Ecology, 97(5), 2016, pp.

1228–1238
© 2016 The Authors. Ecology, published by Wiley Periodicals, Inc., on behalf of the Ecological Society of America.
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in
any medium, provided the original work is properly cited.

Comparing process-­based and constraint-­based approaches for

modeling macroecological patterns
Xiao Xiao,1,2,3,4,8 James P. O’Dwyer,5 and Ethan P. White1,2,6,7
Department of Biology, Utah State University, Logan, UT 84322-5305, USA
1
2
Ecology Center, Utah State University, Logan, UT 84322-5205, USA
School of Biology and Ecology, University of Maine, Orono, ME 04469, USA
3
4
Senator George J. Mitchell Center for Sustainability Solutions, University of Maine, Orono, ME 04469, USA
5
Department of Plant Biology, University of Illinois, Urbana, IL 61801, USA
6
Department of Wildlife Ecology and Conservation, University of Florida, Gainesville, FL, 32611, USA
7
Informatics Institute, University of Florida, Gainesville, FL 32611, USA

Abstract. Ecological patterns arise from the interplay of many different processes,
and yet the emergence of consistent phenomena across a diverse range of ecological systems
suggests that many patterns may in part be determined by statistical or numerical con-
straints. Differentiating the extent to which patterns in a given system are determined
statistically, and where it requires explicit ecological processes, has been difficult. We tackled
this challenge by directly comparing models from a constraint-­based theory, the Maximum
Entropy Theory of Ecology (METE) and models from a process-­based theory, the size-­
structured neutral theory (SSNT). Models from both theories were capable of characterizing
the distribution of individuals among species and the distribution of body size among
individuals across 76 forest communities. However, the SSNT models consistently yielded
higher overall likelihood, as well as more realistic characterizations of the relationship
between species abundance and average body size of conspecific individuals. This suggests
that the details of the biological processes contain additional information for understanding
community structure that are not fully captured by the METE constraints in these systems.
Our approach provides a first step towards differentiating between process-­and constraint-­
based models of ecological systems and a general methodology for comparing ecological
models that make predictions for multiple patterns.
Key words: constraints; maximum entropy theory of ecology; mechanisms; model comparison; neutral
theory; processes.

Introduction many rare species and a few abundant ones, but with
mechanisms ranging from purely statistical to popula-
tion dynamics to resource partitioning (Marquet et al.
Patterns of biodiversity that are aggregated across
2003, McGill et al. 2007). Moreover, many macroeco-
large numbers of individuals often take similar shapes
logical patterns are not independent. For example, the
across ecosystems and taxonomic groups (Brown 1995).
species-­area relationship at small spatial scales can be
Understanding why such patterns seem to be universal,
derived from the shape of the SAD and the level of
for example the skewed distribution of individuals
­intraspecific aggregation (Harte 2011, McGill 2011),
among species (the species abundance distribution)
while the SAD itself can be obtained as a spatially auto-
(Fisher et al. 1943, McGill et al. 2007) and the uneven al-
correlated sample from the regional species pool
location of body size among individuals (the individual
(McGill 2011). This combination of equivalent models
size distribution) (Enquist and Niklas 2001, Muller-­
with different processes and interrelated patterns makes
Landau et al. 2006b), is one of the central pursuits of
determining process using a single pattern challenging
macroecology (Brown 1999, McGill and Nekola 2010).
and instead calls for unified theoretical frameworks that
This task is not trivial because common patterns are
are capable of capturing multiple patterns as well as
often associated with multiple models that have differ-
their intercorrelations with a minimal set of assump-
ent assumptions about mechanisms yet make similar or
tions (Marquet et al. 2014).
even identical predictions (Frank 2014). For example,
Theories that have been proposed for macroecological
more than 20 models exist for the species-­abundance
patterns tend to fall into two conceptually distinct catego-
distribution (SAD) all making realistic predictions with
ries (Brown 1999, Frank 2014). Similar patterns may arise
directly from fundamental ecological processes if the
Manuscript received 27 May 2015; revised 29 September 2015;
accepted 5 November 2015; final version received 2 December same processes dominate across multiple systems.
2015. Corresponding Editor: B. E. Kendall. Theories in this category include the theory of island bio-
8
E-mail: [email protected]. geography (MacArthur and Wilson 1967), which

