here is the hidden area text

Generalized Additive Models for Location, Scale and Shape

Statistical modelling at its best


What is GAMLSS

GAMLSS is a modern distribution-based approach to (semiparametric) regression models, where all the parameters of the assumed distribution for the response can be modelled as additive functions of the explanatory variables.
NEWS: A short course on GAMLSS will be held in Verona, Italy on the 23 and 24 of November.


How to use GAMLSS

The GAMLSS framework of statistical modelling is implemented in a series of packages in R. The packages can be downloaded from the R library, CRAN. The book Flexible Regression and Smoothing: Using GAMLSS in R , shows using practical examples, how GAMLSS models can be fitted. For more recent work and examples see the vignettes. For short courses on GAMLSS see Short Courses


How to learn more about GAMLSS

The book Flexible Regression and Smoothing: Using GAMLSS in R is published by CRC Press. Two others books are in preparation i) Distributions for Location Scale and Shape, the GAMLSS implementation in R iii) Generalised Additive Models for Location Scale and Shape: A Distributional Regression Approach. Draft versions of some of the books and other booklets can be found in Books & Articles The GAMLSS article on the Journal of Statistical Software can be useful (but a bit outdated) for a short introduction. Also some useful material can be found on short course booklets.


What distributions can be used within GAMLSS

GAMLSS provide over 90 continuous, discrete and mixed distributions for modelling the response variable. Truncated, censored, log and logit transformed and finite mixture versions of these distributions can be also used.


What additive terms can be used within GAMLSS

P-splines, Cubic splines, loess smoothing, ridge regression, lasso regression, random effects and varying coefficient models are some of the additive functions provided in the implementation. Appropriate interface is also provided so GAMLSS models can be used in combination with random effect models from the lme() function of the nlme package, smoothers from the gam() function (of package mgcv), the neural network function nnet() (of package nnet) , and decision threes (of package apart).


Who's is using GAMLSS

GAMLSS has been used in a variety of fields including: actuarial science, biology, biosciences, energy economic, genomics, finance, fisheries, food consumption, growth curves centile estimation, marine research, medicine, meteorology, rainfalls, vaccines, e.t.c. The World Health Organisation, the International Monetary Fund and the European Commission are some of international organisations who used GAMLSS.

Blogs and Latest News
  • A new package gamlss.inf is introduced recently to help with the fitting of inflated distributions on the interval [0,1] and zero adjusted distributions on the interval (o, infinity) for a response variable. The inflated and zero adjusted distributions are mixed  (i.e  continuous and discrete) distributions with a continuous part on a real line interval and […]

  • The book `Flexible Regression and Smoothing: Using GAMLSS in R’ is now published by CRC Press. It provides a broad overview of regression and smoothing techniques by focusing on the practical application. It provides an introduction to the  GAMLSS software in R. It includes a comprehensive collection of real data examples, which reflect the range […]

  • Version 4.4-0 1. gamlss i) the term.plot() did not plot (by mistake) a gam object. This is now corrected ii) the vis.lo() is introduced for plotting fitted loess curve iii) VC.test() for Vuong and Clarke test is introduced (use with caution) iv) the function stepVGDAll.A() is introduced. v) Q.stats() function is amended so the number […]

  •  1. package: gamlss i) The function within gamlss() has a line added to prevent the iterative weighs wt to go to Inf. ii) The tp() function within lms() and quantSheets() has changed name and modified slightly iii) The vcoc.gamlss() has the warnings changed and allows if theinverse of the Hessian (R) fails to recalucated […]

  • Version 4.2-7 i) gamlss gamlssML(): now allows the fitting binomial data (sorry it never checked before) and the use of formula in the specification of the model (e.g, y~1) to be consistent with gamlss(). Note that explanatory variables will be ignored if used with gamlssML().  .gamlss.multin.list is now on NAMESPACE  the functions vcov.gamlss() and summary.gamlss() […]

  • This version is released on the 22–6-2013 and it is the first time that robust (sandwich) standard errors are introduce  in gamlss models.  Of course those standard errors apply to parametric GAMLSS models only. When non-parametric smoothing terms are used then  the (sandwich) standard errors can still be used with caution since they are not yet take […]

  •   Version 4.2-5 The most important change in this version of gamlss is the way that the standard errors are calculated. In  previous version the vcov() function was calculated using a final iteration to a non-linear maximisation procedure. This procedure failed in a lot of occasions and the result was that the reported standard errors […]

  • The new version of gamlss is 4.2-0. The following are the changes made:   package gamlss: The functions and prof.term() are improved. The argument step is not anymore compulsory and if not set the argument length is used instead. For most cases there is no need to have a fine grid since the function is approximated using splinefun(). The output is […]

  • The new features in version 2.1-2 are as follows: package gamlss: The function histSmo() is added for density estimation. The function histDist() now has the function gamlssML() as its main fitting function. The fitting function  gamlss() is only used if gamlssML() fails. The function gamlssML() has now an argument start.from. In the function fitDist(), the normal distribution NO() is added to the list of “.realline” so it also appears […]

What did they say
  • All models are wrong but some are useful

    George Box

  • no matter how beautiful your theory, no matter how clever you are or what your name is, if it disagrees with experiment, it’s wrong

    Richard Feynman

  • entities should not be multiplied beyond necessity

    Occam’s Razor

  • The type of statistical inference used may be less important to the conclusions than choosing a suitable model or models in the first place.

    The GAMLSS team

  • Far better an approximate answer to the right question, which is often vague, than an exact answer to the wrong question, which can always be made precise.

    John W. Tukey