likelihood of response (people are more likely to respond if they a priory think the topic is important or relevant to their special(!) situation, if they are emotionally involved, having made "extreme" experiences (good or bad). If the response rate is large, it is plausible to assume that these cases are not that "special", and so is the homogeneity assumption. But this is, to my opinion, entirely out of reach when the response rate is low. I wonder what the difference is to just close both eyes, ignore all warnings and just assume that these are proper random samples?
As usual, it's a gradual thing, and the gradient depends on actual context. However, I still think that *this* case is extreme, so that even without knowing the context, a response rate of only 14% screems loudly that inferences based thereon are worthless. Even just a desciptive summary is likely to be worthless, unless one does not know what reasons the 14% had to respond and what reasons the 86% had not to respond (= to know of what "subpopulation" we talk about).
laziness, voluntary participation, indirect contact. many people sure didn't even see the mail
I agree with David Morse that comparing your sample to existing demographic data can increase confidence that the sample is not biased on those factors, although it cannot be definitive evidence in that regard.
I think you should apply Hansen and Hurwitz (1946) technique for non-respondents.
Just as a comment, I believe we all need to be more careful in the use of the term representative sample. To a sampling statistician, representative sample means a stratified sample see : https://www.bing.com/search?q=W+Cochran+Sampling+Techniques+3rd+ed&pc=U528&form=QBLH&sp=-1&pq=w+cochran+sampling+techniques+3rd+ed&sc=1-36&qs=n&sk=&cvid=EEBF2C2356764781BFB25FE53245EBD4
Also note: https://www.bing.com/search?q=representative+sample&qs=LS&pq=representative&sk=AS1LS1&sc=8-14&cvid=58C94FBF824A4E5BA0B5D54A349C9914&FORM=QBRE&sp=3
In particular, The question about the difference between a random and a representative sample. Words matter and so in science particularly, we must pay careful attention to definitions. Best wishes to all, David Booth
I think it depends on your population size.
what is your population size?
what is your sampling method?
There is no hard and short cut rule that varifie information regarding population. But we take approximately 5% of the population.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size