1228
May 2016 APPROACHES FOR MACROECOLOGY  1229

explains the species richness on islands as the equilibrium

Methods
between immigration and extinction, and the neutral the-
ory of biodiversity (Hubbell 2001), which shows that Theoretical frameworks
demographic stochasticity can lead to community-­level
diversity patterns. Alternatively, patterns may arise as This section briefly outlines the underlying assumptions
emergent statistical phenomena with forms determined of METE and SSNT, and the specification of the two mod-
primarily by some set of numerical constraints on the els under each theory. For mathematical configurations
system (Frank 2014), where processes operate only
­ and derivation of the predicted patterns, see Appendix S1.
indirectly through their effects on the constraints.
­
Theories built on constraints include the feasible set Maximum Entropy Theory of Ecology (METE).—The
(Locey and White 2013), and recent applications of the Maximum Entropy Theory of Ecology (METE) is a the-
maximum ­entropy principle to ecology (Shipley et al. ory built on the maximum entropy principle (MaxEnt;
2006, Dewar and Porté 2008, Harte 2011). Neither of Jaynes 2003). MaxEnt states that the least biased state of
these approaches relies on the operation of specific a system is the one with the highest information entropy
­processes but instead on the fact that many possible (Shannon entropy). Given a set of constraints that the
combinations of processes and states of the system
­ system has to satisfy, this state can be obtained by opti-
­produce similar empirical ­outcomes (Frank 2014). mization using the method of Lagrange multipliers, with
In this study we examined two theoretical frameworks, no tunable parameters aside from the constraints.
the Maximum Entropy Theory of Ecology (METE) Among existing applications of MaxEnt to ecology
(Harte 2011) and the size-­ structured neutral theory (e.g., Shipley et al. 2006, Pueyo et al. 2007, Dewar and
(SSNT) (O’Dwyer et al. 2009), which are two of the most Porté 2008), METE is arguably the most comprehensive,
comprehensive theories in macroecology. Both theories encompassing three distinct branches of ecological pat-
are able to predict two distinct sets of patterns, those of terns – the spatial distributions of individuals and species,
biodiversity as well as body size and energy use. METE is a the distributions of individuals among species and higher
constraint-­based theory, where patterns arise as the most taxonomic ranks, and the allocation of body size and
likely (least biased) state of a community constrained by a energy use at different taxonomic levels. We examined two
set of state variables, such as species richness, the total existing models of METE, ASNE (Harte et al. 2008, Harte
number of individuals, and the total energy consumption 2011), where the acronym stands for Area, Species,
across all individuals. SSNT is an extension of the neutral Number of individuals, and Energy, and the newly devel-
theory of ecology (Hubbell 2001) and is a process-­based oped AGSNE (Harte et al. 2015), where the additional
theory, where the patterns arise as the steady state of a “G” stands for Genera or other higher taxonomic ranks
dynamic system governed by individual birth, death, and (family, order, etc.). In this study, we focused on the non-­
growth in size. Both theories make predictions for multi- spatial patterns in ASNE and AGSNE, which are pre-
ple patterns of biodiversity as well as biomass and energy dicted independently from the spatial patterns. In
use, providing a multifaceted characterization of commu- non-­spatial ASNE, the allocations of individuals and of
nity structure. body size within a community are regulated by three state
We evaluated two existing models of METE and two variables: species richness S, total abundance N, and total
models that we derived for SSNT, to explore whether metabolic rate within the community EMETE. Non-­spatial
community structure in biodiversity and body size can be AGSNE requires an additional input G for a higher taxo-
adequately captured by constraints or processes. One of nomic group, which we took to be the number of genera
the METE models, ASNE (Harte 2011, see Methods within the community.
for details), has been shown in previous studies to
have mixed performance among its predictions Size-­
structured neutral theory (SSNT).—Size-struc-
(Newman et al. 2014, Xiao et al. 2015), while the other tured neutral theory (SSNT) is an extension of Hubbell’s
models have not been thoroughly tested with empirical neutral theory of ecology (NTE; Hubbell 2001). In NTE,
data. Using data from 76 forest communities we macroecological patterns emerge as the steady state of the
examined the models’ ability to characterize three community where individuals go through the processes
major macroecological patterns and compared their of birth, death, and speciation. SSNT introduces a size
performance using a single joint distribution that component into NTE, where the size of each individual
encapsulates these and other p ­ redictions as marginal or increases through time. Ontogenetic growth thus intro-
conditional distributions. Direct comparison of multiple duces variation in individual sizes, and also variation in
models from the two t­heoretical frameworks, using a the average size and total biomass across different spe-
large number of datasets and multiple empirical cies. The structure of the community in SSNT is gov-
patterns, allows strong inference to be made about the erned by the forms and values of the three demographic
relative performance of the models and, by extension, parameters b (birth rate), m (mortality rate), and g (rate
the ability of current constraint-­based and process-­based of growth).
approaches to characterize community-level macro­ We examined two realized models of SSNT. In the sim-
ecological patterns of diversity and body size. plest model (SSNT_N, with “N” for neutral), all three
1230 XIAO XIAO ET AL. Ecology, Vol. 97, No. 5

demographic parameters are assumed to be constant for

Table 1. Analytical forms of the patterns predicted by the four models with interpretations. λ’s and λ’’s are Lagrange multipliers for ASNE and AGSNE, respectively. τ’s are parameters
all individuals regardless of their species identities or other

abundance n (and that the species belongs to a genus with m

)]
individual characteristics. This is called the “completely

species in AGSNE). Note that metabolic rate scales as D2

The average individual metabolic rate within a species with
⋅ e−𝜆2 𝜆2 n + 1 − e−𝜆2 nE 𝜆2 nE + 1

𝜀̄ AGSNE (m,n) ≈[e−𝜆3 mn 𝜆�3 mn + 1 − e−𝜆3 mnE 𝜆�3 mnE + 1 ]∕

neutral case” (O’Dwyer et al. 2009). Note that while the

)
­assumption of b and m being constant holds regardless of

(
the unit used for size, g can only be constant in one particu-

Size-­density relationship (SDR)

lar set of size units, e.g., constant growth in diameter as a

(
function of current diameter does not translate into con-

)
stant growth in cross-­sectional area or volume as a func-

[𝜆�3 mn e−𝜆3 mn − e−𝜆3 mnE ]

)
�
tion of current area or volume. In SSNT_N, we made the

for SSNT_N and SSNT_M. C’s are normalization constants. γ in ΨASNE(D) is defined as γ  =  λ1 + λ2∙D2, and γ’ in ΨAGSNE(D) is defined as γ’ = λ2′ + λ3′∙D2.

�
intentionally naïve assumption that g was constant across

)
individuals when measured as the increase in diameter

[
D (i.e., g (D) = dD = constant).

�
)
dt

𝜀̄ SSNT_M = 𝜏63 + 𝜏62 + 𝜏3 + 1

n𝜆2 e−𝜆2 n −e−𝜆2 nE
In the second model, termed SSNT_M where M stands

(
(

3
for metabolism or metabolic theory, we incorporated

𝜀̄ SSNT_N = 𝜏22 + 𝜏2 + 1
�
1
insights from the metabolic theory of ecology (MTE;

3
2

instead of D.
(
Brown et al. 2004), and made the more realistic assump-

3
2
tion that g was a function of size, while b and m were still

𝜀̄ ASNE (n) =
held constant. Specifically, MTE predicts that a plant’s
growth rate measured as increase in biomass is propor-
tional to the plant’s metabolic rate (Enquist et al. 1999,
West et al. 1999, Muller-­Landau et al. 2006a), which trans-

The probability that a randomly selected individual from

)
lates into constant growth rate when size is measured in

⋅ 1 − (N + 1) e−𝛾N + Ne−𝛾(N+1)

the community has diameter between (D, D  +  ΔD)

units of diameter raised to the power of 2/3, D2/3.

Macroecological patterns.—All four models can

Individual size distribution (ISD)

predict the same set of three major macroecological

­patterns: the species-­abundance distribution (SAD; dis-
tribution of individuals among species), the individual

⋅ D−1∕3
− 𝜆� +𝛾 � m

2
size distribution (ISD; distribution of body size among
)
(1−e−𝛾 � m )

regardless of species identity.

