Comments on: Repeated Measures ANOVA in R

By: Norm Vinson

Norm Vinson — Tue, 25 Jun 2024 13:24:12 +0000

the font is very light coloured. I’m old. it’s tough on my eyes.

By: Jana

Jana — Wed, 24 Jan 2024 13:46:15 +0000

Hi, I am trying to use anova_test with a pre defined planned contrast. But I cant find any way to make it work. Is it possible, and if so, could you please explain how? Thank you ver much in advance!

By: Yamato Miyashita

Yamato Miyashita — Fri, 08 Sep 2023 06:36:26 +0000

I believe the number of pariticpants in the dataset for the the three-way ANOVA example is “12”, not 10.

By: Gauthier Berthomieu

Gauthier Berthomieu — Tue, 20 Jun 2023 13:34:47 +0000

Hi,

Thanks for this awesome work. I have one question regarding the assumptions for a RM Anova:
I often find the info that the normality assumption for linear models is based on the normality of the residuals of the model rather than the normality of the data itself.

That being said, in this tutorial, we run the Shapiro-Wilk test on the data, leading to 12 tests results.

Aren’t we supposed to gather all the model residuals and run the normality test on these pooled residuals ?

Thanks again!

By: Pedro

Pedro — Mon, 12 Sep 2022 08:25:52 +0000

Thanks for this detailed script. In my experiment, I have in randomized complete block design. How can I implement this additional factor in the anova_test script?

By: pedro

pedro — Tue, 06 Sep 2022 16:04:03 +0000

Thanks Kassambara for this nice tutorial! I have one minor issue. In my experiment, I have two factor experiment arranged in block design. How can I include this block as a factor in the anova_test that you had presented? Or it is included in the model somehow?

By: Alice

Alice — Sat, 18 Jun 2022 09:01:50 +0000

Thank you for the tutorial. I am doing a two-way repeated measures anova. I found a main effect of one of my variables. I ran post hoc tests using this script.

# comparisons for treatment variable
selfesteem2 %>%
pairwise_t_test(
score ~ treatment, paired = TRUE,
p.adjust.method = “bonferroni”
)
# comparisons for time variable
selfesteem2 %>%
pairwise_t_test(
score ~ time, paired = TRUE,
p.adjust.method = “bonferroni”
)

However, the n number is scaling – it is double or triple what it should be, meaning that the degrees of freedom are also too high. I think this is because it is collapsing across each factor and assuming that the data is unpaired, when actually the data is paired. Do you know how I can rectify this?

Thanks in advance for your help!

By: Henry

Henry — Wed, 27 Apr 2022 11:52:51 +0000

In reply to Iza. You just need to add a line after you have originally run the ANOVA with the name of your model (here res.aov). This will give you the Mauchly's test result.

By: Simon Reynaert

Simon Reynaert — Thu, 24 Mar 2022 16:44:43 +0000

In reply to Bartek.

Hiya Bartek, you can do outlier capping in this case instead of deleting them (atleast, if you are convinced it was not just a measurement error). More info about that here: http://r-statistics.co/Outlier-Treatment-With-R.html

By: su

su — Tue, 22 Feb 2022 01:44:01 +0000

Thank you for the tutorial and I’ve learn a lot. However, I have a question that I hope you can help with.
In the post-hoc test for 2-way rmAnova with non-significant interaction, the pairwise t-test was applied for treatment and time variables separately. Both variables are within-subjects, so the “paired =TRUE” is assigned.
For example for the t-test for the treatment variable, t-test compares scores of each combination of id*time and for the same combination the score of different treatment are paired. It means that in t-test there are multiple pairs which are actually from the same subject but only with different time variable. However, in some statistical software packages such as SPSS they use marginal means for pairwise t-test used as pos-hoc test for two(or more)-way rmANOVA. It means that the scores for each subject(id) and each treatment averaged across time are used in t-test. I’ve found that the results could be quite different especially if the other variable(in this example, it’s time) have many levels, since the degree of freedom for t-test will be quite difference. I wonder which method is correct, considering that in two-way rmAnova I might be more interested in one variable and the other variable might even be non-significant in the result of Anova.