scispace - formally typeset

Journal ArticleDOI

A length-based hierarchical model of brown trout (Salmo trutta fario) growth and production

01 Jan 2012-Biometrical Journal (WILEY‐VCH Verlag)-Vol. 54, Iss: 1, pp 108-126

TL;DR: A hierarchical Bayesian model is presented to estimate the growth parameters, production, and production over biomass ratio (P/B) of resident brown trout populations by investigating the growth and production of a brown trout population by using data collected in the field from 2005 to 2010.

AbstractWe present a hierarchical Bayesian model (HBM) to estimate the growth parameters, production, and production over biomass ratio (P/B) of resident brown trout (Salmo trutta fario) populations. The data which are required to run the model are removal sampling and air temperature data which are conveniently gathered by freshwater biologists. The model is the combination of eight submodels: abundance, weight, biomass, growth, growth rate, time of emergence, water temperature, and production. Abundance is modeled as a mixture of Gaussian cohorts; cohorts centers and standard deviations are related by a von Bertalanffy growth function; time of emergence and growth rate are functions of water temperature; water temperature is predicted from air temperature; biomass, production, and P/B are subsequently computed. We illustrate the capabilities of the model by investigating the growth and production of a brown trout population (Neste d'Oueil, Pyrenees, France) by using data collected in the field from 2005 to 2010.

Topics: Brown trout (64%), Salmo (56%), Population (52%)

Summary (1 min read)

Jump to: [Introduction][Emergence][Weight][Abundance] and [Conflict of interest]

Introduction

  • Abundance is modeled as a mixture of Gaussian cohorts; cohorts centers and standard deviations are related by a von Bertalanffy growth function; time of emergence and growth rate are functions of water temperature; water temperature is predicted from air temperature; biomass, production, and P/B are subsequently computed.
  • The trend is to construct such statistical models within a Bayesian framework (Congdon, 2006).
  • The authors primary objective is to provide a layout to compute growth parameters of brown trout populations by using accessible data (namely removal sampling and air temperature data).
  • Time scale is daily, spans from January 1st of the year of the first campaign to December 31st of the year of the last campaign, with a total of T days.

Emergence

  • Cardinal temperatures are minimum (ymin), optimum (yopt), and maximum (ymax) temperatures required for growth as well as minimum (y0) and optimum (y1) temperatures required for hatching.
  • CE50 is the critical value leading to the emergence of 50% of the fry.
  • See text for values of multidimensionnal parameters (Li; to; tovio;k).
  • Are illustrated in Fig. 1: abundance and growth submodels depend on common parameters (mo;k and so;k, defined later), growth depends on the time of emergence and growth rate, these quantities further depend on the water temperature, fish biomass is the cross-product of fish weight and abundance, and the combination of growth and biomass parameters lead to production.

Weight

  • Parameters are shape and rate for gamma distributions, expectation and variance for normal and lognormal distributions, and boundaries for uniform distributions.
  • Units which are provided with precisions (e.g. 1=s2l) are units of respective standard deviations (e.g. sl).
  • Standard deviations are related to random errors across campaigns (sl, st, sa, sb, sZ, sz), among individuals (sN, s0), and residual (sm, sT, sW).

Abundance

  • The abundance submodel is briefly presented and is more thoroughly investigated by Ruiz and Laplanche (2010).
  • The motivation to include this additional error term in the model is illustrated and discussed later.
  • Model alternatives are defined whether cohort standard deviations so;k (M1 4) and centers mo;k (M1 3) are constrained with a VBGF and whether growth rate Gt (M1 2) and date of emergence temo;k (M1) are temperature-dependent.
  • Nevertheless, in the aim of illustrating the modeling of temperature-dependent time of emergence and growth rate, following results are computed by using baseline (M1).
  • The model at the current state applies to riverine brown trout (S. trutta fario).

Conflict of interest

  • The authors have declared no conflict of interest.
  • Growth with seasonally varying temperatures – an expansion of the von Bertalanffy growth model.

