pool() on lavaan fit objects gives error: coef() not available on S4 object

[Gootjes]

Before submitting the issue

• Check whether a bug report is the right format for your issue. For questions about using the mice package, see Discussions. For theoretical background about the mice methods, see Flexible Imputation of Missing Data.
• Please make sure that you are using the latest version of mice (e.g. with the packageVersion() or sessionInfo() function).
• Please verify whether the bug has already been addressed (e.g. on GitHub or Stack Overflow.

Describe the bug
The error happens in get.dfcom on this line

mice/R/get.df.R

Line 24 in c924a74

return(as.numeric(max(nobs - length(coef(model)), 1)))

The package lavaan implements coef as an S4 function (compare lavaan::coef to stats::coef), however, this function is not available at this point, so the regular coef function is called on the lavaan fit object, leading to an error.

To Reproduce

library(lavaan)
#> This is lavaan 0.6-15
#> lavaan is FREE software! Please report any bugs.
library(mice)
#> 
#> Attaching package: 'mice'
#> The following object is masked from 'package:stats':
#> 
#>     filter
#> The following objects are masked from 'package:base':
#> 
#>     cbind, rbind
pool(list(sem("mpg ~ cyl", data = mtcars[1:10,]), sem("mpg ~ cyl", data = mtcars[1:20,])))
#> Error in object$coefficients: $ operator not defined for this S4 class

^{Created on 2024-01-14 with reprex v2.0.2}

Expected behavior
That coef() works.

A solution would be to make get.dfcom a S3method, such that get.dfcom.default is the current function, and custom functions are implemented for lavaan and coxph (and in the future others?)

[hanneoberman]

I believe this is not a bug in mice but a problem with the lavaan fit object not being processed as expected by broom::tidy().

library(lavaan)
#> This is lavaan 0.6-15
#> lavaan is FREE software! Please report any bugs.
library(mice)
#> 
#> Attaching package: 'mice'
#> The following object is masked from 'package:stats':
#> 
#>     filter
#> The following objects are masked from 'package:base':
#> 
#>     cbind, rbind
sem("hgt ~ age", data = boys)
#> lavaan 0.6.15 ended normally after 1 iteration
#> 
#>   Estimator                                         ML
#>   Optimization method                           NLMINB
#>   Number of model parameters                         2
#> 
#>                                                   Used       Total
#>   Number of observations                           728         748
#> 
#> Model Test User Model:
#>                                                       
#>   Test statistic                                 0.000
#>   Degrees of freedom                                 0
imp <- mice(boys, print = FALSE)
fit <- with(imp, sem("hgt ~ age"))
pool(fit)
#> Warning in get.dfcom(object, dfcom): Infinite sample size assumed.
#> Warning: The `exponentiate` argument is not supported in the `tidy()` method
#> for `lavaan` objects and will be ignored.
#> Error in `purrr::keep()`:
#> ℹ In index: 1.
#> ℹ With name: estimate.
#> Caused by error in `lavaan::parameterEstimates()`:
#> ! unused arguments (effects = "fixed", parametric = TRUE, exponentiate = FALSE)
#> Backtrace:
#>      ▆
#>   1. ├─mice::pool(fit)
#>   2. │ └─mice:::pool.fitlist(...)
#>   3. │   ├─base::summary(fitlist, type = "tidy", exponentiate = FALSE)
#>   4. │   └─mice:::summary.mira(fitlist, type = "tidy", exponentiate = FALSE)
#>   5. │     ├─... %>% bind_rows()
#>   6. │     └─base::lapply(...)
#>   7. │       ├─generics (local) FUN(X[[i]], ...)
#>   8. │       └─broom:::tidy.lavaan(X[[i]], ...)
#>   9. │         └─... %>% as_tibble()
#>  10. ├─dplyr::bind_rows(.)
#>  11. │ └─rlang::list2(...)
#>  12. ├─tibble::as_tibble(.)
#>  13. ├─dplyr::select(...)
#>  14. ├─broom:::rename2(...)
#>  15. │ └─purrr::keep(dots, ~quo_name(.x) %in% colnames(.data))
#>  16. │   └─purrr:::where_if(.x, .p, ...)
#>  17. │     └─purrr:::map_(.x, .p, ..., .type = "logical", .purrr_error_call = .purrr_error_call)
#>  18. │       ├─purrr:::with_indexed_errors(...)
#>  19. │       │ └─base::withCallingHandlers(...)
#>  20. │       ├─purrr:::call_with_cleanup(...)
#>  21. │       └─purrr (local) .f(.x[[i]], ...)
#>  22. │         └─broom (local) .fn(...)
#>  23. │           ├─quo_name(.x) %in% colnames(.data)
#>  24. │           └─base::colnames(.data)
#>  25. │             └─base::is.data.frame(x)
#>  26. ├─dplyr::mutate(., term = paste(lhs, op, rhs))
#>  27. ├─tibble::rownames_to_column(.)
#>  28. ├─tibble::as_tibble(.)
#>  29. └─base::.handleSimpleError(...)
#>  30.   └─purrr (local) h(simpleError(msg, call))
#>  31.     └─cli::cli_abort(...)
#>  32.       └─rlang::abort(...)
rlang::last_trace()
#> Error: Can't show last error because no error was recorded yet

^{Created on 2024-01-15 with reprex v2.0.2}

[Gootjes]

To clarify, my reprex does not reproduce the error on your machine?

That bug you found in your reprex is reported here: #614

I believe you are not running into the bug I am reporting because the sem models aren't fitted because the data argument isn't provided to lavaan::sem by mice:::with.mids. The calls on the non-fitted models never reach the offending line of code.

print(fit)
# (...)
#> lavaan 0.6.17 did not run (perhaps do.fit = FALSE)?

New reprex that reproduces the bug:

library(lavaan)
#> This is lavaan 0.6-17
#> lavaan is FREE software! Please report any bugs.
library(mice)
#> 
#> Attaching package: 'mice'
#> The following object is masked from 'package:stats':
#> 
#>     filter
#> The following objects are masked from 'package:base':
#> 
#>     cbind, rbind
imp <- mice(boys, print = FALSE)
fit <- as.mira(lapply(1:5, function(i) sem("hgt ~ age", data = complete(imp, i))))
pool(fit)
#> Error in object$coefficients: $ operator not defined for this S4 class

^{Created on 2024-01-15 with reprex v2.0.2}