find_formula fails when applied to lme inside lapply

[iago-pssjd]

I come here from vincentarelbundock/modelsummary#559.
I adapt the reprex

library(insight)
library(nlme)
models <-lapply(c("", " + Sex"), \(.x) lme(as.formula(paste0("distance  ~ age", .x)), random = ~ 1, data = Orthodont))
lapply(models, find_formula)
Error in paste0("distance  ~ age", .x) : object '.x' not found

I would say that this seems to be a duplicate of issue #309, but next line works

find_formula(lme(as.formula(paste0("distance  ~ age", "")), random = ~ 1, data = Orthodont))
$conditional
distance ~ age
<environment: 0x55c02e144f78>

$random
~1
<environment: 0x55c02e144f78>

attr(,"class")
[1] "insight_formula" "list"         

Thanks!

[iago-pssjd]

Maybe solvable changing eval(x$call$fixed) in

insight/R/find_formula.R

Lines 1198 to 1200 in 6fdc601

	find_formula.lme <- function(x, verbose = TRUE, ...) {
	fm <- eval(x$call$fixed)
	fmr <- eval(x$call$random)

with formula(x$terms)?

[strengejacke]

ok, but can we deal with random effects then?

library(insight)
library(nlme)
models <-lapply(c("", " + Sex"), \(.x) lme(as.formula(paste0("distance  ~ age", .x)), random = ~ 1, data = Orthodont))
lapply(models, function(i) formula(i$terms))
#> [[1]]
#> distance ~ age
#> <environment: 0x0000021eb3dc7010>
#> 
#> [[2]]
#> distance ~ age + Sex
#> <environment: 0x0000021eb68257f0>

find_formula(models[[]])
#> $conditional
#> list()
#> <environment: 0x0000021eb70a1520>
#> 
#> attr(,"class")
#> [1] "insight_formula" "list"


m2 <- lme(distance ~ age, random = ~ 1, data = Orthodont)
find_formula(m2)
#> $conditional
#> distance ~ age
#> <environment: 0x0000021eb745e998>
#> 
#> $random
#> ~1
#> <environment: 0x0000021eb745e998>
#> 
#> attr(,"class")
#> [1] "insight_formula" "list"

^{Created on 2022-10-06 with reprex v2.0.2}

[iago-pssjd]

@strengejacke I don't see your point. I only modify the fixed effects part of find_formula.lme

[vincentarelbundock]

@strengejacke I don't see your point. I only modify the fixed effects part of find_formula.lme

One question is: if the random part is something like random=~x1+x2, will your change properly classify those as "random", or will it stuff the "x1" and "x2" variables in the "fixed" part by mistake?

[strengejacke]

Ah ok, didn't test with the PR.

[iago-pssjd]

@vincentarelbundock
Let's see with the example:

library(insight)
library(nlme)
models <-lapply(c("", " + Sex"), \(.x) lme(as.formula(paste0("distance  ~ age", .x)), random = ~ Sex, data = Orthodont))
find_formula(models[[1]])
$conditional
distance ~ age
<environment: 0x55a163099e58>

$random
~Sex
<environment: 0x55a163aea038>

attr(,"class")
[1] "insight_formula" "list"      

In this case, it seems to work. I may try some more complex example.

[vincentarelbundock]

Great!

But I bet the code breaks if the lapply modifies the random component instead of the fixed.

Is that a big deal?

[iago-pssjd]

@vincentarelbundock

if the lapply modifies the random component

I am not sure what do you mean, but maybe you refer if I use the as.formula x paste combination in the random component. Indeed, this part is not solved with this PR, it behaves like until now, the only information about random formula contained in lme object is that in call, which is the used by find_formula.lme, so the issue would persist for the random part.

[iago-pssjd]

I close the issue, since th PR to solve the fixed effects part was merged and I do not see there is any chance to solve the random effects part as described in previous comments above.

[strengejacke]

thanks, we can always open a new issue if needed.