Soul, L. C. & D. F. Wright. 2021. Phylogenetic Comparative Methods, A User's Guide For Paleontologists

Soul and Wright
Recent advances in statistical approaches called phylogenetic

comparative methods (PCMs) have provided paleontologists
with a powerful set of analytical tools for investigating
evolutionary tempo and mode in fossil lineages. However,
attempts to integrate PCMs with fossil data often present
workers with practical challenges or unfamiliar literature. Elements of Paleontology
This Element presents guides to the theory behind and the
application of PCMs with fossil taxa. Based on an empirical
dataset of Paleozoic crinoids, example analyses are presented
to illustrate common applications of PCMs to fossil data,
including investigating patterns of correlated trait evolution
and macroevolutionary models of morphological change.
Phylogenetic
Phylogenetic Comparative Methods: A User’s Guide for Paleontologists

The authors emphasize the importance of accounting
for sources of uncertainty and discuss how to evaluate
model fit and adequacy. Finally, the authors discuss several
promising methods for modeling heterogeneous evolutionary
Comparative Methods:
dynamics with fossil phylogenies. Integrating phylogeny-
based approaches with the fossil record provides a rigorous,
A User’s Guide for
quantitative perspective on understanding key patterns in the
history of life.
Paleontologists
About the Series

The Elements of Paleontology
Series Editor-in-Chief
Colin D. Sumrall Laura C. Soul and
series is a publishing collaboration
between the Paleontological Society
University of
Tennessee David F. Wright
and Cambridge University Press.
The series covers the full spectrum
of topics in paleontology and
paleobiology, and related topics in
the Earth and life sciences of interest
to students and researchers of
paleontology.
Downloaded from https://www.cambridge.org/core. IP address: 189.131.210.246, on 22 Apr 2021 at 04:30:41, subject ISSN
to the2517-780X
Cambridge(online)
Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108894142 ISSN 2517-7796 (print)
Downloaded from https://www.cambridge.org/core. IP address: 189.131.210.246, on 22 Apr 2021 at 04:30:41, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108894142
Elements of Paleontology
edited by
Colin D. Sumrall
University of Tennessee
PHYLOGENETIC
COMPARATIVE METHODS:
A USER’S GUIDE FOR
PALEONTOLOGISTS
Laura C. Soul
The Natural History Museum, London and National
Museum of Natural History, Smithsonian Institution
David F. Wright
American Museum of Natural History
and National Museum of Natural History,
Smithsonian Institution
University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi – 110025, India
79 Anson Road, #06–04/06, Singapore 079906
Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781108794688
DOI: 10.1017/9781108894142
© Laura C. Soul and David F. Wright 2021
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2021
A catalogue record for this publication is available from the British Library.
ISBN 978-1-108-79468-8 Paperback
ISSN 2517-780X (online)
ISSN 2517-7796 (print)
Additional resources for this publication at cambridge.org/soulwright
Cambridge University Press has no responsibility for the persistence or accuracy of

URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Phylogenetic Comparative Methods:
A User’s Guide for Paleontologists
DOI: 10.1017/9781108894142
First published online: April 2021
Laura C. Soul
The Natural History Museum, London and National Museum of Natural History,
David F. Wright
American Museum of Natural History and National Museum of Natural History,
Author for correspondence: Laura C. Soul, [email protected]
Abstract: Recent advances in statistical approaches called phylogenetic

comparative methods (PCMs) have provided paleontologists with a
powerful set of analytical tools for investigating evolutionary tempo
and mode in fossil lineages. However, attempts to integrate PCMs with
fossil data often present workers with practical challenges or unfamiliar
literature. This Element presents guides to the theory behind and the
application of PCMs with fossil taxa. Based on an empirical dataset of
Paleozoic crinoids, example analyses are presented to illustrate
common applications of PCMs to fossil data, including investigating
patterns of correlated trait evolution and macroevolutionary models of
morphological change. The authors emphasize the importance of
accounting for sources of uncertainty and discuss how to evaluate
model fit and adequacy. Finally, the authors discuss several promising
methods for modeling heterogeneous evolutionary dynamics with
fossil phylogenies. Integrating phylogeny-based approaches with the
fossil record provides a rigorous, quantitative perspective on
understanding key patterns in the history of life.
Keywords: phylogenetic comparative methods, paleontology,

macroevolution, R, fossil
ISBNs: 9781108794688 (PB), 9781108894142 (OC)

ISSNs: 2517-780X (online), 2517-7796 (print)
Contents
1 Introduction 1
2 Getting Started: Data and Phylogeny 2
3 Phylogenetic Nonindependence 7
4 Tempo and Mode: Brownian Motion and More 12
5 Incorporating Estimates of Error 20
6 Variation Across Trees 20
7 Model Fit and Model Adequacy 27
8 Post-Hoc Modeling of Discrete Characters 29
9 Modeling Heterogeneous Trait Evolution 36
10 Conclusion 45
References 46
Phylogenetic Comparative Methods: A User’s Guide for Paleontologists 1
1 Introduction
A fundamental prediction of biological evolution is that a species will most
commonly share many characteristics with lineages from which it has recently
diverged, and fewer characteristics with lineages from which it diverged further
in the past. This principle, which results from descent with modification, is one
of the most basic in biology (Darwin 1859). A reconstruction of the relation-
ships between species that is based on similarities and differences in character-
istics – a phylogenetic hypothesis – can be used as a powerful tool to understand
fundamental questions about the history of life (Nunn 2011; Baum and Smith
2013; Harmon 2019).
Phylogenetic comparative methods (PCMs) can be broadly defined as statis-
tical approaches that incorporate information about the shared evolutionary
history of taxa (i.e. their nonindependence) to identify macroevolutionary
patterns or test hypotheses about how those patterns relate to macroevolutionary
drivers, such as climate or biotic interactions. Until recently, there have been
largely separate analytical frameworks for phylogenetic inference (inferring
relationships between taxonomic units using morphological or molecular char-
acter data) and PCMs (testing hypotheses about evolution while treating the
relationships as known). In this Element we only discuss the latter, but other
recent literature focuses on a unified methodological framework that integrates
the two (Warnock and Wright 2020; Wright et al. 2020). Many other types of
information, such as biogeographic data (Matzke and Wright 2016; Landis
2017), can be used in this modeling framework, and as new models are devel-
oped the scope of questions that can be addressed will increase.
There are many reasons (perhaps particularly for paleobiologists) why it
might not be feasible, or of interest, to use this kind of unified framework,
where you must infer phylogenetic relationships as a means to answer other
macroevolutionary questions. Perhaps you have previously estimated
a phylogeny but now want to use it to answer new questions, perhaps you are
interested in combining several smaller phylogenies to generate a supertree, or
perhaps your specific question is not yet answerable in a Bayesian process-
based framework (e.g. Warnock and Wright 2020; Wright et al. 2020). The good
news is that trees constructed in many different ways can be used in PCMs to
make reliable inferences about trait diversification, provided that the tree is
appropriately scaled to time using the stratigraphic record as an extra source of
information (Bapst 2014a; Soul and Friedman 2015; Barido-Sottani, Tiel et al.
2020).
PCMs have been rapidly proliferating in the past five to ten years, and the
kinds of questions they can be used to rigorously answer are now very diverse.
2 Paleontology
Here, we review and demonstrate some of the analytical approaches that can be
applied when you already have a phylogeny in hand. Many of these PCMs are
used to model trait change through time and the relationship between that trait
change and other variables. Most do not model the underlying microevolution-
ary processes occurring in populations or the external drivers that generate the
trait change. Instead, PCMs more commonly use stochastic models to investi-
gate the long-term outcome of evolutionary change. Therefore, as is outlined in
more detail later on, different underlying processes can generate similar patterns
that can be equally well explained by the same model. Careful interpretation of
the results of any analysis is imperative.
We begin by outlining some fundamental approaches that are conceptually
important, and then move on to more complex macroevolutionary models.
Multiple books have been written on this vast topic (see Nunn 2011, Garamszegi
2014, and Harmon 2019, each of which are not focused on the fossil record but
paleontologists might find them useful nonetheless), but we hope to provide you
with a clear, digestible explanation of the theory behind a variety of PCMs, along
with enough information on how to apply them to get started in answering your
own questions. A reader already well versed in these approaches will find a review
of recently published methods, and suggestions for their implementation in
a paleobiological context, in the last third of the Element. A variety of software
has been used to implement PCMs, but the majority of those likely to be of interest
to paleontologists are available in R (e.g. Bapst 2012; Lloyd 2016; Barido-Sottani
et al. 2019). We therefore provide all of our reproducible examples in R. Package
names and inline code examples are in Courier New font.
All data used in the examples in this Element are available on GitHub (https://
github.com/daveyfwright/PCMsForPaleontologists), along with a script to load
and format the data in the R programming environment so that they can be
analyzed immediately, as well as the full R script and annotated script and
outputs from the example analyses. Although the examples in this Element are
intended to provide a guide for implementing comparative analyses in R, we
encourage readers to also follow along directly using the scripts we provide on
GitHub. The example code in the main Element assumes a basic working
knowledge of R, including loading data, inspecting objects, reading help
pages, and object assignment. The materials online do not make this assumption
and provide a detailed step-by-step guide.
2 Getting Started: Data and Phylogeny

Functions in R that can be used to manipulate phylogeny or apply phylogenetic
comparative approaches make use of trees that are in “phylo” format,
originally implemented in the ape package (Paradis et al. 2004). In this section,
we outline features of this format, as well as some important things to remember
when preparing paleontological data for an analysis in R (see also Bapst 2014b).
The dataset we use here is for fossil crinoids (Eucladida, Echinodermata),
a morphologically diverse clade of marine invertebrates with a well-sampled
fossil record. Our use of this dataset is primarily intended to demonstrate the
different tree-based analytical tools that can be applied, rather than to glean
specific inferences about crinoid macroevolution, and we caution readers that
our results should be viewed in this light.
Key packages in R that contain implementations of PCMs that are commonly
applied to fossil data are ape (standard format and processing for phylogenies
in R – Paradis et al. 2004), nlme (fitting Gaussian models, e.g. least-squares
regression – Pinheiro et al. 2019), geiger (a versatile package that performs
and plots many PCMs – Harmon et al. 2008; Pennell et al. 2014), phytools
(additional plotting and simulation functions – Revell 2012), and OUwie
(heterogeneous macroevolutionary model fitting, e.g. Brownian motion or
Early Burst – Beaulieu and O’Meara 2020). There are many others, so it is
valuable to spend time exploring the different tools that might be best suited to
answering a particular question. Once you have gained familiarity with the
above key packages, a very extensive list of all the packages that can be used for
phylogenetic approaches in R can be found on this website: https://cran.r
-project.org/web/views/Phylogenetics.html.
2.1 Phylogeny
For most of the example analyses we use a single phylogeny of eucladid
crinoids that has branch lengths that represent time in millions of years
(Figure 1, called tree in our scripts). It is the maximum clade credibility
tree (MCCT), inferred using a model of morphological character evolution
combined with a process-based model of diversification that allows inference
of ancestor–descendant relationships (this method is distinct from ancestral
state reconstruction – see subsection 4.2; Wright 2017a). The MCCT is the
tree in the posterior distribution with the largest product of clade frequencies
(i.e. probabilities), which represents a point estimate of phylogeny in
a Bayesian context. Our emphasis on a single tree is for illustrative purposes
only; in reality, analyses should always be applied to a set of possible
phylogenies to assess how robust results are to variation in tree topology
and branch lengths (see Section 6). The set of possible phylogenies (often
referred to as a tree set) could be a set of the most parsimonious trees,
a random sample from a Bayesian posterior distribution, repeats of stochastic
4 Paleontology
Figure 1 Phylogenetic tree comprising eighty-two species belonging to the

Eucladida (Crinoidea, Echinodermata). Tree topology and branch lengths
correspond to the maximum clade credibility tree (MCCT) based on a Bayesian
tip dating analysis of morphological character data presented in Wright (2017a).
Branch lengths are scaled to absolute time in units of millions of years. The
MCCT is called tree in our scripts and represents a point estimate of phylogeny.
Where possible, we recommend testing macroevolutionary inferences across
a distribution of trees rather than a point estimate, and our emphasis on the MCCT
is for illustrative purposes only (see subsection 2.1 and Section 6) .
timescaling of a composite topology (Bapst 2014a), etc. The number of trees

to include in the set depends on how variable results are across different trees;
if results are highly variable, a larger set will be needed to characterize the
possible outcomes of the analysis and how common they are. For example,
100, 500, or 1,000 are common set sizes in the published literature. The
phylogenies used for analysis should have branch lengths that are scaled to
time (best practice would be to use a birth-death-sampling approach [Stadler
2010; Gavryushkina et al. 2014; Heath et al. 2014; Wright 2017; Stadler et al.
2018], but if that is not possible, see Hedman 2010, Bapst 2014b, Halliday
and Goswami 2016, or Lloyd 2016 for options and considerations when
choosing an a posteriori timescaling approach).
A tree in phylo format has branches that are called “edges” and branching
points that are called “nodes”. For timescaled phylogenies, the branch (or edge)
is a graphical representation of the amount of time since the lineage leading to
one taxon diverged from its sister lineage, so the length of terminal branches
does not represent the amount of time the actual species or genus at the tip was
extant, only the time since divergence. Each node has a number assigned to it,
beginning with the tips. This format is for both ultrametric (branches all end at
the same time; usually trees of all living taxa) and nonultrametric trees
(branches end at different times; usually trees that include extinct taxa).
A phylo tree in R has four components: (1) a matrix that identifies how the
branches connect, by listing their start and end node numbers, (2) a vector of tip
labels (user defined, usually taxon names), (3) a vector of branch lengths, and
(4) an integer number that is the number of internal nodes. For trees that include
only extinct taxa, the root age of the tree can be set using tree$root.time
<- X, where X is the numerical age (usually reported in units Ma for paleonto-
logical datasets). A root age is required by some analyses, and facilitates good
visualization when plotting a tree. If there is no root age assigned, most
functions will assume the youngest tip ends at the present day; without it, the
tree is in relative time, rather than absolute.
An important first step prior to analysis is to plot the tree (Revell et al. 2018).
In our eucladid example some taxa were inferred to be ancestral to others in the
tree (Figure 1; see subsection 4.2 for clarification of what this means). In R, a
phylo object displays these sampled ancestors as sister to their descendants,
but with zero-length branches (i.e. no inferred change between the node and tip
for the ancestral taxon). For many PCMs, zero-length branches are mathemat-
ically intractable (the reasons relate to division-by-zero issues). Adding a very
short length to each zero-length branch resolves this problem. If you are
concerned about the possibility of this introducing a bias in your own analysis,
you could also drop these tips from the tree and compare the results using each
tree. Note that in the Eucladida dataset there are many inferred sampled ances-
tors, so dropping these tips may represent high information loss. Whether or not
this is the case for different datasets will depend on the group under investiga-
tion. Node labels are required by some PCM packages (e.g. OUwie), so it is best
practice to assign them using a vector the same length as the number of nodes
(as long as you check that the R function you are using doesn’t use them for
anything you aren’t expecting; for example, OUwie uses them to define ances-
tral macroevolutionary regimes – see subsection 9.1). This can be done using
tree$node.label <- rep(1, Nnode(tree)), which gives all nodes
the label “1”.
A very useful basic function to inspect and manipulate phylogenies in R is
vcv. When applied to a timescaled tree this function outputs the phylogenetic
variance-covariance (VCV) matrix of that tree. The VCV matrix is an intuitive
numeric representation of the tree, and is used in the inner workings of many
R functions implementing PCMs. Models of continuous trait change (like
Brownian motion; see Section 4 onwards) differ from one another with respect
to how the variance and covariance of the trait are expected to change through
time, and the VCV matrix is the basis for the statistical expectation under
6 Paleontology
different models. Elements on the diagonal of the matrix give the variance.
When a tree has branch lengths scaled to time (as paleontological trees usually
do), these values give the root-to-tip distance for each taxon on the tree (i.e.
duration). The functions max(diag(vcv(tree))) will output the max-
imum root-to-tip distance (i.e. the duration of the whole tree). The off-diagonal
elements give the covariance, which is the amount of shared variance between
pairs of taxa (i.e. the length of their shared evolutionary history). Figure 2 shows
a small example tree and its associated VCV matrix.
2.2 Trait Data

