How it works | metaConvert

The metaConvert approach

Most effect-size calculators are formula-first: you pick a formula, then supply the inputs it needs. metaConvert is data-first — you record what each study reports, and the applicable formulas follow from the columns you filled in. Nothing to choose per study, and no manual conversion.

Other tools

Formula first

You choose the formula, then fit the data to it.

a formula

chosen by hand

its inputs

supplied to match

one effect size

metaConvert

Data first

You enter the data; the formulas follow from it.

your columns

mean_expmean_sd_expn_exp

every applicable formula

means+SDtOR

Hedges' g = 0.44

The procedure

1

Record each statistic under its predefined column name

Every quantity has a fixed column name that metaConvert recognizes — mean_exp for the experimental-group mean, n_exp for its sample size, student_t for a t value. The column name identifies the quantity, so no formula selection or manual mapping is required.

The Input data page lists every recognized column

mean_expmean of the experimental group

mean_sd_expits standard deviation

n_expits sample size

student_ta t statistic, where reported

# the name identifies the quantity

2

Enter the statistics available; leave the others empty

Each formula has required inputs and, frequently, interchangeable alternatives — a standard deviation or a standard error, for instance. Empty cells are expected: each study is processed independently, using whichever statistics it provides.

✓ required ○ interchangeable alternative · may be left empty

mean_exp	mean_sd_exp	mean_se_exp	n_exp
201.9	88.0	NA	24

# SD supplied, SE left empty — estimation proceeds

3

Applicable formulas are computed; the most reliable is retained

For each study, metaConvert applies every formula the available columns support and retains a single estimate following a predefined hierarchy of reliability. Estimates from means and standard deviations, for example, take precedence over those recovered from a p-value alone.

means + standard deviations rank 1 retained

t statisticrank 3

p-value onlyrank 6

# a slice of the reliability hierarchy — the highest-ranked applicable route is kept

4

Inputs and effect sizes are screened for extraction errors

The metaDETECT framework screens both the statistics you enter and the effect sizes they produce, checking each against more than fifty criteria and flagging implausible results — a confidence interval inconsistent with its estimate, or a standardized mean difference beyond a plausible range — before the data are pooled.

See the metaDETECT criteria, with worked examples

Unusual Large SMD: |g| = 3.05 (threshold: 3) (from means_sd)

Invalid Estimate lies outside its confidence interval

# flags attached per study, prior to pooling

Worked example

These studies report different statistics — group means and SDs, a t statistic, an odds ratio — and one reports more than one usable input. Each is reduced to the same target measure, Hedges' g; where several inputs are available, metaConvert computes the effect size from each and retains the most reliable estimate.

study	mean_exp	mean_sd_exp	n_exp	student_t	or	Hedges' g
Adler 2018	24.6	6.3	38	NA	NA	0.44
Brandt 2019	NA	NA	41	2.31	NA	0.51
Okafor 2020	NA	NA	52	NA	2.18	0.43
Petrov 2021 2 inputs	27.4	7.1	46	2.42	NA	0.49

A single call to summary() returns the right-most column. The target measure is the same for every study; the formula differs according to the statistics each study provides — no per-study formula choice or manual conversion.

Petrov 2021 reports two usable inputs — metaConvert uses both

means + SD → g = 0.49 rank 1 retained

t statistic → g = 0.48 rank 3

metaConvert computes the effect size from every usable input, then retains the most reliable by the hierarchy. Computing both routes is itself a consistency check: here the two estimates agree to within 0.01, and a marked disagreement would be flagged by metaDETECT.

Tutorial