Who to Survey
(Return to Administration )

1. Size of Sample
   Basically you have 3 options: a few, a lot, or everybody.
   1. EVERYBODY: There are compelling reasons to include everybody, so a 100% sample is the most common. Otherwise you have to explain why some were included and some were excluded. Remember also that the cost of including everyone is often much less than the cost of engineering a random sample.
   2. A LOT: Nonetheless, there are at least 2 reasons for choosing less than a 100% sample:
      1. The most common reason is if your company is real big. You don't want to be analyzing a sample of 5000 respondents using the low-tech tools outlined in this book. Sampling only some of a huge number is a good way to get a more managable sample size.
      2. Be forewarned that a big sample can generate a large number of uninteresting but statistically significant differences. The size of the sample acts like a magnifying glass; if you have a big enough sample differences that are strategically trivial can rise to the level of statistical significance.
      3. Another significant reason for less than a 100% sample is if you intend to administer the survey frequently, i.e., more than once a year. You can sample one half for one administration and then the other half for the next administration; no one sees a survey more than once a year but you have data every 6 months. (Obviously the same procedure could produce data 4x or 6x a year if you have enough employees.)
   3. A FEW: The need for great speed may push for a smaller sample. If you have 800 employees and you only have one week to administer the survey, going for a 20% sample might be a good choice.
2. Selecting a random sample
   1. Probably the easiest procedure for selecting a random sample is to generate a list of employees along with a two-digit number from your HR database; the most useful is the day of birth, since that tends to be random and it never changes (don't use month of birth since that is not random). If you want a 50% random sample, take everyone born between the 1st and the 15th day of the month (feel free to adjust the range to get closest to 50%). Then you can tap the other half by using those born between the 16th and the 31st. If you can't get day of birth, take the first letter of their first name, and start down the alphabet until you get 50%.
   2. If you need to ensure representation for all groups along a particular dimension, that calls for a stratified random sample. Use the sampling procedure outlined above but within key demographic groups. For example, you could take the subset of executives and managers and then use day of birth to select half of them. Then take another employee group (such as individual conributors) and apply the rule within that group. This will ensure that you have good representation of all the relevant groups. The same strategy could be applied to divisions or regions, whatever demographic for which you want solid representation.
3. Selecting a targeted sample
   1. For some surveys you will really want the opinion of a subset of the organization, such as just engineers, or just project managers

Back to Top