Written and maintained by CASRAI Editorial Board
Last updated
The Pygmalion effect is the finding that higher expectations from other people lead to improved performance. It is named after the sculptor of Greek myth whose statue came to life, and it is closely tied to the self-fulfilling prophecy and to self-efficacy. This page defines it, then does what most definitions skip: it shows how large the effect has actually been measured to be, how to convert between the effect sizes in the literature, and how to report one.
What is the Pygmalion effect? Definition in one paragraph
The Pygmalion effect is the phenomenon in which higher expectations from others lead to improved performance (FORRT, Open Social Psychology). The classic demonstration is Rosenthal and Jacobson (1968): teachers were told that some students were “late bloomers” destined for large IQ gains, although those students had been selected at random. The students labelled as bloomers showed larger gains than controls, especially in the earlier grade levels (Jussim, Teacher Expectations). The opposite pattern, where low expectations depress performance, is called the Golem effect.
Is there a formula? Converting the reported effect sizes
There is no native formula for the Pygmalion effect itself. It is a phenomenon, and what you calculate is an effect size for the difference between an expectancy group and a comparison group. The literature reports it in two metrics: a correlation (r) and a standardised mean difference (d). To compare them, use the standard conversion for two equal-sized groups:
r = d / √(d² + 4)
- d: the standardised mean difference, the gap between the group means divided by the pooled standard deviation.
- r: the equivalent correlation between group membership and the outcome.
- 4: a constant that applies when the two groups are the same size. Unequal groups need a different constant.
For background on the metrics, see the CASRAI dictionary entry for effect size and the guide to how to interpret Cohen’s d.
Worked example with real numbers
FORRT reports that published effect sizes for teacher expectancy range from d = .11 (Raudenbush, 1984) to d = .70 (Rosenthal and Rubin, 1978). Converting both ends of that range to r:
- Upper end: d = .70. d² = .49. d² + 4 = 4.49. √4.49 = 2.119. r = .70 / 2.119 = .33.
- Lower end: d = .11. d² = .0121. d² + 4 = 4.0121. √4.0121 = 2.003. r = .11 / 2.003 = .055.
The upper figure is a moderate correlation; the lower figure is close to zero. For comparison, a summary of 345 expectancy studies quoted by Jussim (Teacher Expectations) reports an overall effect equal to a correlation of about .30. Naturalistic classroom studies, which measure expectations teachers already hold rather than expectations experimenters induce, typically find self-fulfilling effects of about .1 to .2 as standardised regression coefficients (Jussim, Teacher Expectations).
Interpretation: how to read the numbers
| Source of estimate | Reported size | What it tells you |
|---|---|---|
| Rosenthal and Rubin (1978) | d = .70 (r = .33 by the conversion above) | The upper end of the range FORRT reports for teacher expectancy |
| Raudenbush (1984) | d = .11 (r = .055) | The lower end of the same range, close to no effect |
| Naturalistic classroom studies | .1 to .2 (standardised regression coefficients) | Real classrooms with real expectations show modest effects |
Two points matter for interpretation. First, the spread is wide: depending on the study design, the same phenomenon looks anywhere from negligible to moderate. Second, in naturalistic studies accuracy accounts for most of the link between teacher expectations and student outcomes: teachers’ expectations mostly predict performance because they are partly correct, not because they cause it (Jussim, Teacher Expectations).
Pygmalion effect examples
In the classroom
Teachers who hold high expectations tend to create more learning opportunities, ask richer questions and give more targeted feedback (Brophy and Good, 1970; Rubie-Davies, 2007, cited in FORRT). That is the proposed mechanism: expectations change teacher behaviour, and behaviour changes student outcomes.
Beyond schools
FORRT lists applications of the same idea in management (Eden, 1990), therapy (Jenner, 1990) and nursing homes (Learman et al., 1990).
Assumptions and common mistakes
- Treating the 1968 result as settled. Critiques include Thorndike (1968), who concluded that any observed effects might have been coincidental, and Snow (1969), who noted that 20% of participants were not retested (both cited in FORRT). Jussim notes that five bloomers had extreme gains, and that excluding those five made the difference between bloomers and controls disappear.
- Assuming it replicates reliably. Of earlier attempts to demonstrate a self-fulfilling prophecy, only about one third succeeded (Rosenthal and Rubin, 1978, as reported in Jussim, Teacher Expectations).
- Confusing expectation with accuracy. A teacher who predicts that a strong student will do well has not caused that result.
- Mixing it up with neighbouring effects. A participant adjusting behaviour because they guess the hypothesis is covered in demand characteristics; a researcher’s presence or measurement changing the outcome is the observer effect; a researcher’s expectations skewing what they record is observer bias.
- Combining d and r without converting. Pool them only after converting to one metric.
How to report it (APA example)
Report an effect size and its confidence interval alongside the test result, not a p-value alone. See how to report confidence intervals for the pattern. A template for an expectancy comparison, with your own values in place of the brackets:
Students whose teachers were led to expect gains improved more than controls, d = [value], 95% CI [lower, upper], equivalent to r = [value].
Because expectancy studies are vulnerable to experimenter expectancy, state whether the people collecting outcome data were blind to condition.
How to calculate the d to r conversion in Excel, R or Python
Excel (d in cell A1):
=A1/SQRT(A1^2+4)
R:
d <- 0.70 r <- d / sqrt(d^2 + 4) round(r, 3) # 0.33
Python:
import math d = 0.70 r = d / math.sqrt(d**2 + 4) print(round(r, 3)) # 0.33
Each returns .33 for d = .70 and .055 for d = .11, matching the arithmetic above.
FAQ
What is the Pygmalion effect?
It is the finding that higher expectations from others lead to improved performance. The best-known study is Rosenthal and Jacobson (1968), in which randomly chosen students were labelled as late bloomers.
What is the difference between the Pygmalion effect and the Golem effect?
The Pygmalion effect describes performance rising with high expectations; the Golem effect describes performance falling with low expectations.
How big is the Pygmalion effect?
Estimates range from d = .11 to d = .70 (r about .055 to .33). Naturalistic classroom studies typically find about .1 to .2.
Is the Pygmalion effect real?
Expectations do change how teachers behave, but the original result has been criticised, and only about one third of early attempts to demonstrate a self-fulfilling prophecy succeeded. Treat it as a small and variable effect.
What are examples of the Pygmalion effect?
The best-known example is the Rosenthal and Jacobson (1968) classroom study. FORRT also lists management (Eden, 1990), therapy (Jenner, 1990) and nursing homes (Learman et al., 1990).
Sources
- Rosenthal, R., and Jacobson, L. (1968). Pygmalion in the classroom.
- Rosenthal, R., and Rubin, D. B. (1978). Interpersonal expectancy effects: The first 345 studies. Behavioral and Brain Sciences, 3, 377-386.
- Jussim, L. Teacher Expectations, in The Psychology of Classroom Learning (Anderman and Anderman, eds.), sites.rutgers.edu (https://sites.rutgers.edu/lee-jussim/wp-content/uploads/sites/135/2019/05/TEs-7.doc), date checked 2026-10-29.
- FORRT, Open Social Psychology, “Pygmalion Effect”, forrt.org (https://forrt.org/open-social-psychology/chapter2.html), date checked 2026-10-29.
- Related: publication bias, relevant when asking why early expectancy results looked stronger than later ones.
