Bar Plot Explained: How to Make, Read and Report Bar Charts with Error Bars
A bar plot is the most familiar chart in research, and one of the most frequently misused. It is excellent for comparing counts or summary values across a small number of categories. It is poor at showing distributions, and its error bars are often drawn without saying what they represent. This guide explains when to use a bar plot, how to choose error bars, how to read one critically, and how to report it, with a worked example whose arithmetic you can check.
What is a bar plot?
A bar plot (bar chart) draws one rectangle per category, with the length of each bar proportional to a value: a count, a percentage, or a summary statistic such as a mean. Categories are discrete and have no numeric spacing, which distinguishes a bar plot from a histogram, where bars represent continuous bins of a single variable and touch one another.
When should you use a bar plot?
- Comparing counts or percentages across categories (for example, responses per option).
- Comparing group means or medians when the number of groups is small.
- Showing grouped or stacked comparisons across two categorical variables.
- Presenting results to an audience that needs a quick comparison rather than a distribution.
Bar plot versus other charts. Use a line graph for change over time or ordered conditions, a pie chart only for a few parts of a whole, and a heatmap for matrices. For continuous outcomes in small groups, showing the individual points alongside or instead of bars is more transparent (Weissgerber et al., 2015).
The mathematical foundation
When bars show means, the error bars come from the sample standard deviation s and sample size n:
SE = s / √n
The 95% confidence interval for a mean is:
CI = M ± t(1 − α/2, n − 1) × SE
These three quantities answer different questions. The SD describes how spread out the individual observations are. The SE describes how precisely the mean is estimated, and shrinks as n grows. The 95% CI is roughly 2 SE wide on each side for moderate samples. An SE bar is therefore much shorter than an SD bar, which can make data look tidier than they are. Cumming and Finch (2005) give practical rules for interpreting such bars; a notable one is that overlapping SE bars say very little about statistical significance.
Assumptions and data requirements
- Categorical grouping. One column naming the category; one numeric column (a count, or the raw values from which a summary is computed).
- Zero baseline. Bar length encodes value, so the value axis should start at zero (Cleveland & McGill, 1984; Tufte, 2001).
- Independent groups. Error bars for means assume independent observations. For repeated measures, standard between-subject error bars can be misleading.
- Adequate group size. Means and error bars from very small groups are unstable. With fewer than about 10 observations per group, show the raw data.
- Sensible ordering. Order bars by value or by a natural sequence, not alphabetically by default.
Original research example (illustrative)
Background. A university surveys students at four campuses on overall satisfaction using a 1–5 scale.
Objective and question. Do mean satisfaction scores differ across campuses, and how precisely is each mean estimated?
Data. n = 25 students per campus. The values below are invented for teaching.
Step 1: Summarise each group
| Campus | n | Mean | SD | SE | 95% CI |
|---|---|---|---|---|---|
| A | 25 | 3.8 | 0.9 | 0.18 | [3.43, 4.17] |
| B | 25 | 4.1 | 1.0 | 0.20 | [3.69, 4.51] |
| C | 25 | 3.5 | 0.8 | 0.16 | [3.17, 3.83] |
| D | 25 | 4.4 | 1.1 | 0.22 | [3.95, 4.85] |
Step 2: Check the arithmetic
With n = 25, √25 = 5, so each SE = SD / 5: 0.9/5 = 0.18, 1.0/5 = 0.20, 0.8/5 = 0.16, 1.1/5 = 0.22. The critical t with 24 degrees of freedom is 2.064, so the half-width of the 95% CI for campus A is 2.064 × 0.18 = 0.37, giving 3.8 ± 0.37 = [3.43, 4.17]. Campus D: 2.064 × 0.22 = 0.45, giving 4.4 ± 0.45 = [3.95, 4.85].
Step 3: Choose what the error bars show
For campus D, the three possible bars have very different lengths: SD = 1.10, SE = 0.22 and 95% CI half-width = 0.45. A figure that drew SE bars without saying so would look about five times tighter than one that drew SD bars. Here the aim is to show how precisely each campus mean is estimated, so 95% CIs are the natural choice, and the caption must say so.
Step 4: Interpret
Campus D has the highest mean (4.4) and campus C the lowest (3.5). The intervals for campuses C and D, [3.17, 3.83] and [3.95, 4.85], do not overlap, which suggests a real difference between them. Campuses A and B overlap heavily, so these data cannot separate them. Interval overlap is only a rough guide: a formal comparison (such as ANOVA with follow-up contrasts) is needed for inferential claims, and with ordinal 1–5 data the raw distribution should be checked as well. With only 25 respondents per campus, showing the individual responses would be more informative than bars alone.
