www.highstat.com
Highland Statistics Ltd

On demand: Zero-inflated GAMs and GAMMs for spatial and spatial-temporal data

Online course with on-demand video and live Zoom meetings:

Zero-inflated GAMs and GAMMs for the analysis of spatial and spatial-temporal correlated data using R-INLA

This online course consists of 6 modules representing a total of approximately 40 hours of work. Each module consists of multiple video files with short theory presentations, followed by exercises using real data sets, and video files discussing the solutions. All video files are on-demand and can be watched online, as often as you want, at any time of the day, within a 6 month period.

A discussion board allows for daily interaction between instructors and participants. The course also contains 6 2-hour live web meetings (using Zoom) in which we summarise the theory and the exercises. Attending these live web meetings is optional. We will run the web meetings in different time zones. These Zoom summary sessions will be recorded and you can watch them on the course website. You can also use a Discussion board to ask any questions on the course material. The course fee includes a 1-hour face-to-face video chat with the instructors. You can use this video chat to ask questions about your own data.

A detailed outline of the course is provided below. All exercises consist of a data set, a video describing the data and the questions, R solution code and a video discussing the R solution file.

Course content

We start with a short revision of data exploration and linear regression (frequentist approach). We then explain the basics of Bayesian statistics, posterior distributions, priors, etc. 

In the second part, we explain how to deal with zero-inflated data. We will introduce so-called zero-inflated Poisson (ZIP) models, zero-inflated negative binomial (ZINB) models, zero-altered Poisson (ZAP) models, and zero-altered negative binomial (ZANB) models. We will show how to execute these models in R-INLA.

In the third part of the course, we will introduce generalised additive models (GAM) to model non-linear relationships. We will execute these models in R-INLA.

In the fourth part of the course, we will revise linear mixed-effects models and show how to implement a generalised additive mixed-effects model (GAMM) in R-INLA. We also show how to include an interaction between a smoother and a categorical covariate.

In the fifith part of the course, we will apply zero-inflated GAMs and GAMMs on various spatially correlated data sets.

In the sixth part of the course, we apply GAM and GAMM on spatial-temporal correlated data. We also deal with natural barriers for the spatial correlation (e.g. benthic species that live on a coral reef around an island). We will discuss barrier models; these ensure that spatial correlation seeps around a barrier (in this case an island).

Important: The required knowledge for this course is fairly high. You need to have good knowledge of R, data exploration, linear regression and GLM (Poisson, negative binomial, Bernoulli). And you need to have a working knowledge of mixed-effects models. We strongly recommend that you only join this course if you either attended one of our INLA courses or are familiar with R-INLA.

Keywords: Zero inflation, GLM, GLMM, GAM, GAMM, spatial correlation, spatial-temporal correlation, barrier models, R-INLA.

A detailed outline of the course is provided below. All exercises consist of a data set, a video describing the data and the questions, R solution code and a video discussing the R solution file.

 

Preparation material

As preparation for this course, we provide some revision material (videos) from our other courses on data exploration, linear regression, Poisson GLM, and negative binomial GLM (using frequentist approaches). There is no Zoom summary session for this module.

Module 1: Revision and introduction to R-INLA

  • We start with a revision of data exploration and linear regression, followed by an introduction to Bayesian statistics and R-INLA.
  • An exercise revising multiple linear regression (frequentist approach).
  • A short video explaining basic matrix algebra.
  • A video presentation with a short introduction to Bayesian statistics and the role of priors.
  • A video presentation explaining the basic principles of INLA.
  • One exercise showing how to execute a linear regression model in R-INLA (Bayesian approach).
  • Web meeting 1: A 2-hour web-meeting will be scheduled. In this web meeting, we will summarise module 1.