Did you find this useful? Give us your feedback

...read more

Content maybe subject to copyright    Report

To link to this article: DOI:10.1002/bimj.201100083
http://dx.doi.org/10.1002/bimj.201100083
This is an author-deposited version published in: http://oatao.univ-toulouse.fr/
Eprints ID: 5610
To cite this version: Lecompte, Jean-Baptiste and Laplanche, Christophe A
length-based hierarchical model of brown trout (Salmo trutta fario) growth and
production. (2012) BiometricalJournal, vol. 54 (n°1). pp. 108-126. ISSN 0323-
3847
Open Archive Toulouse Archive Ouverte (OATAO)
OATAO is an open access repository that collects the work of Toulouse researchers an
d
makes it freely available over the web where possible.
Any correspondence concerning this service should be sent to the repository
administrator: staff-oatao@inp-toulouse.fr

A length-based hierarchical model of brown trout (Salmo trutta
fario) growth and production
Jean-Baptiste Lecomte
1,2
and Christophe Laplanche
,1,2
1
Universite
´
de Toulouse; INP, UPS, EcoLab (Laboratoire Ecologie Fonctionnelle
et Environnement); ENSAT, Avenue de l’Agrobiopole, 31326 Castanet Tolosan, France
2
CNRS, EcoLab, 31326 Castanet Tolosan, France
We present a hierarchical Bayesian model (HBM) to estimate the growth parameters, production,
and production over biomass ratio (P/B) of resident brown trout (Salmo trutta fario ) populations.
The data which are required to run the model are removal sampling and air temperature data which
are conveniently gathered by freshwater biologists. The model is the combination of eight submodels:
abundance, weight, biomass, growth, growth rate, time of emergence, water temperature, and pro-
duction. Abundance is modeled as a mixture of Gaussian cohorts; cohorts centers and standard
deviations are related by a von Bertalanffy growth function; time of emergence and growth rate are
functions of water temperature; water temperature is predicted from air temperature; biomass,
production, and P/B are subsequently computed. We illustrate the capabilities of the model by
investigating the growth and production of a brown trout population (Neste d’Oueil, Pyre
´
ne
´
es,
France) by using data collected in the field from 2005 to 2010.
Keywords: Growth; Hierarchical Bayesian model; Production; Removal sampling; Salmo
trutta.
1 Introduction
In order to evaluate, compare, and predict the status of fish populations, freshwater biologists have
considered numerous descriptors of fish populations. Such variables can characterize fish stocks
(Pauly and Moreau 1997; Kwak and Waters 1997; Ruiz and Laplanche 2010) durati on and rates of
success through life-history stages (Hutchings 2002, Klemetsen et al., 2003), phenotypic or geno-
typic traits (Ward, 2002; Shinn, 2010). Descriptors of fish populations can be examined separately,
depending on the population aspect under focus. For instance, the impact of surface water con-
tamination by pesticides on fish can be investigated by assessing DNA da mage to blood cells
(Polard et al., 2011). Aquatic resource management would rather focus on fish stock variables, such
as abundance (number of fish per unit area of stream), biomass (mass of fish per unit area),
production (mass of fish produced per unit area per unit time), or production over biomass ratio
P/B (Pauly and Moreau, 1997). The drawback of examining a single population variable is its
limited, descriptive prospect. A countermeasure would be to relate a population variable to cov-
ariates, e.g. environm ental variables or descri ptors of coexisting populations. As an illustration,
growth of salmonids has been related to environmental factors such as temperature (Mallet et al.,
*
Corresponding author: e-mail: laplanche@gmail.com, Tel: 133-534-323-973, Fax: 133-534-323-901

