I have four samples; each sample has a thermal conductivity value.
I want to show the error bars of each value of the thermal conductivity, so which error should I calculate to plot the error bars, standard deviation or standard error?
I don't agree with Jochen Wilhelm
, error bars are always informative and neccessary. It's not the same to show 1.1 vs 1.3 or 1.1±0.3 vs 1.3±0.3. So yes, you should show your error bars.And I recommend using standard deviation. If you want to be even more informative you could calculate the confidence intervals assuming normal distribution and using student-t table (for you confidence level=alpha and degrees of freedom=N-1). In either way, the figure caption should state what error bars represent.
If the difference is 0.2 and the confidence interval is 0.3 there is no difference, that's the point. By the way, confidence interval of a difference doesn't make sense as a concept, in my example it would be between -0.4 and 0.8, so 0.2±0.3*2 or 0.2±0.6, even less interpretable.
Describing differences might still be doable with 2 samples, but not with many. Then a visual representation is faster and easier to interpret, but you need the error bars (or something that identifies statistical differences, like asterisks based on a t-test).
That 0.42 is the propagation of variance, but okay, if you think a forest plot is easier to interpret than error bars it's up to you. Coming back to the original question, four bars with four error bars is by far the simplest way to interpret the results. No point in showing the differences with confidence intervals and seeing which go through zero to go back to the means and interpret if they are significantly different.
- Badges
- Science topic