Module 2: Introduction to zero-inflated models

  • A video presentation on the Poisson, negative binomial and Bernoulli distributions.
  • A video presentation with a short revision of Poisson, negative binomial and Bernoulli generalised linear models (GLMs).
  • One exercise showing how to execute a Poisson GLM in R-INLA.
  • One exercise showing how to execute a negative binomial GLM in R-INLA.
  • One exercise showing how to execute a Bernoulli GLM in R-INLA.
  • A video presentation explaining models for zero-inflated count data (ZIP, ZINB, ZAP and ZANB models).
  • Two exercises on the analysis of zero-inflated count data using R-INLA.
  • Web meeting 2: A 2-hour web-meeting will be scheduled. In this web meeting, we will summarise module 2.

Module 3: Generalised additive models in R-INLA

  • A video with a theory presentation on generalised additive models (GAM).
  • One exercise showing how to execute a GAM with a Gaussian distribution in R-INLA.
  • One exercise showing the application of a Poisson GAM in R-INLA.
  • One exercise showing the application of a negative binomial GAM in R-INLA.
  • One exercise showing the application of a Bernoulli GAM in R-INLA.
  • Web meeting 3: A 2-hour web-meeting will be scheduled. In this web meeting, we will summarise module 3.

Module 4: GAMM with interactions.

  • A video presentation with a short revision of linear mixed-effects models.
  • One exercise showing how to execute a linear mixed-effects model in R-INLA.
  • One exercise showing how to execute a generalised additive mixed-effects model (GAMM) in R-INLA.
  • A video presentation showing how to implement an interaction term between a smoother and a categorical covariate in a GAMM.
  • One exercise showing how to execute a GAMM with an interaction between a smoother and a categorical covariate in R-INLA.
  • Web meeting 4: A 2-hour web-meeting will be scheduled. In this web meeting, we will summarise module 4.

Module 5: GAM and GAMM applied to spatial correlated data

  • A short video explaining the essential steps of adding spatial correlation to a linear regression model.
  • One exercise showing how to apply a linear regression model with spatial correlation in R-INLA.
  • Three exercises on zero-inflated GAMs and GAMMs with spatial correlation in R-INLA.
  • Web meeting 5: A 2-hour web-meeting will be scheduled. In this web meeting, we will summarise module 5.

Module 6: GAM and GAMM applied to spatial-temporal correlated data. Barrier models.

  • Two exercises on the application of zero-inflated GAMs (and GAMMs) to spatial-temporal correlated data in R-INLA.
  • Theory presentation on barrier models.
  • One exercise showing how to apply a GAM with spatial correlation and a barrier.
  • Web meeting 6: A 2-hour web-meeting will be scheduled. In this web meeting, we will summarise module 6.

 

Free 1-hour face-to-face video meeting: The course fee includes a 1-hour face-to-face video meeting with one or both instructors. The meeting needs to take place within 3 months after the last live zoom meeting. You can discuss your own data but we strongly suggest that the statistical topics are within the content of the course. The 1-hour needs to be used in one session and will take place on a mutually convenient day and time of the day.

Web meetings: Web meetings are hosted on zoom.us. Click here for recommended internet speed (see the text under 'Recommended bandwidth for Webinar Attendees'). We will record the meetings and make them available on the course website.

Discussion Board: You can use the Discussion Board to ask any questions related to the course material. 

Pre-required knowledge: The required knowledge for this course is fairly high. You need to have good knowledge of R, data exploration, linear regression and GLM (Poisson, negative binomial, Bernoulli). And you need to have a working knowledge of mixed-effects models. We strongly recommend that you only join this course if you either attended one of our INLA courses or are familiar with R-INLA. Short revisions are provided. This is a non-technical course.

Cancellation policy: What if you are not able to participate? Once participants are given access to course exercises with R solution codes, pdf files of certain book chapters, pdf files of PowerPoint or Prezi presentations and video solution files, all course fees are non-refundable and non-transferable to another participant.

Copyright: Sharing the access details of the course website or the pdf files of our course material is prohibited. Video files cannot be downloaded, but they can be watched in the same way as on Netflix.