1

individuals regardless of their species identity), and the

(

2∕3 −1)
me

size-­density relationship (SDR; relationship between av-

⋅ (1−e−𝛾 )2

ΨSSNT_N (D) = 𝜏2 e−𝜏2 (D−1)

m=1
e−𝛾

ΨSSNT_M (D) = 23 𝜏3 e−𝜏3 (D

S

erage body size within each species and the abundance

AGSNE2

of the species) (Cotgreave 1993). AGSNE is also able to

ASNE2

predict higher-­order patterns, such as the distribution of

D

ΨAGSNE (D) ≈ C
ΨASNE (D) = C

individuals and body size among genera, which we did

not examine in this study (but see Harte et al. 2015).
Table 1 summarizes the predicted forms of the patterns in
the four models. λ’s in ASNE and AGSNE are Lagrange
multipliers (Jaynes 2003), determined by the state variables
S, N and EMETE in ASNE (see Harte 2011 and Xiao et al.
Species abundance distribution (SAD)

2015 for detailed derivation), and by G, S, N and EMETE in

selected species has abundance n.
) − 𝜆� +(𝜆� +𝜆� )n
)

AGSNE (Harte et al. 2015; Appendix S1). τ’s in the SSNT

− 𝜆� +(𝜆� +𝜆� )n

1−e 1 2 3

The probability that a randomly

2 3

models are ratios of the demographic parameters, and are

e−(𝜆1 +𝜆2 )n

also fully determined by the variables S, N, and a measure

(

𝜏1 n
𝜏1 n
e

1) n
n

of total body size ESSNT, with ESSNT_N = Di and

∑
⋅

1)

ΦSSNT_M (n) = − ln(1−𝜏

AGSNE1 n

ΦSSNT_N (n) = − ln(1−𝜏

ESSNT_M = Di (see Appendix S1). Note that patterns of

∑ 2∕3
1
1
ASNE1 n

body size (ISD and SDR) predicted by the METE models

1

ΦAGSNE (n) ≈ C

are of the same unit as metabolic rates (B), which scales with
ΦASNE (n) ≈ C

size in trees with good approximation as the square of diam-

eter (D): B ∝ D2 (West et al. 1999), wheras the basic unit of
size in SSNT is D. For the purpose of comparison, we con-
verted patterns of size from the models into the same units.
The ISDs in ASNE and AGSNE were converted to unit of
Interpretation

D. The SDRs predicted by these two models do not have

SSNT_M

simple analytical forms in unit of D, so we converted the

SSNT_N
AGSNE
Patterns

ASNE

predictions of SSNT_N and SSNT_M to unit of B (i.e., D2)

instead (Table 1; Appendix S1).
May 2016 APPROACHES FOR MACROECOLOGY  1231

to genus or species, and removed four communities in

Data
(Xiao et al. 2015) from our analysis where >10% of the
We used forest census data to empirically evaluate the individuals were not identified to genus. We also excluded
models. This type of data consistently includes individual individuals that were dead or missing size measurements,
level size measurements, allowing the compilation of large as well as those with sizes below or equal to the specified
numbers of communities with the necessary information threshold, since not all individuals in these size classes
for fitting and evaluating the models. Forest data sample were included in the surveys. Overall the compilation
all individuals of every species down to a certain minimum encompasses 76 communities with 2030 species and
size, thus avoiding issues with not detecting juvenile 378806 individuals from four continents (Asia, Australia,
organisms (other than those below the minimum size), North America, and South America) (Table 2).
which may bias the empirical size distributions. In addi-
tion, determinately growing organisms (e.g., birds and
Analyses
mammals) often exhibit multimodal ISDs (Ernest 2005,
Thibault et al. 2011), whereas the ISDs for trees are in gen- We applied the four models to each empirical commu-
eral monotonically decreasing (Enquist and Niklas 2001, nity, and examined their abilities to characterize
Muller-­Landau et al. 2006b), and therefore consistent community structure in abundance and body size.
­
with the qualitative form predicted by the four models Diameter values in each community were rescaled as
(Table 1). D = Doriginal/Dmin, where Dmin is the diameter of the smallest
We combined the data compiled by (Xiao et al. 2015), individual in the community after the exceptional individ-
which encompassed 60 forest communities worldwide, uals were excluded (see 2. Data), so that D has a minimal
with data on 20 additional communities from (Bradford value of 1 in each community following METE’s assump-
et al. 2014). All communities have been fully surveyed with tion (see Harte 2011). Although SSNT does not make such
species identity and measurement of size (diameter or an assumption on minimal size, it predicts the ISD to be an
equivalent) for each individual above community-­specific exponential distribution (Table 1), the fit of which is unaf-
size thresholds (ranging from 10 to 100 mm). In cases fected by this transformation of D. Multiple branches
where multiple surveys are available for a community, we from the same individual were combined to determine the
used those from the most recent survey unless otherwise basal stem diameter with the pipe model, which preserves
specified. We excluded individuals that were not identified the total area as well as the metabolic rate of the branches

Table 2. Summary of datasets.

Area of Individual Number

Dataset Description Plots (ha) of Plots Survey Year References

CSIRO Tropical rainforest 0.5 20 1985–2012* 1

Serimbu Tropical rainforest 1 1 1995† 2–5
La Selva Tropical wet forest 2.24 5 2009 6, 7
Eno-­2 Tropical moist forest 1 1 2000–2001 8
BCI Tropical moist forest 50 1 2010 9–11
DeWalt Bolivia forest plots Tropical moist forest 1 2 N/A 12
Luquillo Tropical moist forest 16 1 1994–1996‡ 13, 14
Sherman Tropical moist forest 5.96 1 1999 15–17
Cocoli Tropical moist forest 4 1 1998 15–17
Western Ghats Wet evergreen/moist/dry 1 34 1996–1997 18
deciduous forests
UCSC FERP Mediterranean mixed 6 1 2007 19
evergreen forest
Shirakami Beech forest 1 2 2006 4, 5, 20
Oosting Hardwood forest 6.55 1 1989 21, 22
North Carolina forest plots Mixed hardwoods/pine 1.3–5.65 5 1990–1993§ 23–25
forest