In our example analyses we use a variety of PCMs to explore patterns in two
continuous traits measured for each species in the Eucladida tree. These are
Shape (calyx shape; the natural log of the length/width ratio of the calyx) and
Figure 2 (a) Hypothetical phylogeny with branch lengths scaled to absolute

time. (b) Variance-covariance matrix that describes the tree. The values along
the diagonal of the matrix represent the amount of time from the root to each tip
in the tree, which corresponds to their expected variance in a Brownian motion
model. In contrast, the off-diagonal values represent the shared amount of
evolutionary history for pairs of tips, which corresponds to their evolutionary
covariance. Note that the diagonal values in the matrix represent terminal
branches whereas the off-diagonal values represent internal branches.
Density (filtration fan density; the natural log of the approximate number of
proximal feeding appendages an individual of the species has). We also briefly
demonstrate analysis of a discrete trait Complexity (calyx complexity; the
number of plates interrupting the posterior interray). We store all the traits in an
object called alltraits for ease of use in R. Some analyses implemented in
R require that the tree and data have exactly the same taxa; for these we use
a tree (called prunedTree in code examples) in which taxa not present in the
trait data have been removed from the tree. We use an estimated standard error
of 0.03 for Shape and 0.12 for Density, based on an average across species
for which there is more than one specimen.
3 Phylogenetic Nonindependence
Felsenstein (1985) was the first to outline the problem of the nonindependence of
species trait data, and an algorithm to account for that problem, which he called
phylogenetic independent contrasts (PIC). The most common evolutionary ques-
tion that this problem (and proposed solution) is applied to involves the relation-
ship between two traits. The extreme case of the problem of nonindependence of
species data is shown in the original paper (Felsenstein 1985, figure 1), and
a related example is provided in Figure 3 based on Nunn and Barton (2001).
An early divergence within a clade leads to two groups of taxa that have quite
different values for morphological traits between those two groups, and more
similar trait values within each group, for both of the traits under investigation.
When a linear regression model or correlation test is used, this will result in
a strong relationship being inferred, but effectively a line is being fit to two
points – the two groups within the clade, and no such strong relationship in fact
exists. The early split means that species in the two parts of the tree have been
evolving separately from one another for a long time, and so have had a long time
to accumulate differences between them; species in the same part of the tree are
more similar to each other because of their recent common ancestor and long
shared evolutionary history. Regression analysis assumes that individual data
points are statistically independent from one another; this assumption is violated
by species data because of the shared evolutionary history.
PIC analysis is an intuitive approach to this problem. It takes the average trait
value of sister clades and weights this by the amount of time since their most
recent common ancestor (i.e. the amount of time they have been evolving
separately). Most researchers now use phylogenetic generalized least squares
(PGLS) instead of PIC because it is a more flexible likelihood-based model
framework (PIC can be shown to be a special case of PGLS; see Garland and
Ives 2000; Blomberg et al. 2012). After a nonphylogenetic ordinary least-
8 Paleontology
Figure 3 How species nonindependence may influence trait correlations and how
phylogenetic independent contrasts address the issue. (a) Raw values for two
traits (X and Y) plotted alongside the underlying phylogenetic relationships
among species. Two subclades are identified and labeled by open and filled
circles, corresponding to species with smaller (open) vs. larger (filled) trait values.
The dashed line represents a best-fit linear model using ordinary least-squares
(OLS) regression. (b) Phylogenetic independent contrasts estimated as the
standardized difference between species trait values for each internal node in the
tree. Note that the y-intercept goes through the origin, and the difference in slope.
The contrast with the largest absolute value (gray circle) represents the
evolutionary shift in trait values between the two subclades. Adapted from Nunn
and Barton (2001).
squares (OLS) regression is performed on species data, there will be

a phylogenetic signal that dictates how far each datapoint is from the regression
line (its residual). We say therefore that there is phylogenetic structure in the
residuals of the regression. PGLS incorporates information from the phylogeny
into the regression model to adjust the regression line so that the residuals are
normally distributed, rather than structured, rendering the analysis statistically
valid.
Fitting both an OLS and PGLS regression line can be done with the same
R function, gls. Generalized least-squares regression is a standard approach to
regression that can be used when the datapoints are not independent from one
another. When this is the case, the residuals of a standard regression will have
a structure that is caused by the relationship between the datapoints.
Generalized least-squares regression is a very flexible framework that allows
you to supply a correlation structure for the residuals of the regression that can
be derived from any potential source of nonindependence. For example, here we

are using phylogeny, so we supply the phylogenetic VCV matrix to define the
expected structure in the residuals, but, in another example, because some traits
are known to vary systematically across space (e.g. Wagner and Marcot 2010),
a matrix of expected spatial covariance could also be used.
There has been repeated discussion in the literature about whether
researchers should or should not “correct for phylogeny” in a particular
analysis by using approaches like PIC (Felsenstein 1985; Harvey et al.
1995; Rohlf 2006; Westoby et al. 2016) and a concern about “overcorrect-
ing.” The original formulation and explanatory figure for PIC has led to this
terminology, which in turn may have led to some misunderstandings.
Although it is true that without considering phylogeny, a comparative ana-
lysis of two species variables might be statistically invalid, it might be helpful
to think of PIC (and PGLS) as incorporating the additional useful information
that phylogeny provides into the estimation of the relationships between
traits, rather than as a correction.
In the context of analyses like regression or ANOVA that can be used to
understand trait correlations and adaptation, if there is a phylogenetic signal in
the residuals from a model fit, then the resulting relationship derived is statis-
tically invalid because the assumption of independent datapoints has been
violated. Phylogenetic signal in any of the individual traits under investigation
does not necessarily mean the residuals of the regression will have
a phylogenetic structure, or vice versa. Revell (2010) provides a thorough
explanation of this issue. If in doubt, rather than using the function
corBrownian shown in Example Analysis 1, you can use corPagel to
define the correlation structure. This will jointly estimate lambda, which is
a measure of the phylogenetic signal in the residuals. As lambda gets closer
to 0, the estimated coefficients in PGLS will converge on those estimated using
OLS. Think about what that means in terms of model assumptions between OLS
and PGLS. Just because “nonphylogenetic” methods like OLS do not incorpor-
ate information about evolutionary relationships does not mean they do not
make assumptions about evolutionary change. In fact, the biological assump-
tions underlying OLS for comparative analyses are mathematically and con-
ceptually equivalent to using a phylogeny, and assuming that it is a star
phylogeny (i.e. all branches simultaneously diverge from a single node).
Thus, both OLS and PGLS are based on models. It is important to keep in
mind that all models are wrong, but some are more wrong than others. We
advocate for using model goodness-of-fit tests like the Akaike information
criterion (AIC) and Bayesian information criterion (BIC), or, where possible,
going further and investigating model adequacy (see Section 7).
10 Paleontology
EXAMPLE ANALYSIS 1 – REGRESSION

Required packages: nlme; ape
To load our example data, download it from github.com/daveyfwright/
PCMsForPaleontologists and then follow the instructions in the file
“Data_loading_script.R.” To follow along, use the script in the file
“full_script_PaleoPCM.R,” or to see output without following along, see the
file “example analyses.pdf.” We include some of the output in the first two
example analyses.
Is calyx shape correlated with fan density? To answer this, we could
calculate the Pearson correlation coefficient, or perform an OLS regression.
Always plot the data, in this case Shape and Density, from the
dataframe allTraits:
plot(allTraits$Shape, allTraits$Density,
pch = 19, main = “”,
xlab = “Calyx shape ln(L/W)”,
ylab = “Fan density”)
The standard test of correlation between two continuous variables is the
Pearson correlation coefficient:
cor.test(allTraits$Shape, allTraits$Density)
Run this line of code and you should get the following output, which
indicates shape and fan density are negatively correlated with p=0.02.
data: allTraits$Shape and allTraits$Density

t = −2.4878, df = 63, p-value = 0.01551
alternative hypothesis: true correlation is not
equal to 0
95 percent confidence interval:
−0.50606523 −0.05952433
sample estimates:
cor
−0.2990814
The standard regression is OLS:

ols <- gls(Density ~ Shape,
data = allTraits, method = “ML”)
Look at the output:

summary(ols)
Generalized least squares fit by maximum likelihood
Model: Density ~ Shape
Data: allTraits
AIC BIC logLik
175.9948 182.518 −84.99741
Coefficients:
Value Std.Error t-value p-value
(Intercept) 3.561713 0.2087974 17.058224 0.0000
Shape −0.683057 0.2745674 −2.487756 0.0155
An OLS regression gives a slope of −0.68.

Perform PGLS regression to see if there is a difference:
tip.heights <- diag(vcv(phy=prunedTree))
cor.BM <- corBrownian(phy=prunedTree)
pgls <- gls(Density ~ Shape,
correlation = cor.BM,
weights = varFixed(~tip.heights),
data = allTraits,
method = “ML”)
Note that this model uses two additional arguments that were not needed
for OLS: (1) correlation, which defines the expected structure in the
residuals, in this case based on the phylogeny and a Brownian motion
model of evolution (explained in the next section), calculated using the
function corBrownian; (2) weights, which is a modification to
account for the different root-to-tip distances in our nonultrametric
tree – important for paleontologists. Phylogenies that have extinct taxa
in them and are scaled to time are referred to as “nonultrametric” because
the tips end at different times. This is in contrast to “ultrametric” trees
where all the tips end simultaneously at the present day, as is the case with
trees of extant taxa.
Now that we have fit both OLS and PGLS, we can compare them
using the AIC to see which is a better model for the data. AIC values
can be used to evaluate the statistical fit of two or more candidate
models, where the “best-fit” model is defined by having the minimum
AIC score. To minimize overfitting, AIC provides a compromise
12 Paleontology
between a model’s goodness of fit to the data (i.e. the likelihood) and

its “complexity” (i.e. number of parameters). For example, when
there is low or no phylogenetic signal in the residuals of the regres-
sion, OLS would likely be preferred because it has fewer parameters.
We can check the AIC scores of our data using the output of the gls
function that we stored earlier in the objects ols and pgls.
summary(ols)$AIC
[1] 175.9948
summary(pgls)$AIC
[1] 212.8831
OLS has a lower AIC score, and we can therefore interpret our results as
suggesting that OLS is a better model for the data. This in turn indicates
that the phylogeny (assuming a Brownian motion model of evolution; see
next section) does not explain much of the structure of the residuals of the
regression. For further explanation, see the end of Section 3. It is useful to
report both the OLS and PGLS regressions and AIC scores in
a publication to allow readers to see the relative model fit.
It is possible to execute PGLS assuming a different model of evolution
on the phylogeny, rather than Brownian motion. Such alternative models
are discussed in the next section.
4 Tempo and Mode: Brownian Motion and More

Explicitly modeling morphological character change provides a framework with
which we can quantitatively test hypotheses about macroevolution. Through
models, we can link evolutionary histories of groups to different biotic or abiotic
factors, and link patterns in morphological evolution to concepts like conver-
gence, constraint, evolutionary radiations, and the adaptive landscape. In this
section we will explore the basic models that form the foundation of many PCMs.
The underlying principle here is to use mathematical models that describe how we
expect a trait to change along lineages in a phylogeny, if evolution is operating in
a particular way. The process of “model fitting” is essentially comparing the real
data that we have for a trait at the tips of the phylogeny and adjusting the
parameter values within the model (like the rate of evolution) to get the closest
match possible between our real data and the expected possible trait values.
Usually we are testing to find which model out of a set of candidate models is
the one that is the best fit for our data (although see model adequacy in Section 7).
4.1 Brownian Motion

Brownian motion (sometimes known as a “random walk” model of evolu-
tion) is a model that can be used to describe the expected variance of a trait
along branches of a phylogeny through time. It is called a random walk
because the direction and amount of change in each time step is independent
of the direction and amount of change in the previous time step. The average
amount of change expected in each time step (the step rate, referred to by the
model parameter sigma σ) is constant over time, so the trait variance across
the tips of the phylogeny increases linearly over time. The simplest
Brownian motion model assumes no inherent directionality of trait change
(i.e. in each time step the trait value is equally likely to increase or decrease,
so the expected net change is zero). In addition to the step rate σ, another
component of the model is the phylogenetic trait mean (referred to by the
model parameter theta θ), which is the estimated trait value for the ancestral
node of the phylogeny.
Trait values of lineages wander around the phylogenetic mean through
time. Under Brownian motion, the longer a branch’s duration on the tree,
the greater the possible divergence in trait change from the original value.
However, since the direction of change under Brownian motion is equally
likely to increase or decrease the trait value, the most probable value at any
time is still the starting value. The range of possible trait values that
a lineage could have can be visualized as an expanding cone through
time; for Brownian motion the cone expands linearly and the center of the
cone isn’t expected to move through time (see figure 3.1 in Harmon 2019).
When thinking about a whole clade, this pattern of change along lineages
leads to a strong expectation – normally distributed trait values at the tips,
with the standard deviation of that distribution dependent on the rate of
evolution (the Brownian motion step rate σ). All else being equal, the
morphological disparity of a clade will increase as a cone through time
under Brownian motion (see figure 1 in Wesley-Hunt 2005; Erwin 2007).
The less time that has elapsed since two branches split from a common
ancestral branch, the more similar they are expected to be, because there has
been less time for them to accumulate individual change and difference
between them. We emphasize that the name “random walk,” which is
commonly used for this model, could cause confusion. On macroevolution-
ary timescales Brownian motion is unlikely to be random in the sense that it
would result entirely from genetic drift (the true “random” microevolution-
ary process). Instead, the change over time is most likely as a result of the
combination of microevolutionary forces, including selection, acting on the
14 Paleontology
trait along with other mechanisms (e.g. drift, mutation, migration, etc.; see
Hansen and Martins 1996). Importantly, the strength or direction of selec-
tion on a lineage might fluctuate over time as a result of different causes, for
example adaptation to a stochastically varying environment, resulting in
trait values sometimes increasing and sometimes decreasing, and therefore
producing a statistically random pattern. Many underlying microevolution-
ary mechanisms are consistent with the macroevolutionary model for
Brownian motion.
Despite the many-to-one mapping between microevolutionary processes
and macroevolutionary outcomes, Brownian motion is almost always used
in macroevolutionary model fitting because it can be thought of as a neutral
model for trait evolution over long timescales. Moreover, because the
equations underlying more complex models of trait evolution (e.g. Early
Burst, Ornstein-Uhlenbeck; see subsection 4.2) often reduce to Brownian
motion when their additional parameters approach zero, it is used as a kind
of null model. It is often a useful starting point to compare the actual pattern
of trait evolution to the simple Brownian motion expectation. If the best-fit
model is Brownian motion, then no additional events or mechanisms (like
a change in the rate of evolution or long-term constraints on trait change) are
required to explain the pattern of macroevolution over time across the whole
phylogeny. For other helpful explanations of Brownian motion on phyloge-
nies see Nunn (2011) or Harmon (2019). See also the figures and equations
in Butler and King (2004) for a clear, in-depth explanation of this and other
models.
4.2 Ancestral State Reconstruction

