Distinguish precision of the mean from variation across settings

Identify which quantity each output describes

The metafor predict.rma documentation, checked 7 November 2026, distinguishes confidence limits for an average from conditional prediction limits for true effects. Its assumptions and approximations matter. Have the analyst identify the applicable model and target population before interpreting either output.

Create a two-row audit from a supplied output extract. Preserve the original labels, units, direction, interval level and model version. Record which field supplied each endpoint instead of copying only the numbers. A decorative diamond and a horizontal line can look interchangeable after a chart is restyled. The source labels must survive that edit so the reviewer can tell whether the presentation has changed the statistical meaning.

Keep unavailable information explicit. If the extract contains only one interval, do not infer the other from its width or fabricate a matching pair. Ask for the missing output or remove the unsupported comparison. This is an interpretation check on an approved analysis, not permission to manufacture results or to select a more favorable estimation method.

Read a pair with a positive center but different boundaries

Borenstein’s methodological explanation, checked 7 November 2026, separates precision of the mean from dispersion of effects. The confidence interval is not a substitute for a prediction interval when describing effect variation. The following numbers are original illustrative outputs, not calculations from published studies.

In fixture A, the reported mean improvement is four percentage points. The mean-confidence row runs from two to six; the prediction row runs from minus three to eleven. Both have midpoint four, but their full widths are four and fourteen respectively. Calculate those widths independently. Correct arithmetic should not tempt the editor to choose the smaller interval simply because it makes the slide cleaner.

Using zero as the direction boundary for this exercise, the first row is wholly positive and the second crosses zero. Write those two observations beside their respective labels. A claim that all comparable implementations improve would outrun the second output. That does not erase the favorable estimated mean; it identifies a different claim that the supplied evidence does not establish. No probability for a particular future campaign is calculated here.

Test what changed in a revised output

Now inspect invented fixture B from a proposed update: the mean remains four, the mean-confidence limits are 3.5 and 4.5, and the prediction limits remain minus three and eleven. Its first width is one rather than four; its second remains fourteen. These stipulated values isolate an interpretation problem. They do not purport to show what a particular dataset or software procedure must produce after more studies arrive.

Compare the two rows separately. The estimate of the mean is more precise in this fixture, while the displayed prediction range has not narrowed. Reject the draft sentence that the results have become uniform. Replace it with a description of the actual changed output and retain the unchanged row. Ask the analyst whether additional uncertainty or model changes affect that description before it reaches the final report.

Confirm that the update uses the same effect direction, units and intended comparison. A change in sign convention or a different outcome cannot be interpreted as improved precision by comparing widths alone. Preserve the previous extract alongside the new one so another reviewer can reconstruct the revision instead of relying on the editor’s recollection.

Repair the presentation at the output boundary

Check each chart label, caption and executive sentence against the audit. Does the sentence name the mean, effect variation, or an actual business forecast? If the draft has silently changed targets, correct that sentence rather than altering the analysis output to fit it. A missing prediction row should be restored when the claim requires it, or the claim should be narrowed if that output is unavailable.

Keep the analyst’s scope statement next to the comparison. A new setting outside the model’s intended evidence population requires separate examination. The presence of a prediction interval does not certify that similarity or repair biased study inputs. Document any unresolved scope question as a limitation rather than assigning it a reassuring number.

Close with the paired output extract, independent width and boundary checks, corrected wording and the remaining analyst questions. The completed artifact establishes which reported result supports which sentence. It does not choose a rollout, replace a model-validity review or promise that one future result will fall inside a particular range.

Make the reasoning concrete.

An invented editorial update replaces fixture A with fixture B. The designer keeps only the narrower row and labels the result more uniform across implementations.

The editor restores the paired comparison and changes the claim. No actual analysis, business rollout or forecast is performed.

Take the questions into your next brief.

Use these prompts in your own notes. Nothing is submitted from this page.

1. Label

   Which original output field and target belong to each row?

2. Arithmetic

   Are the endpoints, midpoints and widths copied and calculated correctly?

3. Boundary

   Which labeled row crosses the stipulated zero boundary?

4. Revision

   What changed in each row between the two supplied fixtures?

5. Sentence

   Which exact output supports each retained claim?

Watch for the wrong conclusion.

• Selecting the narrowest interval without preserving its label.
• Calling a more precise mean evidence of uniform effects.
• Inferring a missing interval from another reported width.

Sources & scope

Original adult business scenarios are explicitly fictional. Methodological statements and applications remain source-derived. Qualified analysts and review teams own formal assessment. No clinical guidance, actual research operations or publication occurs.

• metafor: Predicted Values for rma Objects →Primary technical documentation · Supports: Supports the distinct output targets and conditional prediction interpretation.
• How to understand and report heterogeneity in a meta-analysis: The difference between I-squared and prediction intervals →Primary methodological explanation · Supports: Supports the distinction between precision of a mean and dispersion of effects; all supplied fixture values and arithmetic are original.