Notes: 1Bradford et al. 2014, 2Kohyama et al. 2001, 3Kohyama et al. 2003, 4Lopez-­Gonzalez et al. 2009, 5Lopez-­Gonzalez et al.
2011, 6Baribault et al. 2011a, 7Baribault et al. 2011b, 8Pitman et al. 2005, 9Condit 1998a, 10Hubbell et al. 1999, 11Hubbell et al.
2005, 12DeWalt et al. 1999, 13Zimmerman et al. 1994, 14Thompson et al. 2002, 15Condit 1998b, 16Pyke et al. 2001, 17Condit et al.
2004 18Ramesh et al. 2010, 19Gilbert et al. 2010, 20Nakashizuka et al. 2003, 21Reed et al. 1993, 22Palmer et al. 2007, 23Peet and Chris-
tensen 1987, 24McDonald et al. 2002, 25Xi et al. 2008.
*We chose the most recent survey in each plot before documented disturbances.
†One plot has a more recent survey in 1998, however it lacks species ID.
‡We chose Census 2 because information for multiple stems is not available in Census 3, and the unit of diameter is unclear in
Census 4.
§We chose survey individually for each plot based on expert opinion to minimize the effect of hurricane disturbance.
1232 XIAO XIAO ET AL. Ecology, Vol. 97, No. 5

(Ernest et al. 2009). Predictions of the models in each respectively. Note that it is possible for the coefficient of
community were obtained with the variables S, N,
­ determination to be negative, which indicates that the pre-
and  EMETE = i D2i for ASNE, G, S, N, and EMETE for diction is worse than taking the geometric mean of the
∑
AGSNE, S, N, and ESSNT_N = i Di for SSNT_N, and ­observed values. Finally, we examined if the empirical pat-
∑
S, N, and ESSNT_M = i Di for SSNT_M.
∑ 2∕3
terns were significantly different from the models’ predic-
As an overall measure of model performance, we define tions by bootstrap analysis (Clauset et al. 2009, Connolly
the joint distribution P(n, D1, D2, …, Dn) as the probability et al. 2009, Xiao et al. 2015), where we generated random
that a species randomly selected from the community has samples from the predicted patterns and quantified their
abundance n, while individuals within the species have deviation from the predictions (predi’s) using both R2 and
diameter Di’s with i ranging from 1 to n. This distribution the Kolmogorov–Smirnov statistic, which were compared
combines all three macroecological patterns, where the with empirical deviations (Appendix S2).
SAD is the marginal distribution of n with Di’s integrated Python Code to fully replicate our analyses is
out from P(n, D1, D2, …, Dn), the ISD is the marginal ­deposited in the Dryad Digital Repository (doi:10.5061/
­distribution of Di, and the SDR is the expectation of the dryad.93ct6).
conditional distribution of Di given n. The form of this
joint distribution predicted by each of the four models
Results
is listed in Table 3 (see Appendix S1 for derivations).
We first compared the performance of the four models The log-­likelihoods of the joint distribution P(n, d1, d2,
using the likelihood of P(n, D1, D2, …, Dn) in each com- …, dn) of the SSNT models are higher than those of the
munity, then examined each of the three macroecological METE models in all 76 communities (Fig. 1), which
patterns individually. To quantify the predictive power ­implies that SSNT models consistently do a better job
of the models, we converted the SAD and the ISD into characterizing the overall community structure in the allo-
rank values, where the abundance of species or the diam- cations of individuals and of body size. Comparing
eter of individuals were ranked from the highest to the ­models under the same theoretical framework, the log-­
lowest, and the value at each rank was compared to the likelihood of AGSNE is higher than that of ASNE in all 76
models’ predictions. For example, for the SAD we com- communities, while the log-­ likelihood of SSNT_M is
pared the predicted versus observed abundances of the higher than SSNT_N in 59.
most abundant species in the community, the second Further examination of individual patterns show that
most abundant species, all the way down to the least the models predict nearly identical forms for the SAD (i.e.,
abundant species (Harte 2011, White et al. 2012, Xiao upper-­truncated log-­series in ASNE, near log-­series in
et al. 2015). For the SDR, we compared the observed AGSNE (Harte et al. 2015), untruncated log-­series in
average metabolic rate (D′2) within each species to those SSNT_N and SSNT_M; see Table 1), which not surpris-
expected from the models. The explanatory power of the ingly translates into equally good performance when eval-
models for each pattern was quantified using the coeffi- uated with empirical data (Fig. 2, first column). All four
cient of determination R2: models are also able to characterize the ISD reasonably
∑
[log10 (obsi ) − log10 (predi )]2 well with high predictive power (R2ASNE = 0.89,
R2 = 1 − i � �2 R2AGSNE = 0.90, R2SSNT_N = 0.86, R2SSNT_M = 0.96). Three of
∑ the models show systematic deviations for the largest indi-
i log10 (obsi ) − log10 (obsi )
viduals, however, with the two METE models tending to
over predict the size of the largest individuals and SSNT_N
where obsi and predi were the ith value of abundance or tending to under predict (Fig. 2, second column).
size (diameter for the ISD, metabolic rate for the SDR) in The discrepancy of the two sets of models lies mainly in
the observed and predicted ranked distributions, their predictions of the correlation between individual body

Table 3. Joint distribution P(n, D1, D2, …, Dn) for the four models. Z in AGSNE is a constant. See Table 1 for the interpretation
of the other symbols and parameters, and Appendix S1 for derivations.

Model Predicted joint distribution

n 2
1 2n𝜆2 Di e−𝜆2 nDi
e−(𝜆1 +𝜆2 )n
∏
ASNE
( )
PASNE n,D1 ,D2 , … ,Dn =
CASNE1 n i=1
e−𝜆2 n − e−𝜆2 nEMETE
n
) ( 2G )n ∏ t � � � 2
[ΦAGSNE (n)]1−n [1 − (S + 1) tS + StS+1 ] where t Di = e−(𝜆1 +𝜆2 n+𝜆3 nDi )
( ( )
AGSNE PAGSNE n,D1 ,D2 , … ,Dn = Di 2
ZS i=1 (1 − t)
n
1 𝜏1n ∏
SSNT_N 𝜏2 e−𝜏2 ⋅(Di −1)
( )
PSSNT_N n,D1 ,D2 , … ,Dn = − ( )
log 1 − 𝜏1 n i=1
n
1 𝜏1n ∏ 2 −𝜏3 (D2∕3
SSNT_M i −1) ⋅ D−1∕3
( )
PSSNT_N n,D1 ,D2 , … ,Dn = − ( ) ⋅ 𝜏3 e i
log 1 − 𝜏1 n i=1
3
May 2016 APPROACHES FOR MACROECOLOGY  1233

pattern is observed for the SDR predictions for the

METE models. In contrast, despite the low predictive
power, SDR predictions for the two SSNT models
suggest that the majority of the communities are
indistinguishable from random samples of the predicted
pattern (Figs. B3, B4). This implies that SSNT’s
prediction of no correlation between species abundance
and individual body size is more or less accurate.

