Test to validate difference in survey respondent answers
- June 2, 2010 at 9:18 pm #53472
I recently faciliated an internal survey where business partners could provide feedback on each other on such things as partnering. I’m segmenting the responses by the groups that responded (e.g. for the client managers what were the scores from the consultants vs. sales support). If I’m looking at averages, what is the appropriate test to use to see if there truly are differences in the scores between the respondent groups or if it is just normal variation in the scores? Thank you!0June 3, 2010 at 7:00 pm #190281
For surveys I do not put any faith at all in a shift of the average. In fact I do not bother calculating the average.
I use the Chi-Squared test for the analysis, selecting a confidence limit that is suitable for the survey.
Beware that the Chi-Squared test needs around 5 to 10 responses per category of the response. Check your software for the minimum for your value (it varies).
I personally do not know, yet, what to do with a survey I’m starting to look at. The responses could have been rating category 1 through 5, but as it happens out of the 50+ responses thus far nobody has responded with a value of 3. That causes a divide by zero error …. still looking at that. Just sharing so you can be aware of a problem in your analysis.0June 3, 2010 at 7:44 pm #190282June 8, 2010 at 9:23 pm #190305
Thank you for your input, this is very helpful. Of course the audience for my survey wants to see averages but I’m also trying to provide them with better analytics to help them understand the results and how to interpret them. Thanks again.0June 9, 2010 at 7:28 am #190310
You must be logged in to reply to this topic.