How to make a bar plot with the StatFAS Bar Plot Maker
Open statfas.com/bar-plot-maker.html.
- Prepare your data. One column for category labels and one for values (or raw observations, if the tool summarises them). Use a header row.
- Load the data and map columns. Choose the category and value columns.
- Set options. Typical choices are bar orientation, grouping or stacking, error-bar type, colours, axis labels and title.
- Review the figure. Check that the axis starts at zero, labels are readable and the error-bar type is stated.
- Download. Export the plot at a resolution suitable for publication.
statfas-bar-plot-maker-interface.pngAlt text: "StatFAS bar plot maker interface before data entry."
Caption: The Bar Plot Maker on StatFAS.com.
statfas-bar-plot-example-data.png · Alt: "Campus satisfaction data loaded in the StatFAS bar plot maker."Screenshot 3:
statfas-bar-plot-settings.png · Alt: "Error bar and colour settings in the StatFAS bar plot maker."Screenshot 4:
statfas-bar-plot-result.png · Alt: "Finished bar plot with 95% confidence interval error bars." Use only genuine captures of the tool.Reading a bar plot
- Bar length. Compare lengths from a common baseline. Mistake: reading differences off a truncated axis.
- Error bars. First find out what they represent. Mistake: assuming SE bars show the spread of the data.
- Overlap. Non-overlapping 95% CIs suggest a difference; overlapping ones do not prove there is none.
- Sample size. A bar based on five observations looks identical to one based on five hundred. Report n in the caption.
- Ordering and colour. Ensure that colour carries meaning and remains distinguishable in greyscale and for colour-blind readers.
APA 7 reporting example (based on the illustrative data)
Figure 1 shows mean satisfaction by campus (n = 25 per campus). Mean satisfaction was highest at Campus D (M = 4.40, SD = 1.10) and lowest at Campus C (M = 3.50, SD = 0.80). Error bars represent 95% confidence intervals.
Essential elements: a numbered figure, a self-contained caption that states what the bars and error bars show, the sample size per group, and axis labels with units. Reference the figure in the text and report the corresponding statistical test separately.
Common mistakes and troubleshooting
- Truncated value axis. It exaggerates differences. Start at zero.
- Unlabelled error bars. Always state SD, SE or CI in the caption.
- Using bars for tiny samples. Show the individual data points.
- 3-D effects and decoration. They distort length judgements and add nothing.
- Too many categories or series. Past about 6–8 bars or colours, a table or a different chart is clearer.
- Stacked bars for comparison. Only the bottom segment shares a baseline; use grouped bars if segments must be compared.
- Alphabetical ordering. Sort by value when there is no natural order.
Limitations and alternatives
A bar of means discards the distribution, skew and outliers. Alternatives include box plots, violin plots and dot plots with individual points. For trends over time or across ordered conditions, use a line graph. To summarise categorical shares, see the pie chart guide and its limits. For summary numbers to accompany the figure, use the descriptive statistics calculator.
Frequently asked questions
Should error bars show SD, SE or a confidence interval?
Use SD for the spread of the data, and SE or a 95% CI for the precision of a mean. State which one in the caption.
Should a bar plot's y-axis start at zero?
Yes. Bar length encodes value, so truncation exaggerates differences.
Can I use a bar plot for Likert data?
Yes, but the mean of an ordinal scale is a simplification. A stacked or diverging bar of response percentages often shows more.
When is a bar plot a poor choice?
For small samples of continuous data, where individual points or distributions are more informative.
Conclusion
A good bar plot is honest about what it shows: a zero baseline, labelled error bars, the sample size and a sensible ordering. When you are ready to draw one from your own data, the StatFAS Bar Plot Maker provides a browser-based route, and the exported figure should be checked by you before publication.
References
- Cleveland, W. S., & McGill, R. (1984). Graphical perception: Theory, experimentation, and application to the development of graphical methods. Journal of the American Statistical Association, 79(387), 531–554.
- Cumming, G., & Finch, S. (2005). Inference by eye: Confidence intervals and how to read pictures of data. American Psychologist, 60(2), 170–180.
- Tufte, E. R. (2001). The visual display of quantitative information (2nd ed.). Graphics Press.
- Weissgerber, T. L., Milic, N. M., Winham, S. J., & Garovic, V. D. (2015). Beyond bar and line graphs: Time for a new data presentation paradigm. PLoS Biology, 13(4), Article e1002128.