alexpiehler/gamlss_vs_mgcv_simulation

This is the repository for a simulation study in which the performance of the R packages gamlss and mgcv are compared for location-scale modeling with zero-inflated Poisson data

Code for the simulation study: gamlss vs. mgcv with zero-inflated Poisson data

This is the repository for a simulation study in which the performance of the R packages gamlss (Rigby et al. 2005) and mgcv (Wood et al. 2016) are compared for location-scale modeling with zero-inflated Poisson data. In particular the algorithms, which estimate the smoothing parameters of the spline-functions, are compared. This simulation study extends the work of Wood et al. (2016) by not only modeling the rate of the zero-inflated Poisson distribution through non-linear functions of covariates but by also explicitly modeling the probability of zero-inflation through non-linear functions of covariates.

It is found that the algorithm for estimating smoothing parameters in the packaged mgcv outperforms the algorithm gamlss in speed and convergence rate. While the algorithm in the gamlss package is less stable, especially for low mean and highly dispersed count data, it on average yields comparable results to the algorithm in the mgcv package.

All results are reproducible and the R-code can be found in simulation_study.Rmd.

---

Rigby, R. A. & Stasinopoulos, D. M. (2005), Generalized additive models for location, scale and shape, Journal of the Royal Statistical Society 54(3), 507-554.

Wood, S. N., Pya, N. & Säfken, B. (2016), Smoothing parameter and model selection for general smooth models, Journal of the American Statistical Association 111(516), 1548{1563.