Setting ROPE limits

[nahorp]

Hello!

I'm wondering if there is some function in bayestestR (or the wider easystats universe) that is able to create ROPE limits based on Kruschke's recommendation (±0.1; half of Cohen's recommendation of a small effect)? Reading the documentation, I see that there are defaults for certain model types and if the model doesn't fall into those types, then the default is an absolute -0.1 and 0.1. If no such function exists, any advice on how I could achieve this myself would be great!

For context, the model is a GLMM (student's t distribution; identity link) with 2 categorical predictors (A with 2 levels and B with 3 levels). Thank you for reading :)

[nahorp]

Would appreciate any input/advice with this. Thank you!

[strengejacke]

You mean rope()? Resp. rope_range()?

[nahorp]

You mean rope()? Resp. rope_range()?

Ah, rope_range().

[nahorp]

Anything @strengejacke or the team? :)

[mattansb]

Hi @nahorp the default behavior is to do just this, assuming the x predictors are unit-scaled. Is this not what you're getting? Please attach a reprex if not.

I generally recommend finding your own ROPE and not using defaults.

[nahorp]

Hey @mattansb!

the default behavior is to do just this, assuming the x predictors are unit-scaled

What are 'unit-scaled x predictors'? And how do I do that/ensure they are that in my GLMM?

[strengejacke]

I think the issue is that the ROPE is dependent on the SD of y, and despite your coefficient might be identical (see Sepal Width and SepalW10 in the table at the bottom), the ROPE and equivalence test results differ. Thus,whether +/- 0.1 SD reflects an appropriate region of practical equivalence, also depends on the unit of your predictors.

library(easystats)
data(iris)
m1 <- lm(Sepal.Length ~ Sepal.Width + Species, data = iris)
equivalence_test(m1)
#> # TOST-test for Practical Equivalence
#> 
#>   ROPE: [-0.08 0.08]
#> 
#> Parameter            |       90% CI |   SGPV | Equivalence |      p
#> -------------------------------------------------------------------
#> (Intercept)          | [1.64, 2.86] | < .001 |    Rejected | > .999
#> Sepal Width          | [0.63, 0.98] | < .001 |    Rejected | > .999
#> Species [versicolor] | [1.27, 1.64] | < .001 |    Rejected | > .999
#> Species [virginica]  | [1.78, 2.11] | < .001 |    Rejected | > .999

d <- iris
d$Sepal.Length10 <- 10 * d$Sepal.Length
d$SepalW10 <- 10 * d$Sepal.Width
m2 <- lm(Sepal.Length10 ~ SepalW10 + Species, data = d)
equivalence_test(m2)
#> # TOST-test for Practical Equivalence
#> 
#>   ROPE: [-0.83 0.83]
#> 
#> Parameter            |         90% CI |   SGPV | Equivalence |      p
#> ---------------------------------------------------------------------
#> (Intercept)          | [16.39, 28.63] | < .001 |    Rejected | > .999
#> SepalW10             | [ 0.63,  0.98] | 0.570  |    Rejected | 0.409 
#> Species [versicolor] | [12.73, 16.44] | < .001 |    Rejected | > .999
#> Species [virginica]  | [17.81, 21.12] | < .001 |    Rejected | > .999

compare_parameters(m1, m2)
#> Parameter            |                m1 |                   m2
#> ---------------------------------------------------------------
#> (Intercept)          | 2.25 (1.52, 2.98) | 22.51 (15.21, 29.82)
#> Species (versicolor) | 1.46 (1.24, 1.68) | 14.59 (12.37, 16.80)
#> Species (virginica)  | 1.95 (1.75, 2.14) | 19.47 (17.49, 21.44)
#> Sepal Width          | 0.80 (0.59, 1.01) |                     
#> SepalW10             |                   |  0.80 ( 0.59,  1.01)
#> ---------------------------------------------------------------
#> Observations         |               150 |                  150

^{Created on 2023-05-17 with reprex v2.0.2}