Know-How Exchange : Research/Metrics

The sample size you need depends on how you wish to track satisfaction.

For example, do you want to understand satisfaction generally, or by department, by physician, by illness/condition, or by some other patient characteristic? The more "slices" you would like to measure, the larger the total sample you will need.

The rest of this post will apply for each slice you wish to measure. If you want to measure satisfaction for each of 4 departments, the sample size / sampling guidelines apply for patients of each individual department.

Assuming that you have at least 1,000 patients per quarter (or whatever other reporting period you select), you should consider 125 to 150 to be the absolute minimum number of completed surveys you need. At this sampling level, the margin of error on the measurements will be relatively large, but will give you a general feel for patient sentiment.

A more reasonable sample would be in the 200 to 300 range. This sample size will provide better precision, and will give you the flexibility to compare sub-groups within the data. As the sample size increases, the precision gets better, as does the ability to compare sub-groups within the data.

There are diminishing returns, however, and there is rarely any reason in commercial market research to collect samples of more than 800 to 1000 respondents. At these sample sizes, the incremental costs of data collection and processing tend to outweigh the incremental benefit of a larger sample.

If you have fewer than 1,000 patients per quarter, apply the "finite population correction" to the sample size formula to obtain a desirable sample size. This can be found in any decent basic statistics / market research text. Basically, the "finite population correction" scales down the needed sample size to account for the fact that the overall population is small.

In terms of sampling procedure, a simple random sample from all patients (or all patients within each slice you are measuring) is perfectly adequate.

If you want to be methodologically pure, the easiest thing to do is to get a list of all the patients in excel, assign a random number to each, sort the list by the random number, and pick the first x, where x is the sample size you've selected. Then you need to get those people to answer the survey (e.g. "no" is not an option). Good luck.

What is done in the real world is that the entire list of patients (or a random subset deemed large enough to find the required number of people to complete the survey) is contacted until the required number of surveys have been completed. This is what's technically known as a convenience sample, and what is done for just about every commercial survey conducted. There's nothing wrong with this approach, so long as some reasonable precautions are taken to avoid a sample that is somehow oddly skewed.

One caution - if you decide to sample by department (or any other split) and collect the same number of surveys for each, you cannot simply calculate an "overall" score by looking at the whole sample. This is because you have most likely over-represented some areas and under-represented others, as they did not each handle the same number of patients. This is easily fixed by weighting the data to get "overall" measures.

Hope this helps. Feel free to contact me directly if you have follow-up questions.