1999) and stream flow (Jensen and Johnsen 1999; Daufresne and Renault, 2006). Descriptors of a
population can also be examined jointly. The examination of multiple variables has a more potent,
explicative prospect and could improve our understanding of population dynamics (Nordwall et al.
2001).
Brown trout ( Salmo trutta) is indigenous to Eurasia. Brown trout has been introduced to non-
Eurasian freshwaters for fishing purposes. It can either grow in oceans and migrate to freshwaters
for reproduction (S. trutta trutta), or live in lakes (S. trutta lacustris), or be stream-resident (S. trutta
fario). As a result of such adaptation capabilities, brown trout has successfully colonized fresh-
waters to a world-wide distribution (Elliott, 1994; Klemetsen et al., 2003). Although ecologically
variable, brown trout is demanding in terms of habitat and water quality. As a result, brown trout is
a relevant bioindicator of the quality of freshwaters at a global scale (Lagadic et al 1998; Wood
2007). Moreover, a fundamental environmental variable driving brown trout life history is tem-
perature (Jonsson and Jonsson, 2009), hence using brown trout as a bioindicator of climate change.
Temperature affects grow th (as aforementioned) as well as life-stage timing (Webb and McLay,
1996; Armstrong et al., 2003). Key life-stages of brown trout are egg laying (oviposition), hatching
of larvae, emergence of fry, reproduction of adults, and death. In our case, we will focus on the
effect of temperature on growth and on time between oviposition and emergence of riverine brown
trout (S. trutta fario). We will also consider several variables characterizing stocks (abundance,
biomass, production).
Abundance of riverine fish species is conveniently assessed through removal sampling: (i) a reach is
spatially delimited (later referred to as a stream section), (ii) fractions of fish are successively removed
of the section by electrofishing and counted, (iii) captured fish are released altogether (Lobo
´
n-Cervia
´
,
1991). Abundance is typically computed by using the method suggested by Carle and Strub (1978): the
main statistical assumption was that the probability of capturing fish (referred to in the following as
catchability) would be equal for all fish. Some contributions have shown, however, that the use of
more advanced statistical models is recommendable in the aim of lowering estimation bias (Peterson
et al., 2004; Riley and Fausch, 1992). The trend is to construct such statistical models within a
Bayesian framework (Congdon, 2006). Recent hierarchical Bayesian models (HBMs) relate abun-
dance to environmental covariates (Rivot et al., 2008; Ebersole et al., 2009), include heterogeneity of
the catchability (Ma
¨
ntyniemi et al., 2005; Do-razio et al., 2005; Ruiz and Laplanche, 2010), and can
handle multiple sampling stream sections (Wyatt, 2002; Webster et al., 2008; Laplanche, 2010). The
reason of popularity of HBMs over the last decade is their ability to handle complex relationships
(multi-level, non-linear, mixed-effect) between variables with heterogeneous sources (relationships,
data, priors) of knowledge. Freshwater biologists can statistically relate multiple descriptors of fish
populations together with covariates within a single HBM framework.
We present an HBM of riverine brown trout growth and production. Our primary objective is to
provide a layout to compute growth parameters of brown trout populations by using accessible data
(namely removal sampling and air temperature data). Our second objective is to use such a layout to
compute interval estimates of brown trout production. To fulfill the first objective, we extend the
abundance model of Ruiz and Laplanche (2010) with a growth module. The abundance model
performs a multimodal decomposition of length-abundance plots. Our growth module constrains
parameters of the multimodal decomposition with a growth function. The interest is twofold: to
guide the decomposition of length-abundance plots with a growth function and to use length-
abundance plots to estimate growth parameters. Ruiz and Laplanche (2010) also created a module
which computes fish biomass. To fulfill the second objective, we extend their biomass module with a
production model. The overall model is the combination of eight HBMs, three (abundance, weight,
biomass) related to the contribution of Ruiz and Laplanche (2010) and five to growth and pro-
duction (growth, growth rate, emergence, temperature, production), which are presented succes-
sively. We show the capability of the overall model to estimate the growth parameters and the
production of a brown trout population with a data set collected in the field. We discuss possible
extensions of the current model.

