Written and maintained by CASRAI Editorial Board
Last updated
John Hattie is an education researcher best known for Visible Learning (2008), a synthesis of more than 800 meta-analyses of achievement. His method puts every influence on learning on one scale, Cohen's d, and ranks them. This guide explains what that scale measures, how to read the ranking, which edition of it to cite, and what statisticians have criticised.
What is John Hattie best known for?
Hattie is best known for Visible Learning: A Synthesis of Over 800 Meta-Analyses Relating to Achievement (Routledge, 2008), which collects effect sizes from existing meta-analyses and ranks the influences they study, from the student and the home to the school, the teacher and teaching strategies. The Visible Learning site describes the ranking as a way of comparing influences on achievement by their effect size (Cohen's d).
The ranking has been revised several times, so the number you see depends on the edition:
| Edition | Influences ranked |
|---|---|
| Visible Learning (2009) | 138 |
| Visible Learning for Teachers (2011) | 150 |
| The Applicability of Visible Learning to Higher Education (2015) | 195 |
| Current list (December 2017 update) | 250-plus (252 in the table body) |
The same site says the research is now based on nearly 1,200 meta-analyses, up from about 800 when the book appeared. Cite the edition you actually used: a rank number from the 138-item list is not comparable with a rank from the 252-item list.
What is the 0.40 hinge point?
The hinge point is 0.40, the average effect size across all the interventions Hattie studied. He judges each influence against it: above 0.40 is read as more effective than the typical influence, below as less. The figure is an average of the data set, not a threshold derived from a theory of how much learning is "enough". That distinction drives most of the criticism below.
How is a Hattie effect size calculated?
The unit is Cohen's d, the standardized mean difference: the difference between two group means divided by their pooled standard deviation. For background on the statistic itself see effect size, and for the small-sample corrected version used in meta-analysis see standardized mean difference and Hedges' g.
Formula. d = (M1 − M2) / SDpooled, where M1 and M2 are the group means and SDpooled = √[((n1 − 1)SD1² + (n2 − 1)SD2²) / (n1 + n2 − 2)].
Illustration with made-up numbers (not data from any study). Suppose a taught group has M1 = 78.0, SD1 = 9, n1 = 25 and a comparison group has M2 = 72.0, SD2 = 11, n2 = 25.
- Pooled variance = (24 × 81 + 24 × 121) / 48 = 4,848 / 48 = 101
- SDpooled = √101 = 10.05
- d = (78.0 − 72.0) / 10.05 = 0.60
A d of 0.60 means the taught group's mean sits 0.60 pooled standard deviations above the comparison group's. In a meta-analysis, a study-level d like this is then combined with others, which is where the comparability problems in the criticisms section arise (see inverse-variance weighting in meta-analysis for how a pooled estimate is normally built).
What are some Visible Learning strategies?
The Visible Learning ranking lists teaching strategies alongside student, home, school and teacher influences. These are published d values from the December 2017 version of the ranking on the Visible Learning site; the rank is the position among all 252 influences.
| Influence | Rank | d (Dec 2017) |
|---|---|---|
| Collective teacher efficacy | 1 | 1.57 |
| Self-reported grades | 2 | 1.33 |
| Teacher estimates of achievement | 3 | 1.29 |
| Response to intervention | 5 | 1.29 |
| Jigsaw method | 7 | 1.20 |
| Classroom discussion | 15 | 0.82 |
| Scaffolding | 16 | 0.82 |
| Deliberate practice | 17 | 0.79 |
| Summarization | 18 | 0.79 |
| Feedback | 32 | 0.70 |
Two points matter when quoting this list. First, several top-ranked items (self-reported grades, teacher estimates of achievement) describe what already predicts achievement, not something a teacher does as a strategy. Second, values move between updates: teacher estimates of achievement read 1.62 in August 2017 and 1.29 in December 2017 on the same site.
How to read Hattie's effect sizes
| Reading | What it supports | What it does not support |
|---|---|---|
| d above 0.40 | The influence scored above the average of the studies in Hattie's data set | A claim that it is "the desired effect zone" in any absolute sense |
| d between 0 and 0.40 | The influence scored below that average, with a positive difference between groups | A claim that it does nothing |
| d below 0 | The comparison group scored higher in the pooled studies | A causal claim without checking study designs |
| Rank position | Order within one edition of the list | Comparison with rank numbers from another edition |
Cohen's own 1988 conventions (0.2 small, 0.5 medium, 0.8 large) were offered as a fallback when nothing field-specific is known, not as a rule, and the 0.40 hinge point is a different thing again: an observed average. Treat any one d as a standardized difference between specific groups on a specific measure, then ask what the groups, duration and outcome measure were before comparing it with another d. A confidence interval around d, which a meta-analysis can supply, shows how precisely it is estimated (see confidence interval).