Discussion
In this study, we compared the performance of METE
(Harte 2011) and SSNT (O’Dwyer et al. 2009), two of the
most comprehensive theories to date in macroecology,
­using two realized models for each. Both theories attempt
to unify multiple aspects of community structure under a
single theoretical framework, predicting patterns of biodi-
versity as well as patterns of energy consumption and
body size. Using data from 76 forest communities world-
wide, we showed that the two models of SSNT consistently
Fig. 1. Comparison of the log-­ likelihood (l) of the joint provide a better characterization of overall community
distribution P(n, d1, d2, …, dn) for the four models in each of the structure than the two models of METE (Fig. 1). This dis-
76 forest communities. l of AGSNE, METE_N, and METE_M
are compared with that of ASNE, which has the lowest parity results primarily from the ability of SSNT_N and
likelihood in all communities. The diagonal line is the one-to-­ SSNT_M to more accurately characterize the relationship
one line. For better visualization, l is transformed to −log(−l), between species abundance and body size distributions
which is a monotonic transformation that does not change the
position of the points with respect to the diagonal line. Note that within species, whereas the predictions of ASNE and
values of l depend on the number of individuals within the AGSNE on this relationship deviate from empirical pat-
community, and thus are not comparable across communities. terns (Newman et al. 2014, Xiao et al. 2015).
By comparing multiple competing models on multiple
size and species abundance. The two METE models both predictions simultaneously using an extensive set of data,
predict that individual body size depends on the species that our study achieves the strongest level of model evaluation
the individual belongs to – in ASNE average individual suggested by McGill et al. (2006), and provides insights
body size is negatively correlated with species abundance into the role of the underlying mechanisms of the theories.
(Harte 2011), whereas in AGSNE it is negatively correlated In METE, the macroecological patterns arise as the most
with both species abundance and number of species within likely state of the system assuming that the system is con-
genus (Harte et al. 2015). The SDR predicted by ASNE has strained by a small number of state variables. METE
been shown to be unrealistic in plant communities makes no explicit assumptions about ecological processes,
(Newman et al. 2014, Xiao et al. 2015), and our results show leaving their influence to operate indirectly through their
that AGSNE improves ASNE’s prediction only marginally potential effects on the values of the state variables. In
(R2ASNE = −2.11, R2AGSNE = −2.00). SSNT, on the other hand, SSNT, patterns emerge directly from the interactions of
predicts that there is no correlation between individual size the demographic processes including birth, death, and
and species characteristics, leading to better, but still far growth. The fact that SSNT performs better than METE
from good, agreement with empirical data for the SDR suggests that the demographic processes contain meaning-
(R2SSNT_N = 0.09, R2SSNT_M = 0.02). These results are robust ful information that helps to characterize the patterns, the
when the two models are examined in each of the 76 com- effect of which is not currently captured by the state-­
munities individually (Fig. 3) – the four models yield nearly variable based models.
identical R2 values for the SAD and comparable R2 values Although the differences between the models are
for the ISD across communities (with SSNT_M having the ­important, the fact that all four models are capable of ad-
highest predictive power on average), while the two SSNT equately characterizing the shapes of the SAD and the
models consistently outperforms the two METE models for ISD across a large number of communities with simple
the SDR. assumptions and limited inputs is impressive. Moreover,
The bootstrap analysis (Appendix S2) shows that the the maximum likelihood parameters for SSNT_N and
discrepancy between the models’ predictions and the SSNT_M are also fully determined by S, N, and E (see
observations for the ISD is almost ubiquitously higher Appendix S1), so that these variables serve as summary
than expected from random sampling in all four models. statistics for the demographic parameters. These results
This suggests that none of the models is able to fully imply that METE and SSNT contain overlapping infor-
capture the observed variation in the size distributions of mation. Although these demographic processes explain
individuals, despite their high predictive power. A similar a higher proportion of the variation in the empirical
1234 XIAO XIAO ET AL. Ecology, Vol. 97, No. 5

Fig. 2. Comparison of the performance of the four models for each of the three macroecological patterns. Each point in the
subplot represents the abundance of one species in a community for the SAD, the diameter of one individual in a community for the
ISD, and the average metabolic rate (squared diameter) within one species in a community for the SDR. The colors represent
density of the points, where warmer (redder) colors correspond to denser regions. The diagonal line represents the one-­to-­one line
between the predicted values and the observed values.
May 2016 APPROACHES FOR MACROECOLOGY  1235

Fig. 3. Comparison of R2 values for the three macroecological patterns predicted by the four models in each of the 76 forest
communities. R2 of AGSNE, METE_N, and METE_M are compared with that of ASNE. The diagonal line is the one-­to-­one line.