2 Materials and methods
2.1 Notations and measured variables
A single section (of area A,m
2
) of a stream populated with S. trutta far io is sampled by electrofishing.
Electrofishing is spread over several years in a series of O campaigns with J
o
removals per campaign
(index over campaigns is o 2f1; ...; Og). The number of campaign(s) per year as well as the number of
removal(s) per campaign can be variable. Time scale is daily, spans from January 1st of the year of the
first campaign to December 31st of the year of the last campaign, with a total of T days. Times of
campaigns are noted t
o
(day). Let C
o;j
be the number of fish caught during removal j 2f1; ...; J
o
g at
time t
o
and C
o
¼
P
j
C
o;j
be the total number of fish caught at time t
o
.Thelengthandweightofthe
caught fish are measured and are noted L
o;j;f
(mm) and W
o;j;f
(g) (f 2f1; ...; C
o;j
g). As done by Ruiz and
Laplanche (2010), fish are grouped by length class: D
l
(mm) is the length class width, I is the number of
length classes, [(i 1)D
l
, iD
l
[ are the classes (with [a,b[ denoting an interval including the lower limit a
and excluding the upper limit b), and L
i
5 (i1/2) D
l
are the class centers. Given class centers L
i
and fish
lengths L
o;j;f
, the number of fish of length class i caught during removal j at time t
o
can be computed and
is labeled C
o;i;j
(i 2f1; ...; Ig). The air and water temperatures on day t are noted y
a
t
and y
w
t
, respec-
tively (t 2f1; ...; T g).ThemeasuredvariablesoftheHBMarefishlengthandweight(L
o;j;f
and W
o;j;f
),
catch (C
o;i;j
), stream section area (A),airandwatertemperatures(y
a
t
and y
w
t
). Constant (known)
parameters, free (unknown) parameters of interest, and remaining (unknown) nuisance parameters are
provided together in Tables 1–3, respectively.
2.2 Model structure
As briefly introduced, the overall HBM is the combination of eight submodels. Submodels are
interconnected as a consequence of sharing subsets of parameters, connections between submodels
Table 1 Values of constant parameters.
Parameter Notation Value Unit
Abundance
Area A 574.5 m
2
Length class center L
i
–mm
Growth
Date of campaign t
o
day
Growth rate
Cardinal temperature y
min
3.6 1C
Cardinal temperature y
max
19.5 1C
Cardinal temperature y
opt
13.1 1C
Emergence
Cardinal temperature y
0
2.8 1C
Cardinal temperature y
1
22.4 1C
Date of oviposition t
ovi
o;k
day
Critical value CE
50
76.2 day
Cardinal temperatures are minimum (y
min
), optimum (y
opt
), and maximum (y
max
) temperatures required for growth as well as
minimum (y
0
) and optimum (y
1
) temperatures required for hatching. CE
50
is the critical value leading to the emergence of
50% of the fry. See text for values of multidimensionnal parameters (L
i
; t
o
; t
ovi
o;k
).

are illustrated in Fig. 1: abundance and growth submodels depend on common parameters (m
o;k
and
s
o;k
, defined later), growth depends on the time of emergence and growth rate, these quantities
further depend on the water temperature, fish biomass is the cross-product of fish weight and
abundance, and the combination of growth and biomass parameters lead to production. Submode ls
are also connected to sub sets of measured variables: Fish weight (W
o;j;f
) is predicted from fish length
(L
o;j;f
), water temperature (y
w
t
) is predicted from air tempe rature (y
a
t
), and abundance is related to
removal sampling catch (C
o;i;j
) plus stream section area (A). The model is structured into five levels :
campaign (o 2f1; ...; Og), day (t 2f1; ...; Tg), length class (i 2f1; ...; Ig), removal (j 2f1; ...; J
o
g),
plus an additional level, cohort (k 2f1; ...; Kg), which is defined later. The temperature submodel is
dealt with in Appendix A (Supporting Infomation), remai ning submodels are successively presented
below.
Table 2 Distributions of free parameters (priors).
Parameter Notation Prior Unit
Abundance
Abundance l Gamma(0.001,0.001) per m
2
Cohort proportion t
0
k
Uniform(0,1)
Slope a Normal(0,1000) per mm
Intercept b Normal(0,1000)
Precision 1=s
2
l
Gamma(0.001,0.001) per m
2
Precision 1=s
2
t
Gamma(0.001,0.001)
Precision 1=s
2
a
Gamma(0.001,0.001) per mm
Precision 1=s
2
b
Gamma(0.001,0.001)
Growth
Optimal growth rate G
opt
Lognormal(7.25, 0.30) per day
Asymptotic length L
N
Lognormal(6.23,0.30 ) mm
Length on t
em
1;1
m
0
k
jL
1
Eq. (9) mm
Precision 1=s
2
1
Gamma(0.001,0.001) mm
Precision 1=s
2
0
Gamma(0.001,0.001) mm
Precision 1=s
2
m
Gamma(0.001,0.001) mm
Temperature
Maximum water temperature a
y
Normal(0,1000) 1C
Inflection b
y
Normal(0,1000) 1C
Inflection g
y
Normal(0,1000) per 1C
Minimum water temperature m
y
Normal(0,1000) 1C
Precision 1=s
2
y
Gamma(0.001,0.001) 1C
Weight
Allometric Z Gamma(0.001,0.001) g/L
Allometric z Normal(0,1000)
Precision 1=s
2
Z
Gamma(0.001,0.001) g/L
Precision 1=s
2
z
Gamma(0.001,0.001)
Precision 1=s
2
W
Gamma(0.001,0.001)
Parameters are shape and rate for gamma distributions, expectation and variance for normal and lognormal distributions,
and boundaries for uniform distributions. Units which are provided with precisions (e.g. 1=s
2
l
) are units of respective
standard deviations (e.g. s
l
). Standard deviations are related to random errors across campaigns (s
l
, s
t
, s
a
, s
b
, s
Z
, s
z
),
among individuals (s
N
, s
0
), and residual (s
m
, s
T
, s
W
).

