nonlinear regression prediction in R

Question

I am confused by this warning message as I try to fit my data with a nonlinear regression model by using the drc package and drm function.

I have

 N_obs <- c(1, 80, 80, 80, 81, 82, 83, 84, 84, 95, 102, 102, 102, 103, 104, 105, 105, 109, 111, 117, 120, 123, 123, 124, 126, 127, 128, 128, 129, 130)

 times <- c(3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32)

The model is

  model.drm <- drm(N_obs ~ times, data = data.frame(N_obs = N_obs, times = times), fct = MM.2())

and the warnings come from predictions

  preds <- predict(model.drm, times = times, interval = "confidence", level = 0.95)

There were 30 or more warnings (use warnings() to see the first 50)
> warnings()
Warning messages:
1: In (tquan * sqrt(varVal + sumObjRV)) * c(-1, 1) :
Recycling array of length 1 in array-vector arithmetic is deprecated.
Use c() or as.vector() instead.
2: In (tquan * sqrt(varVal + sumObjRV)) * c(-1, 1) :
Recycling array of length 1 in array-vector arithmetic is deprecated.
Use c() or as.vector() instead.

3: In (tquan * sqrt(varVal + sumObjRV)) * c(-1, 1) :
Recycling array of length 1 in array-vector arithmetic is deprecated.
Use c() or as.vector() instead.

I have been trying to change data inputs by using as.vector(times), c(times),etc., but still cannot get rid of the warnings. Could someone help me identify the problem? Thank you!!

Answer 1

I re-ran your analysis with the sample data provided, and I can reproduce your warnings. Here's a summary:

1. Fit a Michaelis-Menten model of the form f(x; d, e) = d * (1 + e/x)^-1 .

    # Fit a 2 parameter Michaelis-Menten model library(drc); fit <- drm( formula = N_obs ~ times, data = data.frame(N_obs = N_obs, times = times), fct = MM.2()) 

2. Based on the model fit, predict the response for the original times . Note you can omit the newdata argument here, because in that case predict will simply use the fitted values (which are based on times ).

    # Predictions pred <- as.data.frame(predict( fit, newdata = data.frame(N_obs = N_obs, times = times), interval = "confidence", level = 0.95)); pred$times <- times; 

3. Visualise data and predictions.

    library(tidyverse); data.frame(times = times, N_obs = N_obs) %>% ggplot(aes(times, N_obs)) + geom_point() + geom_line(data = pred, aes(x = times, y = Prediction)) + geom_ribbon( data = pred, aes(x = times, ymin = Lower, ymax = Upper), alpha = 0.4); 

The model fit seems reasonable, and I would say that the warnings can be safely ignored (see details).

Details

I had a look at the drc source-code, and the warning originates from line 201 of predict.drc.R :

retMat[rowIndex, 3:4] <- retMat[rowIndex, 1] + (tquan * sqrt(varVal + sumObjRV)) * c(-1, 1)

In that line, an array of dimension 1 is added to a numeric vector.

Here is a simple example to reproduce the warning:

arr <- array(5, dim = 1);
arr + c(1, 2);
#[1] 6 7
#Warning message:
#In arr + c(1, 2) :
#  Recycling array of length 1 in array-vector arithmetic is deprecated.
#  Use c() or as.vector() instead. 

Note that the result is still correct; it's just that R doesn't like the addition of a one-dimensional array and a vector, and prefers instead adding proper scalars and vectors, or vectors and vectors.