The Hattie Effect Size Calculator calculates Cohen’s d from pre-test and post-test means and standard deviations to gauge educational impact.
Report an issue
Spotted a wrong result, broken field, or typo? Tell us below and we’ll fix it fast.
What Is a Hattie Effect Size Calculator?
A Hattie effect size calculator estimates the standardized impact of an educational intervention, typically using Cohen’s d or Hedges’ g. Effect size is a unitless measure that expresses how much an outcome changed relative to the variability in the data. John Hattie popularized the use of effect sizes across thousands of studies and introduced a “hinge point” of 0.40 as a practical benchmark.
This calculator follows those conventions. It converts raw differences between groups or time points into a standardized metric that can be compared across grades, subjects, and tests. Results can be interpreted alongside Hattie’s hinge point to guide action in schools.
How the Hattie Effect Size Method Works
Effect size compares a difference in means to the spread of scores. The spread is measured by a standard deviation, which reflects how far scores vary around the average. For two independent groups, the typical reference spread is the pooled standard deviation. For repeated measures, the correlation between pre and post matters because the same students are measured twice.
- Compute the mean difference relevant to your design (between groups or pre vs. post).
- Estimate the appropriate standard deviation reference (pooled across groups or adjusted for within-student correlation).
- Divide the difference by that reference spread to obtain Cohen’s d.
- Apply a small-sample correction to get Hedges’ g, which is less biased in small samples.
- Estimate a standard error and 95% confidence interval to express uncertainty.
The result is a standardized number that is comparable across different tests and grade levels. Positive values indicate improvement in the treatment or from pre to post. Negative values indicate a decline or that the control outperformed the treatment.
Hattie Effect Size Formulas & Derivations
Below are core formulas used by the calculator. Terms are defined on first use. For an independent-groups design, Cohen’s d standardizes the mean difference by a pooled standard deviation. For repeated measures, the standardization accounts for the pre–post correlation.
- Independent groups (treatment t vs. control c): d = (M_t − M_c) / s_p, where s_p = sqrt(((n_t − 1)s_t^2 + (n_c − 1)s_c^2) / (n_t + n_c − 2)).
- Small-sample correction (Hedges’ g): g = J × d, where J = 1 − 3 / (4df − 1) and df = n_t + n_c − 2.
- Repeated measures (single group, pre vs. post) with correlation r: d_rm = (M_post − M_pre) / (s_av × sqrt(2(1 − r))), where s_av = sqrt((s_post^2 + s_pre^2) / 2).
- Alternative repeated measure using change-score SD: d_z = (M_post − M_pre) / s_diff, where s_diff is the standard deviation of change scores.
- Approximate variance for g (independent groups): Var(g) ≈ (n_t + n_c) / (n_t n_c) + g^2 / (2(n_t + n_c − 2)), so SE(g) = sqrt(Var(g)), and 95% CI = g ± 1.96 × SE(g).
- Useful conversions: from correlation r, d = 2r / sqrt(1 − r^2); from t-statistic, d = t × sqrt(1/n_t + 1/n_c) (independent groups).
These relationships link design choices, inputs, and the final standardized result. Hedges’ g is recommended for reporting because it reduces bias when sample sizes are modest. Confidence intervals help judge precision and compare effects across studies and contexts.
Inputs, Assumptions & Parameters
The calculator accepts common statistics from classroom studies and program evaluations. It supports both independent groups and repeated measures. Whenever possible, include the pre–post correlation for within-student designs for the most accurate standardization.
- Group means: M_t and M_c for treatment and control, or M_pre and M_post for repeated measures.
- Standard deviations: s_t and s_c, or s_pre and s_post; optionally s_diff for change scores.
- Sample sizes: n_t and n_c, or n for repeated measures.
- Pre–post correlation r (if available) to adjust for the dependency in repeated measures.
- Confidence level (e.g., 95%) to compute intervals around the effect size estimate.
Effect sizes are unitless, but the inputs must reflect the same scale within a comparison. Extreme outliers, heavy-tailed distribution shapes, or severe ceiling/floor effects can distort standard deviations and inflate or deflate the result. If variances differ greatly between groups, consider Glass’s Δ, which uses the control SD, or use a robust alternative.
How to Use the Hattie Effect Size Calculator (Steps)
Here’s a concise overview before we dive into the key points:
- Select your study type: independent groups or repeated measures.
- Enter the sample sizes and means for all relevant groups or time points.
- Enter the standard deviations; add the pre–post correlation if you have it.
- Choose your confidence level and whether to report Cohen’s d or Hedges’ g.
- Click Calculate to compute the standardized effect and confidence interval.
- Review notes about assumptions and data suitability shown beneath the result.
These points provide quick orientation—use them alongside the full explanations in this page.
Case Studies
An eighth-grade math program was piloted in two classes. Control: n_c = 30, M_c = 68, s_c = 10. Treatment: n_t = 28, M_t = 75, s_t = 12. Pooled SD s_p ≈ 11.01, Cohen’s d ≈ (75 − 68) / 11.01 = 0.64. Hedges’ g with df = 56 gives J ≈ 0.987, so g ≈ 0.63. The estimated SE ≈ 0.27, so the 95% CI is roughly 0.10 to 1.16. This suggests a meaningful positive effect, comfortably above the 0.40 hinge point.