Figures (8)
Citations
More filters

Journal ArticleDOI
Abstract: We present a Bayesian hierarchical model to estimate the abundance and the biomass of brown trout (Salmo trutta fario) by using removal sampling and biometric data collected at several stream sections. The model accounts for (i) variability of the abundance with fish length (as a distribution mixture), (ii) spatial variability of the abundance, (iii) variability of the catchability with fish length (as a logit regression model), (iv) spatial variability of the catchability, and (v) residual variability of the catchability with fish. Model measured variables are the areas of the stream sections as well as the length and the weight of the caught fish. We first test the model by using a simulated dataset before using a 3-location, 2-removal sampling dataset collected in the field. Fifteen model alternatives are compared with an index of complexity and fit by using the field dataset. The selected model accounts for variability of the abundance with fish length and stream section and variability of the catchability with fish length. By using the selected model, 95% credible interval estimates of the abundances at the three stream sections are (0.46,0.59), (0.90,1.07), and (0.56,0.69) fish/m2. Respective biomass estimates are (9.68, 13.58), (17.22, 22.71), and (12.69, 17.31) g/m2.

15 citations


Journal ArticleDOI
TL;DR: This work developed density-environment relationships for three different life stages of brown trout that show the limiting effects of hydromorphological variables at habitat scale and proposed a new approach for the identification of optimal flows using the limiting factor approach and the evaluation of basic ecological relationships.
Abstract: The correct definition of ecosystem needs is essential in order to guide policy and management strategies to optimize the increasing use of freshwater by human activities. Commonly, the assessment of the optimal or minimum flow rates needed to preserve ecosystem functionality has been done by habitat-based models that define a relationship between in-stream flow and habitat availability for various species of fish. We propose a new approach for the identification of optimal flows using the limiting factor approach and the evaluation of basic ecological relationships, considering the appropriate spatial scale for different organisms. We developed density-environment relationships for three different life stages of brown trout that show the limiting effects of hydromorphological variables at habitat scale. In our analyses, we found that the factors limiting the densities of trout were water velocity, substrate characteristics and refugia availability. For all the life stages, the selected models considered simultaneously two variables and implied that higher velocities provided a less suitable habitat, regardless of other physical characteristics and with different patterns. We used these relationships within habitat based models in order to select a range of flows that preserve most of the physical habitat for all the life stages. We also estimated the effect of varying discharge flows on macroinvertebrate biomass and used the obtained results to identify an optimal flow maximizing habitat and prey availability.

13 citations


Journal ArticleDOI
TL;DR: A hierarchical Bayesian model for the resident brown trout life cycle was built to assess the relative influence of local and general determinants of mortality, and could help to predict monthly juvenile and adult mortality under scenarios of global warming and changes in shel...
Abstract: Successful management and protection of wild animal populations relies on good understanding of their life cycles. Because population dynamics depends on intricate interactions of biological and ecological processes at various scales, new approaches are needed that account for the variability of demographic processes and associated parameters in a hierarchy of spatial scales. A hierarchical Bayesian model for the resident brown trout (Salmo trutta) life cycle was built to assess the relative influence of local and general determinants of mortality. The model was fitted to an extensive data set collected in 40 river reaches, combining abundance and environmental data (hydraulics, water temperature). Density-dependent mortality of juveniles increased at low water temperatures and decreased with shelter availability. High water temperature increased density-dependent mortality in adults. The model could help to predict monthly juvenile and adult mortality under scenarios of global warming and changes in shel...

