Effect sizes for categorical variable in brm cumulative logit model


#1

Hi,

I haven’t been able to find the answer to this anywhere, but it should be simple:

I’m running an ordinal regression model in brms using cumulative(“logit”).

I am using a categorical variable which I have set with ‘factor’.

I would like to know what the effect size would be for the reference value for this variable.

Regards,
Gareth


#2

Can you please be more specific on what you are trying to achieve for instance by showing use the code you are using?


#3

‘Ratings’ is collected on a 7-point Likert scale

Data$Height <- 1/2*scale(Data$Height)
Data$Chord <- factor(Data$Chord)
Data$Number <- factor(Data$Number)

mdlchord <- brm(Ratings ~ 1 + Height + I(Height^2) + Chord + (1 |Number),
prior = c(set_prior (“student_t(3,0,2.5)”, class = “b”)),
data = Data, family = cumulative(“logit”), save_all_pars = TRUE,
warmup = 1000, iter = 4000, init = 0, cores = 4)

For the population level effects I get

Population-Level Effects:
Estimate Est.Error l-95% CI u-95% CI Eff.Sample Rhat
Intercept[1] -2.49 0.18 -2.85 -2.12 166 1.03
Intercept[2] -1.06 0.18 -1.42 -0.70 164 1.03
Intercept[3] 0.01 0.18 -0.35 0.37 166 1.03
Intercept[4] 0.70 0.18 0.34 1.06 167 1.03
Intercept[5] 1.86 0.18 1.50 2.23 166 1.03
Intercept[6] 3.19 0.18 2.83 3.56 166 1.03
Height -0.06 1.96 -3.84 3.94 4000 1.00
IHeightE2 -0.07 1.94 -3.99 3.82 4000 1.00
Chord2 -1.18 0.30 -1.77 -0.58 4000 1.01
Chord3 -0.29 0.28 -0.83 0.27 437 1.01
Chord4 -0.46 0.28 -1.01 0.07 475 1.01
Chord5 -0.02 0.27 -0.55 0.50 413 1.01
Chord6 -0.24 0.29 -0.82 0.32 384 1.01
Chord7 -0.04 0.28 -0.57 0.50 551 1.01
Chord8 0.39 0.28 -0.13 0.94 4000 1.01
Chord9 0.55 0.30 -0.02 1.13 4000 1.01
Chord10 -0.52 0.29 -1.07 0.06 589 1.01

I’m looking for the effect sizes of the chords after taking account of Height.
What is my effect size for Chord1?

Thanks


#4

It seems to me that the effect sizes for the other chords are shown relative to Chord1? In which case I can assign it 0. I only am not sure of that because I would have expected a rather different result for where Chord1 fits amongst the others.


#5

The regression coefficients of the chords are indeed relative for Chord1, as by default R using dummy coding for factors.


#6

Ok, thanks.

I should have immediately accepted this most simple answer, but needed to confirm because of the unexpected result it gave.