More often than not, you will want to not only examine the results from the overall population, but also understand the differences between key demographic subgroups within the population. For example, you might want to understand the differences between males and females or senior managers and regular employees. If you plan to look at distinct subgroups such as these, you should perform a stratified random sample. In a nutshell, this means you will need to select a separate random sample from each of the subgroups rather than just taking a single random sample from the entire group.

Sponsored Links:

The process is slightly more time consuming and will require you to survey a greater number of people overall, but this technique can be very valuable. If you want to conduct a stratified random sample, think carefully about the single most relevant demographic division that can be made between people within your population. It is probably not practical to conduct a stratified random sample on more than one demographic category as the process becomes much more complex and you will ultimately end up needing to survey almost the entire population if any of the subgroups are very small. In other words, if you want to look at age and position, you will need to look at each position/age combination and you might find very small numbers of people within some of these areas.

Statistical Accuracy - Confidence and Error
In order to understand random sampling, you need to become familiar with a couple of basic statistical concepts.

• Error - This is that "plus or minus X%" that you hear about. What it means is that you feel confident that your results have an error of no more than X%.
  
• Confidence - This is how confident you feel about your error level. Expressed as a percentage, it is the same as saying if you were to conduct the survey multiple times, how often would you expect to get similar results.

These two concepts work together to determine how accurate your survey results are. For example, if you have 90% confidence with an error of 4%, you are saying that if you were to conduct the same survey 100 times, the results would be within +/- 4% of the fist time you ran the survey 90 times out of 100.

If you are not sure what sort of error you can tolerate and what level of confidence you need, a good rule of thumb is to aim for 95% confidence with a 5% error level.

Error is also referred to as the "confidence interval" and Confidence is also known as "Confidence Level." In order to avoid confusion, these concepts will simply be referred to as "Error" and "Confidence" in this article.

Determining the "Correct" Sample Size
Determining the "correct" sample size requires 3 pieces of information

1. The size of your population
2. Your desired error level (e.g. 5%)
3. Your desired level of confidence (e.g. 95%)

Final Steps - Putting it All Together
Once you have determined how many people you need from either your population as a whole or from each subgroup within your population, you simply need to determine a way to randomly select the specified number of people from each group. There are many wrong ways to go about this. Whatever technique you use, be sure that you really are selecting people at random and not accidentally giving preference to anybody for any reason. An easy and fast way to randomly select people is to use MS Excel. The steps to make the random selection are as follows:

• Copy and paste a list of every person in the group into a single column. You can use names, email addresses, employee numbers, or whatever.
• In a second column, fill the entire column with Excel's "Randomize" function. The exact value of each cell should be "=rand()" (do not include the quotation marks). Only fill the cells next to where you pasted the group info in step #1.
• Sort both columns by the "Randomize" column. It does not matter whether you sort them in ascending or descending order.
• Scroll down to the row number of the group size. Everybody from this row up is a part of your sample (see important note below regarding response rates).