Audit a Fixed-Effect Inverse-Variance Calculation

Prepare comparable estimates

Create one row per independent estimate intended for the synthesis. Record the study, comparison, outcome, time point, effect scale, estimate, and standard error. Verify that multiple publications or shared comparator groups have not introduced dependent rows disguised as separate studies.

A fixed-effect calculation needs a defensible common-effect interpretation. Mathematical compatibility is insufficient: two estimates can share units while addressing different populations or questions. Establish the grouping decision in the review plan before using the pooled number to support a conclusion.

Understand what determines the weights

For an estimate with standard error SE, the inverse-variance weight is 1/SE². The weighted mean divides the sum of weight times estimate by the sum of weights. Its conventional standard error is the square root of one divided by that weight sum.

Cochrane's meta-analysis methods chapter describes generic inverse-variance synthesis and its assumptions. A large weight reflects the supplied precision, not a certificate of study quality. An incorrectly small SE can dominate the answer even when the arithmetic is flawless.

Recompute a fictional result

Suppose two independent estimates on the same additive scale are 1 and 3, each with SE 1. Each weight is one, giving a pooled estimate of 2 and SE approximately 0.707107. A normal 95% confidence interval is approximately 0.6141 to 3.3859.

Use the inverse-variance pooling tool to reproduce these invented inputs. Both studies contribute 50% of the weight. This tiny example demonstrates the calculation; it cannot establish that a common-effect model is appropriate or support a clinical conclusion.

Audit the scale and precision

This tool is presented for difference-scale estimates. Do not enter raw risk ratios. Ratio synthesis ordinarily requires a log-scale specification and an appropriate implementation that preserves that interpretation. Mixing raw and log estimates is invalid, and any exponentiation for reporting must transform confidence limits consistently. Keep a separate record of the scale rather than relying on unlabeled numbers.

The meta-analysis weight worksheet should distinguish reported SEs from reconstructed ones. Check interval levels, conversion methods, units, and rounding when recovering precision from another statistic. Retain the original source values.

Inspect each weight alongside the study design and risk-of-bias information. A sensitivity analysis can reveal dependence on an influential study, but it should not become a search for exclusions that make the desired result appear. Investigate heterogeneity separately and do not treat a non-significant heterogeneity test as proof of a common effect.

Finish with the input table, method, assumption statement, normalized weights, and unresolved comparability questions. The calculator supplies a limited numerical audit; it does not select the synthesis model, estimate every source of uncertainty, or replace substantive review judgment.

Source checked on 2 November 2026; fictional calculation independently recomputed.