FigureCheck
Choosing error bars: SD, SEM or confidence interval?
Choose and label standard deviation, standard error, or confidence-interval error bars according to the scientific question.
Use SD to describe variability among observations, SEM to describe uncertainty in an estimated mean, and a confidence interval to show a range from a stated estimation procedure. Always label which quantity the bars represent.
The choice follows the claim
If the graph is meant to show how variable individual measurements are, standard deviation is usually the relevant summary. If it is about precision of an estimated mean, SEM may be relevant, although a confidence interval is often easier to interpret. A confidence interval is not a fixed probability that the true value lies inside one computed interval.
Error bars are not a significance test
Overlap or non-overlap of plotted bars is not a universal hypothesis test. The interpretation depends on what the bars represent, sample size, study design, and dependence between measurements. Use a suitable statistical comparison and report its effect size when the question is about group differences.
Worked example: repeated plant heights
A class measures 20 plants in each treatment. Plot mean plus or minus SD to show plant-to-plant variability. Plot a 95% confidence interval when emphasizing uncertainty in each treatment mean. State the sample size and interval method in the caption.
Common mistakes to avoid
- Leaving the error-bar type unstated
- Using SEM to make variable data look more consistent
- Declaring significance from visual overlap alone
Open an editable example and apply the idea to a real graph.
Sources and further reading
These links support the scientific principles summarized above. Always follow the reporting rules required by your course, laboratory, journal, or field.