survey

Got this survey question from an employee wellness organization:

Are you worried about you or your employees contracting Coronavirus (COVID-19)?

1. Not worried

2.

3. Worried

4.

5. Very worried

There’s a fundamental flaw with how the question is constructed. Suppose you’re very worried about your employees getting COVID-19, but not worried about getting it yourself. How would you answer this survey item, which combines both questions into one? We should only ask one question… per question.

This also may serve as an example of how Likert scales can be poorly applied. Likert midpoints, when used, usually represent a neutral response (in this case, answer number 3 would be something like “Neither worried or unworried,” or a better option — since it may not be possible to be neither worried or unworried — may have been to include an even number of response choices with no midpoint).

Here, the survey providers essentially offer 4 levels of worry and 1 level of not-worried. They might be able to adjust for this in their analysis, but more typically survey providers generate a mean average score, which will be meaningless in this non-linear scale.

This reinforces what most of us have learned about survey design, and serves as a reminder to consider, when we read results of surveys on important topics (like public health or employer sentiment), how the data was collected.