This works <div class="highlight highlight-source-r notranslate position-relativ

This is also why these two give the same result: <p dir

<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/us

<a class="user-mention notranslate" data-hovercard-type="user" data-hover

Seems it works: <div class="highlight highlight-source-r notranslate position-rela

<div class="highlight highlight-source-r notranslate position-relative overflow-auto" dir="auto" dat

Problem with as.factor in formula,about easystats/modelbased

mattansb commented on July 20, 2024

The values of the "at" argument should type/value match the original data, and it internally dows any transformations etc.
In this case, gear is numeric (kinda), so:

model <- lm(mpg ~as.factor(gear), data = mtcars)

emmeans::ref_grid(model, at = list(gear = c(3,4,5)))
#> 'emmGrid' object with variables:
#>     gear = 3, 4, 5

^{Created on 2020-05-19 by the reprex package (v0.3.0)}

This is also why these two give the same result:

m1 <- lm(mpg ~ wt, data = mtcars)
m2 <- lm(mpg ~ scale(wt), data = mtcars)


emmeans::emmeans(m1, ~ wt, at = list(wt = 3))
#>  wt emmean    SE df lower.CL upper.CL
#>   3   21.3 0.552 30     20.1     22.4
#> 
#> Confidence level used: 0.95
emmeans::emmeans(m2, ~ wt, at = list(wt = 3))
#>  wt emmean    SE df lower.CL upper.CL
#>   3   21.3 0.552 30     20.1     22.4
#> 
#> Confidence level used: 0.95

^{Created on 2020-05-19 by the reprex package (v0.3.0)}

from modelbased.

strengejacke commented on July 20, 2024

This is also why these two give the same result:

I think this is because the association is actually the same, it does not change when you scale your input (the coefficients do, not the "slope"). With predict() you get also identical results:

m1 <- lm(mpg ~ wt, data = mtcars)
m2 <- lm(mpg ~ scale(wt), data = mtcars)
library(ggeffects)

ggpredict(m1, "wt [3]")
#> 
#> # Predicted values of mpg
#> # x = wt
#> 
#> x | Predicted |   SE |         95% CI
#> -------------------------------------
#> 3 |     21.25 | 0.55 | [20.17, 22.33]

ggpredict(m2, "wt [3]")
#> 
#> # Predicted values of mpg
#> # x = wt
#> 
#> x | Predicted |   SE |         95% CI
#> -------------------------------------
#> 3 |     21.25 | 0.55 | [20.17, 22.33]

^{Created on 2020-05-19 by the reprex package (v0.3.0)}

from modelbased.

DominiqueMakowski commented on July 20, 2024

but this is annoying because it makes changing type in the formula fail with emmeans with the new implementation of its wrapper:

modelbased/R/utils_emmeans_wrapper.R

Lines 68 to 83 in a852fe8

    
           .emmeans_wrapper <- function(model, levels = NULL, fixed = NULL, modulate = NULL, transform = "response", length = 10, ...) { 
        
             levels <- c(levels, modulate) 
        
             fixed <- c(fixed, modulate) 
        
             # Get emmeans refgrid 
        
             at <- insight::get_data(model)[levels] 
        
             at <- sapply(at, visualisation_matrix, length = length, simplify = FALSE) 
        
             suppressMessages(refgrid <- emmeans::ref_grid(model, at = at)) 
        
             # Run emmeans 
        
             means <- emmeans::emmeans(refgrid, levels, by = fixed, transform = transform, ...) 
        
             means 
        
           }

And the only way to easily accommodate would be in insight? So that it returns non-modified data?

Worst case scenario we can just do nothing, since modifying types in formulas is not recommended anyway, but...

from modelbased.

mattansb commented on July 20, 2024

@strengejacke I have no idea how predict works - I am only an emmeans expert! 😅🤪

from modelbased.

strengejacke commented on July 20, 2024

@strengejacke I have no idea how predict works - I am only an emmeans expert! 😅🤪

;-)

They work pretty similar - though not exactly the same.

from modelbased.

strengejacke commented on July 20, 2024

So that it returns non-modified data?

