Choose the Right Figure
A decision guide for selecting the right figure type for your data and research goal.
Start with what you want to show. The right figure depends on your data structure, your sample size, and what aspect of the data you are trying to communicate.
This page is about choosing a figure family, not about declaring a scientific conclusion automatically. Licklider uses the structure of your table and the figure type you choose to guide the next step, but you remain responsible for whether the figure and matched analysis fit your study design.
What you get from this choice
Choosing a figure in Licklider does more than change the visual style. It also determines what kind of output is likely to appear next.
| Figure family | Typical output in Licklider |
|---|---|
| Group comparison | A comparison figure plus matched statistical guidance. Depending on the analysis you run, Licklider may show the selected test, p-value or significance annotation, and for supported tests effect-size and 95% CI summaries. |
| Relationship and regression | A scatter or regression-oriented figure. Regression views can add a fitted line and confidence band, and the corresponding analysis can surface model summaries separately from the figure. |
| Time and longitudinal | A line-based figure showing change over an ordered sequence or over time, often with group summaries and error bars. |
| Specialized scientific | A figure tied to a specific analysis family, such as Kaplan-Meier, ROC, volcano, forest, or diagnostic plots. Some of these also surface analysis-specific summaries when the matching analysis is run. |
| Part to whole | A composition figure showing how a total is divided across categories. |
| Multivariate | An exploratory figure for clustering, dimensionality reduction, or pattern discovery across many variables. These views are descriptive and are not, by themselves, an inferential verdict. |
Step 1: What do you want to show?
| Goal | Figure type |
|---|---|
| Compare groups on a continuous outcome | Group comparison |
| Show the distribution of values within groups | Group comparison |
| Show the relationship between two continuous variables | Relationship and regression |
| Show change over time or across an ordered sequence | Time and longitudinal |
| Show survival or time-to-event data | Specialized scientific |
| Show proportions or parts of a whole | Part to whole |
| Show associations in multivariate data | Multivariate |
Group comparison
Use when your outcome is continuous and you are comparing two or more groups.
| Figure | Best for | Avoid when |
|---|---|---|
| Strip plot | n ≤ 8 per group; every point should be visible | n is large — overplotting makes it unreadable |
| Box plot | n ≥ 5 per group; comparing medians and spread | n < 5 — the box cannot be drawn meaningfully |
| Violin plot | n ≥ 10 per group; distribution shape matters | n < 10; multimodal or discrete data — KDE is unreliable |
| Group comparison mean and SEM | Parametric summaries; mean ± SEM, SD, or CI | Non-normal data where median is more appropriate |
| Dot plot | Small n; showing individual points alongside summary | Large n |
| Histogram | Single continuous variable; showing distribution shape | Comparing across groups simultaneously |
How Licklider chooses among these: Licklider uses sample size and distribution properties as inputs. For small n (roughly 8 or fewer per group), it favors strip plots. For medium n it favors box plots, often with individual points overlaid. For large, dense datasets it may suggest violin. Multimodal or discrete distributions reduce the score for violin plots. These are suggestions — the final choice is yours.
The n thresholds on this page are product heuristics for readability and chart eligibility, not universal scientific cutoffs. They help Licklider avoid plots that are visually misleading or unstable, but they do not by themselves prove that a method is statistically appropriate.
Relationship and regression
Use when you are examining the association between two or more continuous variables.
| Figure | Best for |
|---|---|
| Scatter plot | Exploring the relationship between two variables; overlaying a regression line |
| Regression plot | Showing a fitted model with confidence band |
| Bland-Altman plot | Agreement between two measurement methods |
Time and longitudinal
Use when the x-axis is time or another ordered sequence.
| Figure | Best for |
|---|---|
| Line chart | Group means over time; multiple groups as separate lines with error bars |
| Spaghetti plot | Individual trajectories over time (not yet available) |
Specialized scientific
These figures are used for specific analysis types and are typically generated alongside a matched analysis.
| Figure | Used with |
|---|---|
| Kaplan-Meier curve | Survival analysis |
| ROC curve | Logistic regression / binary classifier evaluation |
| Volcano plot | Differential expression or similar ranked-effect analyses |
| Forest plot | Meta-analysis or multi-study effect size summaries |
| Q-Q plot | Normality check (Diagnostic) |
| Residual plot | Regression diagnostics |
Part to whole
Use when showing how a total is divided across categories.
