Dot Plot (Estimation Plot)

The Gardner-Altman estimation plot is the most information-dense two-group comparison display available. The upper panel shows all individual data points for both groups; the lower panel shows the effect estimate (mean difference, median difference, or Cohen's d) with its 95% CI as a point estimate and error bar. It is the visual equivalent of reporting both the p-value and the effect size —and makes the CI the primary display rather than the p-value.

STEP 1 —When to Use

Use an estimation plot when: you want to make the effect size and CI the primary visual message; you are comparing two groups and want to show both the raw data and the estimated difference; you are submitting to a journal that encourages estimation over significance-only reporting (Nature Methods, eLife, PLoS Biology).

Cumming estimation plots (multi-group extension) can handle >=3 groups with pairwise differences shown in the lower panel.

[Image placeholder: Gardner-Altman plot. Upper panel: two strip plots (Group A and Group B, n=12 each) with individual points, means shown as gap lines. Lower panel: mean difference = 4.2 units with 95% CI [1.8, 6.6], shown as a point with vertical CI line on a separate y-axis calibrated for the difference scale. The two panels share the same x-axis position.]

Gardner-Altman plot: raw data (top) with mean difference and 95% CI in a dedicated panel (bottom). The CI —not the p-value —is the primary visual message.

STEP 2 —Journal Requirement

The lower panel y-axis must be labeled with the effect metric: "Mean difference (Group A ~Group B)" with units.
The 95% CI must be shown (error bar on the point estimate) and the CI method must be stated (bootstrap or analytical).
n per group must be stated.
If a p-value is also reported, it must accompany —not replace —the effect size and CI.

STEP 3 —Licklider Implementation

Input

Continuous outcome and binary group variable (two-group) or multi-level group (Cumming plot)
Effect metric: mean difference (default), median difference, Cohen's d, or ratio of means
CI method: bootstrap (5000 iterations, BCa correction, default) or analytical

Output

Upper panel: individual data points (beeswarm layout) per group, with mean or median shown as a gap line
Lower panel: effect estimate as a filled circle, 95% CI as a vertical line, zero reference line
Bootstrap distribution shown as half-violin in the lower panel (optional)
Shared x-axis connecting upper and lower panels
p-value annotation optional (displayed below effect estimate)

[Image placeholder: Licklider Gardner-Altman plot. Upper: beeswarm strip plots for 2 groups (n=14 each), gap lines for means, individual points color-coded by group. Lower: mean difference = 3.8 (95% BCa CI [1.4, 6.1]), bootstrap distribution shown as orange half-violin, zero reference line dashed. p-value annotation: p = 0.003.]

Licklider estimation plot: individual data + gap lines (top), bootstrap-derived effect estimate with half-violin CI distribution (bottom).

STEP 4 —Draft Output (Draft / Needs review)

Data are presented as a Gardner-Altman estimation plot. Upper panel:
individual data points for each group (n = 14 per group). Lower panel:
mean difference (Group A ~Group B) = 3.8 units (95% BCa bootstrap CI
[1.4, 6.1]; 5000 bootstrap iterations). Two-sample t-test: t(26) = 3.2,
p = 0.003. Cohen's d = 1.21 (95% CI [0.39, 2.01]).