Sampling method

I have a population of 20000 employees all working in the same building. My task is to get there opinions of the food in cafe restaurant.

However I am unsure of the sampling method that I would use. I think it is fair to say my population of interest is not the 20000 but the people that actually use the restaurant. So it would take a long time to determine the people that have been to the cafe for food from the sample 20000 workers.

If it was assumed that all the 20,000 workers used the cafe I could use Stratified Sampling technique as I would have a data frame. However I don't think I could do this. I was just wondering would it be best to carry out a simple random first to see if our assumption was valid.

Any suggestions would be greatly appreciated.

First, you need to define your task. Do you care about all opinions of the cafe, or just the opinions of those that use it regularly, or just those who have eaten there once, even if it was two years ago? Those who haven't actually tried the food may have opinions -- based on hearsay -- about the food none-the-less, and you may care about those opinions. And you might care about appearance, cleanliness, speed-of-service, or other things besides just the food. (Perhaps you care if 25% of the people who don't eat there say that they don't eat there because it looks crowded or the lines are too long. And perhaps people who ate there but more than six months ago may not have tasted your current menu.)

Second, you can certainly take a sample of the 20,000 workers and survey them about the food, with the first question being, "Have you eaten at the cafeteria in the last six months?" and end the survey if the answer is "No". But if your survey is very short and to the point, perhaps you could get at least one more piece of information even then, like "Why don't you eat there? A) to expensive, B) what people say about the food quality, ..."

If you care about current customers only, you could offer rewards to them to fill out a survey. ("\$3 off any sandwich if you fill out a survey.)

We haven't even gotten to how you would survey: email, people in the hallway with clipboards, mail.

You mention stratified sampling. What would you consider stratifying on, and why?

• Thank you for the reply. I was going to split it into 2 subgroups. National citizens and foreign workers as it is a multinational company. The question is very vague but I would assume it's anyone that is currently in the company who has an opinion on the food. I just thought a sample of 20,000 would be too large to sample and you would certainly not expect 20,000 replies. Feb 23, 2016 at 20:04
• @Eamonn: You need to distinguish between "sample" and "population". 20,000 is your population. You will sample many fewer than that. Feb 23, 2016 at 20:55
• as I mentioned in the question. I don't believe my population is 20,000 as I am only interested on people that have an opinion. So since I don't have a sampling frame I can't use random sampling/Stratifying . So my question is could I carry out systematic sampling on the population of 20,000 and exclude the members that don't have an opinion. Or will this somehow lead to the sample been biased. Feb 23, 2016 at 22:00