add ellipsis to FUNouter

[reikookamoto]

This is a follow-up to #1224

Part 1: Resolving the error message

This is the full traceback that I get when calling hypotheses(m, "b1 = b2"), as shown in the minimum reproducible example in the original issue.

9. do.call("FUN", args) at get_se_delta.R#102
8. func(x) at get_jacobian.R#29
7. (function (func, x, eps = NULL)
   {
   baseline <- func(x)
   inner_loop <- function(chunk, ...) { ...
   }
6. do.call(get_jacobian_fdforward, numderiv) at get_jacobian.R#11
5. get_jacobian(func = function (x)
   {
   names(x) <- names(coefs)
   model_tmp <- set_coef(model, x, ...) ...
   }
4. do.call("get_jacobian", args) at get_se_delta.R#112
3. get_se_delta(model = structure(list(obj = list(par = c(beta = 1, beta = 1, beta = 1, beta = 1, betadisp = 0, theta = 0), fn = function (x = last.par[lfixed()],
   ...
   }
2. do.call("get_se_delta", args) at hypotheses.R#344
1. hypotheses(m, "b1 = b2")

• My understanding:
  • At 8, the function func() refers to inner() defined in get_delta_se.R
  • func() adds an element called newparams to the argument list args
  • At 9, "FUN" refers to FUNouter() defined in hypotheses.R
  • FUNouter() only expects the arguments model and hypothesis, but the additional newparams argument is passed, leading to the error
• Possible resolution:
  • To fix this, I updated the definition of FUNouter() by adding an ... to allow for additional arguments
  • After making this change, I ran devtools::check() and it passed with no errors, warnings, or notes

Part 2: Standard error returned is NA

When I call hypotheses(m, "b1 = b2") after updating FUNouter(), I get the following output:

Estimate Std. Error  z Pr(>|z|)  S  2.5 %  97.5 %
    0.634         NA NA       NA NA    NA     NA

The returned data frame gives a sensible estimate, but the remaining columns are all NA.

• My understanding:
  • "J" matrix returned on line 112 of get_se_delta.R is a matrix where all entries are 0
  • The se calculated on line 136 is also 0, which gets overwritten by NA_real_ in the following line
  • Consequently, the computation of the z-statistic, p-value, and confidence interval in get_ci() all return NA
• Hypothesis (no pun intended):
  • Could this be related to precision and floating-point arithmetic issues (i.e., subtractions involving very small differences) happening in the inner_loop() function defined in get_jacobian_fdforward() within get_jacobian.R?

This is pushing my limited knowledge of matrices and the delta method, so I'm not sure if the NA values are expected behaviour or if they indicate a deeper issue.

Thanks for your help!