I'm not sure how to detect if data is non-modified? But I think that emmeans either requires numerics, if really "numeric" (including numeric factor levels), and else the character version (see below).

levels <- "gear"
data(mtcars)
mtcars$gear <- factor(sjmisc::rec(mtcars$gear, rec = "3=a;4=b;5=c"))
model <- lm(mpg ~gear, data = mtcars)

at <- insight::get_data(model)[levels]
at <- sapply(at, modelbased::visualisation_matrix, simplify = FALSE)

emmeans::ref_grid(model, at = at)
#> 'emmGrid' object with variables:
#>     gear = a, b, c

^{Created on 2020-05-19 by the reprex package (v0.3.0)}

Thus, check if factor at has numeric levels:
is.factor(x) && !anyNA(suppressWarnings(as.numeric(levels(x))))

and if so, convert to numeric before passing to sapply(vismatrix).

from modelbased.

strengejacke commented on July 20, 2024

Does that work? (I hope so...)

from modelbased.

strengejacke commented on July 20, 2024

Seems it works:

data(mtcars)
levels <- "gear"
model <- lm(mpg ~as.factor(gear), data = mtcars)

at <- insight::get_data(model)[levels]
at <- sapply(at, modelbased::visualisation_matrix, simplify = FALSE)

.numeric_levels <- function(x) {
  is.factor(x) && !anyNA(suppressWarnings(as.numeric(levels(x))))
}

at <- lapply(at, function(i) {
  if (.numeric_levels(i)) {
    as.numeric(levels(i))
  } else {
    i
  }
})

emmeans::ref_grid(model, at = at)
#> 'emmGrid' object with variables:
#>     gear = 3, 4, 5

^{Created on 2020-05-19 by the reprex package (v0.3.0)}

from modelbased.

DominiqueMakowski commented on July 20, 2024

data(mtcars)
levels <- "gear"
data <- mtcars
data$gear <- as.factor(data$gear)
model <- lm(mpg ~gear, data = data)

at <- insight::get_data(model)[levels]
at <- sapply(at, modelbased::visualisation_matrix, simplify = FALSE)

.numeric_levels <- function(x) {
  is.factor(x) && !anyNA(suppressWarnings(as.numeric(levels(x))))
}

at <- lapply(at, function(i) {
  if (.numeric_levels(i)) {
    as.numeric(levels(i))
  } else {
    i
  }
})

emmeans::ref_grid(model, at = at)
#> Error in `contrasts<-`(`*tmp*`, value = contrasts.arg[[nn]]): contrasts apply only to factors

^{Created on 2020-05-19 by the reprex package (v0.3.0)}

Doesn't work then it the regular case, need to find a way to conditionally run this fix

from modelbased.

DominiqueMakowski commented on July 20, 2024

Because I think the issue is that the pulled variable from the model's data (through insight::get_data()) does not match the type of the variable in the model, as the latter has been changed "in the formula", which doesn't affect the data in the model.

from modelbased.

DominiqueMakowski commented on July 20, 2024

You might have a more clever idea than my patch 😁

from modelbased.

Problem with as.factor in formula about modelbased HOT 11 CLOSED

Comments (11)

Related Issues (20)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	.emmeans_wrapper <- function(model, levels = NULL, fixed = NULL, modulate = NULL, transform = "response", length = 10, ...) {

	levels <- c(levels, modulate)
	fixed <- c(fixed, modulate)

	# Get emmeans refgrid
	at <- insight::get_data(model)[levels]
	at <- sapply(at, visualisation_matrix, length = length, simplify = FALSE)

	suppressMessages(refgrid <- emmeans::ref_grid(model, at = at))

	# Run emmeans
	means <- emmeans::emmeans(refgrid, levels, by = fixed, transform = transform, ...)

	means
	}