A commonly used analysis in paleobiological studies is ancestral state recon-
struction, for both continuous and discrete traits. This approach is a way of
averaging trait values at tips, to make an estimate of what the trait value at
a particular node might have been. Generally these estimates are very imprecise,
and this large uncertainty around estimates means that as a standalone approach,
especially for continuous traits, it is unlikely to provide many useful insights. To
visualize this imprecision from one candidate ancestral value reconstruction
approach, see the function plotTraitgram in paleotree. Furthermore,
these approaches rely strongly on using a model of evolution to make the
estimates; usually Brownian motion is assumed. It is possible to use other
models (for example, Ornstein-Uhlenbeck; see next subsection) by weighting
the average values according to the expectations of the model, but rarely are the
adequacy of those models for the data assessed in advance. If the model
assumed is not in fact a very good model for the data, then the ancestral state
estimates will also not be accurate. For discrete traits the situation is a little
better because possible states are constrained (see Section 8 for more detail). It
is especially problematic to perform ancestral state reconstruction using
a phylogeny of living taxa, without fossils; directional trait change through
time or heterogeneity in tempo and mode across lineages will be missed. This
will lead to inaccurate estimates of ancestral values (Finarelli and Flynn 2006).
In other parts of this Element, we refer to estimating “sampled ancestors”;
this is unrelated to ancestral state reconstruction. Sampled ancestors are actual
fossil specimens from which we can take measurements, and through probabil-
istic methods it is possible to estimate which of these is likely to be from
a lineage whose descendants are also represented by at least one other actual
fossil specimen in the dataset (Bapst 2013b, 2013a; Bapst and Hopkins 2017;
Gavryushkina et al. 2017; Wright and Toom 2017). This is important because it
is likely that many paleontological datasets contain at least one ancestor–
descendant pair (Foote 1996), and ignoring this can distort the phylogenetic
topology, branch lengths, and outcome of downstream analyses (Bapst 2014a;
Soul and Friedman 2017). Sampled ancestors provide an actual trait value at
a particular point in time along a lineage. In contrast, reconstructed ancestral
states provide an imprecise estimate of a trait value for a hypothetical
unsampled population from which two sister lineages descended.
EXAMPLE ANALYSIS 2 – FITTING A BROWNIAN MOTION MODEL
Required packages: phytools; geiger

A useful starting point in any phylogenetic investigation of morphology is
to visualize the tree and trait plotted as a traitgram in “phylomorphospace.”
This means that the y-axis (see Figure 4) is meaningful, rather than when we
usually plot trees where the branching order and position of the tips through
time are important, but the position of the tips across the width of the tree is
arbitrary. This can be done using the function phenogram from the package
phytools. Traitgram and phenogram are words that are used interchange-
ably in the literature (Ackerly 2009). Positions of the tips on the y-axis are the
measured trait values for taxa; the positions of the nodes are estimated
assuming Brownian motion and therefore should not be considered rigorous
estimates of ancestral trait values. At this stage they are just for visualization
(see subsection 4.2). The function requires that the tree and trait have match-
ing taxa and the trait vector has names corresponding to the tree tip labels (i.e.
taxon names).
16 Paleontology
Figure 4 Phenograms, or traitgrams, help visualize patterns of trait

evolution by projecting a phylogenetic tree into “trait space” (y-axis) and
time (x-axis). The upper and lower diagrams were constructed using
a model of Brownian motion to project the eucladid maximum clade
credibility tree into the trait space of filtration fan density (upper) and
calyx shape (lower), respectively.
Traitgram of Density
phenogram(prunedTree, Density, ftype = “off”)
The argument ftype controls how the taxon names are displayed.
Traitgram of Shape
phenogram(tree, Shape, ftype = “off”)
Now we will fit Brownian motion models to both of the continuous traits

using the function fitContinuous from the geiger package.
densityBM <- fitContinuous(phy = prunedTree,
dat = Density,
model = “BM”,
meserr = 0.12)
shapeBM <- fitContinuous(phy = tree,

dat = Shape,
model = “BM”,
meserr = 0.03)
We can then look at the outputs densityBM and shapeBM, which show all
the estimated parameter values like the step rate (the output “sigsq”) and
the phylogenetic mean (the output “z0”).
densityBM
GEIGER-fitted comparative model of continuous data

fitted ‘BM’ model parameters:
sigsq = 0.012168
z0 = 3.701690
model summary:
log-likelihood = −66.448830
AIC = 136.897659
AICc = 137.091208
free parameters = 2
shapeBM
GEIGER-fitted comparative model of continuous data

fitted ‘BM’ model parameters:
sigsq = 0.003723
z0 = −0.206883
model summary:
log-likelihood = −33.232143
AIC = 70.464285
AICc = 70.616184
free parameters = 2
4.3 Beyond Brownian Motion

It is reasonable to think that a simple Brownian motion model of evolution
might not be an adequate description of patterns of morphological vari-
ation over millions of years. As paleontologists we can think of many
examples when patterns in the fossil record suggest periods of constraint,
18 Paleontology
or rapid increases in morphological disparity. Fortunately, there is

a straightforward framework that extends the Brownian motion model
and can be used to model these kinds of patterns. The most commonly
implemented alternative model is one that can be used to describe stabil-
izing selection – the Ornstein-Uhlenbeck model (Hansen 1997; Butler and
King 2004; Hunt and Carrano 2010; Hunt 2012). This model includes
three parameters. The first two are the step rate and the trait optimum
identical to a Brownian motion model, but there is an additional param-
eter – alpha α. There is still a random walk component of trait change
(included within the step rate), but each change is biased toward getting
closer to an optimum trait value, and the strength of that bias or attraction
toward the trait optimum is α. This is sometimes referred to as the elastic
band parameter, because the further a trait value is from the optimum, the
higher the strength of attraction to the optimum. The parameter α can be
converted to phylogenetic half-life, which might be more straightforward
to interpret. This is the time taken for a trait to evolve halfway to the new
optimum and is calculated using ln(2)/α. The original application of
Ornstein-Uhlenbeck in evolutionary biology was to paleontological data,
highlighting the historic importance of paleobiology in establishing quan-
titative approaches to understanding macroevolution. It was applied to try
to understand whether there had been an adaptive optimum in the evolution
of tooth crown height in extinct equids (Lande 1976).
When modeling evolutionary change in a single trait along a lineage, the
trait optimum might really represent an optimum value of high fitness for the
organism (Hunt and Carrano 2010; Hunt 2012). Although this model was
originally developed to describe stabilizing selection, when we are fitting an
Ornstein-Uhlenbeck model to the evolution of a whole clade, the trait “opti-
mum” likely represents some long-term average of trait optima that have
varied through time. A useful metaphor for this could be a peak on
a macroevolutionary landscape (see e.g. Boucher et al. 2018). The simple
version of Ornstein-Uhlenbeck described here is sometimes referred to as
a “single-peak” model, but further methods for modeling more complex
macroevolutionary landscapes and/or shifts between peaks are available (dis-
cussed in subsection 9.1). It is also possible to include an extra parameter,
which is the starting value for the trait at the root of the tree, which may be
different to the trait optimum.
Brownian motion is a special case of the Ornstein-Uhlenbeck model where
α=0. Unlike for Brownian motion, for Ornstein-Uhlenbeck if α6¼0, the vari-
ance does not increase linearly with time. After a particular time (related to α),
the variance remains the same (see also figure 1 in Slater 2013). Traits
evolving under Ornstein-Uhlenbeck are therefore less tightly linked to the

structure of the phylogeny than traits evolving under Brownian motion. In
other words, they will have less phylogenetic signal, and on average each tip
value observed will be less similar to its nearest relatives than it would be
expected to be if evolving under Brownian motion. An additional observation
about Ornstein-Uhlenbeck is that if you have two sister tips, A and B, if the
durations of the terminal branches leading to them are different (as for non-
ultrametric trees), then the covariance of A with B will not be the same as the
covariance of B with A, because covariance declines (i.e. phylogenetic signal
is lost) with time since divergence.
If thought about further, it is possible to see how distinguishing one generat-
ing process from another might be difficult; a trait evolving under Brownian
motion that is highly labile and therefore does not have a strong phylogenetic
signal could show a similar trait distribution to the expectation under Ornstein-
Uhlenbeck, as could a trait evolving under Brownian motion fit to a phylogeny
that has errors in the relationships between taxa (for further explanation see
Section 5). It is therefore important when fitting Ornstein-Uhlenbeck and other
more complex models and interpreting the results to consider whether the
quality of the input data is adequate to distinguish between these different
scenarios. Despite these caveats, when extinct taxa are included in phylogenies
that have been extensively researched to produce well-established phylogenetic
relationships, the differences between expectations under different models are
large enough that it should be possible to interpret model-fitting results with
some confidence (Soul and Friedman 2015). An important advantage of includ-
ing extinct taxa in a phylogeny is that it constrains possible trait change through
time in ways that make it possible to distinguish between different models with
confidence (Ho and Ané 2014b).
Aside from Ornstein-Uhlenbeck, examples of other models you may have
heard of are “Early Burst” and “ACDC” (Blomberg et al. 2003; Harmon et al.
2010). These are models where the step rate σ changes exponentially through
time. They are usually fit to data to try and understand the nature of adaptive
radiations, for which theory suggests that you should see a high rate of morpho-
logical evolution early in the clade’s history, or following a dispersal or extinc-
tion event, as morphological disparity accumulates through filling of new
niches. The high early rate is followed by a period of stability characterized
by a lower rate of morphological change (Slater 2013).
Another example of an alternative model is Brownian motion with
a directional trend. This model is otherwise identical to Brownian motion
except that each step change is consistently biased in one direction. This
doesn’t mean the change is only in that direction, but that there is a higher
20 Paleontology
probability of a change one way than the other. It is not possible to detect
evolutionary patterns consistent with directional Brownian motion with a tree
of only living taxa; this model can only be applied to a tree that contains fossil
data (Slater et al. 2012).
5 Incorporating Estimates of Error

5.1 Error in the Phylogeny
If there are errors in the tree topology (relative to actual evolutionary relation-
ships), this will bias the analysis toward support for an Ornstein-Uhlenbeck
model (Cooper et al. 2016), and in fact will generally bias support toward
models with a low phylogenetic signal. This is because topological error
reduces the apparent phylogenetic signal, making it appear as if trait change
is less tightly linked to the structure of the phylogeny than the Brownian
motion expectation. This effect cannot be corrected for within the analysis and
is therefore something to be aware of when interpreting model-fitting outputs,
especially for trees with fossil taxa.
5.2 Error in the Trait Data

The standard error of the sample and error in the trait measurements can also
introduce bias. Noise from error can increase the apparent trait variance,
leading to incorrect parameter estimates in model-fitting analyses, and a bias
toward models with a lower phylogenetic signal, as is the case with error in
the phylogenetic relationships. Often in paleontology we are only collecting
measurements from a small number of specimens representing each taxon,
which means that the standard error could be large. Silvestro et al. (2015)
show with simulation that incorporating an estimate of error should be
a priority, and that not doing so can bias toward Ornstein-Uhlenbeck as
the best-fit model even when trait change was simulated under Brownian
motion. Including an estimate of the standard error in any model-fitting
analysis is therefore important (for a more thorough discussion of this in
a paleontological context see Hunt and Carrano 2010). There are implemen-
tations of PCMs in R that allow inclusion of error estimates that make use of
a variety of methods (e.g. Revell 2012). In Example Analysis 3 we have
incorporated an estimate of standard error using the argument SE in the
fitContinuous function.
6 Variation Across Trees

In the previous example analyses, each method is only applied to one tree
topology. However, as mentioned in Section 1, it is extremely improbable
EXAMPLE ANALYSIS 3 – SIMULATION AND FITTING MODELS OF CONTINUOUS

