ONLINE COURSE – Advancing in R (ADVR01)

https://www.prstats.org/course/advancing-in-r-advr01/

25th - 29th March 2024

Please feel free to share!

COURSE DETAILS - This course is designed to provide attendees with a comprehensive understanding of statistical modelling and its applications in various fields, such as ecology, biology, sociology, agriculture, and health. We cover all foundational aspects of modelling, including all coding aspects, ranging from data wrangling, visualisation and exploratory data analysis, to generalized linear mixed models, assessing goodness-of-fit and carrying out model comparison.

Course description
This course is designed to provide attendees with a comprehensive understanding of
statistical modelling and its applications in various fields, such as ecology, biology, sociology,
agriculture, and health. We cover all foundational aspects of modelling, including all coding
aspects, ranging from data wrangling, visualisation and exploratory data analysis, to
generalized linear mixed models, assessing goodness-of-fit and carrying out model
comparison.

Data wrangling
For data wrangling, we focus on tools provided by R's tidyverse. Data wrangling is the art of
taking raw and messy data and formatting and cleaning it so that data analysis and
visualization may be performed on it. Done poorly, it can be a time consuming, laborious,
and error-prone. Fortunately, the tools provided by R's tidyverse allow us to do data
wrangling in a fast, efficient, and high-level manner, which can have dramatic consequence
for ease and speed with which we analyse data. We start with how to read data of different
types into R, we then cover in detail all the dplyr tools such as select, filter, mutate, and
others. Here, we will also cover the pipe operator (%>%) to create data wrangling pipelines
that take raw messy data on the one end and return cleaned tidy data on the other. We
then cover how to perform descriptive or summary statistics on our data using dplyr’s
group_by and summarise functions. We then turn to combining and merging data. Here, we
will consider how to concatenate data frames, including concatenating all data files in a
folder, as well as cover the powerful SQL-like join operations that allow us to merge
information in different data frames. The final topic we will consider is how to “pivot” data
from a “wide” to “long” format and back using tidyr’s pivot_longer and pivot_wider
functions.

Data visualisation
For visualisation, we focus on the ggplot2 package. We begin by providing a brief overview
of the general principles data visualization, and an overview of the general principles behind
ggplot. We then proceed to cover the major types of plots for visualizing distributions of
univariate data: histograms, density plots, barplots, and Tukey boxplots. In all of these
cases, we will consider how to visualize multiple distributions simultaneously on the same
plot using different colours and "facet" plots. We then turn to the visualization of bivariate
data using scatterplots. Here, we will explore how to apply linear and nonlinear smoothing
functions to the data, how to add marginal histograms to the scatterplot, add labels to
points, and scale each point by the value of a third variable. We then cover some additional
plot types that are often related but not identical to those major types covered during the
beginning of the course: frequency polygons, area plots, line plots, uncertainty plots, violin
plots, and geospatial mapping. We then consider more fine grained control of the plot by
changing axis scales, axis labels, axis tick points, colour palettes, and ggplot "themes".
Finally, we consider how to make plots for presentations and publications. Here, we will
introduce how to insert plots into documents using RMarkdown, and also how to create
labelled grids of subplots of the kind seen in many published articles.

Generalized linear models
Generalized linear models are generalizations of linear regression models for situations
where the outcome variable is, for example, a binary, or ordinal, or count variable, etc. The
specific models we cover include binary, binomial, and categorical logistic regression,
Poisson and negative binomial regression for count variables, as well as extensions for
overdispersed and zero-inflated data. We begin by providing a brief overview of the normal
general linear model. Understanding this model is vital for the proper understanding of how
it is generalized in generalized linear models. Next, we introduce the widely used binary
logistic regression model, which is is a regression model for when the outcome variable is
binary. Next, we cover the binomial logistic regression, and the multinomial case, which is
for modelling outcomes variables that are polychotomous, i.e., have more than two
categorically distinct values. We will then cover Poisson regression, which is widely used for
modelling outcome variables that are counts (i.e the number of times something has
happened). We then cover extensions to accommodate overdispersion, starting with the
quasi-likelihood approach, then covering the negative binomial and beta-binomial models
for counts and discrete proportions, respectively. Finally, we will cover zero-inflated Poisson
and negative binomial models, which are for count data with excessive numbers of zero
observations.

Mixed models
We will focus primarily on multilevel linear models, but also cover multilevel generalized
linear models. Likewise, we will also describe Bayesian approaches to multilevel modelling.
We will begin by focusing on random effects multilevel models. These models make it clear
how multilevel models are in fact models of models. In addition, random effects models
serve as a solid basis for understanding mixed effects, i.e. fixed and random effects, models.
In this coverage of random effects, we will also cover the important concepts of statistical
shrinkage in the estimation of effects, as well as intraclass correlation. We then proceed to
cover linear mixed effects models, particularly focusing on varying intercept and/or varying
slopes regression models. We will then cover further aspects of linear mixed effects models,
including multilevel models for nested and crossed data data, and group level predictor
variables. Towards the end of the course we also cover generalized linear mixed models
(GLMMs), how to accommodate overdispersion through individual-level random effects, as
well as Bayesian approaches to multilevel levels using the brms R package.

Model selection and model simplification
Throughout the course we consider the fundamental issue of how to measure model fit and
a model’s predictive performance, and discuss a wide range of other major model fit
measurement concepts like likelihood, log likelihood, deviance, and residual sums of
squares. We thoroughly explore nested model comparison, particularly in general and
generalized linear models, and their mixed effects counterparts. We discuss out-of-sample
generalization, and introduce leave-one-out cross-validation and the Akaike Information
Criterion (AIC). We also cover general concepts and methods related to variable selection,
including stepwise regression, ridge regression, Lasso, and elastic nets. Finally, we turn to
model averaging, which may represent a preferable alternative to model selection.

Please email [log in to unmask] with any questions.

########################################################################

To unsubscribe from the ENVIRONMENTAL-SCIENCE-ENGAGEMENT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=ENVIRONMENTAL-SCIENCE-ENGAGEMENT&A=1

This message was issued to members of www.jiscmail.ac.uk/ENVIRONMENTAL-SCIENCE-ENGAGEMENT, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/