BUG - ED() Fails for Exponential Decay Models with Two Fixed Parameters

[hreinwald]

The Problem

Calling ED() on exponential decay models (EXD.2, EXD.3, AR.2, AR.3, W1.x, W2.x) fitted with two fixed parameters — leaving only a single parameter to be estimated — throws an "incorrect number of dimensions" error. It is currently impossible to retrieve ED values from these models in this configuration. See the code example below:

Code Example

# Reproducible example for testing

x <- data.frame(
  conc = c(1e+01, 1e+00, 1e-01, 1e+04, 1e+03, 1e+02, 1e+01, 1e+00,
           1e-01, 1e+04, 1e+03, 1e+02, 1e+01, 1e+00, 1e-01, 1e+04,
           1e+03, 1e+02, 0e+00, 0e+00, 0e+00, 0e+00, 0e+00, 0e+00,
           0e+00, 0e+00, 0e+00, 0e+00, 0e+00, 0e+00, 0e+00, 0e+00,
           0e+00, 0e+00, 0e+00, 0e+00),
  yield = c(15083.677, 142275.764, 197718.468, 0.000, 0.000, 0.000,
            67265.046, 197718.468, 266206.515, 28129.019, 5299.695,
            47697.033, 0.000, 139014.428, 178150.455, 0.000, 0.000,
            28129.019, 181411.790, 142275.764, 329394.891, 156544.107,
            230331.770, 256422.508, 112923.744, 187934.462, 321649.219,
            158582.442, 189157.384, 109662.408, 155321.106, 158582.442,
            311865.213, 259683.844, 152059.728, 178150.455)
)

##  Parameters ##
drm.formula = as.formula("yield ~ conc")
lowerl = 0 
# Compute the mean of the control group (conc == 0) to use as the upper limit in the model fitting
x %>%
  filter(conc == 0) %>%
  pull(yield) %>%
  mean(na.rm = TRUE) -> upperl


## Fitting with one fixed parameter (lower limit) and one free parameter (upper limit)
res = drm(
  formula = drm.formula, data = x,
  fct = EXD.3(fixed = c(lowerl, NA, NA)),      # <- this works
  na.action = na.omit 
)

plot(res, type = "all")
ED(res, c(10, 20,50), interval = "tfls")


## Fitting with one fixed parameter (lower limit) and one free parameter (upper limit)
res2 = drm(
  formula = drm.formula, data = x,
  fct = EXD.3(fixed = c(lowerl, upperl, NA)),  # <- this leads to the BUG in the ED() call
  na.action = na.omit 
)

plot(res2, type = "all")
try( 
  ED(res2, c(10, 20,50), interval = "tfls")  # <<< ERROR HERE
)

Two issues inside ED.drc:

1. Variance-covariance matrix not coerced to a proper matrix. When only one parameter is estimated, vcov() returns a 1×1 object that can collapse to a scalar or unnamed numeric of length 1.
   ED.drc performs matrix operations (%*%, t(), indexing) that assume vcov always has two dimensions. A scalar lacks these dimensions, causing the error.

2. Named gradient vectors cause dimension mismatches. Model-specific derivative functions (e.g., for EXD or W1) can return named numeric vectors. When passed into matrix multiplication, R can interpret the names as an implicit dimension attribute, producing conformability or dimension errors in the delta-method calculation.

Suggested Fix

Two small defensive changes in ED.drc:

# 1. Always coerce vcov to a proper matrix
vcMat <- as.matrix(vcov(object))

# 2. Strip names from gradient vectors before matrix algebra
g <- unname(g)