| Figure | Best for |
|---|---|
| Stacked bar chart | Comparing composition across groups |
| Pie chart | Showing a single composition (use sparingly) |
| Donut chart | Showing a single composition with a label in the center |
Multivariate
Use when exploring patterns across many variables at once. These figures require a minimum number of numeric columns.
| Figure | Minimum columns | Best for |
|---|---|---|
| Heatmap | 3+ numeric | Pairwise comparisons; expression matrices |
| PCA biplot | 3+ numeric | Dimensionality reduction; sample clustering |
| Parallel coordinates | 3+ numeric | Comparing profiles across variables |
| Hierarchical clustering heatmap | 5+ numeric | Unsupervised grouping by expression profile |
These minimum-column requirements are current product eligibility rules, not a claim that fewer columns are never scientifically analyzable.
What Licklider can and cannot determine automatically
Licklider can use column types, row counts, sample size per group, and some distribution or pattern checks to suggest a figure. That reduces obvious mismatches such as proposing a violin plot for very small groups or a multivariate figure when there are too few numeric columns.
Licklider cannot infer every design fact from the table alone. Hidden pairing, repeated measures without a declared ID, nested or clustered sampling, pseudoreplication, multiple-comparisons intent, batch confounding, and causal interpretation still require your judgment.
If those design facts are wrong or missing, the suggested figure may look reasonable while the matched statistical test is wrong for your study design.
Some specialized figures also depend on required columns and pattern detection. If time and event columns, label and score columns, or effect and confidence-interval columns are not recognized, Licklider may not suggest Kaplan-Meier, ROC, or forest plots automatically even when those figures would be scientifically appropriate.
How to change the figure
If Licklider's initial suggestion is not what you want, you can change it in two ways:
- Inspector — use the chart type control when a figure is active. Column mappings that are compatible between chart types are carried over automatically.
- Chat — describe the figure you want directly. For example: "Show this as a violin plot" or "Use a strip plot instead."
Figures and statistical tests
Choosing a group comparison figure triggers a statistical test automatically. This coupling is meant to reduce drift between the story told by the figure and the analysis run in the background.
For two-group comparisons, Licklider uses the data and the design information currently available to choose between a parametric and a non-parametric route. For example, selecting a box plot for a two-group comparison may run Welch's t-test or Mann-Whitney U depending on the data.
The outputs you may see include the selected test, significance annotations or omnibus results, and for supported tests effect-size and confidence-interval summaries in the statistical results surfaces. A figure can still be appropriate even when a given statistical result is not available for that figure or analysis path.
This automation does not guarantee that the selected test is scientifically correct for your design. If pairing, repeated-measures structure, clustering, or the planned comparison family is wrong or undeclared, change the test via the Inspector or Chat.
For guidance on which test to use — see Choose the Right Test.
What this page does not cover
- Detailed documentation for each figure type — see Figures and Visualization
- How to configure axis labels, colors, and style — see the individual figure page
- Choosing the right statistical test — see Choose the Right Test
- Common workflows combining figures and tests — see Common Workflows by Dataset Shape
Design rationale and references
This page treats figure selection as structured guidance rather than as an automatic verdict. That is why Licklider uses readability- and data-shape heuristics for visual suggestions, keeps the final figure choice user-controllable, and links group-comparison figures to matched statistical guidance so that the visualization and the analysis do not drift apart.
The sample-size and column-count thresholds on this page are implementation heuristics for chart eligibility and display quality. They are not presented as universal scientific thresholds. Where the page refers to statistical-method selection, the rationale should be read together with the matched test-selection guidance.
- Hintze, J. L., & Nelson, R. D. (1998). Violin Plots: A Box Plot-Density Trace Synergism. The American Statistician, 52(2), 181-184.
- Fay, M. P., & Proschan, M. A. (2010). Wilcoxon-Mann-Whitney or t-test? On assumptions for hypothesis tests and multiple interpretations of decision rules. Statistics Surveys, 4, 1-39.