Criticisms of the method
The most detailed statistical critique is by Pierre-Jerome Bergeron, of the University of Ottawa, published in the McGill Journal of Education in 2017. Its main points, quoted or paraphrased from the paper:
- d is not a universal unit. The paper argues that "Cohen's d (Hattie's measure of effect size) simply cannot be used as a universal measure of impact." Averaging d across studies with different durations, populations and outcome measures produces a number that does not describe any single real situation.
- The 0.40 classification has no justification. Bergeron writes that "Above 0.4 represents the desired effect zone. There is no justification for this classification."
- Common language effect size (CLE) errors. The paper reports that a calculation error in the CLE values, noticed in 2012 by Norwegian researchers (Topphol, 2012), was "flagrant to the point of giving negative probabilities or probabilities superior to 100%." A probability cannot be below 0 or above 1, so a CLE outside that range signals a calculation fault, not a large effect.
Bergeron's paper is one critique, and its wording is strong; weigh it alongside the Visible Learning site's own account of the ranking. The practical lesson is method-neutral: do not treat a ranked list of pooled d values as a ranking of what will work in your classroom.
How to cite Visible Learning and the ranking (APA)
When you quote a Hattie value, name the edition or update date of the ranking and the date you retrieved it, and state the comparison and outcome measure behind any d you report. Reference-list patterns:
- Hattie, J. (2008). Visible learning: A synthesis of over 800 meta-analyses relating to achievement. Routledge.
- Bergeron, P.-J. (2017). How to engage in pseudoscience with real data: A criticism of John Hattie's arguments in Visible Learning from the perspective of a statistician (L. Rivard, Trans.). McGill Journal of Education, 52(1). https://mje.mcgill.ca/article/view/9475
In text, write something like: "Feedback had d = 0.70 in the December 2017 version of the Visible Learning ranking (Visible Learning, 2017 update, retrieved 2026-10-29), with the full URL in the reference list."
How to calculate Cohen's d in Excel, R or Python
Excel, with means in B1 and B2, standard deviations in C1 and C2, and sample sizes in D1 and D2:
=(B1-B2)/SQRT(((D1-1)*C1^2+(D2-1)*C2^2)/(D1+D2-2))R:
m1 <- 78; m2 <- 72; s1 <- 9; s2 <- 11; n1 <- 25; n2 <- 25
sp <- sqrt(((n1-1)*s1^2 + (n2-1)*s2^2) / (n1+n2-2))
(m1 - m2) / sp # 0.597Python:
from math import sqrt
m1, m2, s1, s2, n1, n2 = 78, 72, 9, 11, 25, 25
sp = sqrt(((n1-1)*s1**2 + (n2-1)*s2**2) / (n1+n2-2))
print((m1 - m2) / sp) # 0.597All three return 0.597, which rounds to the 0.60 worked above. For meta-analysis in R see meta-analysis in R with metafor.
Frequently asked questions
What is John Hattie best known for?
The Visible Learning synthesis of meta-analyses on achievement and its ranked list of influences, compared using Cohen's d with a 0.40 hinge point.
What are some Visible Learning strategies?
Items in the ranking include feedback (d = 0.70), deliberate practice (0.79), summarization (0.79), scaffolding (0.82), classroom discussion (0.82) and the jigsaw method (1.20), all December 2017 values; see the table above for the caveats.
What is the 80/20 rule in teaching?
The 80/20 rule (the Pareto principle) is a general idea about uneven effects. It is not part of Hattie's published ranking or method, which is based on effect sizes and the 0.40 hinge point, so do not attribute it to him.
Is 0.40 a good effect size in education?
It is the average of the influences in Hattie's data set. Bergeron's critique argues there is no justification for treating everything above it as the "desired effect zone".
Sources
- Visible Learning, "Hattie Ranking: 252 Influences And Effect Sizes Related To Student Achievement", https://visible-learning.org/hattie-ranking-influences-effect-sizes-learning-achievement/ (checked 2026-10-29).
- Bergeron, P.-J. (2017). McGill Journal of Education, 52(1), https://mje.mcgill.ca/article/view/9475 (checked 2026-10-29).
- Hattie, J. (2008). Visible learning: A synthesis of over 800 meta-analyses relating to achievement. Routledge.
- Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum.
