Forest Plot

What a forest plot shows, what columns are required, how the null effect reference line is set, and where the current support boundary sits.

A forest plot displays effect size estimates and their confidence intervals for multiple studies, subgroups, or predictors in a single figure. Each row represents one estimate, shown as a point with a horizontal line extending to the confidence interval bounds.

The figure is a visualization layer for already computed estimates. It helps readers compare direction, magnitude, and uncertainty across rows, but it does not by itself generate or validate the underlying statistical model.

What the figure shows

Effect size points and CI bars

Each row shows one effect estimate as a filled marker, with a horizontal bar extending from the lower to the upper bound of the confidence interval. Wider bars indicate greater uncertainty; narrower bars indicate more precise estimates.

Row labels

When a label column is provided, each row is labeled on the y-axis with the study name, subgroup, or predictor name.

Null effect reference line

A vertical dashed line marks the null effect value — the value that would indicate no difference between groups. The default null value is 1, appropriate for odds ratios, hazard ratios, and relative risks. For mean differences and standardized mean differences, the null value is 0.

This distinction matters because ratio measures encode "no effect" as multiplicative equality (1), while difference measures encode "no effect" as additive equality (0).

To use a different null value, specify it in the Chat:

  • "Use a null effect of 0 for this forest plot"

Licklider can suggest a sensible default from the effect measure, but it does not automatically guarantee that the imported effect column is on the intended scale. If an odds ratio is supplied as if it were a mean difference, or vice versa, the reference line can be misleading even if the figure renders cleanly.

Required columns

A forest plot requires three numeric columns:

  • Effect column — the point estimate for each row (odds ratio, hazard ratio, mean difference, or similar)
  • CI lower column — the lower bound of the confidence interval
  • CI upper column — the upper bound of the confidence interval

An optional label column provides the row names shown on the y-axis. Without it, rows are numbered from the top.

Licklider detects these columns automatically from common naming patterns such as effect, or, hr, rr, ci_low, and ci_up, and ci_hi.

Automatic detection is a convenience, not a semantic guarantee. You should still verify that the chosen effect and CI columns refer to the same model, use the same scale, and correspond to the same row labels.

How to read the figure

Read a forest plot row by row:

  1. Identify the point estimate and whether it lies to the left or right of the null line.
  2. Check whether the confidence interval crosses the null line; if it does, the data remain compatible with little or no effect on that scale.
  3. Compare interval width across rows to judge which estimates are more or less precise.

The visual comparison is often the main value of the figure. Forest plots are especially good at showing how a result changes across predictors, subgroups, or related analyses without forcing the reader to scan a dense coefficient table.

What is not currently shown

A pooled effect diamond — the traditional diamond shape at the bottom of a meta-analysis forest plot representing the combined estimate — is not currently available. The figure shows individual estimates only.

That means this page should not be read as a full meta-analysis display. It supports row-wise effect comparison, not pooled meta-analytic synthesis.

Use cases

Forest plots in Licklider are a general-purpose figure for any data with an effect and confidence interval structure. Common uses include:

  • Displaying hazard ratios from a Cox regression model with multiple predictors
  • Comparing effect sizes across subgroups of a study
  • Summarizing results from multiple related analyses side-by-side

The figure does not automatically connect to Cox regression output — the effect, CI lower, and CI upper columns must be present in the dataset.

Design rationale and references

Licklider treats the forest plot as a general-purpose effect-and-interval display because the core visual grammar is useful well beyond formal meta-analysis. The figure works whenever you already have one effect estimate and one confidence interval per row and want to compare them in a compact visual form [1, 2].

The null line defaults to 1 for ratio measures and 0 for difference measures because those are the standard no-effect values on their respective scales [1, 2]. Making that reference explicit helps readers avoid the common mistake of interpreting all forest plots as if they used the same effect metric.

The current public scope excludes a pooled diamond on purpose. A pooled summary estimate implies additional modeling choices about weighting, heterogeneity, and meta-analytic synthesis that go beyond a simple row-wise forest display.

Methodological foundations

  1. Harrer, M., Cuijpers, P., Furukawa, T. A., & Ebert, D. D. (2021). Doing Meta-Analysis With R: A Hands-On Guide (ch. 6). Chapman & Hall/CRC. -> Clear reference for reading forest plots, effect scales, and the role of the null line.

  2. Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2009). Introduction to Meta-Analysis. Wiley. -> Standard reference for forest-plot interpretation and the distinction between ratio and difference measures.

Current support boundary

  • Forest plot in Licklider visualizes supplied effect estimates and confidence intervals; it does not by itself fit Cox, logistic, or meta-analytic models.
  • Licklider does not automatically guarantee that the detected effect column, CI bounds, and label column come from the same model output or the same statistical scale.
  • Licklider does not automatically infer whether a ratio measure should be shown on a transformed scale, nor does this page claim full support for every measure-specific convention.
  • A pooled meta-analytic diamond is not currently shown, so the figure should not be interpreted as a complete pooled meta-analysis display.
  • The figure is therefore best used to compare already computed row-wise estimates, not as proof that those estimates were generated under the correct model assumptions.

What this page does not cover