6 citations


Journal ArticleDOI
Abstract: Growth is a fundamental ecological process of stream-dwelling salmonids which is strongly interrelated to critical life history events (emergence, mortality, sexual maturity, smolting, spawning). The ability to accurately model growth becomes critical when making population predictions over large temporal (multi-decadal) and spatial (meso) scales, e.g., investigating the effect of global change. Body length collection by removal sampling is a widely-used practice for monitoring fish populations over such large scales. Such data can be efficiently integrated into a Hierarchical Bayesian Model (HBM) and lead to interesting findings on fish dynamics. We illustrate this approach by presenting an integrated HBM of brown trout (Salmo trutta) growth, population dynamics, and removal sampling data collection processes using large temporal and spatial scales data (20 years; 48 sites placed along a 100 km latitudinal gradient). Growth and population dynamics are modelled by ordinary differential equations with parameters bound together in a hierarchical structure. The observation process is modelled with a combination of a Poisson error, a binomial error, and a mixture of Gaussian distributions. Absolute fit is measured using posterior predictive checks, which results indicate that our model fits the data well. Results indicate that growth rate is positively correlated to catchment area. This result corroborates those of other studies (laboratory, exploratory) that identified factors besides water temperature that are related to daily ration and have a significant effect on stream-dwelling salmonid growth at a large scale. Our study also illustrates the value of integrated HBM and electrofishing removal sampling data to study in situ fish populations over large scales.

5 citations


19 Feb 2015
Abstract: In Italy, the Water Framework Directive introduced major changes to water management, by making water resources more efficient and enforceable, so, this study aims to highlight some of the potential implications of its implementation for lake management. In this respect, the Life+ INHABIT project was crucial in launching the monitoring plan of lake macroinvertebrates, standardized at national level. Quantile regression analysis was focused on nine lakes located in two Italian regions (Piedmont and Sardinia). The studied lakes cover a wide trophic spectrum, from oligotrophy to hyper-trophy. In particular, all Sardinians lakes had higher trophic conditions than Piedmontese lakes, reaching only a meso-eutrophic state. The lakes were sampled through the national standardised protocol taking three replicates for each of the three recognized lake zones. The study lakes showed high chemical variability with conductivity varying between 53 and 561 μS/cm, pH between 6.5 and 9.1, and alkalinity between 14 and 398 mg/L. The bottom sediments were characterized by fine sand (range 5199%), followed by silt (0.6-35%) and clay (0 – 28%). Lake Habitat Survey was also applied and the results of its synthetic Indices highlighted higher values in natural lakes and lower in the reservoirs. In all the lakes, macroinvertebrates mainly consisted of Dipteran chironomids and oligochaetes characterized by relative abundance up to 80% and >90%, respectively. In order to link macroinvertebrate community characteristics to their habitat we measured twenty-one variables that represent chemical, physical and morphological characteristics of the 2. Quantile regression as predictive for lake macroinvertebrate biodiversity __________________________________________________________ 10 environment and for each variable the range of variation was provided to clarify the domain of application. Using quantile regression to evaluate limiting responses, we selected sampling depth and oxygen percent saturation (oxygen content) as the two variables that best explain the number of taxa present and the diversity of a site. The authors also stress the importance of the combination of selected variables in structuring lacustrine macroinvertebrate communities. We provide models able to predict the potential of community diversity as a function of environmental characteristics. All of this evidence, can help water managers in deciding to initially invest resources in those lakes where the biological communities were expected to be impaired and to identify which lakes could be in pristine or in near-pristine conditions.

2 citations


References
More filters

Book
19 Jun 2013
TL;DR: The second edition of this book is unique in that it focuses on methods for making formal statistical inference from all the models in an a priori set (Multi-Model Inference).
Abstract: Introduction * Information and Likelihood Theory: A Basis for Model Selection and Inference * Basic Use of the Information-Theoretic Approach * Formal Inference From More Than One Model: Multi-Model Inference (MMI) * Monte Carlo Insights and Extended Examples * Statistical Theory and Numerical Results * Summary

35,811 citations


Journal ArticleDOI
Abstract: The principles governing the application of the conceptual model technique to river flow forecasting are discussed. The necessity for a systematic approach to the development and testing of the model is explained and some preliminary ideas suggested.