patterns, their effects are likely to at least be partially Blonder et al. 2014), while patterns that show spatial or
channeled through the constraints. taxonomical variation, such as the patterns of body size,
Our study is one step towards the goal of disentangling are more likely to be tied to ecological processes (Blonder
the effects of different mechanisms on macroecological et al. 2014).
patterns. Though we have adopted model comparison for One weakness prevalent across all four models is their
stronger inference, we do not advocate rejecting the theo- inability to characterize the SDR, the relationship
retical framework (METE) with the poorer fitting models between species abundance and the body size of individ-
or its underlying constraint-­based view as a potential uals within species, despite their success in inde-
­explanation for patterns. There are three reasons for be- pendently predicting the distribution of individuals
ing cautious about over interpreting these results. First, among species and the allocation of body size among
METE and SSNT are general theories built on first princi- individuals. Our results agree with previous studies
ples, whereas our conclusions are limited to their current showing that the SDR exhibits significant variation at
realized models based on specific assumptions. Models the local scale (Lawton 1990, Cotgreave 1993), not
under the same theoretical framework yield different pre- strongly abundance-­dependent as the METE models
dictions with different assumptions and inputs, which can predict. The prediction of the SSNT models that the
be evaluated with empirical data and improved with addi- SDR results from random draws is more in line with
tional information. This is demonstrated in our study by empirical observations (Appendix S2), but they too lack
comparing the two models from the same theory – with an predictive power (Fig. 2). Although part of the variation
additional constraint G, AGSNE has consistently higher may result from the limitation of data we used, e.g., spe-
likelihood than ASNE (Fig. 1), while incorporating infor- cies having different growth rates were surveyed at dif-
mation from the metabolic theory in SSNT_M eliminates ferent life stages, it could also indicate species-­specific
the systematic deviation in the predicted ISD (Fig. 2). size biases in resource use (White et al. 2007). One poten-
Future models will likely be developed with alternative tial remedy that may improve the predicted SDR as well
implementations leading to new and/or improved as lead to additional predictions is to take an approach
predictions. alternative to the two that we have addressed, and model
Second, our inference is limited by the scope of the macroecological patterns by directly stacking models of
data. Though the models have the potential to be applied individual species. This approach has shown promise in
to a wide variety of systems, we focused exclusively on predicting other patterns, such as biodiversity across
trees, where data of full surveys are readily available with space (Guisan and Rahbek 2011, D’Amen et al. 2015).
species identity and body size for all individuals. Similar models could potentially be developed to model
Although our results are consistent across forest commu- the abundance and body size of species based on their
nities of different types and sizes (Fig. 3), it remains to be traits, and to obtain the macroecological patterns from
seen if they can be generalized to other taxa. Previous the species-­level predictions. Such models will likely sac-
studies suggest that the size distributions predicted by rifice parsimony for accuracy, however, and require a lot
ASNE and AGSNE may be more accurate when applied more parameters than the models that we examined.
to invertebrates (Harte 2011, Harte et al. 2015). Third, Another potentially fruitful route to push the two the-
patterns that can be unified under the same theoretical ories forward is to unify the constraint-­and the process-­
framework do not necessarily have to arise from the same based approaches, which have generally been adopted by
underlying mechanism. Indeed, there is increasing evi- distinct theories but do not necessarily have to be mutu-
dence that the SAD is driven by statistical properties of ally exclusive. Results of our study imply that part of the
the system (White et al. 2012, Locey and White 2013, effects of the demographic processes propagate through
1236 XIAO XIAO ET AL. Ecology, Vol. 97, No. 5

the constraints, while other studies (e.g., Haegeman and Literature Cited
Etienne 2010) state that different configurations for the Baribault, T. W., R. K. Kobe and A. O. Finley. 2011a. Data
same set of constraints can often be tied to (and may from: tropical tree growth is correlated with soil phosphorus,
eventually be informed from) process-­based mechanistic potassium, and calcium, though not for legumes. Dryad
models. Indeed, an exciting new model is being developed Digital Repository http://dx.doi.org/10.5061/dryad.r9p7.
where the maximum entropy principle is combined with Baribault, T. W., R. K. Kobe, and A. O. Finley. 2011b. Tropical
demographic processes to characterize not only the tree growth is correlated with soil phosphorus, potassium,
and calcium, though not for legumes. Ecological Monographs
steady state but also temporal dynamics of a system 82:189–203.
(Umemura and Harte 2015). The attempts to model eco- Blonder, B., L. Sloat, B. J. Enquist, and B. McGill. 2014.
logical systems completely with constraints or processes Separating macroecological pattern and process: comparing
may thus represent two extremes of a continuous spec- ecological, economic, and geological systems. PLoS ONE
trum, along which multiple models exist that lean 9:e112850.
towards one approach or the other, yet all provide ade- Bradford, M. G., H. T. Murphy, A. J. Ford, D. Hogan, and
D. J. Metcalfe. 2014. Long-­term stem inventory data from
quate characterization of the system if properly formu-
tropical rain forest plots in Australia. Ecology 95:2362.
lated. We look forward to future studies that combine Brown, J. H. 1995. Macroecology. University Of Chicago Press,
new theoretical development with strong empirical tests Chicago, Illinois, USA.
to further elucidate the entangled effects of constraints Brown, J. H. 1999. Macroecology: progress and prospect. Oikos
versus biological processes in structuring ecological 87:3–14.
systems. Brown, J. H., J. F. Gillooly, A. P. Allen, V. M. Savage, and
G. B. West. 2004. Towards a metabolic theory of ecology.
Ecology 85:1771–1789.
Acknowledgments
Clauset, A., C. R. Shalizi, and M. E. J. Newman. 2009. Power-­
We thank B. Blonder and J. Harte for their constructive law distributions in empirical data. SIAM Review 51:661–703.
reviews, and members of the Weecology Lab for feedback on Condit, R. 1998a. Tropical forest census plots. Springer-Verlag
this research. R. K. Peet provided data for the North Carolina and R G. Landes Company, Berlin, Germany, and Georgetown,
forest plots. The Serimbu (provided by T. Kohyama) and Texas.
Shirakami (provided by T. Nakashizuka) datasets were obtained Condit, R. 1998b. Ecological implications of changes in drought
from the PlotNet Forest Database. The eno-­2 plot (provided patterns: shifts in forest composition in Panama. Climatic
by N. Pitman) and DeWalt Bolivia (provided by S. ­DeWalt) Change 39:413–427.
datasets where obtained from SALVIAS. The BCI forest Condit, R., S. Aguilar, A. Hernández, R. Pérez, S. Lao,
dynamics research project was made possible by National G. Angehr, S. P. Hubbell, and R. B. Foster. 2004. Tropical for-
­Science Foundation grants to S. P. Hubbell: DEB-­0640386, est dynamics across a rainfall gradient and the impact of an El
DEB-0425651, DEB-­0346488, DEB-­0129874, DEB-­00753 Niño dry season. Journal of Tropical Ecology 20:51–72.
102, DEB-9909347, DEB-­ 9615226, DEB-­ 9615226, DEB-­ Connolly, S. R., M. Dornelas, D. R. Bellwood, and T. P.
9405933, DEB-9221033, DEB-­9100058, DEB-­8906869, DEB-­ Hughes. 2009. Testing species abundance models: a new
8605042, DEB-­ 8206992, DEB-­ 7922197, support from the bootstrap approach applied to Indo-­ Pacific coral reefs.
Center for Tropical Forest Science, the Smithsonian Tropical Ecology 90:3138–3149.
Research Institute, the John D. and Catherine T. MacArthur Cotgreave, P. 1993. The relationship between body size and
Foundation, the Mellon Foundation, the Small World Insti- population abundance in animals. Trends in Ecology &
tute Fund, and numerous private individuals, and through Evolution 8:244–248.
the hard work of over 100 people from 10 countries over the D’Amen, M., A. Dubuis, R. F. Fernandes, J. Pottier,
past two decades. The UCSC Forest Ecology Research Plot L. Pellissier, and A. Guisan. 2015. Using species richness and
was made possible by National Science Foundation grants to functional traits predictions to constrain assemblage predic-
G. S. Gilbert (DEB-­0515520 and DEB-­084259), by the Pepper-­ tions from stacked species distribution models. Journal of
Giberson Chair Fund, the University of California, and the Biogeography 42:1255–1266.
hard work of dozens of UCSC students. These two projects are DeWalt, S. J., G. Bourdy, L. R. ChÁvez de Michel, and
part the Center for Tropical Forest Science, a global network of C. Quenevo. 1999. Ethnobotany of the Tacana: Quantitative
large-­scale demographic tree plots. The Luquillo Experimental inventories of two permanent plots of Northwestern Bolivia.
Forest Long-­Term Ecological Research Program was supported Economic Botany 53:237–260.
by grants BSR-­8811902, DEB 9411973, DEB 0080538, DEB Dewar, R. C., and A. Porté. 2008. Statistical mechanics unifies
0218039, DEB 0620910 and DEB 0963447 from NSF to the In- different ecological patterns. Journal of Theoretical Biology
stitute for Tropical Ecosystem Studies, University of Puerto Rico, 251:389–403.
and to the International Institute of Tropical Forestry USDA Enquist, B. J., and K. J. Niklas. 2001. Invariant scaling relations
Forest Service, as part of the Luquillo Long-­Term Ecological across tree-­dominated communities. Nature 410:655–660.
Research Program. Funds were contributed for the 2000 census Enquist, B. J., G. B. West, E. L. Charnov, and J. H. Brown.
by the Andrew Mellon foundation and by Center for Tropical 1999. Allometric scaling of production and life-­history varia-
Forest Science. The U.S. Forest Service (Dept. of Agriculture) tion in vascular plants. Nature 401:907–911.
and the University of Puerto Rico gave additional support. We Ernest, S. K. M. 2005. Body size, energy use, and community
also thank the many technicians, volunteers and interns who structure of small mammals. Ecology 86:1407–1413.
have contributed to data collection in the field. This research Ernest, S. K. M., E. P. White, and J. H. Brown. 2009. Changes
was supported by a CAREER award from the U.S. National in a tropical forest support metabolic zero-­sum dynamics.
Science Foundation (DEB-­0953694) and by the Gordon and Ecology Letters 12:507–515.
Betty Moore Foundation’s Data-­Driven Discovery Initiative Fisher, R. A., A. S. Corbet, and C. B. Williams. 1943. The rela-
through Grant GBMF4563, both to E. P. White. tion between the number of species and the number of
May 2016 APPROACHES FOR MACROECOLOGY  1237

