Effect Sizes for ANOVAs"

options(knitr.kable.NA = "")
options(digits = 2)
knitr::opts_chunk$set(comment = ">", warning = FALSE)


.eval_if_requireNamespace <- function(...) {
  pkgs <- c(...)
  knitr::opts_chunk$get("eval") && all(sapply(pkgs, requireNamespace, quietly = TRUE))

knitr::opts_chunk$set(eval = .eval_if_requireNamespace("effectsize", "afex"))


In the context of ANOVA-like tests, it is common to report ANOVA-like effect sizes. These effect sizes represent the amount of variance explained by each of the model's terms, where each term can be represented by 1 or more parameters.

For example, in the following case, the parameters for the treatment term represent specific contrasts between the factor's levels (treatment groups) - the difference between each level and the reference level (obk.long == 'control').

data(obk.long, package = "afex")
# modify the data slightly for the demonstration:
obk.long <- obk.long[1:240 %% 3 == 0, ]
obk.long$id <- seq_len(nrow(obk.long))

m <- lm(value ~ treatment, data = obk.long)


But we can also ask about the overall effect of treatment - how much of the variation in our dependent variable value can be predicted by (or explained by) the variation between the treatment groups. Such a question can be answered with an ANOVA test:


As we can see, the variance in value (the sums-of-squares, or SS) has been split into pieces:

We can now ask what is the percent of the total variance in value that is associated with treatment. This measure is called Eta-squared (written as $\eta^2$):

$$ \eta^2 = \frac{SS_{effect}}{SS_{total}} = \frac{72.23}{72.23 + 250.96} = 0.22 $$

and can be accessed via the eta_squared() function:

options(es.use_symbols = TRUE) # get nice symbols when printing! (On Windows, requires R >= 4.2.0)

eta_squared(m, partial = FALSE)

Adding More Terms

When we add more terms to our model, we can ask two different questions about the percent of variance explained by a predictor - how much variance is accounted by the predictor in total, and how much is accounted when controlling for any other predictors. The latter questions is answered by the partial-Eta squared ($\eta^2_p$), which is the percent of the partial variance (after accounting for other predictors in the model) associated with a term:

$$ \eta^2_p = \frac{SS_{effect}}{SS_{effect} + SS_{error}} $$ which can also be accessed via the eta_squared() function:

m <- lm(value ~ gender + phase + treatment, data = obk.long)

eta_squared(m, partial = FALSE)

eta_squared(m) # partial = TRUE by default

(phase is a repeated-measures variable, but for simplicity it is not modeled as such.)

In the calculation above, the SSs were computed sequentially - that is the SS for phase is computed after controlling for gender, and the SS for treatment is computed after controlling for both gender and phase. This method of sequential SS is called also type-I test. If this is what you want, that's great - however in many fields (and other statistical programs) it is common to use "simultaneous" sums of squares (type-II or type-III tests), where each SS is computed controlling for all other predictors, regardless of order. This can be done with car::Anova(type = ...):

eta_squared(car::Anova(m, type = 2), partial = FALSE)

eta_squared(car::Anova(m, type = 3)) # partial = TRUE by default

$\eta^2_p$ will always be larger than $\eta^2$. The idea is to simulate the effect size in a design where only the term of interest was manipulated. This terminology assumes some causal relationship between the predictor and the outcome, which reflects the experimental world from which these analyses and measures hail; However, $\eta^2_p$ can also simply be seen as a signal-to-noise- ratio, as it only uses the term's SS and the error-term's SS.[^in repeated-measure designs the term-specific residual-SS is used for the computation of the effect size].

(Note that in a one-way fixed-effect designs $\eta^2 = \eta^2_p$.)

Adding Interactions

Type II and type III treat interaction differently. Without going into the weeds here, keep in mind that when using type III SS, it is important to center all of the predictors; for numeric variables this can be done by mean-centering the predictors; for factors this can be done by using orthogonal coding (such as contr.sum for effects-coding) for the dummy variables (and NOT treatment coding, which is the default in R). This unfortunately makes parameter interpretation harder, but only when this is does do the SSs associated with each lower-order term (or lower-order interaction) represent the SS of the main effect (with treatment coding they represent the SS of the simple effects).

# compare
m_interaction1 <- lm(value ~ treatment * gender, data = obk.long)

# to:
m_interaction2 <- lm(
  value ~ treatment * gender,
  data = obk.long,
  contrasts = list(
    treatment = "contr.sum",
    gender = "contr.sum"

eta_squared(car::Anova(m_interaction1, type = 3))
eta_squared(car::Anova(m_interaction2, type = 3))

If all of this type-III-effects-coding seems like a hassle, you can use the afex package, which takes care of all of this behind the scenes:

m_afex <- aov_car(value ~ treatment * gender + Error(id), data = obk.long)


Other Measures of Effect Size

Unbiased Effect Sizes

These effect sizes are unbiased estimators of the population's $\eta^2$:



Both $\omega^2$ and $\epsilon^2$ (and their partial counterparts, $\omega^2_p$ & $\epsilon^2_p$) are unbiased estimators of the population's $\eta^2$ (or $\eta^2_p$, respectively), which is especially important is small samples. Though $\omega^2$ is the more popular choice [@albers2018power], $\epsilon^2$ is analogous to adjusted-$R^2$ [@allen2017statistics, p. 382], and has been found to be less biased [@carroll1975sampling].

