Sampling is the process of measuring a small amount
in order to get a picture of what the whole looks like. You use it when you conduct a survey for customer satisfaction, or to get customer input for new products. Auditors use it to review your company’s financials or conduct due-diligence. When leaders walk around and talk to employees and get a “feeling” about morale, they are “sampling”.
How do you know your sample size is big enough?
You’ve heard the expression that one data point does not make a trend. Yet over sampling costs a lot of time and money and doesn’t necessarily give you a better result. So how many samples do you need to give you a trend with confidence that your conclusions are valid?
Statistics, Probabilities, Relevance, Confidence Levels, Sampling Methods
, etc. are the words you’ll hear from your research department, or consultants hired to figure this out. There are a lot of professionals out there who do this for a living, and for good reason. Sampling is a science, and if you want to maximize your confidence, and minimize your expense, it’s best to consult a statistician to aid you. Having said that ….
Here are a few rules of thumb.
Create a more homogenous population.
The closer you can come to a group that is similar and in your space, the better your result. This will seem obvious but is often performed poorly. For example, if you want feedback on a product that works on an Apple Mac, you’ll want your population to be Mac users. If you’re selling something that targets small businesses between 5 and 25 employees, you’ll want to make sure your survey doesn’t include home office or larger companies. If you’re trying to figure out why something about your product isn’t working the way customers expect, you’ll want to first figure out if there is something in common about those customers who have complained about it. Creating a smaller relevant “population” will improve the quality of your result.
With the caveats that statisticians would give, your sample size should be 1/SQRT(N). Take your populations size, or the smaller universe from above, then calculate the square root, and that’s how many you should sample in order to give you about a 95% confidence level that the result will hold true to the greater population. In the above examples, if you identified 1000 companies with 5 to 25 employees, then you would sample SQRT(1000) = 32 (rounded up). If you had 10,000 you would sample a 100.
Sampling can give you a very fast and inexpensive way to get an idea about which way to go, or what the issue might be, or what might be wrong. It isn’t fail-proof, but given just a little thought, most of the time, it will yield a result in the right direction. Spend just a few minutes thinking about how you can use sampling in other ways in your business, with your customers, and employees.