CHARACTER CHANGE
Required packages: ape; phytools; geiger

Simulations are a very useful tool to understand the behavior of different
models and analyses (Barido-Sottani, Saupe et al. 2020). We can use
simulation to get a better sense of what trait evolution on a phylogeny
might look like under different macroevolutionary models.
The function rtree in ape is a very basic tree generation tool that
gives nonultrametric trees with a particular number of tips, but doesn’t use
a divergence rate. For more sophisticated true simulation tools that can be
used to generate trees under particular diversification or preservation
scenarios, look at the packages FossilSim and paleotree. Other
modeling packages (like OUwie) usually include tools that can be used to
simulate characters on a tree under the particular models that the package
implements.
Generate a tree with fifty tips

simtree <- rtree(50)
Simulate a trait evolving under a Brownian motion model of evolution on
that tree and plot it using the function bmPlot from phytools
Figure 5 A single simulation of the changing values of one trait along the
lineages of a phylogeny.
22 Paleontology
bmsim <- bmPlot(tree = simtree,

type=”BM”,
anc=0,
sig2=0.01)
This function allows you to visualize a single character and how it

changes through time along branches of the tree, in discrete time steps,
as the clade diversifies (see Figure 5). Try changing the step rate using
the argument sig2 to see how that affects the trait distribution through
time.
The output of bmPlot contains all the information needed to make the
visualization; the simulated trait values at the tips, the tree, and all the trait
values at each time step. We can also use this information to plot the trait
values at the tips in a histogram. This matches the data we would have in
reality and would be modeling. We would not be able to see the full trait
history as in the visualization we just made, as we rarely know how a trait
value has changed through time in all the lineages that make up a clade
(although there are many studies of trait change through time in single
lineages).
Plot a histogram of the trait values at the tips of the tree

hist(bmsim$x[1:50],
xlab = “Trait values”,
main = “Simulated values at tips”)
Other packages can also be used to simulate continuous trait change. The
functions below simulate tip values for Brownian motion and Ornstein-
Uhlenbeck using the ape package. Try changing the model parameters
sigma and alpha (the step rate and constraint parameters). Try making
the optimum value theta different from the starting trait value at the root.
Use phenogram to visualize what effect these changes have, and to become
more familiar with the kinds of patterns that will be best fit by the various
different models.
BM <- rTraitCont(phy = simtree,
model = “BM”,
sigma = 0.1)
phenogram(tree = simtree,
x=BM,
ftype = “off”)
OU <- rTraitCont(phy = simtree,

model = “OU”,
sigma = 0.1,
alpha = 0.1,
theta = 0,
root.value = 0)
phenogram(tree = simtree,
x=OU,
ftype = “off”)
Below are scripts that can be used to fit models to real data and understand
the output. The argument SE is used to input an estimate of the standard
error (see Section 5 for further explanation).
First we will fit Ornstein-Uhlenbeck models to each trait using
a function from the package geiger:
densityOU <- fitContinuous(phy = prunedTree, dat =
Density, model = “ OU”, SE = 0.12)
shapeOU <- fitContinuous(phy = tree, dat = Shape,
model = “OU”, SE = 0.03)
Then we will fit an Early Burst model:
densityEB <- fitContinuous(phy = prunedTree, dat =
Density, model = “EB”, SE = 0.12)
shapeEB <- fitContinuous(phy = tree, dat = Shape,
model = “EB”, SE = 0.03)
Look at the parameter values for each model fit. Are they biologically
reasonable? Checking parameter estimates is an important step, particularly
because for Ornstein-Uhlenbeck models the different parameter estimates
can interact with each other (Hunt 2012). Check that z0, the starting trait
value, is within the range of tip values, and when investigating small changes
on million-year timescales both alpha and sigsq should be small, on the
order of 0.01 or less.
When you fit an Ornstein-Uhlenbeck model using geiger it should
output a warning about using the VCV method. For our purposes, this
warning can be safely ignored but deserves further comment. In the original
version of this function, to fit Ornstein-Uhlenbeck quickly, the algorithm
rescaled the branch lengths rather than operating on the VCV matrix directly
(Slater 2013, 2014). This was problematic for nonultrametric trees because
this particular kind of rescaling changed the covariance between tips in a way
24 Paleontology
that assumed the input tree was ultrametric. The current version of this and
other functions modify the VCV matrix directly to avoid the issue. Almost all
PCMs (especially commonly used ones) can be applied to nonultrametric
trees using R implementations, but this is a good reminder that it is important
to check that none of the mechanics of the implementation are problematic
when applied to paleontological trees, particularly with newly released
method implementations.
Once the parameter estimates for the various models have been checked
to make sure they are reasonable, the next step is to look at the AICcs
(Akaike information criterion for small sample sizes) to find out which
model is the best fit for the data and tree. The AICc is part of the output of
the fitContinuous function that we used earlier, so it can be accessed
using densityBM$opt$aicc.
These results indicate that Brownian motion is the best-fit model for
Density but has a marginal AICc difference. Ornstein-Uhlenbeck is the
best-fit model for Shape, with a larger AICc difference. Although most
of the parameter estimates are biologically reasonable and broadly similar
across different models, the exponential decline parameter in both Early
Burst model fits is extremely small. In fact, it is at the lower boundary
limits set by geiger. The maximum-likelihood estimate for this param-
eter is effectively zero for both traits and therefore the Early Burst model
reduces to simple Brownian motion. However, note that the cost of this
additional, unnecessary parameter leads to a decrease in model support.
that a single phylogeny (e.g. the maximum a posteriori tree from a Bayesian
analysis, or a single most parsimonious tree) perfectly represents evolution-
ary “truth.” Thus, it is important to get some measure of the robustness of
results to variation in topology and branch lengths. We strongly recommend
accounting for variation across phylogenetic hypothesis regardless of the
methodological approach used to infer the tree topology and branch lengths;
using a point estimate of the tree is never recommended. A simple way to
understand how sensitive a particular result is to variation across trees is to
run the analysis over a set of different trees. In Example Analysis 4 we
randomly select 100 trees from the posterior distribution output by a joint
estimation analysis. Other commonly used sizes for tree sets are 500 and
1,000, but really it is arbitrary and at the researcher’s discretion as to how
large a set of trees they need to include in order to fully characterize the
variation in results.
EXAMPLE ANALYSIS 4 – APPLICATION TO MORE THAN ONE TREE

Required packages: ape; geiger
Here, we load the post-burnin posterior distribution of 1,000 trees and select
a subsample of 100 of them, then make sure they all have a root age and no
zero-length branches as described in subsection 2.1. Multiple trees are stored
together in an object of the class “multiphylo,” which can be treated as
a list of trees. There are faster ways to run analyses over many trees (for
example, using apply functions, or functions from the package dplyr, both
of which we recommend), but here we use for loops so you can see clearly
what is being done.
Load the post-burnin posterior distribution of trees:

trees <- read.tree(”Eucladid_trees.tre”)
Select 100 trees:

treeset <- trees[sample(1:1000, 100, replace = FALSE)]
The approximate age of the youngest taxon in the dataset is 268.8 Ma:
last.time <- 268.8
Adjust the root time; because all taxa are extinct, the for loop means that the
root time for every individual tree in the tree set gets adjusted in the same way:
for (i in 1:length(treeset)){
treeset[[i]]$root.time <- max(diag(vcv(treeset
[[i]]))) + last.time
}
Any of the previous analyses can be run over the entire tree set. For example,
here we fit a Brownian motion model to all 100 trees using a for loop, to
characterize variation in results.
Set up a vector to store results:

sigmas <- vector(length=100)
Run the model-fitting analysis on all 100 trees and store the evolutionary rate
for each tree in the vector:
for(i in 1:100) {
currentTree <- treeset[[i]]
fitBM <- fitContinuous(phy = currentTree,
dat = Shape,
model = “BM”,
SE = 0.03)
26 Paleontology
sigmas[i] <- fitBM$opt$sigsq

}
In this example we did not save the full output from the model fits, but rather
just the parameter estimate for the step rate in each model fit. Decide
beforehand what part of the results you need to access, and store them in
vectors if they are single integers or lists if you want to store a more
complex output.
Finally, view a histogram of all the estimates for the evolutionary step
rate across the different tree topologies that we saved to the vector:
hist(sigmas)
If the phylogeny was not generated with a Bayesian inference approach that
produces a posterior distribution of trees, other methods are available to account
for uncertainties in tree topology or node ages. For example, if phylogenetic
relationships were inferred using parsimony methods, one could select
a random subset of the most parsimonious trees generated by the analysis
(Lloyd et al. 2012). Similarly, you could choose to investigate the sensitivity
of your results across a set of timescaled trees from the output of a stochastic
timescaling algorithm (e.g. Bapst 2013a) to understand the effect of branch-
length variation on your analysis.
Once analyses have been run over an entire distribution of trees, the results
should be summarized appropriately. Unfortunately, no universally agreed-upon
way to do this has been developed. Nevertheless, many reasonable approaches are
possible for visualizing and summarizing the results. For example, displaying
median values with the variation around them (as long as it is made clear that this
variation is separate from the error or confidence intervals of the results them-
selves), showing model support across trees in a bar plot (figure 6 or figure 2 in
Halliday and Goswami 2016), or showing a representative result and discussing
how robust it is have all been used in the literature and are improvements over
reporting a point estimate alone (for more examples see Hopkins and Smith 2015;
Clarke et al. 2016; Soul and Friedman 2017). Where appropriate for the analysis,
we suggest reporting the distribution of p-values across a sample of trees in the
form of its frequency distribution and summary statistics. For example, some
authors have summarized results of phylogenetic statistical tests in terms of either
the mean p-value estimated across the distribution of trees and/or the proportion
of samples where the test statistic exceeds some a priori threshold of statistical
significance (Soul and Benson 2017; Cole et al. 2019). A subset of PCM
implementations do have built-in ways of summarizing across sets of trees (see
for example rjmcmc in geiger), but this is rare. The most important thing is to
choose an approach that makes the variation in results clear, as reporting only the
average is misleading unless there is truly no variation across trees. Note that if
there is a large amount of variation in relationships across topologies (i.e. the
phylogeny is poorly constrained), then in model-fitting exercises there is likely to
be a bias toward support for models with a low phylogenetic signal, for the
reasons discussed in section 4.3. The combination of variation across topologies
and support for a low-signal model should lead to skepticism about whether there
is enough information in the data to reliably distinguish between models.
7 Model Fit and Model Adequacy

The previous Example Analyses 1–4 involved fitting models of evolution to
data by estimating the maximum-likelihood parameter values for a particular
model given the data, then comparing the AIC scores of each model fit to see
which was the best-fit model, and what that might tell us about evolution of the
clade in question. However, just because one model is a better fit than the other
models that were tested, it does not necessarily mean that it is actually a good
model for the data (Pennell et al. 2015; Voje et al. 2018). No model can
encompass all the different processes that lead to the actual pattern of trait
change, but the important thing to consider is whether a particular model
captures the evolutionary dynamics to the extent that it allows you to answer
the question you are interested in addressing. In this section we explore how to
approach testing model adequacy. There is not a single test that can say
definitively, “this model is good enough,” but there are ways to understand
how much information is in a dataset, how reliable the outcome of a model-
fitting exercise is, or how confident we should be in the answer to an evolution-
ary question derived from PCMs. Often, finding the ways in which a model does
not capture aspects of the data (i.e. model misspecification) can provide extra
information about evolutionary processes.
There are a variety of reasons that simple models of trait evolution applied
across a whole clade might not be good models for the underlying dynamics
(see e.g. Polly 2019). Furthermore, it may be the case that there is simply not
enough information in your dataset to allow you to identify an adequate
model, especially if the model you are interested in assessing produces
similar patterns of trait evolution to other candidate models (Boettiger et al.
2012; Slater and Pennell 2014). With simple models like linear regression,
taking steps such as plotting the data, applying goodness-of-fit tests, or
inspecting the residuals can help establish whether the model is adequate.
For more complex approaches (i.e. most PCMs) visualization alone does not
28 Paleontology
often help (Revell et al. 2018). There are no well-established model adequacy
testing approaches for PCMs on nonultrametric (i.e. fossil) trees. Even for
analyses on trees of living taxa, model adequacy testing is still relatively
uncommon, despite growing recognition of its importance within the PCM
literature.
Throughout this Element we mention the possibility of the dataset not
containing enough information to make reliable inferences based on model-
fitting outcomes. Simulation is one approach that can be used to test whether or
not this is the case. For example, posterior predictive simulation can be used to
compare the relative adequacy of two candidate models and allow greater power
to be able to distinguish between them (see Slater and Pennell 2014). A more
general simulation framework can be used to investigate the power to reliably
identify a particular model. We suggest the following steps as a straightforward
addition to model-fit comparisons, which are made possible by the wide range
of simulation tools now available in R:
(1) Find the best-fit model (or models if they have similar goodness of fit).
(2) Identify the fitted model parameters.
(3) Use these parameters to repeatedly simulate trait change under the best-fit
model(s) on the original phylogeny.
(4) Perform model fitting and parameter estimation on each of the simulated
datasets.
(5) Record how frequently the model that was used for simulation is correctly
identified.
If the model that was used for simulation is not often identified as the best-fit
model, or the fitted parameter values vary widely across simulation runs, then
this suggests there is not enough information in the dataset to reliably inform
inference, and the original outcome should be interpreted cautiously.
For a subset of phylogenetic comparative approaches it is possible to use
standardized phylogenetic contrasts to investigate model adequacy and identify
types of model misspecification (Pennell et al. 2015). One statistical condition
of the phylogenetic contrasts of observations of a continuous trait on
a phylogeny, under some models, is that they are independent and identically
distributed (IID). This is true for nonultrametric trees when a Brownian motion
model of evolution is assumed, as well as for many related Gaussian models,
even where the mode or rate of evolution changes across the tree, but it is not
true for any kind of Ornstein-Uhlenbeck model on a nonultrametric tree
(although it is possible to perform a scalar transformation of the branch lengths
that results in IID contrasts; see Ho and Ané 2014a). If the contrasts are IID,
then various test statistics on the contrasts can be used. The procedure (outlined
in detail in Pennell et al. 2015; see table 1 of that paper in particular for details of
the types of model misspecification that can be identified with different test
statistics) is similar to that used earlier, except that the test statistic of interest is
applied to the contrasts of the real data and the contrasts of each of the simulated
datasets. If the test statistic value for real data falls outside of the distribution of
the test statistic across the simulated datasets, then there is some degree of
misspecification, or inadequacy, of the model. For example, if the mean of the
squared contrasts is not within the simulated distribution, this indicates that the
overall rate of evolution is over- or underestimated.
8 Post-Hoc Modeling of Discrete Characters

The majority of models we have discussed thus far are appropriate only for
continuously valued traits. However, discrete morphological characters have
been particularly important in paleontology because they comprise the bulk of
primary data for inferring phylogenies of fossil taxa, including both parsimony
and model-based approaches (see Wiley and Lieberman 2011; Wright 2019;
Warnock and Wright 2020) and investigating patterns of morphological dispar-
ity (Lloyd 2016). In addition to intrinsically discrete traits (e.g. presence/
absence), many other kinds of morphological, paleoecological, and taphonomic
traits are often categorized into discrete character states for practical reasons
(e.g. species occurrences in reef vs. non-reef settings).
Table 1 Model-fitting results for analyses of trait evolution in eucladid crinoids

described in Example Analysis 3
Trait Model logL AICc Parameter estimates

Filtration fan BM −66.449 137.09 σ2 = 0.012, z0 = 3.70
density OU −66.250 138.89 α = 0.002, σ2 = 0.013, z0 =
3.71
EB −66.449 139.29 a = −0.000001, σ2 = 0.012, z0
= 3.70
Calyx shape BM −33.209 70.57 σ2 = 0.0037, z0 = −0.207
OU −29.337 64.98 α = 0.0094, σ2 = 0.0050, z0 =
−0.333
EB −33.209 72.73 a = −0.000001, σ2 = 0.0037,
z0 = −0.207
Models with the best fit according to AICc scores (Akaike information criterion for small
sample sizes) are indicated in bold. BM = Brownian motion; OU = Ornstein-Uhlenbeck;
EB = Early Burst.
30 Paleontology
EXAMPLE ANALYSIS 5 – MODEL ADEQUACY

Required packages: ape; OUwie
In Example Analysis 3 we found that Ornstein-Uhlenbeck was the
best-fit model out of those that we tested for the Shape data. Here we
examine whether that Ornstein-Uhlenbeck model was actually a good
model, and reliably identifiable from the data. To do this we will need to
simulate under an Ornstein-Uhlenbeck model of evolution; the package
OUwie has a function that does this.
First, extract the parameter estimates from the previous model-fitting
output:
alpha <- shapeOU$opt$alpha
sigma.sq <- shapeOU$opt$sigsq
theta <- shapeOU$opt$z0
Prepare the data so that they are in the format that can be read by OUwie
(see the OUwie help file for further details).
shapeData <- data.frame(species = names(Shape),
regime = 1,
shape = Shape)
Use the function replicate to repeatedly simulate under Ornstein-

