How do overlapping error bars affect significance?

Overlap of error bars suggests the differences may not be statistically significant, but you should perform a formal comparison using the uncertainties to be certain.

Think of each data point as a target with a fuzzy circle around it showing how uncertain you are. If the fuzzy circles touch or sit on top of each other, it means the true values might be the same, so you cannot be confident there is a real difference just by looking. If the circles do not touch, that is a hint there might be a real difference, but you still need to check with a proper statistical test or by calculating the combined uncertainty. Error bars often show one standard deviation or some confidence interval, and you must know which one is plotted because that changes how far apart values need to be to seem different.

A common mistake is assuming any overlap means there is no difference, which is not always true because the amount of overlap and the type of error bar matter; the fix is to compute a formal test or the uncertainty of the difference. Another mistake is treating error bars from independent samples as if they were directly comparable without combining uncertainties, and the tip is to calculate the uncertainty of the difference using propagation rules. Some people forget to check what the error bars represent, and the simple remedy is to always label or ask whether the bars show standard deviation, standard error, or a confidence interval.

For more videos and resources on interpreting uncertainty and error bars, check out w w w dot gumball degree dot com.

Interpreting Overlapping Error Bars

1. Think of fuzzy circles
2. Overlap means possible same value
3. Nonoverlap hints real difference
4. Perform formal statistical test
5. Calculate combined uncertainty
6. Know which error bar plotted

Common Mistakes Interpreting Error Bars

1. Assuming any overlap implies no difference
2. Comparing independent error bars without combining uncertainties
3. Ignoring what error bars represent

errorbars, overlappingerrorbars, datavisualization, statisticalsignificance, uncertainty, confidenceinterval, standarderror, dataanalysis, graphinterpretation, experimentalerrors