Skip to contents

Fit a single growth-curve model by nonlinear least squares

Source: R/growth_models.R
fit_growth_model.Rd

Fits logistic_curve(), gompertz_curve() or richards_curve() to a cumulative-incidence series by nonlinear least squares (minpack.lm::nlsLM()), which is markedly more robust to the identification issues of these S-shaped curves than base stats::nls().

Usage

fit_growth_model(
  t,
  y,
  model = c("logistic", "gompertz", "richards"),
  beta_fixed = 1.5,
  start = NULL,
  lower = NULL,
  upper = NULL
)

Arguments

t

Numeric vector of time indices (1, 2, ..., n, relative to the start of the fitting window).

y

Numeric vector of cumulative counts, same length as t.

model

One of "logistic", "gompertz", "richards".

beta_fixed

If not NULL (Richards only), the shape parameter is held fixed at this value instead of estimated (improves convergence on short/fast-saturating phases, as used throughout the simulation study).

start

Optional named list overriding the default starting values y0, K, gamma (and beta for Richards).

lower, upper

Optional named lists overriding the default box constraints.

Value

An object of class growth_fit: a list with model, the fitted nls object, coef (named coefficient vector), fitted (fitted values at t), resid, mse (mean squared error, Equation for EQM in the manuscript), aic, and n_par (number of estimated parameters, used by the AIC-based ensemble weights).