individuals in a random sample of an animal population. Marquet, P. A., et al. 2014. On Theory in Ecology. BioScience
Journal of Animal Ecology 12:42–58. 64:701–710.
Frank, S. A. 2014. Generative models versus underlying symme- McDonald, R. I., R. K. Peet, and D. L. Urban. 2002.
tries to explain biological pattern. Journal of Evolutionary Environmental correlates of aak decline and red maple
Biology 27:1172–1178. ­increase in the North Carolina piedmont. Castanea 67:84–95.
Gilbert, G. S., E. Howard, B. Ayala-Orozco, M. Bonilla- McGill, B. J. 2011. Linking biodiversity patterns by autocorrelated
Moheno, J. Cummings, S. Langridge, I. M. Parker, J. Pasari, random sampling. American Journal of Botany 98:481–502.
D. Schweizer, and S. Swope. 2010. Beyond the tropics: forest McGill, B. J., and J. C. Nekola. 2010. Mechanisms in macroeco-
structure in a temperate forest mapped plot. Journal of logy: AWOL or purloined letter? Towards a pragmatic view
Vegetation Science 21:388–405. of mechanism. Oikos 119:591–603.
Guisan, A., and C. Rahbek. 2011. SESAM -­a new framework McGill, B. J., et al. 2007. Species abundance distributions: mov-
integrating macroecological and species distribution models ing beyond single prediction theories to integration within an
for predicting spatio-­ temporal patterns of species assem- ecological framework. Ecology Letters 10:995–1015.
blages. Journal of Biogeography 38:1433–1444. McGill, B. J., B. A. Maurer, and M. D. Weiser. 2006. Empirical
Haegeman, B., and R. S. Etienne. 2010. Entropy maximization evaluation of neutral theory. Ecology 87:1411–1423.
and the spatial distribution of species. The American Muller-Landau, H. C., et al. 2006a. Testing metabolic ecology
Naturalist 175:E74–E90. theory for allometric scaling of tree size, growth and mortal-
Harte, J. 2011. Maximum entropy and ecology: a theory of ity in tropical forests. Ecology letters 9:575–588.
abundance, distribution, and energetics. Oxford University Muller-Landau, H. C., et al. 2006b. Comparing tropical forest
Press, Oxford, UK. tree size distributions with the predictions of metabolic ecol-
Harte, J., T. Zillio, E. Conlisk, and A. B. Smith. 2008. Maximum ogy and equilibrium models. Ecology Letters 9:589–602.
entropy and the state-­variable approach to macroecology. Nakashizuka, T., et al. 2003. Monitoring dynamics of beech for-
Ecology 89:2700–2711. ests with different structure in Shirakami Mountains. Tohoku
Harte, J., A. Rominger and W. Zhang. 2015. Integrating macro- Journal of Forest Science 8:67–74.
ecological metrics and community taxonomic structure. Newman, E. A., M. E. Harte, N. Lowell, M. Wilber and
Ecology Letters 18:1068–1077. J. Harte. 2014. Empirical tests of within-­and across-­species
Hubbell, S. P. 2001. The unified neutral theory of biodiversity energetics in a diverse plant community. Ecology
and biogeography. Princeton University Press. 95:2815–2825.
Hubbell, S. P., R. B. Foster, S. T. O’Brien, K. E. Harms, O’Dwyer, J. P., J. K. Lake, A. Ostling, V. M. Savage, and
R. Condit, B. Wechsler, S. J. Wright, and S. Loo de Lao. J. L. Green. 2009. An integrative framework for stochastic,
1999. Light-­ gap disturbances, recruitment limitation, and size-­structured community assembly. Proceedings of the
tree diversity in a neotropical forest. Science 283:554–557. National Academy of Sciences of the United States of
Hubbell, S. P., R. Condit and R. B. Foster. 2005. Barro America 106:6170–6175.
Colorado forest census plot data. Palmer, M. W., R. K. Peet, R. A. Reed, W. Xi, and P. S. White.
Jaynes, E. T. 2003. Probability theory: the logic of science. 2007. A multiscale study of vascular plants in a North Carolia
Cambridge University Press, Cambridge, UK. Piedmont forest. Ecology 88:2674.
Kohyama, T., E. Suzuki, T. Partomihardjo, and T. Yamada. Peet, R. K., and N. L. Christensen. 1987. Competition and tree
2001. Dynamic steady state of patch-­mosaic tree size struc- death. BioScience 37:586–595.
ture of a mixed dipterocarp forest regulated by local crowd- Pitman, N. C. A., C. E. Cerón, C. I. Reyes, M. Thurber, and
ing. Ecological Research 16:85–98. J. Arellano. 2005. Catastrophic natural origin of a species-­
Kohyama, T., E. Suzuki, T. Partomihardjo, T. Yamada, and poor tree community in the world’s richest forest. Journal of
T. Kubo. 2003. Tree species differentiation in growth, recruit- Tropical Ecology 21:559–568.
ment and allometry in relation to maximum height in a Pueyo, S., F. He, and T. Zillio. 2007. The maximum entropy
Bornean mixed dipterocarp forest. Journal of Ecology formalism and the idiosyncratic theory of biodiversity.
91:797–806. Ecology Letters 10:1017–1028.
Lawton, J. H. 1990. Species richness and population dynamics Pyke, C. R., R. Condit, S. Aguilar, and S. Lao. 2001. Floristic
of animal assemblages. Patterns in body size: abundance composition across a climatic gradient in a neotropical low-
space. Philosophical Transactions of the Royal Society B: land forest. Journal of Vegetation Science 12:553–566.
Biological Sciences 330:283–291. Ramesh, B. R., M. H. Swaminath, S. V. Patil, R. Pélissier, P. D.
Locey, K. J., and E. P. White. 2013. How species richness and Venugopal, S. Aravajy, C. Elouard, and S. Ramalingam.
total abundance constrain the distribution of abundance. 2010. Forest stand structure and composition in 96 sites along
Ecology Letters 16:1177–1185. environmental gradients in the central Western Ghats of
Lopez-Gonzalez, G., S. L. Lewis, M. Burkitt, T. R. Baker and India. Ecology 91:3118.
O. L. Phillips. 2009. ForestPlots.net Database. www.forest- Reed, R. A., R. K. Peet, M. W. Palmer, and P. S. White. 1993.
plots.net. Date of extraction [06, 07, 2012]. Scale dependence of vegetation-­environment correlations: a
Lopez-Gonzalez, G., S. L. Lewis, M. Burkitt, and O. L. Phillips. case study of a North Carolina piedmont woodland. Journal
2011. ForestPlots.net: a web application and research tool to of Vegetation Science 4:329–340.
manage and analyse tropical forest plot data. Journal of Shipley, B., D. Vile, and É. Garnier. 2006. From plant traits to
Vegetation Science 22:610–613. plant communities: a statistical mechanistic approach to bio-
MacArthur, R. H. and E. O. Wilson. 1967. The Theory of diversity. Science 314:812–814.
Island Biogeography. Princeton University Press, Princeton, Thibault, K. M., E. P. White, A. H. Hurlbert, and S. K. M.
New Jersey, USA. Ernest. 2011. Multimodality in the individual size distribu-
Marquet, P. A., J. A. Keymer and H. Cofre. 2003. Breaking the tions of bird communities. Global Ecology and Biogeography
stick in space: of niche models, metacommunities and pat- 20:145–153.
terns in the relative abundance of species. Pages 64–86 in Thompson, J., N. Brokaw, J. K. Zimmerman, R. B. Waide,
T. M. Blackburn, and K. J. Gaston, editors. Macroecology: E. M. Everham, D. J. Lodge, C. M. Taylor, D. García-
concepts and Consequences. Blackwell Science, Oxford. Montiel, and M. Fluet. 2002. Land use history, environment,
1238 XIAO XIAO ET AL. Ecology, Vol. 97, No. 5