Uhlenbeck with the same parameters and store the simulated trait values
for the tips of the tree. The function replicate is a convenient
shortcut; it does the same thing as a for loop but is faster when you
want to repeat the same operation many times. The simulated tip values
are output in the variable X by the function, so this is the element we
need to store.
simtrait <- replicate(100,
OUwie.sim(phy = tree,
data = shapeData[,1:2],
root.age = tree$root.time,
alpha = c(alpha,alpha),
sigma.sq = c(sigma.sq,sigma.sq),
theta = c(theta,theta),
theta0 = theta
)
$X)
Set up the simulated trait values into vectors with names that can be read by
the model-fitting function.
form.simtrait = list()
for(i in 1:100) {form.simtrait[[i]] <- simtrait[,i]
names(form.simtrait[[i]]) <- names
(Shape)
}
Fit the single-peak Ornstein-Uhlenbeck model, which was the best-fit model
for our data before out of the models we tested (warning: if you do this over
100 iterations, it will take a long time).
OUfits <- lapply(X = form.simtrait,
FUN = fitContinuous,
phy = tree,
model = “OU”)
Fit a single-rate Brownian motion model to the same sets of simulated trait
data:
BMfits <- lapply(X = form.simtrait,
FUN = fitContinuous,
phy = tree,
model = “BM”)
If you run this analysis using the eucladid tree, you should get several
warning messages during the Ornstein-Uhlenbeck model fitting to simulated
data, a first indication that the resulting model fit is unlikely to be reliable. If
you inspect the output model-fit AICcs and estimated parameters, you should
find that Ornstein-Uhlenbeck is the preferred model over Brownian motion,
but that the estimated parameters for Ornstein-Uhlenbeck are not similar to
the parameters used for simulation, and are quite high. This suggests that the
best-fit Ornstein-Uhlenbeck model is unlikely to be adequately capturing
variation in the trait data. A next step might be to investigate whether a single-
peak Ornstein-Uhlenbeck model is missing heterogeneity in rate or mode of
evolution (see Section 9).
In practical terms, if the discrete trait of interest is included directly in

a model-based phylogenetic inference analysis, it is possible to jointly estimate
transition rates between states and ancestral states with other parameters in
a Bayesian context (Warnock and Wright 2020). However, it is far more
common to investigate patterns of discrete trait evolution across
a phylogenetic tree (or set of trees) after conducting a phylogenetic inference
32 Paleontology
analysis. Such post-hoc modeling typically involves the estimation of ancestral

states (i.e. the trait values at the nodes of the tree), which can then be used to
characterize the timing and phylogenetic position transitions between states. In
a similar way to the approach we showed earlier for continuous traits, fitting
alternative models of discrete trait evolution in R can be done with functions in
the geiger package.
One of the methods most widely used to estimate ancestral states for
discrete characters is parsimony. Given a tree depicting evolutionary rela-
tionships, parsimony-based reconstructions map a trait onto the tree such that
the number of evolutionary changes is minimized (Maddison and Maddison
2020). Parsimony methods do not incorporate information regarding branch
durations, so changes are equally likely to occur on short and long branches.
Notably, parsimony-based reconstructions cannot always be unambiguously
determined, so multiple parsimonious reconstructions are sometimes pos-
sible, which can lead to equivocal results. Beyond parsimony methods,
modeling discrete characters is most commonly performed using
a variation of the Mk model described by Lewis (2001), which involves
a single parameter: the rate of evolutionary change. In contrast with parsi-
mony, the Mk model involves estimating branch lengths, and allows for the
possibility of multiple changes to occur along a branch, such that the total
number of evolutionary transitions can be greater than the minimum required
under parsimony. In its simplest form, the Mk model assumes changes are
equally likely between character states, and these states have identical equi-
librium frequencies. However, in a Bayesian context these assumptions can
be relaxed to accommodate a wide variety of possible model variations,
including rate variation among characters (Wagner and Marcot 2010;
Wright 2019), unequal transition frequencies between state changes
(Wright et al. 2016), and even whether ecological vs. nonecological trait
partitions evolve at different intrinsic rates (Wright et al. 2020).
Although simulation-based studies show Bayesian methods using the Mk
model often outperform parsimony in phylogenetic inference (Wright and Hillis
2014; O’Reilly et al. 2016; Puttick et al. 2019), both methods perform well
when reconstructing ancestral states given a tree (Gascuel and Steel 2014; see
next paragraph), and the decision to choose parsimony vs. model-based
approaches may depend on your research question. For example, a researcher
might be interested to show that a particular phylogenetic hypothesis requires
some minimal threshold of evolutionary changes; in such cases, parsimony
methods may be preferred because they always prefer the minimum number of
changes. In that sense, parsimony is biased to underestimate the total number of
character changes when the rate of evolution is high; however, the extent to
which a researcher can use that bias to their advantage may be useful for some
types of questions (i.e. “this character changed at least X times across the
phylogeny”). Moreover, sometimes it may be of interest to compare the min-
imum number of changes required between different phylogenetic hypotheses
(regardless of how the trees were inferred). Similarly, parsimony methods (and
related approaches) for discrete character evolution can also be used to study
rate variation and patterns of character evolution (Wagner 2012), and inform
rate distributions in model-based phylogenetic methods (Harrison and Larsson
2015; Wright 2017a).
Unlike reconstructing ancestral states for continuous traits (see subsection
4.2), methods to reconstruct ancestral states for discrete characters are
remarkably robust and provide far more accurate estimates than their con-
tinuous trait counterparts. For example, a simulation study by Gascuel and
Steel (2014) shows both parsimony and model-based methods of discrete
ancestral state reconstruction perform well, have nearly the same accuracy,
and are robust to sampling bias and model misspecification. As discussed for
continuous traits, the fidelity of methods to reconstruct ancestral state values
for discrete characters improves when fossil data are incorporated into the
tree. Puttick (2016) used simulations to examine the impact of including
fossil taxa in analyses of discrete trait evolution in extant lineages, and
showed that including both fossil and extant taxa consistently outperformed
analyses with extant taxa alone. In practice, there could be errors in coding
derived from incomplete or fragmentary fossil material, or perhaps incorrect
assignment of a habitat based on incorrect paleoenvironmental reconstruc-
tions (e.g. freshwater vs. marine). Puttick (2016) therefore also investigated
the impact of including a proportion of fossil taxa having incorrect discrete
trait values and found that analyses of fossil and extant taxa still outper-
formed extant-only analyses at estimating ancestral states, even when up to
75 percent of fossil taxa were incorrectly coded. These results highlight the
important role of fossil taxa for estimating ancestral states for discrete
characters, and encourage the inclusion of as many fossil taxa as possible,
even when there is uncertainty in their underlying traits.
Another popular model-based method for post-hoc modeling of discrete
character evolution is a Bayesian approach called stochastic character map-
ping (Nielsen 2002; Revell 2013). In this method, discrete trait evolution is
simulated an arbitrarily large number of times across a tree, and the resulting
character histories are sampled proportional to their posterior probability,
using Markov-chain Monte Carlo, to generate the posterior distribution.
Although the output can be difficult to visualize for traits with more than
two states (Revell 2013), a benefit of stochastic character mapping is that the
34 Paleontology
EXAMPLE ANALYSIS 6 – MODELS FOR DISCRETE TRAIT CHANGE

Required packages: ape; phytools; geiger
In this example we will fit three models of discrete character
evolution to the MCCT for eucladid crinoids to investigate the evolu-
tion of calyx complexity, an important trait related to growth patterns
and rates of development (Kammer 2008; Wright 2015; Ausich et al.
2020). Calyx complexity is defined as the number of plates interrupt-
ing the posterior interray (more plates = increased “complexity”).
This trait typically varies from zero to three among most Paleozoic
eucladid taxa, and is therefore a multistate character with four discrete
states.
First, let’s fit a model with a single-rate parameter describing all possible
transition rates between character states. For example, if a character has two
discrete states (e.g. number of posterior plates), labeled “1” and “2,” this
single-parameter model assumes the transition rate between states “1” ➔
“2” is equal to the backward transition between “2” ➔ “1.” The same logic
can be extended to multistate characters, such as our example of calyx
complexity. This model is specified as equal rates, or “ER” for the model
argument in the fitDiscrete function.
ER <- fitDiscrete(tree, dat = Complexity,
model = “ER”)
In contrast with the simplistic equal-rates model, we can also specify

a more complex model where each possible rate is a unique parameter. For
example, a different rate is estimated for transitions between states “1” ➔ “2”
and “2” ➔ “1,” etc. In geiger, this model is specified as all rates different, or
“ARD” for the model argument in the fitDiscrete function.
ARD <- fitDiscrete(tree, dat = Complexity,
model = “ARD”)
For discrete traits with more than two states, we might predict that certain
kinds of traits evolve along an order sequence of changes between states.
For example, based on developmental patterns, some crinoid researchers
consider that calyx complexity should be an ordered character; that is,
transitions should be restricted so that they can only occur between
consecutive states (e.g. Wright 2015, 2017). In this case you can use the
meristic model to model transition rates.
MER <- fitDiscrete(tree, dat = Complexity,
model = “meristic”)
As in previous examples, models can be compared using the AICcs.

ER[[4]]$aicc
ARD[[4]]$aicc
MER[[4]]$aicc
Here, we have found that the meristic model corresponding to ordered

state transitions is the best-fit model for the data.
It is also straightforward to reconstruct ancestral states, and the asso-
ciated uncertainty, under an equal-rates model. phytools has useful
functions for plotting the likelihood of a given state onto the tree.
First, estimate ancestral states using the function ace from the ape
package:
anc.char <- ace(Complexity, tree, type = “discrete”)
Ladderize the tree for better visualization and then plot in the usual
way:
plot(ladderize(tree), show.tip.label = FALSE)
Add pie charts of character state likelihoods to the nodes:

nodelabels(pie = anc.char$lik.anc,
piecol = c(”lightblue”, “darkblue”,
“salmon”, “darkred”),
cex = 0.5
)
uncertainty in patterns of character evolution can readily be accounted for by

examining the posterior distribution of character histories. As might be
expected, this method presents a challenge for any attempts to summarize
results across a distribution of time-calibrated trees (see Section 6 for ideas
regarding how to summarize results across tree distributions). Stochastic
character mapping can be conducted in R using the make.simmap function
in the package phytools.
A common feature of models discussed thus far, for both continuous and
discrete traits, is that the underlying dynamics of trait evolution do not
change over time and/or among clades. Relative to the literature on discrete
traits, modeling heterogeneous dynamics of trait evolution is far more devel-
oped for continuous characters, which is more thoroughly discussed in
Section 9. However, an important, promising family of methods for investi-
gating heterogeneous dynamics in discrete character evolution was
36 Paleontology
developed by Lloyd et al. (2012). Using discrete character matrices (such as

those constructed for phylogeny estimation or disparity analyses), these
approaches use likelihood ratio tests to evaluate whether a given phylogeny
is consistent with single-rate vs. multi-rate models, where two (or more)
rates may arise from either temporal shifts in evolutionary dynamics and/or
along specific branches. This method attempts to account for patterns of
character sampling and completeness, and has been applied to a variety of
macroevolutionary questions in a variety of fossil groups, including lungfish
(Lloyd et al. 2012), echinoids (Hopkins and Smith 2015), Mesozoic mam-
mals (Close et al. 2015), and Paleozoic crinoids (Wright 2017b). Relevant
functions for performing these kinds of analyses are available in the
R package Claddis and further described by Lloyd (2016).
9 Modeling Heterogeneous Trait Evolution

In this section we provide descriptions and some introductory scripts for
a variety of more complex, or recently published, approaches. These approaches
deal with modeling continuous trait evolution where the tempo or mode varies
through time or across lineages (for examples of approaches that can be applied
to discrete trait evolution to model heterogeneous change, see Section 8). This is
not an exhaustive review of all the approaches available, but a summary of some
of those that might be of most interest to paleontologists, and references where
they have been applied in a paleobiological context.
When starting out with more complex models it is appropriate to be cautious.
There are several publications that highlight situations where phylogenetic
comparative approaches do not behave as expected or intended (Freckleton
2009; Uyeda et al. 2018). It is very important to ensure that any method applied
is appropriate for the question and data, and that it is applied correctly. This is
likewise true of all the more straightforward analyses we explained earlier, but it
can be more difficult to identify problems as models become more complex.
Like all statistical methods, PCMs have assumptions, biases, and limitations
(Cooper et al. 2016). Because PCMs comprise a diverse ensemble of models,
we cannot provide a comprehensive guide or checklist of assumptions pertain-
ing to particular methods here. Instead, we emphasize it is critical for
researchers to carefully consider the statistical assumptions of a given PCM
(and their biological implications) when designing a study or interpreting
results. We reiterate that this advice is general to all biological sciences and
not an issue specific to PCMs.
In fact, many criticisms of PCM approaches can be addressed by a researcher
(1) ensuring they fully understand the method being implemented, as well as the
data it is applied to, to identify possible biases or features of the data that could
lead to misleading results (e.g. sampling rate; Soul and Friedman 2017); (2)
taking important steps such as checking that output parameter estimates are
biologically reasonable – if in doubt, simulation can be a practical way to try and
characterize the behavior of a particular method when applied to data similar to
your own (Barido-Sottani, Saupe et al. 2020); and (3) not overinterpreting
results. Finally, we note that many potential issues with PCM-based studies
are alleviated or otherwise improved when fossil taxa are included (Slater et al.
2012; Ho and Ané 2014b; Hunt and Slater 2016).
9.1 Multi-Regime Ornstein-Uhlenbeck Models

