-1

I have a continuous response with three categorical variables, no random effects. I have fit a gamlss object using the zero-adjusted gamma distribution. Link to data here because it's large. I've can't get the data into a more standard distribution and I'd like to avoid using a hurdle-type analysis approach if possible for easier interpretation.

data_1 <- read.csv('data_1.csv', sep = ',')
library(gamlss)
mZAGA <- gamlss(cent ~ treat * cat * season, family = ZAGA(), data = data_1)

I want to compare the significance of the main effecs and interactions within the model, which is not a feature directly available in gamlss it seems. This paper describes converting a gamlss object into an nlme::lme object (https://arxiv.org/abs/1810.03085, page 17) which seems like a good approach, since lme models are more widely accepted by stats packages.

Once the mixed GAMLSS is fitted and stored in an R object, the function getSmo() can be used to transform it to an lme object, so that all methods used with mixed models fitted via lme() become available, such as analysis of variance via the anova() function.

However the getSmo documentation isn't helping me with this conversion and trial-and-erroring is getting me nowhere. How can I use my coefficients (mu's, sigma and nu) to build an lme object? Is this something I can even do while retaining the zero-adjusted gamma distribution from gamlss?

Tormey R
  • 11
  • 3

1 Answers1

-1

If you have no random effects, why are you using lme?

You can apply backward elimination from your model for mu: treat * cat * season, and then from (treat+cat+season)**2 in gamlss.

See Chapter 11: Model selection techniques, pages 377-415, in 'Flexible Regression and Smoothing: Using GAMLSS in R’ M. D. Stasinopoulos, R. A. Rigby, G. Z. Heller, V. Voudouris and F. De Bastiani. Chapman and Hall/CRC, Boca Raton, 2017

Paperback (2020): www.routledge.com/9781138197909

Note that in a zero adjusted gamma distribution, Y ~ ZAGA(mu, sigma,nu), in gamlss, Y has value 0 with probability nu, and Y ~ GA(mu, sigma) with probability (1-nu).

See page 455 and also Chapter 9: Mixed Distributions (pages 177 - 206) of

‘Distributions for Modeling Location, Scale, and Shape: Using GAMLSS in R’ R. A. Rigby, M. D. Stasinopoulos, G. Z. Heller and F. De Bastiani. Chapman and Hall/CRC, Boca Raton, 2019

Paperback (2021) www.routledge.com//9780367278847

In your gamlss model, you are just modelling mu.

Robert
  • 1
  • 1