# Pool your studies and draw the forest plot

> Upload a table with one row per study. Yila computes the effect sizes, fits a random-effects model, draws the forest plot with the heterogeneity statistics and tests a moderator, with the R script to reproduce it.

URL: https://yila.ai/playbooks/forest-plot

## What you get

- **Forest plot** (PNG + SVG): Every study with its counts, weight and estimate on a log axis, the pooled diamond, the prediction interval and the heterogeneity line.
- **Bubble plot** (PNG + SVG): The meta-regression on your moderator, each study as a point with the fitted line and its 95% confidence band.
- **Results table** (CSV): Per-study estimates, confidence intervals and weights, the pooled estimate with its prediction interval, and the meta-regression coefficients.
- **The script** (R script): The complete analysis in R with metafor, so a reviewer can rerun every number from your table.

## How it works

1. **Upload the study table and name the measure**: Give Yila one row per study: event counts and totals per arm, or an effect size with its standard error. Say which measure to pool and which study characteristic might explain differences.
2. **Yila checks the table and sets up the model**: It reads the columns, computes the effect sizes and fits the random-effects model. Say how each arm is coded, cases and non-cases or cases and totals, so the group sizes are right; zero-event studies are reported, not silently adjusted.
3. **Get the figures, the table and the script**: The forest plot and the bubble plot are rendered, checked for overlaps and saved with the results table and the script. Ask for subgroups, a funnel plot or a different estimator in the same conversation.

## Made for

- **A systematic review**: The main forest plot and the heterogeneity statistics a PRISMA review reports, ready for the results section.
- **A thesis chapter**: A pooled analysis whose script and outputs a committee can check line by line.
- **Checking a published meta-analysis**: Recompute a pooled estimate from the trial counts in its tables before you cite it.

## Questions

### What does my data need to look like?

One row per study. For binary outcomes, the number of events and the number of participants (or non-events) in each arm; for continuous outcomes, means, SDs and sizes; or an effect size with its standard error if you already have them. Say in the request how the columns are coded.

### Fixed or random effects?

Random effects by default, estimated with REML, because studies usually differ in population and design. If you have a reason for a common-effect model, ask for it and Yila reports both.

### What do Q, I² and τ² tell me?

Q tests whether the studies disagree more than chance allows, I² is the share of the variation that comes from real differences between studies, and τ² is the size of that between-study variance on the effect scale. The prediction interval turns τ² into the range you would expect for a new study.

### Can it explain the heterogeneity?

Yes, with a meta-regression on a study-level variable such as latitude, dose or year, shown as a bubble plot. With few studies Yila says how little a meta-regression can prove.

### Risk ratio, odds ratio or mean difference?

Any measure metafor supports: RR, OR and risk difference for binary outcomes, mean difference and SMD for continuous ones, and others such as hazard ratios from reported estimates. Name it in the request.

### Can I add subgroups or a funnel plot?

Yes, in the same conversation. Subgroup estimates, a funnel plot with Egger's test or a leave-one-out analysis are added to the same script.

### Where does the example data come from?

From the 13 BCG vaccine trials collected by Colditz et al. (1994) and distributed with the open-source metafor package as dat.bcg. Yila never uses users' data as examples.

---

- [Pool your studies and draw the forest plot](https://yila.ai/playbooks/forest-plot)
- [llms.txt](https://yila.ai/llms.txt)