Ornstein-Uhlenbeck models have three or four parameters that define an “adaptive
peak” on a phenotypic landscape (sensu Simpson 1944). In the context of phylo-
genetic paleobiology, this peak can also be thought of as a macroevolutionary
regime, or macroevolutionary landscape (Boucher et al. 2018). For some period of
time, the evolution of a set of taxa can be characterized using single estimates for
each model parameter, and though the actual mechanisms generating the observed
patterns may vary through time, they have the same outcome. This model is
consistent with Simpson’s concept of the adaptive landscape. Often, we would
not expect that a clade would remain in the same macroevolutionary regime
(i.e. where the same trait value is optimal and the strength of α [the constraint
parameter] remains the same) over the many tens of millions of years of its
evolutionary history, or that all branches of a clade would experience the same
regime. We might also be interested in how the regime a clade experiences relates
to factors like environmental change, or whether multiple nonsister lineages
experienced the same regime, which can result in convergence.
It is possible to model shifts in the evolutionary regime through time, or along
lineages, by estimating new parameter values for different sections of the
phylogeny. The package OUwie has model-fitting and trait simulation functions
for this purpose. OUwie has several options for models; you can allow the
strength of the selection parameter α, the trait optimum θ, or the evolutionary
rate σ to shift. You can also fix or vary different combinations of these for each
regime. The function requires a priori definition of clade or time regime shift
points; it is therefore suitable for hypothesis testing (Gearty and Payne 2020).
When choosing shift points for an analysis, they should be related to specific
hypotheses about factors that may have affected the evolutionary history of the
clade (e.g. a novel adaptation, dispersal to a new region, or environmental
changes). Thus, we emphasize it is critical to have in-depth knowledge of the
taxonomic group under investigation to meaningfully address evolutionary
38 Paleontology
questions with these methods. If you are not an expert on the group you want to
investigate, we encourage you to consider reaching out to collaborate with one.
A variety of other packages can also be used to investigate multi-regime models
(including multivariate multi-regime models) of trait evolution, including
mvMorph, mvSLOUCH, OUCH, geiger, and ape.
Multi-regime Ornstein-Uhlenbeck models can be very complex and are
therefore often computationally difficult to fit. It is especially important to
check for convergence when estimating parameters (Gearty and Payne
2020) and that the parameter values are sensible. For example, for
OUwie, ensure that the results show estimated values for θ (theta_0,
the value at the root of the tree) that are within or close to the known range
of trait values, and that the step rate is not unreasonably high. If it is not
clear what could be considered a reasonable rate, it should usually be
possible to find rate estimates in the literature for the same taxon, or for
similarly preserved taxa.
9.2 Character-Dependent Continuous Trait Change

A further question that can be answered with multi-regime Ornstein-Uhlenbeck
models is how a particular discrete character has influenced the evolution of other
continuous traits. OUwie can be used to test this. If the discrete trait is confined to
particular clades, then the same approach as in Example Analysis 7 can be used to
define regimes. If the discrete trait is distributed over the tree rather than being
confined to particular clades, then you can use any ancestral state reconstruction
approach to reconstruct the state at each node (see Section 8), and then use the
different estimated states to define the regimes on the tree. Uncertainty in the
ancestral states estimated can be appropriately accounted for using the package
SIMMAP. The additional layer of uncertainty derived from estimating the ances-
tral states under a particular model of evolution should be taken into account
when interpreting results. Price (2019) provides a worked example for paleonto-
logical data (see also Price et al. 2015).
9.3 Heterogeneous Rates on Individual Branches

It is possible to model change of one or more continuous traits on a phylogeny
with rate differences between and/or along each branch, and a range of models
has been developed to do so (O’Meara et al. 2006; Hunt 2013; Revell 2013;
Landis and Schraiber 2017). Markov chain Monte Carlo (MCMC) approaches
are computationally well suited for use to fit this kind of highly heterogeneous
model. One example model and Bayesian implementation is AUTEUR
(Eastman et al. 2011), which is implemented with the function rjmcmc.bm
EXAMPLE ANALYSIS 7 – MULTI-REGIME ORNSTEIN-UHLENBECK

Required packages: OUwie
Most of this example script serves to correctly set up the input objects.
First, set up a dataframe and tree in the format that they can be read in. OUwie
has a very specific format required for the input data. It requires a three-
column dataframe where the first column contains the taxon names,
the second contains the regime each tip belongs to (at the start the regime
labels are placeholders for when we define them according to a hypothesis
later), and the third column contains the trait data. You can optionally give
a fourth column with the standard error of the trait measurements (this means
an estimate of the intraspecific error if you have multiple specimens, plus
measurement error; see Section 5 on error) – this is advisable whenever
possible. The tree must have internal node labels corresponding to the regime
the node falls within. At the time of publication it is the case that to prevent the
function outputting an error message, there must be two or more regimes,
even if a single-regime model is subsequently fit.
Here we use the same crinoid data for the Density trait that we have
been using previously. The possible regime shift points defined further
down are based on DFW’s prior knowledge as a crinoid systematist and
taxonomist familiar with their morphology and ontogeny, derived from
previous experience and knowledge of the published literature.
Set up a dataframe for the trait we are using, with three columns as
specified earlier:
densityData <- data.frame(species = names(Density),
regime = 1,
density = Density)
Add node labels to the phylogeny we have been using

prunedTree$node.label <- sample(c(1,2),
Nnode(prunedTree),
replace = TRUE)
Once the data and phylogeny are correctly formatted, we can estimate
parameters for single-rate Brownian motion and Ornstein-Uhlenbeck
models, similarly to how we demonstrated in Example Analysis 3
using the alternative R package geiger. The root age of the tree must
be provided, and it can be helpful to provide starting values for the
parameters to aid convergence. Ideally the algorithm should be started
from a variety of values to make sure that it converges on the same
parameter estimates, and not local optima.
40 Paleontology
By default, the algorithm does not estimate theta_0, the trait value, at
the root as part of the model. Instead, the optimum at the root is dropped,
which can stabilize estimates of the adaptive optimum trait value. If there is
reason to include estimating theta_0, be aware that sometimes estimates
of theta_0 and alpha can interact (Ho and Ané 2014b), so it is
important to check for sensible parameter estimates after the model fit.
First, fit single-regime Brownian motion and Ornstein-Uhlenbeck to
the Density trait to provide a null against which more complex models
can be compared.
densityBM <- OUwie(phy = prunedTree, data = densityData,
model = “BM1”,
quiet = TRUE,
root.age = tree$root.time)
densityOU <- OUwie(phy = prunedTree, data = densityData,
starting.vals = c(0.01,0.01),
quiet = TRUE,
model = “OU1”,
Now we can specify the node labels on the tree according to the hypothe-
sized regimes. There are two subclades within the Eucladida tree that seem
to show many modifications associated with building a dense filtration fan
(e.g. many branches per arm or many pinnules on branches). OUwie can be
used to test whether this high fan density was associated with a different
trait mean or rate of evolution. In this example we will define the tip and
node regimes ourselves beforehand because it involves more than one
clade. If we hypothesize a rate shift in only one clade, it is easiest to use
the “clade” argument within the OUwie function rather than defining
the regimes beforehand.
Find the node defining the first clade of interest:
mrca1 <- getMRCA(prunedTree, c(”Blothrocrinus”,
“Hydreionocrinus”))
Make a vector of all nodes and tips that descend from it:
first <- c(mrca1, getDescendants(prunedTree, mrca1))
Do the same for the second clade:

mrca2 <- getMRCA(prunedTree, c(”Pirasocrinus”,
“Zeacrinites”))
second <- c(mrca2, getDescendants(prunedTree, mrca2))

Combine the two vectors to a single one, specifying all nodes and tips
that we hypothesize belong to the “high-density fan” regime:
desc <- union(first, second)
Separate out node and tip vectors and label the nodes of the phylogeny
with regimes accordingly:
tips <- prunedTree$tip.label[desc[desc<=Ntip
(prunedTree)]]
nodes <- desc[desc>Ntip(prunedTree)]-Ntip
(prunedTree)
treeD <- prunedTree
treeD$node.label <- rep(1, Nnode(treeD))
treeD$node.label[nodes] <- rep(2, length(nodes))
Modify the trait dataframe to specify the correct regime labels for tips:
regimetableD <- densityData
regimetableD[tips,2] <- 2
Now that the data and phylogeny are prepared, we can fit a variety of
multi-regime models to the data. Any combination of model parameters
can be fixed or allowed to change between regimes, using the “model”
argument (refer to the documentation for OUwie for a full explanation of
all the options for this argument and what they mean). Here we did not
include estimating theta_0. Depending on the number of taxa in the
dataset and the variation in trait values, very complex models such as one
where all parameter values change (“OUMVA”) may not be well fit to the
data or generate reliable parameter estimates. In that situation, variation
across different possible topologies will be useful to assess how reliable
the result is. Below is code to fit a single-regime Brownian motion model,
and an Ornstein-Uhlenbeck model with two regimes that have different
trait means. The other models can be fit by modifying the model argument.
See the full script online for the code to do this. The results of fitting all of
the available models are shown in Table 2.
high_densityBMS <- OUwie(phy = treeD, data =

regimetableD,
model = “BMS”,
quiet = TRUE,
42 Paleontology
high_densityOUM <- OUwie(phy = treeD, data =

regimetableD,
starting.vals = c(0.01,0.01),
model = “OUM”
quiet = TRUE,
After parameter estimates are checked, AICcs can be compared to those for
the simple model. Although OUwie may indicate that a reliable solution to the
model fit has been reached, it is common for more complex models like
OUMVA to have unrealistic estimates of parameters, indicating the algorithm
did not effectively fit the model. In fact, this is the case with the eucladid data we
are using, where very high model support for OUMVA only occurs in the case
where parameter estimates are biologically unrealistic (Table 2).
Figure 6 Mean and variation in Akaike weights for seven candidate

models of continuous trait evolution. Note that although the mean
weight for complex models is low, for some trees within the set,
complex models have high weights. These often represent fits with
unrealistic model parameters.
A single-rate Brownian motion model is the best-fit model but it is
only marginally better than the next best model, or in fact than most of the
other models (Table 2). It is likely the case that there is not enough
information in our dataset to be able to meaningfully distinguish between
these models. Figure 6 is an example of how to summarize variation
across a set of trees; we show the median and variance of Akaike weights
for each model when it has been fit to data combined with 100 trees drawn
randomly from the posterior distribution.
Table 2 Model-fitting results for multi-regime Ornstein-Uhlenbeck
analyses
Model logL AICc Parameter estimates

BM −66.449 137.09 σ2 = 0.012, θ = 3.70
OU −66.250 138.89 α = 0.002, σ2 = 0.013, θ = 3.71
BMS −64.831 138.33 σ21 = 0.011, σ22 = 0.012, θ1 = 3.68, θ2 =
5.48
OUM −64.442 137.55 α = 0.0023, σ2 = 0.0023, θ1 = 3.68, θ2 =
5.51
OUMV −64.474 139.96 α = 0.0016, σ21 = 0.0373, σ22 = 0.0354, θ1
= 3.68, θ2 = 5.50
OUMA −64.421 139.86 α1 = 0.0027, α2 = 0.0053, σ2 = 0.0126, θ1
= 3.68, θ2 = 5.03
OUMVA −64.580 142.62 α1 = 2.4 × 10−8, α2 = 7.3 × 10−8, σ21 =
0.0114, σ22 = 0.0119, θ1 = 3.68, θ2 =
5.20
in geiger. This method has been applied to paleontological data (Anderson

et al. 2013; Benson and Choiniere 2013; Soul and Benson 2017; Ruta et al.
2019). A further example, trait MEDUSA, can be applied to multivariate data
and is implemented using maximum likelihood in the package motmot
(Thomas and Freckleton 2012; Puttick et al. 2020), and has also been applied
to datasets of fossil taxa (Brocklehurst and Brink 2017; Button et al. 2017;
Ruta et al. 2019). Both of these packages can also be used to fit and compare
simpler phylogenetic models. Other methods can be used to infer rates for
multivariate data, rather than to a single trait (Adams 2014). Diniz-Filho et al.
(2015) use an approach based on phylogenetic eigenvectors to detect non-
stationarity (i.e. significant shifts) in evolutionary rates in the skulls of thero-
pod dinosaurs. Levy processes can be used to model jumps in evolutionary
rate along branches in a phylogeny (Landis and Schraiber 2017), although this
has not yet been implemented for nonultrametric trees.
9.4 Other New Approaches

There are a range of other phylogenetic models and rate estimation approaches
that have been, or could be, applied to trees of fossil taxa. Often they can be
44 Paleontology
used to model trait change in groups of lineages, characters, or periods of time,

without having to specify the groupings in advance. This can be particularly
useful for exploratory analyses of trait change, rather than for hypothesis
testing as in our previous example analyses. At the time of writing, many of
these methods are relatively new and have not yet been applied in a broad
range of different scenarios. This means both that there are exciting opportun-
ities for new discoveries and that in some cases the behavior of the methods
across different areas of parameter space is not yet well characterized and
results should be interpreted cautiously. Below is a subset of new approaches
that have been applied to fossil data, or where they have not yet been applied to
paleontological phylogenies, they have a software implementation that allows
for it.
(a) The Fokker-Planck-Kolmogorov (FPK) model (Boucher et al. 2018;

Blomberg et al. 2020) is a flexible modeling framework that can be
used to estimate the shape of a macroevolutionary landscape that has
one or more optima, using a parameter called the “evolutionary poten-
tial.” This framework is distinct from the earlier multi-peak Ornstein-
Uhlenbeck models where switches between macroevolutionary regimes
are modeled; instead, it fits a model of trait change in a single macroevo-
lutionary regime, but that regime can have multiple optima, and be
bounded or unbounded.
(b) Mitov et al. (2019) introduced an approach for fitting mixed Gaussian
phylogenetic models, which infers regimes that can be any combination
of mode or rate for each model parameter across the phylogeny (for
implementation see Mitov et al. 2019).
(c) Parins-Fukuchi (2020) introduces a modeling framework for change in
disparity of continuous traits partitioned into integrated suites.
(d) The matching competition model (Drury et al. 2016, modified for applica-
tion to nonultrametric trees in Manceau et al. 2017) is designed to model the
effect of biotic interactions on trait evolution and is implemented in
RPanda (Morlon et al. 2016).
(e) As a more explicit alternative to convergence inferred via multi-peak
Ornstein-Uhlenbeck models, Stayton (2015) presents phylogenetic metrics
designed for hypothesis tests of convergence. Speed and Arbuckle (2017)
provide a review of other recent approaches to quantifying convergence
using phylogeny.
(f) Relative rates of morphological evolution and ancestral character states can
be estimated using the threshold model (Revell 2014, applied to paleonto-
logical data in Sallan et al. 2018).
10 Conclusion
As models for the “Tree of Life,” phylogenetic trees are science’s best attempt to
represent a very complex process – the diversification of life through time – in
a tractable way. Phylogenies and comparative methods allow us to answer
questions about how evolution works that could not otherwise be answered,
but they are obviously a simplification of real evolutionary processes. There is
a large suite of PCMs available for paleontologists to apply in order to under-
stand their group of interest, and most of these are now easily implemented
using R and other software. These methods are very powerful tools for infer-
ence, but are not always intuitive. For this reason, they can seem intimidating,
and misunderstandings are common. We hope that we have provided a practical
entry point for those looking to begin using PCMs in their own work, while
emphasizing the importance of careful interpretation of results and a good
understanding of the relationship between the quality of the data you put in
and the reliability of the information you get out. Phylogenies are themselves
hypotheses, and uncertainty in the ages of fossils and relationships between the
taxa they represent, combined with uncertainties in model fit, should always be
considered when interpreting results and should always be clearly represented
in published work. Caveats aside, this is an exciting time to be a paleontologist,
as recent advances in computational and mathematical tools are allowing us to
address some of evolutionary biology’s biggest questions in a rigorous, quanti-
tative way.
Supporting Materials
Supplemental data, full scripts for analyses, and trees are available at: github
.com/daveyfwright/PCMsForPaleontologists.
References
Ackerly D. 2009. Conservatism and diversification of plant functional traits:
Evolutionary rates versus phylogenetic signal. Proc. Natl. Acad. Sci. 106:
19699–19706.
Adams D. C. 2014. Quantifying and comparing phylogenetic evolutionary rates
for shape and other high-dimensional phenotypic data. Syst. Biol. 63:
166–177.
Anderson P. S. L., Friedman M., Ruta M. 2013. Late to the table: Diversification
of tetrapod mandibular biomechanics lagged behind the evolution of
terrestriality. Integr. Comp. Biol. 53: 197–208.
Ausich W. I., Wright D. F., Cole S. R., Sevastopulo G. D. 2020. Homology of
posterior interray plates in crinoids: A review and new perspectives from
phylogenetics, the fossil record and development. Palaeontology. 63: 525–545.
Bapst D. W. 2012. Paleotree: An R package for paleontological and phylogen-
etic analyses of evolution. Methods Ecol. Evol. 3: 803–807.
Bapst D. W. 2013a. A stochastic rate-calibrated method for time-scaling phy-
logenies of fossil taxa. Methods Ecol. Evol. 4: 724–733.
Bapst D. W. 2013b. When can clades be potentially resolved with morphology?
PLoS One. 8: e62312.
Bapst D. W. 2014a. Assessing the effect of time-scaling methods on
phylogeny-based analyses in the fossil record. Paleobiology. 40:
331–351.
Bapst D. W. 2014b. Preparing palaeontological datasets for phylogenetic com-
parative methods. In: Garamszegi L. Z., editor. Modern phylogenetic com-
parative methods and their application in evolutionary biology. Berlin,
Heidelberg: Springer-Verlag. pp. 515–544.
Bapst D. W., Hopkins M. J. 2017. Comparing cal3 and other a posteriori
time-scaling approaches in a case study with the pterocephaliid trilobites.
Paleobiology. 43: 49–67.
Barido-Sottani J., Pett W., O’Reilly J. E., Warnock R. C. M. 2019. FossilSim:
An R package for simulating fossil occurrence data under mechanistic
models of preservation and recovery. Methods Ecol. Evol. 10: 835–840.
Barido-Sottani J., Saupe E., Smiley T. M., Soul L. C., Wright A. M.,
Warnock R. C. M. 2020. Seven rules for simulations in paleobiology.
Paleobiology. 46(4): 435–444.
Barido-Sottani J., Tiel N. van, Hopkins M. J., Wright D. F., Stadler T.,
Warnock R. C. M. 2020. Ignoring fossil age uncertainty leads to inaccurate
References 47
topology and divergence times in time calibrated tree inference. Frontiers in