17,307 citations


Journal ArticleDOI
TL;DR: The focus is on applied inference for Bayesian posterior distributions in real problems, which often tend toward normal- ity after transformations and marginalization, and the results are derived as normal-theory approximations to exact Bayesian inference, conditional on the observed simulations.
Abstract: The Gibbs sampler, the algorithm of Metropolis and similar iterative simulation methods are potentially very helpful for summarizing multivariate distributions. Used naively, however, iterative simulation can give misleading answers. Our methods are simple and generally applicable to the output of any iterative simulation; they are designed for researchers primarily interested in the science underlying the data and models they are analyzing, rather than for researchers interested in the probability theory underlying the iterative simulations themselves. Our recommended strategy is to use several independent sequences, with starting points sampled from an overdispersed distribution. At each step of the iterative simulation, we obtain, for each univariate estimand of interest, a distributional estimate and an estimate of how much sharper the distributional estimate might become if the simulations were continued indefinitely. Because our focus is on applied inference for Bayesian posterior distributions in real problems, which often tend toward normality after transformations and marginalization, we derive our results as normal-theory approximations to exact Bayesian inference, conditional on the observed simulations. The methods are illustrated on a random-effects mixture model applied to experimental measurements of reaction times of normal and schizophrenic patients.

12,022 citations


"A length-based hierarchical model o..." refers methods in this paper

  • ...Convergence was investigated by using the ANOVA-type diagnosis described by Gelman and Rubin (1992) with three chains....

    [...]


Journal ArticleDOI
Abstract: Summary. We consider the problem of comparing complex hierarchical models in which the number of parameters is not clearly defined. Using an information theoretic argument we derive a measure pD for the effective number of parameters in a model as the difference between the posterior mean of the deviance and the deviance at the posterior means of the parameters of interest. In general pD approximately corresponds to the trace of the product of Fisher's information and the posterior covariance, which in normal models is the trace of the ‘hat’ matrix projecting observations onto fitted values. Its properties in exponential families are explored. The posterior mean deviance is suggested as a Bayesian measure of fit or adequacy, and the contributions of individual observations to the fit and complexity can give rise to a diagnostic plot of deviance residuals against leverages. Adding pD to the posterior mean deviance gives a deviance information criterion for comparing models, which is related to other information criteria and has an approximate decision theoretic justification. The procedure is illustrated in some examples, and comparisons are drawn with alternative Bayesian and classical proposals. Throughout it is emphasized that the quantities required are trivial to compute in a Markov chain Monte Carlo analysis.

10,825 citations


Book
01 Jan 2007
TL;DR: The R Book is the first comprehensive reference manual for the R language, including practical guidance and full coverage of the graphics facilities, and introduces the advantages of the R environment, detailing its applications in a wide range of disciplines.
Abstract: The high-level language of R is recognized as one of the most powerful and flexible statistical software environments, and is rapidly becoming the standard setting for quantitative analysis, statistics and graphics. R provides free access to unrivalled coverage and cutting-edge applications, enabling the user to apply numerous statistical methods ranging from simple regression to time series or multivariate analysis. Building on the success of the authors bestselling Statistics: An Introduction using R, The R Book is packed with worked examples, providing an all inclusive guide to R, ideal for novice and more accomplished users alike. The book assumes no background in statistics or computing and introduces the advantages of the R environment, detailing its applications in a wide range of disciplines. Provides the first comprehensive reference manual for the R language, including practical guidance and full coverage of the graphics facilities. Introduces all the statistical models covered by R, beginning with simple classical tests such as chi-square and t-test. Proceeds to examine more advance methods, from regression and analysis of variance, through to generalized linear models, generalized mixed models, time series, spatial statistics, multivariate statistics and much more. The R Book is aimed at undergraduates, postgraduates and professionals in science, engineering and medicine. It is also ideal for students and professionals in statistics, economics, geography and the social sciences.

6,722 citations


"A length-based hierarchical model o..." refers methods in this paper

  • ...Samples are processed by using R (Crawley, 2007)....

    [...]

  • ...2 BUGS simulations Submodels, to the exception of the emergence submodel which is executed with R (Crawley, 2007), are implemented by using OpenBUGS, open source version of WinBUGS (Lunn et al....

    [...]

  • ...2.9.2 BUGS simulations Submodels, to the exception of the emergence submodel which is executed with R (Crawley, 2007), are implemented by using OpenBUGS, open source version of WinBUGS (Lunn et al., 2009; Ntzoufras, 2009)....

    [...]