Generalized Eta2

Partial Eta squared aims at estimating the effect size in a design where only the term of interest was manipulated, assuming all other terms are have also manipulated. However, not all predictors are always manipulated - some can only be observed. For such cases, we can use generalized Eta squared ($\eta^2_G$), which like $\eta^2_p$ estimating the effect size in a design where only the term of interest was manipulated, accounting for the fact that some terms cannot be manipulated (and so their variance would be present in such a design).

eta_squared(m_afex, generalized = "gender")

$\eta^2_G$ is useful in repeated-measures designs, as it can estimate what a within-subject effect size would have been had that predictor been manipulated between-subjects [@olejnik2003generalized].

Cohen's f

Finally, we have the forgotten child - Cohen's $f$. Cohen's $f$ is a transformation of $\eta^2_p$, and is the ratio between the term-SS and the error-SS.

$$\text{Cohen's} f_p = \sqrt{\frac{\eta^2_p}{1-\eta^2_p}} = \sqrt{\frac{SS_{effect}}{SS_{error}}}$$

It can take on values between zero, when the population means are all equal, and an indefinitely large number as the means are further and further apart. It is analogous to Cohen's $d$ when there are only two groups.


When Sum-of-Squares are Hard to Come By

Until now we've discusses effect sizes in fixed-effect linear model and repeated-measures ANOVA's - cases where the SSs are readily available, and so the various effect sized presented can easily be estimated. How ever this is not always the case.

For example, in linear mixed models (LMM/HLM/MLM), the estimation of all required SSs is not straightforward. However, we can still approximate these effect sizes (only their partial versions) based on the test-statistic approximation method (learn more in the Effect Size from Test Statistics vignette).


fit_lmm <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)

anova(fit_lmm) # note the type-3 errors

F_to_eta2(45.8, df = 1, df_error = 17)

Or directly with `eta_squared() and co.:


Another case where SSs are not available is when using Bayesian models...

For Bayesian Models

An alternative route to obtaining effect sizes of explained variance, is via the use of the posterior predictive distribution (PPD). The PPD is the Bayesian expected distribution of possible unobserved values. Thus, after observing some data, we can estimate not just the expected mean values (the conditional marginal means), but also the full distribution of data around these values [@gelman2014bayesian, chapter 7].

By sampling from the PPD, we can decompose the sample to the various SSs needed for the computation of explained variance measures. By repeatedly sampling from the PPD, we can generate a posterior distribution of explained variance estimates. But note that these estimates are conditioned not only on the location-parameters of the model, but also on the scale-parameters of the model! So it is vital to validate the PPD before using it to estimate explained variance measures.

Let's fit our model:


m_bayes <- stan_glm(value ~ gender + phase + treatment,
  data = obk.long, family = gaussian(),
  refresh = 0

We can use eta_squared_posterior() to get the posterior distribution of $eta^2$ or $eta^2_p$ for each effect. Like an ANOVA table, we must make sure to use the right effects-coding and SS-type:

pes_posterior <- eta_squared_posterior(m_bayes,
  draws = 500, # how many samples from the PPD?
  partial = TRUE, # partial eta squared
  # type 3 SS
  ss_function = car::Anova, type = 3,
  verbose = FALSE


  rope_range = c(0, 0.1), test = "rope"

Compare to:

m_ML <- lm(value ~ gender + phase + treatment, data = obk.long)

eta_squared(car::Anova(m_ML, type = 3))

For Ordinal Outcomes

When our outcome is not a numeric variable, the effect sizes described above cannot be used - measured based on sum-of-squares are ill suited for such outcomes. Instead, we must use effect sizes for ordinal ANOVAs.

In R, there are two functions for running ordinal one way ANOVAs: kruskal.test() for differences between independent groups, and friedman.test() for differences between dependent groups.

For the one-way ordinal ANOVA, the Rank-Epsilon-Squared ($E^2_R$) and Rank-Eta-Squared ($\eta^2_H$) are measures of association similar to their non-rank counterparts: values range between 0 (no relative superiority between any of the groups) to 1 (complete separation - with no overlap in ranks between the groups).

group_data <- list(
  g1 = c(2.9, 3.0, 2.5, 2.6, 3.2), # normal subjects
  g2 = c(3.8, 2.7, 4.0, 2.4), # with obstructive airway disease
  g3 = c(2.8, 3.4, 3.7, 2.2, 2.0) # with asbestosis




For an ordinal repeated measures one-way ANOVA, Kendall's W is a measure of agreement on the effect of condition between various "blocks" (the subjects), or more often conceptualized as a measure of reliability of the rating / scores of observations (or "groups") between "raters" ("blocks").

# Subjects are COLUMNS
(ReactionTimes <- matrix(
    398, 338, 520,
    325, 388, 555,
    393, 363, 561,
    367, 433, 470,
    286, 492, 536,
    362, 475, 496,
    253, 334, 610
  nrow = 7, byrow = TRUE,
  dimnames = list(
    paste0("Subject", 1:7),
    c("Congruent", "Neutral", "Incongruent")




Try the effectsize package in your browser

Any scripts or data that you put into this service are public.

effectsize documentation built on Sept. 14, 2023, 5:07 p.m.