Ecology and Evolution, 8: 183
Baum D. A., Smith S. D. 2013. Tree thinking: An introduction to phylogenetic
biology. Greenwood Village, CO: Roberts.
Benson R. B. J., Choiniere J. N. 2013. Rates of dinosaur limb evolution provide
evidence for exceptional radiation in Mesozoic birds. Proc. R. Soc. B Biol.
Sci. 280: 20131780.
Blomberg S. P., Garland T., Ives A. R. 2003. Testing for phylogenetic signal in
comparative data: Behavioral traits are more labile. Evolution. 57: 717–745.
Blomberg S. P., Rathnayake S. I., Moreau C. M. 2020. Beyond Brownian
motion and the Ornstein-Uhlenbeck process: Stochastic diffusion models
for the evolution of quantitative characters. Am. Nat. 195: 145–165.
Blomberg S. P., Lefevre J. G., Wells J. A., Waterhouse M. 2012. Independent
contrasts and PGLS regression estimators are equivalent. Syst. Biol. 61:
382–391.
Boettiger C., Coop G., Ralph P. 2012. Is your phylogeny informative?
Measuring the power of comparative methods. Evolution. 66: 2240–2251.
Boucher F. C., Démery V., Conti E., Harmon L. J., Uyeda J. 2018. A general
model for estimating macroevolutionary landscapes. Syst. Biol. 67: 304–319.
Brocklehurst N., Brink K. S. 2017. Selection towards larger body size in both
herbivorous and carnivorous synapsids during the Carboniferous. Facets. 2:
68–84.
Butler M. A., King A. A. 2004. Phylogenetic comparative analysis: A modeling
approach for adaptive evolution. Am. Nat. 164: 683–695.
Button D. J., Barrett P. M., Rayfield E. J. 2017. Craniodental functional evolu-
tion in sauropodomorph dinosaurs. Paleobiology. 43: 435–462.
Clarke J. T., Lloyd G. T., Friedman M. 2016. Little evidence for enhanced
phenotypic evolution in early teleosts relative to their living fossil sister
group. Proc. Natl. Acad. Sci. 113: 11531–11536.
Close R. A., Friedman M., Lloyd G. T., Benson R. B. J. 2015. Evidence for a
mid-Jurassic adaptive radiation in mammals. Curr. Biol. 25: 2137–2142.
Cole S. R., Wright D. F., Ausich W. I. 2019. Phylogenetic community paleo-
ecology of one of the earliest complex crinoid faunas (Brechin Lagerstätte,
Ordovician). Palaeogeogr. Palaeoclimatol. Palaeoecol. 521: 82–98.
Cooper N., Thomas G. H., FitzJohn R. G. 2016. Shedding light on the “dark
side” of phylogenetic comparative methods. Methods Ecol. Evol. 7: 693–699.
Darwin C. R. 1859. On the origin of species by means of natural selection.
London: John Murray.
48 References
Diniz-Filho J. A. F., Alves D. M. C. C., Villalobos F., Sakamoto M.,

Brusatte S. L., Bini L. M. 2015. Phylogenetic eigenvectors and nonstationar-
ity in the evolution of theropod dinosaur skulls. J. Evol. Biol. 28: 1410–1416.
Drury J., Clavel J., Manceau M., Morlon H. 2016. Estimating the effect of
competition on trait evolution using maximum likelihood inference. Syst.
Biol. 65: 700–710.
Eastman J. M., Alfaro M. E., Joyce P., Hipp A. L., Harmon L. J. 2011. A novel
comparative method for identifying shifts in the rate of character evolution on
trees. Evolution. 65: 3578–3589.
Erwin D. H. 2007. Disparity: Morphological pattern and developmental
context. Palaeontology. 50: 57–73.
Felsenstein J. 1985. Phylogenies and the comparative method. Am. Nat. 125:
1–15.
Finarelli J. A., Flynn J. J. 2006. Ancestral state reconstruction of body size in the
Caniformia (Carnivora, Mammalia): The effects of incorporating data from
the fossil record. Syst. Biol. 55: 301–313.
Foote M. 1996. On the probability of ancestors in the fossil record.
Freckleton R. P. 2009. The seven deadly sins of comparative analysis. J. Evol.
Biol. 22: 1367–1375.
Garamszegi L. Z. 2014. Modern phylogenetic comparative methods and their
application in evolutionary biology. Berlin, Heidelberg: Springer-Verlag.
Garland T., Ives A. R. 2000. Using the past to predict the present: Confidence
intervals for regression equations in phylogenetic comparative methods. Am.
Nat. 155: 346–364.
Gascuel O., Steel M. 2014. Predicting the ancestral character changes in a tree is
typically easier than predicting the root state. Syst. Biol. 63: 421–435.
Gavryushkina A., Welch D., Stadler T., Drummond A. 2014. Bayesian infer-
ence of sampled ancestor trees for epidemiology and fossil calibration. PLoS
Comput. Biol. 10: e1003919.
Gavryushkina A., Heath T. A., Ksepka D. T., Stadler T., Welch D.,
Drummond A. J. 2017. Bayesian total-evidence dating reveals the recent
crown radiation of penguins. Syst. Biol. 66: 57–73.
Gearty W., Payne J. L. 2020. Physiological constraints on body size distribu-
tions in Crocodyliformes. Evolution. 74: 245–255.
Halliday T. J. D., Goswami A. 2016. The impact of phylogenetic dating method
on interpreting trait evolution: A case study of Cretaceous-Palaeogene
eutherian body-size evolution. Biol. Lett. 12: 6–12.
Hansen T. F. 1997. Stabilising selection and the comparative analysis of
adaptation. Evolution. 51: 1342–1351.
References 49
Hansen T. F., Martins E. P. 1996. Translating between microevolutionary

process and macroevolutionary patterns: The correlation structure of inter-
specific data. Evolution. 50: 1404–1417.
Harmon, Luke. 2019. “Phylogenetic Comparative Methods: Learning from
Trees.” EcoEvoRxiv. May 20. doi:10.32942/osf.io/e3xnr.
Harmon L. J., Weir J. T., Brock C. D., Glor R. E., Challenger W. 2008.
GEIGER: Investigating evolutionary radiations. Bioinformatics. 24:
129–131.
Harmon L. J., Losos J. B., Davies T. J., Gillespie R. G., Gittleman J. L.,
Jennings B. W., Kozak K. H., McPeek M. A., Moreno-Roark F., Near T. J.,
Purvis A., Ricklefs R. E., Schluter D., Schulte II J. A., Seehausen O.,
Sidlauskas B. L., Torres-Carvajal O., Weir J. T., Mooers A. Ø. 2010. Early
bursts of body size and shape evolution are rare in comparative data.
Evolution. 64: 2385–2396.
Harrison L. B., Larsson H. C. E. 2015. Among-character rate variation distri-
butions in phylogenetic analysis of discrete morphological characters. Syst.
Biol. 64: 307–324.
Harvey P. H., Read A. F., Nee S. 1995. Further remarks on the role of phylogeny
in comparative ecology. J. Ecol. 83: 733.
Heath T. A., Huelsenbeck J. P., Stadler T. 2014. The fossilized birth–death
process for coherent calibration of divergence-time estimates. Proc. Natl.
Acad. Sci. 111: E2957–E2966.
Hedman M. M. 2010. Constraints on clade ages from fossil outgroups.
Ho L. S. T., Ané C. 2014a. A linear-time algorithm for Gaussian and
non-Gaussian trait evolution models. Syst. Biol. 63: 397–408.
Ho L. S. T., Ané C. 2014b. Intrinsic inference difficulties for trait evolution with
Ornstein-Uhlenbeck models. Methods Ecol. Evol. 5: 1133–1146.
Hopkins M. J., Smith A. B. 2015. Dynamic evolutionary change in
post-Paleozoic echinoids and the importance of scale when interpreting
changes in rates of evolution. Proc. Natl. Acad. Sci. U.S.A. 112:
3758–3763.
Hunt G. 2012. Measuring rates of phenotypic evolution and the inseparability of
tempo and mode. Paleobiology. 38: 351–373.
Hunt G. 2013. Testing the link between phenotypic evolution and speciation:
An integrated palaeontological and phylogenetic analysis. Methods Ecol.
Evol. 4: 714–723.
Hunt G., Carrano M. T. 2010. Models and methods for analyzing phenotypic
evolution in lineages and clades. Paleontol. Soc. Pap. 16: 245–269.
50 References
Hunt G., Slater G. 2016. Integrating paleontological and phylogenetic

approaches to macroevolution. Annu. Rev. Ecol. Evol. Syst. 47: 189–213.
Jeremy M. Beaulieu and Brian O'Meara (2020). OUwie: Analysis of
Evolutionary Rates in an OU Framework. R package version 2.5. https://
CRAN.R-project.org/package=OUwie.
Kammer T. W. 2008. Paedomorphosis as an adaptive response in pinnulate
cladid crinoids from the Burlington limestone (Mississippian, Oseadean) of
the Mississippi Valley. In: Webster, G. D., Maples, C. D., editors.
Echinoderm paleobiology. Bloomington, IN: University of Indiana Press.
pp. 177–195.
Lande R. 1976. Natural selection and random genetic drift in phenotypic
evolution. Evolution. 30: 314.
Landis M. J. 2017. Biogeographic dating of speciation times using paleogeo-
graphically informed processes. Syst. Biol. 64: 307–324.
Landis M., Schraiber J. G. 2017. Pulsed evolution shaped modern vertebrate
diversity. Proc. Natl. Acad. Sci. U.S.A. 114: 13224–13229.
Lewis P. O. 2001. A likelihood approach to estimating phylogeny from discrete
morphological character data. Syst. Biol. 50: 913–925.
Lloyd G. T. 2016. Estimating morphological diversity and tempo with discrete
character-taxon matrices: Implementation, challenges, progress, and future
directions. Biol. J. Linn. Soc. 118: 131–151.
Lloyd G. T., Wang S. C., Brusatte S. L. 2012. Identifying heterogeneity in rates
of morphological evolution: Discrete character change in the evolution of
lungfish (Sarcopterygii; Dipnoi). Evolution. 66: 330–348.
Maddison D. R., Maddison W. P. 2020. MacClade 4. http://macclade.org/
macclade.html.
Manceau M., Lambert A., Morlon H. 2017. A unifying comparative phylogen-
etic framework including traits coevolving across interacting lineages. Syst.
Biol. 66: 551–568.
Matzke N. J., Wright A. 2016. Inferring node dates from tip dates in fossil
Canidae: The importance of tree priors. Biol. Lett. 12: 1–4.
Mitov V., Bartoszek K., Stadler T. 2019. Automatic generation of evolutionary
hypotheses using mixed Gaussian phylogenetic models. Proc. Natl. Acad.
Sci. 116: 16921–16926.
Morlon H., Lewitus E., Condamine F. L., Manceau M., Clavel J., Drury J. 2016.
RPANDA: An R package for macroevolutionary analyses on phylogenetic
trees. Methods Ecol. Evol. 7: 589–597.
Nielsen R. 2002. Mapping mutations on phylogenies. Syst. Biol. 51: 729–739.
Nunn C. L. 2011. The comparative approach in evolutionary anthropology and
biology. Chicago: University of Chicago Press.
References 51
Nunn C. L., Barton R. A. 2001. Comparative methods for studying primate