and tree composition in a tropical forest. Ecological using a simple maximum entropy model. Ecology
Applications 12:1344–1363. 93:1772–1778.
Umemura, K. and J. Harte. 2015. Dynamic extension of the Xi, W., R. K. Peet, J. K. Decoster, and D. L. Urban. 2008.
maximum entropy theory of ecology (METE) using a master Tree damage risk factors associated with large, infrequent
equation approach. Poster presented at the 100th Ecological wind disturbances of Carolina forests. Forestry 81:
Society of America Annual Meeting. 317–334.
West, G. B., J. H. Brown, and B. J. Enquist. 1999. A general Xiao, X., D. J. McGlinn, and E. P. White. 2015. A strong test of
model for the structure and allometry of plant vascular sys- the Maximum Entropy Theory of Ecology. The American
tems. Nature 400:664–667. Naturalist 185:E70–E80.
White, E. P., S. K. M. Ernest, A. J. Kerkhoff, and B. J. Enquist. Zimmerman, J. K., E. M. III Everham, R. B. Waide,
2007. Relationships between body size and abundance in D. J. Lodge, C. M. Taylor, and N. V. L. Brokaw. 1994.
ecology. Trends in Ecology & Evolution 22:323–330. Responses of tree species to hurricane winds in subtropical
White, E. P., K. M. Thibault, and X. Xiao. 2012. Characterizing wet forest in Puerto Rico: Implications for tropical tree life
species abundance distributions across taxa and ecosystems histories. The Journal of Ecology 82:911–922.

Supporting Information
Additional supporting information may be found in the online version of this article at http://onlinelibrary.wiley.com/
doi/10.1890/15-0962.1/suppinfo