Dear Ms. Olivia Chung, you could try the construction of confidence intervals:

Margin of Error (ME):

ME = 1,96(sd/sqrdroot(n))

Upper Bound:

Average + ME

Lower Bound:

Average - ME

However, the nature of your variable [standardized between 0-10] influences the construction of an upper bound greater than 10. It's impossible to score beyond 10, but it's statistically possible to calculate limits upon that. You could avoid this with greater samples.

if you are aiming:

- to compare variability between your scores, I suggest the coefficient of variation of each variable: cv = (sd/average)*100.

- to show the "best ranked", I suggest the third quartile [separates the worst 75% from the best 25%].

- to show central tendency measures: median [separates the worst 50% from the best 50%] and mode [most frequent].

I don't feel you're doing anything wrong. Mean & SD summarize data to fit an idealized normal distribution. Reality is more messy. You have a ceiling effect. If you're trying to summarize your data with greater precision, maybe also calculate skew, since it's so strong (e.g., -2.13 for overall, which you can calculate in Excel with =SKEW(D3:D10)). Hope this helps some, Olivia! ~ Kevin

Olivia Chung I think you can still define the standard deviation in the usual way, just that you'll have some additional inequalities for different statistics, check out this note: http://www.stat.cmu.edu/~siva/705/lec3.pdf

If you are using the SD to define the error bar, I think you can clip the error bars so that the variable remains in the range between [0,10]