What this means
A reading fluency initiative was tracked pre to post with the same students. n = 35, M_pre = 50, s_pre = 9; M_post = 55, s_post = 11; r = 0.70. Average SD s_av ≈ 10.05 and adjustment sqrt(2(1 − r)) ≈ 0.775. Cohen’s d_rm ≈ 5 / (10.05 × 0.775) = 0.64; Hedges’ g ≈ 0.63 with small-sample correction. The effect is near two-thirds of a standard deviation and above 0.40, indicating likely practical gains.
What this means
Accuracy & Limitations
Effect sizes depend on the quality of the inputs and whether assumptions hold. They assume approximately normal distributions, reasonable variance stability, and accurate measurement. They also summarize average differences and cannot reveal who benefits most or least.
- Small samples increase uncertainty and can bias d upward; Hedges’ g reduces, but does not remove, bias.
- Non-normal or skewed score distributions can alter the standard deviation and distort the standardized result.
- Inequality of variances suggests using alternative standardizations (for example, Glass’s Δ) or robust methods.
- Clustered designs (students within classes) need multilevel adjustments for standard errors and confidence intervals.
- Publication or selection bias can inflate average reported effects in meta-analyses.
Use confidence intervals and study design checks to judge reliability. When possible, corroborate with additional outcomes, subgroup analyses, and sensitivity checks. Consider the educational context, not just the number.
Units and Symbols
Effect size is standardized, so it is unitless, but your inputs must share the same units within each comparison. The table below lists common symbols and how to interpret them within this statistics calculator.
| Symbol | Name | Units or Notes |
|---|---|---|
| d | Standardized mean difference | Unitless (standard deviations) |
| g | Bias-corrected effect size | Unitless; recommended for reporting |
| M | Mean score | Same units as the test or outcome |
| SD (s) | Standard deviation | Same units as the test; reflects variability |
| n | Sample size | Count of students or observations |
| r | Pre–post correlation | Unitless; −1 to +1 |
Read across each row to match a symbol to its meaning and units. When converting or combining statistics, ensure each symbol reflects the same group or time frame so the result is coherent.
Tips If Results Look Off
If the effect size seems too large, too small, or the confidence interval is huge, first check your inputs. Many issues stem from a misplaced decimal, swapped groups, or a missing correlation for repeated measures. The distribution of scores and the presence of outliers can also affect results.
- Verify means and SDs come from the same scale and testing window.
- Confirm n values match the reported SDs and means.
- For pre–post designs, add r; without it, effects may be overstated.
- Inspect histograms or boxplots for outliers or strong skew.
- Try Glass’s Δ if treatment variability is very different from control.
When in doubt, compute the effect two ways, such as using pooled SD and control SD, and compare. Large discrepancies can signal a variance problem that needs a design-based remedy.
FAQ about Hattie Effect Size Calculator
What is the difference between Cohen’s d and Hedges’ g?
Cohen’s d is the raw standardized mean difference, while Hedges’ g applies a small-sample correction (J) to reduce upward bias. For moderate or small samples, g is preferred for reporting.
What is Hattie’s 0.40 hinge point?
It is a practical benchmark from meta-analytic summaries suggesting that effects at or above 0.40 are generally associated with worthwhile educational impact. It is a guide, not a rule.
Can effect sizes be negative?
Yes. A negative value means the control outperformed the treatment or that post scores were lower than pre scores. Direction depends on how you define the contrast.
Do I need the pre–post correlation for repeated measures?
It is strongly recommended. Accounting for the correlation prevents overstating the effect and yields a more accurate standardization for within-student designs.
Hattie Effect Size Terms & Definitions
Effect Size
A standardized measure of the magnitude of a difference, usually expressed in standard deviations so results can be compared across studies.
Pooled Standard Deviation
A combined estimate of spread from two groups, weighted by their sample sizes, used to standardize the mean difference in independent designs.
Hedges’ g
A bias-corrected version of Cohen’s d that applies a correction factor based on degrees of freedom to improve accuracy in smaller samples.
Hinge Point
A reference value of about 0.40 popularized by Hattie, suggesting effects at or above this level are educationally meaningful in many contexts.
Confidence Interval
A range around the estimated effect size that likely contains the true effect, often computed at the 95% level to express uncertainty.
Standard Error
The estimated variability of the effect size estimator, used to build confidence intervals and to weight studies in meta-analysis.
Distribution
The pattern of scores or values for a variable, including its shape, center, and spread; assumptions about distribution affect standardization.
Meta-Analysis Weight
The influence assigned to a study’s effect size, often proportional to the inverse of its variance so more precise studies count more.
Sources & Further Reading
Here’s a concise overview before we dive into the key points:
- Visible Learning: Hattie’s rankings and effect sizes overview
- Coe, R. (2002). It’s the effect size, stupid: What effect size is and why it is important
- Morris, S. B. (2008). Estimating effect sizes from pretest–posttest–control group designs
- Hedges, L. V. (1981). Distribution theory for Glass’s estimator of effect size and related estimators
- Cochrane Handbook: General principles for calculating effect sizes
- metafor Project: Standardized mean difference (SMD) notes and formulas
These points provide quick orientation—use them alongside the full explanations in this page.