Did you try contacting them again, and maybe even a third time to get more responses? Keep track of the number of times they were contacted as that will be important when you analyze the data.

How much a problem this is depends on the details. After months of reminders you have 75 responses. fifty were returned on first notice 24 on the second notice, and 1 on the sixth notice. There are five questions and there is no difference between the initial fifty responses and the 24 returned on the second try. You might not have much trouble in this case. Alternatively the survey is 80 questions, you got 4 responses on the first try, 5 on the second, 4 on the third, ... , and after more tries you have 75 and you did not keep track. In addition, you find that 45 of these surveys were not completely filled out, so you will have to impute some missing data and some surveys will have to be discarded. In this case it might be very difficult to get this published. It may be a useful example of survey problems for a class in statistical methods or experimental design.

Maybe talk to a statistician at your university. There may be ways to refocus and still get something useful if the problems are too great to achieve the original goals.

What do you mean "normally, I would need a sample size of 132"? Do you mean based on grant requirements, power analyses, etc.? Also, this seems like a really high response rate for company founders. Can you say more about how you contacted them and got that many to agree? I guess a similar sample for my area is when we get principals to take part in studies and we have to do a lot to get response rates to be high.

Stephen -

So you didn't pick a random sample, and I expect you don't have a model-based sample either. You tried for a census, but you mention a "margin of error." Are you assuming your nonresponse is missing completely at random, and thus assuming your responses constitute a random sample? That would be quite an assumption, even with a small amount of nonresponse, much less such a large nonresponse.

Perhaps you can get a better idea as to the composition of your population if you stratify, possibly look at propensity groups.

But I'm sure I don't know exactly what is going on here. I don't understand the term "spin-offs." That sounds like something to do with network sampling, but your question does not seem to indicate that.

Best wishes - Jim

Maybe wait to see how many participants you actually get - have you done a power analysis to determine that this is your minimum sample size?

It depends on what your RQs are - a sample of 72 is not worthless, and from this sample you might still be able to report interesting findings.

This is not a sample from which you can obtain anything close to reliable inference, at least not without a lot more information. Results could be very misleading. It isn't a design-based sample, and it isn't a model-based sample. It isn't randomized, or balanced, or anything as far as I can see. One should not assume too much.

With a sample of 132, a population of 196, and a mention of margin of error, it sounds like perhaps a simple random sample was assumed from an infinite population.  But since the population is finite, and relatively small, consideration of the finite population correction (fpc) factor would have greatly reduced the required sample size.  However, this would all hinge on the sample being drawn with equal probabilities of selection for every member of the population.  So this does not work.

I assumed (possibly incorrectly) that this is about companies that have formed as a result of research that was done (at least in part) at the university. So it is a spin-off from a research lab.

There are always non-responders, and a risk that the survey is by its nature producing a sub-population where inferences about the sub-population do not apply beyond that population. The strategy for dealing with this is to contrast surveys that have required different levels of persuasion to acquire. If there is a sub-population one would expect that there would be a difference of some sort between people who filled out the survey immediately and those who did it reluctantly. Based on this outcome one then argues that the results do or do not apply to the sub-population that refused to cooperate.

Hello!

Thanks for the numerous responses. This survey is only a small part of my master dissertation, which has to be handed in at the end of May. I mainly focus on data retrieved from Orbis but now also have a small questionairre with 4 small questions to every spin-off founder.

The assumption of Timothy is correct, spin-offs are companies affiliated with universities. I contacted every founder of all 196 spin-offs founded in Flanders (northern region of Belgium) during 2001-2017. I did this by sending an email to the company address and by contacting the founder via LinkedIn.

I get to the required sample size of 132 by power analysis. At the moment I have 23 responses after 4 days, I will send a reminder after 1 week and then close the survey.

I will discuss this with my promotor this afternoon. Since this is not the main topic of my research, I hope I can still mention the responses of the founders.

That answer makes a huge difference.

In a thesis it is allowed to discuss one or more projects that are failures (in my opinion). The important part is what you learned from the failure. With 4 questions and 23 completed surveys you can analyze the data. The next part is to interpret the analysis. The discussion is then about what failed, why, and how to change things to increase the chances of success, and within this to also talk about what the data might say. Even if it turns out to be unpublishable, this project could still be an outstanding success.

Because you contacted all 196 spin-offs, your study is more of a census with about a 30% non-response rate. You might recontact those not responding, then test that group against the first responders. If the two groups are close, you could make the assumption that the remaining non responders are similar and include them with the rest. If they are significantly different, then it becomes more difficult.

To

Timothy A Ebert

very aptly summarized the scenario.

well done