The best way to gather data is to collect it from everyone in your population when that is possible. Then there is no sampling error, no question about representativeness or bias. We have data on everybody.

It is not always feasible to collect data from every client you see, everyone who receives your service, or every participant in your planned activities. Therefore, you will need to sample when:

1. List all the possible people who qualify to be in the sample. This is called the sampling frame.

2. Use an unbiased process for selecting members of the sample. A table of random numbers is often used. If one is not available, you could systematically select every n^th name in the sample frame to get the sample size you want. If you are limited to being only able to collect data from 2, 3, 5 or even 10 people, it may be best to use a judgment sample. In this case, you use your best judgment to get a good cross-section of your clients into the sample.

3. The size of your sample is determined by a number of factors. These include:

   • how large your population is

   • how accurate you want your estimate of the population based on the sample to be

   • what you can afford

   As a rule of thumb, you want your sample to be at least 30 if you are randomly drawing people from a large sampling frame. This will allow diversity to occur in your sample. If your sampling frame is large, say 500 or more, a sample size of 30 will give you a rough estimate of the population’s responses if you could get them. The attached table is a simple reference for determining sample size that will yield +/- 5% error in the responses you get on most questionnaires. In the table, N is the population size and S is the sample size. Thus, if you wanted to sample 9,000 client records, the appropriate sample size would be 368. If you say, "How can I possibly collect data from 368 records?", the matter becomes not one of precision or accuracy but one of feasibility. What can you do? Would 100 be manageable? Or 50? The smaller the sample size, the less accurate the estimate of what you are trying to measure.

4. There is one more consideration. Sometimes you know that the response rate on a questionnaire is not going to be good. You figure you will be lucky to get 30% responding and the question will become, "Who responded? Did we get a response bias?" To answer this, you should compare the sample you did get against known characteristics of the population, such as number of males and females, average age, and geographical distribution.

   If you expect a low response rate, say 30%, is it better to start with a large sample, say 1,000, so you can end up with 300? Or, is it better to go with a small sample, say 35, and try to get all of them? In most cases, the smaller sample size with a 100% response rate is likely to give you a better estimate of what you are trying to measure.

Click Here: Table for Determining Sample Bias from a Given Population