Likert scale data - displaying variance with standard deviation

[sputnik0110]

Hello!

I conduct small sample size research studies (for usability research) and as a component of the study sessions, I typically administer a survey with a series of 5-point Likert questions to gauge attitudes.

E.g: "I found this system to be complex."
1=strongly disagree to 5=strongly agree

When analyzing and presenting the data, I've used the Median to calculate the central tendency and then the standard deviation to represent variance. An issue I've had with this method is that the standard deviation can exceed the maximum scale value - so the median score for a given question could be 5 (max score possible), with a standard deviation of 2...resulting in the error bars on the graph extending past the max value. Something seems wrong with that

I've done quite a bit of web searching on Lickert scales, analysis, etc. and there seems to be a divergent set of opinions on this so I'm looking for some advice please!

Thank you,
Jon-Eric

[CowboyBear]

Standard deviation represents the average distance of observations from the mean. So using SD along with median rather than mean in this way doesn't necessarily work too well. Using upper and lower quartile along with median (and perhaps min and max) might make more sense as a depiction of central tendency and variability. I wouldn't describe either SD or LQ/UQ as "error bars" though.