1
$\begingroup$

I'm researching behavior of a certain group of people through a survey. I have access to the entire population which is 8000 and plan to email a survey, and collected data through Qualtrics, an online survey website. Is there a downside to surveying the entire population or should I do formal calculations given a 95% confidence interval and an anticipated response rate of say 20%, which will require me to send out the survey to about 2000. But, as all the data is electronic and therefore all the statistics will be done by a compute, isn't it ideal to survey the entire population?

$\endgroup$
4
  • 7
    $\begingroup$ Technically it's not called a "survey" at that point, it's a "census". $\endgroup$ – Daniel R. Collins Sep 22 '16 at 2:40
  • $\begingroup$ Call it what you want; does the name actually invalidate my question? $\endgroup$ – Ivan Velev Sep 22 '16 at 3:08
  • 1
    $\begingroup$ Is there any way you can make them all answer the questions? Because if not, you will still only sample the part of the population which is willing to cooperate. $\endgroup$ – skymningen Sep 22 '16 at 8:20
  • 1
    $\begingroup$ It kind of matters what you call it, a survey is something specific and many of the assumptions behind survey methods do not hold anymore (actually they start being problematic earlier, for example if you sample, say, half of the population). It's not necessarily bad but you need to adjust your computations accordingly. I won't be able to provide much details but it might be worth asking about this on the stats SE website. Also, is this “entire population” the only thing you are interested in or can you think of a theoretical population that's larger (e.g. future members of the group, etc.)? $\endgroup$ – Relaxed Sep 22 '16 at 8:20
6
$\begingroup$

Usually the size of the survey is governed by a cost-benefit analysis. With interviews, for example, the cost scales linearly with the number of participants, so you want to survey only as many as needed to get the confidence you want. With an online survey, your marginal cost may be zero or close to it, so if the additional information is of use, go for it. Of course, only a (possibly nonrepresentative) fraction of the population will respond.

$\endgroup$
2
$\begingroup$

If the cost and time are not problems for your study, use all of the population.

Based on the complete population, you have less error than the sampling based ones.

Also, you can compute your error (technically your standard error) and report the confidence interval based on the population.

For further details you can see the following textbooks, Cochran book, sampling wiley.

$\endgroup$
2
$\begingroup$

If you cannot count or measure the whole population, be certain that you take a random sample. And, if your sample size, n, is greater than or equal to 10% of the population, use a finite population adjustment of the standard deviation of the mean (the standard error). That is multiply it by

sqrt{(N-n)/(N-1},

where the size of your entire population is N.

As pointed out by Brick, you will get some non-responses in either case.

In either case your population is not what you seek, but rather the population of ,,, which responded. That will have to be made clear in your report. And, if you attempt to get samples from the entire population, you do not need to make a finite correction because the non-responders constitute another population.

$\endgroup$
1
  • $\begingroup$ Valuable point about the responding subpopulation not necessarily representing the whole. $\endgroup$ – Aaron Brick Sep 22 '16 at 20:18
1
$\begingroup$

If Qualtrics will charge you the same amount to send out 2,000 or 8,000 surveys, you may as well send it to everyone. But be aware that if you have data for the whole population, it will change the interpretation of statistical test results somewhat. With full-population data you are no longer generalizing results to the population from a sample but instead describing that population precisely. This may or may not matter for your purposes in this case, but is worth being aware of.

Also, as a general piece of advice, terminology matters in technical fields like statistics. Using the wrong term doesn't necessarily invalidate a question but can make it much harder to understand, let alone answer in a way that is helpful to you. Corrections and requests for clarification shouldn't prompt a harsh response.

$\endgroup$
1
$\begingroup$

If you can survey the entire population, you should do it. If you have data from everyone in the population, statistics will become much and much simpler. There is no such thing as confidence interval. You will just need to compute the population mean, population variance etc.

However, you'll most likely get missing data (not everybody will respond to you).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy