Which Chart Should I Use? Data Visualization Selection Guide
The best chart is the one that makes the analytical question easy to answer without distorting scale, uncertainty, or context. Start with the task, then choose the visual form.
Chart selection table
| Question | Strong starting chart | Avoid when |
|---|---|---|
| Compare categories | Bar chart or dot plot | Categories are ordered time points |
| Show a distribution | Histogram, density plot, box plot, or violin plot | Averages are the real decision target without distribution context |
| Examine a relationship | Scatterplot | Both variables are categorical |
| Show change over time | Line chart | Time points are unordered or very sparse |
| Show uncertainty | Point-range or interval plot | Intervals have different definitions that are not labeled |
| Show composition | Stacked bar for a small number of parts | Many categories make comparisons difficult |
| Show geography | Choropleth or symbol map | Location is not part of the question |
| Monitor stability | Control chart | Data are not time ordered or the chart type is wrong for the measure |
Comparing categories
Use bars from a common zero baseline when length encodes magnitude. Dot plots are compact and often make close comparisons easier. Sort categories meaningfully unless an established order matters.
Avoid three-dimensional bars. Perspective changes apparent size without adding information.
Showing distributions
A histogram reveals shape but depends on bin width. A box plot summarizes median, quartiles, and flagged extremes but can hide multimodality. A violin or density plot shows shape but may imply more precision than a small sample supports.
For important comparisons, combine raw or jittered observations with a clear summary when sample size permits.
Showing relationships
Use a scatterplot for two quantitative variables. Add a fitted line only when its model and uncertainty are relevant. Encode a third variable with facets or restrained color before using size or three-dimensional position.
Correlation does not establish causation. Read the statistical test selection guide before attaching inferential claims.
Showing time
Line charts work because adjacent time points are ordered. Use a consistent time interval, identify gaps, and distinguish actual observations from forecasts. For multiple series, direct labels or small multiples often outperform a crowded legend.
Forecasts should show prediction intervals and the point where observed data end. See the ARIMA forecasting checklist.
Showing uncertainty
Use points with confidence or credible intervals for estimates. Label the interval type and level. Do not assume readers know whether an error bar represents a standard deviation, standard error, confidence interval, or prediction interval.
Showing composition
Stacked bars work for a small number of categories, especially when totals and proportions both matter. Most segments lack a common baseline, so exact comparison is difficult. Use grouped bars, dots, or a table when the reader must compare every component precisely.
Choosing color and scales
- Use a sequential palette for ordered magnitude.
- Use a diverging palette only around a meaningful midpoint.
- Use a categorical palette for unordered groups.
- Do not rely on color alone to communicate a distinction.
- Explain logarithmic scales prominently.
- Avoid truncating a bar-chart axis in a way that exaggerates differences.
A five-question chart audit
- What exact question should the viewer answer?
- Is the visual encoding appropriate for the variable types?
- Are scales, denominators, units, sample sizes, and uncertainty clear?
- Can the chart be understood without relying on color alone?
- Does every decorative element earn its space?
Use the DataStatPro visualization tools to create statistical figures, then verify labels, scales, legends, accessibility, and export dimensions before publication.
Frequently asked questions
Should I use a bar chart or line chart?
Use a bar or dot chart for discrete category comparisons. Use a line chart when the horizontal axis is ordered, usually time, and connecting adjacent points is meaningful.
Are pie charts always wrong?
No, but they are difficult for precise comparison, especially with many or similar slices. A sorted bar chart is often easier to read.
When should I include error bars?
Include them when uncertainty or variability is important. State exactly what they represent and how they were calculated.
What is the best chart for correlation?
A scatterplot is the standard starting point for two quantitative variables because it reveals direction, form, clusters, and unusual observations.