adaptation and allometry. Evol. Anthropol. 10: 81–98.
O’Meara B. C., Ané C., Sanderson M. J., Wainwright P. C. 2006. Testing for
different rates of continuous trait evolution using likelihood. Evolution. 60:
922–933.
O’Reilly J. E., Puttick M. N., Parry L., Tanner A. R., Tarver J. E., Fleming J.,
Pisani D., Donoghue P. C. J. 2016. Bayesian methods outperform parsimony
but at the expense of precision in the estimation of phylogeny from discrete
morphological data. Biol. Lett. 12: 20160081.
Paradis E., Claude J., Strimmer, K. 2004. APE: Analyses of phylogenetics and
evolution in R language. Bioinformatics. 20: 289–290.
Parins-Fukuchi C. 2020. Detecting mosaic patterns in macroevolutionary
disparity. Am. Nat. 195: 129–144.
Pennell M. W., Fitzjohn R. G., Cornwell W. K., Harmon L. J. 2015. Model
adequacy and the macroevolution of angiosperm functional traits. Am. Nat.
186: E33–E50.
Pennell M. W., Eastman J. M., Slater G. J., Brown J. W., Uyeda J. C.,
FitzJohn R. G., Alfaro M. E., Harmon L. J. 2014. Geiger V2.0: An expanded
suite of methods for fitting macroevolutionary models to phylogenetic trees.
Bioinformatics. 30: 2216–2218.
Pinheiro J., Bates D., DebRoy S., Sarkar D. 2019. nlme: Linear and nonlinear
mixed effects models. R package version 3: 1–140.
Polly P. D. 2019. Spatial processes and evolutionary models: A critical review.
Palaeontology. 62: 175–195.
Price S. A. 2019. State-dependent diversification of traits. http://treethinkers
.org/tutorials/state-dependent-diversification-of-traits.
Price S. A., Friedman S. T., Wainwright P. C. 2015. How predation shaped fish:
The impact of fin spines on body form evolution across teleosts. Proc. R. Soc.
B Biol. Sci. 282. https://doi.org/10.1098/rspb.2015.1428.
Puttick M. N. 2016. Partially incorrect fossil data augment analyses of discrete
trait evolution in living species. Biol. Lett. 12: 20160392.
Puttick M. N., Ingram T., Clarke M., Thomas G. H. 2020. MOTMOT: Models of
trait macroevolution on trees (an update). Methods Ecol. Evol. 11: 464–471.
Puttick M. N., O’Reilly J. E., Pisani D., Donoghue P. C. J. 2019. Probabilistic
methods outperform parsimony in the phylogenetic analysis of data simu-
lated without a probabilistic model. Palaeontology. 62: 1–17.
Revell, L.J. (2010), Phylogenetic signal and linear regression on species data.
Methods in Ecology and Evolution, 1: 319–329.
Revell L. J. 2012. phytools: An R package for phylogenetic comparative
biology (and other things). Methods Ecol. Evol. 3: 217–223.
52 References
Revell L. J. 2013. Two new graphical methods for mapping trait evolution on
phylogenies. Methods Ecol. Evol. 4: 754–759.
Revell L. J. 2014. Ancestral character estimation under the threshold model
from quantitative genetics. Evolution. 68: 743–759.
Revell L. J., Schliep K., Valderrama E., Richardson J. E. 2018. Graphs in
phylogenetic comparative analysis: Anscombe’s quartet revisited. Methods
Ecol. Evol. 9: 2145–2154.
Rohlf F. J. 2006. A comment on phylogenetic correction. Evolution. 60: 1509.
Ruta M., Krieger J., Angielczyk K. D., Wills M. A. 2019. The evolution of the
tetrapod humerus: Morphometrics, disparity, and evolutionary rates. Earth
Environ. Sci. Trans. R. Soc. Edinburgh. 109: 351–369.
Sallan L., Friedman M., Sansom R. S., Bird C. M., Sansom I. J. 2018. The
nearshore cradle of early vertebrate diversification. Science. 464: 460–464.
Silvestro D., Kostikova A., Litsios G., Pearman P. B., Salamin N. 2015.
Measurement errors should always be incorporated in phylogenetic compara-
tive analysis. Methods Ecol. Evol. 6: 340–346.
Simpson G. G. 1944. Tempo and mode in evolution. New York: Columbia
University Press.
Slater G. J. 2013. Phylogenetic evidence for a shift in the mode of mammalian
body size evolution at the Cretaceous-Palaeogene boundary. Methods Ecol.
Evol. 4: 734–744.
Slater G. J. 2014. Correction to “Phylogenetic evidence for a shift in the mode of
mammalian body size evolution at the Cretaceous-Palaeogene boundary,”
and a note on fitting macroevolutionary models to comparative paleonto-
logical data sets. Methods Ecol. Evol. 5: 714–718.
Slater G. J., Pennell M. W. 2014. Robust regression and posterior predictive
simulation increase power to detect early bursts of trait evolution. Syst. Biol.
63: 293–308.
Slater G. J., Harmon L. J., Alfaro M. E. 2012. Integrating fossils with molecular
phylogenies improves inference of trait evolution. Evolution. 66: 3931–3944.
Soul L. C., Benson R. B. J. 2017. Developmental mechanisms of macroevolu-
tionary change in the tetrapod axis: A case study of Sauropterygia. Evolution.
71: 1164–1177.
Soul L. C., Friedman M. 2015. Taxonomy and phylogeny can yield comparable
results in comparative palaeontological analyses. Syst. Biol. 64: 608–620.
Soul L. C., Friedman M. 2017. Bias in phylogenetic measurements of extinction
and a case study of end-Permian tetrapods. Palaeontology. 60: 169–185.
Speed M. P., Arbuckle K. 2017. Quantification provides a conceptual basis for
convergent evolution. Biol. Rev. 92: 815–829.
References 53
Stadler T. 2010. Sampling-through-time in birth-death trees. J. Theor. Biol. 267:

396–404.
Stadler T., Gavryushkina A., Warnock R. C. M., Drummond A. J., Heath T. A.
2018. The fossilized birth-death model for the analysis of stratigraphic range
data under different speciation modes. J. Theor. Biol. 447: 41–55.
Stayton C. T. 2015. The definition, recognition, and interpretation of convergent
evolution, and two new measures for quantifying and assessing the signifi-
cance of convergence. Evolution. 69: 2140–2153.
Thomas G. H., Freckleton R. P. 2012. MOTMOT: Models of trait macroevolu-
tion on trees. Methods Ecol. Evol. 3: 145–151.
Uyeda J. C., Zenil-Ferguson R., Pennell M. W. 2018. Rethinking phylogenetic
comparative methods. Syst. Biol. 67: 1091–1109.
Voje K. L., Starrfelt J., Liow L. H. 2018. Model adequacy and micro-
evolutionary explanations for stasis in the fossil record. Am. Nat. 191:
509–523.
Wagner P. J. 2012. Modelling rate distributions using character compatibility:
implications for morphological evolution among fossil invertebrates. Biol.
Lett. 8: 143–146.
Wagner P. J., Marcot J. D. 2010. Probabilistic phylogenetic inference in the
fossil record: current and future applications. Paleontol. Soc. Pap. 16:
189–211.
Wagner, P.J. and Marcot, J.D., 2013. Modelling distributions of fossil sampling
rates over time, space and taxa: assessment and implications for macroevo-
lutionary studies. Methods in Ecology and Evolution, 4(8), pp.703–713.
Warnock R. C. M., Wright A. M. 2020. Understanding the tripartite approach to
Bayesian divergence time estimation. EcoEvoRxiv. https://doi.org/10.32942
/osf.io/4vazh.
Wesley-Hunt G. D. 2005. The morphological diversification of carnivores in
North America. Paleobiology. 31: 35–55.
Westoby M., Leishman M., Lord J. 2016. Further remarks on phylogenetic
correction. J. Ecol. 83: 727–729.
Wiley E. O., Lieberman B. S. 2011. Phylogenetics: Theory and practice of
phylogenetic systematics. New York: John Wiley & Sons.
Wright A. M. 2019. A systematist’s guide to estimating Bayesian phylogenies
from morphological data. Insect Syst. Divers. 3: 2.
Wright A. M., Hillis D. M. 2014. Bayesian analysis using a simple likelihood
model outperforms parsimony for estimation of phylogeny from discrete
morphological data. PLoS One. 9: e109210.
54 References
Wright A. M., Lloyd G. T., Hillis D. M. 2016. Modeling character change

heterogeneity in phylogenetic analyses of morphology through the use of
priors. Syst. Biol. 65: 602–611.
Wright A. M., Wagner P. J., Wright D. F. 2020. Testing character evolution
models in phylogenetic paleobiology: A case study with Cambrian echino-
derms. EcoEvoRxiv. https://doi.org/10.32942/osf.io/ykzg5.
Wright D. F. 2015. Fossils, homology, and phylogenetic paleo-ontogeny:
A reassessment of primary posterior plate homologies among fossil and
living crinoids with insights from developmental biology. Paleobiology. 41:
570–591.
Wright D. F. 2017a. Bayesian estimation of fossil phylogenies and the evolution
of early to middle Paleozoic crinoids (Echinodermata). J. Paleontol. 91:
799–814.
Wright D. F. 2017b. Phenotypic innovation and adaptive constraints in the
evolutionary radiation of palaeozoic crinoids. Sci. Rep. 7: 1–10.
Wright D. F., Toom U. 2017. New crinoids from the Baltic region (Estonia):
Fossil tip-dating phylogenetics constrains the origin and Ordovician–Silurian
diversification of the Flexibilia (Echinodermata). Palaeontology. 60:
893–910.
Acknowledgments
This Element stems from materials originally developed for a 2019
Paleontological Society sponsored workshop titled “Quantitative Methods in
Phylogenetic Paleobiology.” We thank Vera Korasidis and Melanie J. Hopkins
for providing feedback on an early version of this Element, as well as Will Gearty
and David Bapst for thorough reviews that improved the manuscript’s quality and
clarity. LCS acknowledges support from the Smithsonian National Museum of
Natural History Deep Time Initiative. DFW acknowledges support from the
Gerstner Scholars Fellowship and the Gerstner Family Foundation, the Lerner-
Gray Fund for Marine Research, and the Richard Gilder Graduate School,
American Museum of Natural History, as well as a Norman Newell Early
Career Grant from the Paleontological Society.
Editor-in-Chief
Colin D. Sumrall
University of Tennessee
About the Series

The Elements of Paleontology series is a publishing collaboration between the
Paleontological Society and Cambridge University Press. The series covers the full
spectrum of topics in paleontology and paleobiology, and related topics in the Earth and
life sciences of interest to students and researchers of paleontology.
The Paleontological Society is an international nonprofit organization devoted
exclusively to the science of paleontology: invertebrate and vertebrate paleontology,
micropaleontology, and paleobotany. The Society’s mission is to advance the study of the
fossil record through scientific research, education, and advocacy. Its vision is to be
a leading global advocate for understanding life’s history and evolution. The Society has
several membership categories, including regular, amateur/avocational, student, and
retired. Members, representing some 40 countries, include professional paleontologists,
academicians, science editors, Earth science teachers, museum specialists, undergraduate
and graduate students, postdoctoral scholars, and amateur/avocational paleontologists.
Elements in the Series

Beyond Hands On: Incorporating Kinesthetic Learning in an Undergraduate
Paleontology Class
David W. Goldsmith
Confronting Prior Conceptions in Paleontology Courses
Margaret M. Yacobucci
Utilizing the Paleobiology Database to Provide Educational Opportunities for
Undergraduates
Rowan Lockwood, Phoebe A. Cohen, Mark D. Uhen and Katherine Ryker
Equity, Culture, and Place in Teaching Paleontology: Student-Centered Pedagogy
for Broadening Participation
Christy C. Visaggi
Computational Fluid Dynamics and its Applications in Echinoderm Palaeobiology
Imran A. Rahman
Understanding the Tripartite Approach to Bayesian Divergence Time Estimation
Rachel C. M. Warnock and April M. Wright
The Stratigraphic Paleobiology of Nonmarine Systems
Steven Holland and Katharine M. Loughney
Echinoderm Morphological Disparity: Methods, Patterns, and Possibilities
Bradley Deline
Functional Micromorphology of the Echinoderm Skeleton
Przemyslaw Gorzelak
Disarticulation and Preservation of Fossil Echinoderms: Recognition of Ecological-
Time Information in the Echinoderm Fossil Record
William I. Ausich
Crinoid Feeding Strategies: New Insights From Subsea Video And Time-Lapse
David Meyer, Margaret Veitch, Charles G. Messing, Angela Stevenson
Phylogenetic Comparative Methods: A User’s Guide for Paleontologists
Laura C. Soul and David F. Wright
A full series listing is available at: www.cambridge.org/EPLY

Soul, L. C. & D. F. Wright. 2021. Phylogenetic Comparative Methods, A User's Guide For Paleontologists

Uploaded by

Copyright:

Available Formats

Soul, L. C. & D. F. Wright. 2021. Phylogenetic Comparative Methods, A User's Guide For Paleontologists

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Soul, L. C. & D. F. Wright. 2021. Phylogenetic Comparative Methods, A User's Guide For Paleontologists

Uploaded by

Copyright:

Available Formats

Soul and Wright

Recent advances in statistical approaches called phylogenetic

Phylogenetic Comparative Methods: A User’s Guide for Paleontologists

About the Series

Cambridge University Press is part of the University of Cambridge.

Cambridge University Press has no responsibility for the persistence or accuracy of

Abstract: Recent advances in statistical approaches called phylogenetic

Keywords: phylogenetic comparative methods, paleontology,

ISBNs: 9781108794688 (PB), 9781108894142 (OC)

2 Getting Started: Data and Phylogeny 2

4 Tempo and Mode: Brownian Motion and More 12

5 Incorporating Estimates of Error 20

6 Variation Across Trees 20

7 Model Fit and Model Adequacy 27

8 Post-Hoc Modeling of Discrete Characters 29

9 Modeling Heterogeneous Trait Evolution 36

2 Getting Started: Data and Phylogeny

Figure 1 Phylogenetic tree comprising eighty-two species belonging to the

timescaling of a composite topology (Bapst 2014a), etc. The number of trees

2.2 Trait Data

Figure 2 (a) Hypothetical phylogeny with branch lengths scaled to absolute

squares (OLS) regression is performed on species data, there will be

be derived from any potential source of nonindependence. For example, here we

EXAMPLE ANALYSIS 1 – REGRESSION

data: allTraits$Shape and allTraits$Density

The standard regression is OLS:

Look at the output:

An OLS regression gives a slope of −0.68.

between a model’s goodness of ﬁt to the data (i.e. the likelihood) and

4 Tempo and Mode: Brownian Motion and More

4.1 Brownian Motion

4.2 Ancestral State Reconstruction

EXAMPLE ANALYSIS 2 – FITTING A BROWNIAN MOTION MODEL

Required packages: phytools; geiger

Figure 4 Phenograms, or traitgrams, help visualize patterns of trait

Now we will ﬁt Brownian motion models to both of the continuous traits

shapeBM <- ﬁtContinuous(phy = tree,

GEIGER-ﬁtted comparative model of continuous data

GEIGER-ﬁtted comparative model of continuous data

4.3 Beyond Brownian Motion

or rapid increases in morphological disparity. Fortunately, there is

evolving under Ornstein-Uhlenbeck are therefore less tightly linked to the

5 Incorporating Estimates of Error

5.2 Error in the Trait Data

6 Variation Across Trees

EXAMPLE ANALYSIS 3 – SIMULATION AND FITTING MODELS OF CONTINUOUS

Required packages: ape; phytools; geiger

Generate a tree with ﬁfty tips

bmsim <- bmPlot(tree = simtree,

This function allows you to visualize a single character and how it

Plot a histogram of the trait values at the tips of the tree

OU <- rTraitCont(phy = simtree,

EXAMPLE ANALYSIS 4 – APPLICATION TO MORE THAN ONE TREE

Load the post-burnin posterior distribution of trees:

Select 100 trees:

Set up a vector to store results:

sigmas[i] <- ﬁtBM$opt$sigsq

7 Model Fit and Model Adequacy

8 Post-Hoc Modeling of Discrete Characters