2
18
Iovaldi ML. Tamaño del efecto. Rev Argent Cir 2023;115(3):217-219
Acá se ve un tamaño del efecto moderado y
muestra (50 y 40%) – h de Cohen
un valor p no signiꢂcaꢁvo. El resultado puede darse
a la inversa: un tamaño del efecto bajo y un valor p
signiꢂcaꢁvo.
[1] 0.2013579 # el tamaño del efecto es bajo
De manera que, el hallazgo de un valor p
Para proporciones hay otras esꢁmaciones estadísticamente significativo no es para festejar
como la h de Cohen, con la misma interpretación, V de como si hubiéramos hallado un “Aleph” en el só-
5
,9
10
Cramer, etc. Otro ejemplo en R, library (pwr) con una tano , sino para interpretarlo con cautela y pro-
línea de script
fundizar el análisis de nuestros datos con otras
herramientas estadísticas. El valor p aislado está
sobrevaluado y conviene incluir el cálculo del ta-
maño del efecto en los estudios de investigación
clínica.
library (pwr)
ES.h (0.5,0.4) ### proporciones 0.5 y 0.4 en cada
■
ENGLISH VERSION
The following is an answer from Microsoꢀ Bing effect size, a d of 0.5 is considered a medium effect size,
2
GPT chat to the quesꢁon: what is effect size?
and a d of 0.8 or larger is considered a large effect size .
If the result is negaꢁve, although it is mathemaꢁcally
correct, it is not used to facilitate its interpretaꢁon. It
is also known as standardized mean difference because
the units of the result is the number of standard
deviaꢁons in which the means differ. Standardized mean
difference is a common measure in meta-analyses.
“
Effect size is a measure of the strength of a
phenomenon in staꢀsꢀcs. It is a descripꢀve staꢀsꢀc
that conveys the esꢀmated magnitude of a
relaꢀonship without making any statement about
whether the apparent relaꢀonship in the data
1
reflects a true relaꢀonship in the populaꢀon . Effect
size refers to a way of quanꢀfying the magnitude of
A simple example in R5,6,7,8
With only relevant data
2
difference between two groups .
There are three ways to measure effect size,
depending on the type of analysis one is performing:
1
. Standardized mean difference 2. Correlaꢀon
mean(data1) # mean group 1
2
coefficient 3. Odds raꢀo .”
[
1] 41.21997
mean(data2) # mean group 2
1] 61.2
Citaꢁons 3 and 4 are suggested but not
referenced in the answer.
[
sd_pooled(data1, data2)
deviaꢁon
# pooled standard
Arꢁꢂcial intelligence is fashionable in the
medical literature and will become a topic of discussion
in scienꢁꢂc seꢄngs. However, my idea is to leave
arꢁꢂcial intelligence for another occasion and take this
opportunity to introduce the concept and importance
of the effect size in clinical research studies in contrast
to reporꢁng the p-value, which is a measure of the
probability that the results are due to chance and only if
the study is properly designed in terms of applicability.
From the chat response, I highlight the concept
of the magnitude of the difference between two groups
because a low p-value may coincide with an effect size
without clinical relevance.
[
1] 27.63153
t.test(data2, data1) # Student’s t-test of the 2
samples without staꢀsꢀcal significance
t = 1.6169, df = 17.724, p-value = 0.1236 # Welch
Two Sample t-test
Calculaꢀon of Cohen’s d according to the formula
mean(data2)
-
mean(data1))/ sd_pooled
(
[
data1,data2)
1] 0.7230879
There are different measures for effect size to
compare 2 means, The most common effect sizes are
Cohen’s d, Hedges’ g and Glass’ delta .
3
I leave relaꢁve risk, odds raꢁos, correlaꢁon Calculaꢀon of Cohen’s on R, library (effectsize)
coefficients, etc., for another occasion because they
are beyond the scope of this editorial and I prefer to
introduce concepts slowly.
cohens_d(data2, data1)
Cohen’s d | 95% CI
---------------------------------
d formula is mean 1 – mean 2/pooled standard
4
deviaꢁon (SDpooled) . A d of 0.2 is considered a small
0.72 | [0.37, 1.92]