Influence of prior distributions and random effects on count regression models

implications for estimating standing dead tree abundance

Research output: Contribution to journalArticle

4 Citations (Scopus)

Abstract

The presence and abundance of standing dead trees (SDTs) in forests is typically characterized by an excess number of zeros and high variation. The variability inherent in SDT data naturally leads to the assessment of novel quantitative methods to represent SDT populations and their role in contributing to forest ecosystem structure. This analysis assessed the performance of count regression methods fit with Bayesian mixed-effects models that estimate SDTs (all dead trees with a diameter at breast height ≥ 12.7cm found on plots 0.07-ha in size) on over 17,000 forest inventory plots across the US Lake States (Michigan, Minnesota, and Wisconsin). Random effects models that used the independent variables basal area, mean annual temperature, and plot ownership (i.e., publicly- or privately-owned) as fixed effects and forest type as a random effect outperformed a method which used fixed-effects, alone. Standard and zero-inflated negative binomial models were effective in accounting for overdispersion present in the data (variance/mean ratio for SDT counts was 72.6), whereas Poisson count models were not. Random effects were calibrated on a new population of SDTs. Employing informative prior distributions from the developed model led to improved estimates of SDT abundance by reducing root mean square error by seven percent. This analysis highlights an approach that uses existing models for representing population averages while calibrating their local random effects with new data to arrive at improved estimates of SDTs across the region.

Original languageEnglish (US)
Pages (from-to)145-160
Number of pages16
JournalEnvironmental and Ecological Statistics
Volume22
Issue number1
DOIs
StatePublished - Jan 1 2015

Fingerprint

Prior distribution
Random Effects
Regression Model
Count
Fixed Effects
Negative Binomial Model
Estimate
ecosystem structure
Influence
distribution
effect
Random effects
Regression model
Mixed Effects Model
Overdispersion
forest inventory
Random Effects Model
basal area
Zero
forest ecosystem

Keywords

  • Biomass
  • Carbon
  • Coarse woody debris
  • Forest inventory
  • Markov chain Monte Carlo (MCMC)
  • Zero-inflation

Cite this

@article{31b9ecdd42a84572970053e42fba4d9d,
title = "Influence of prior distributions and random effects on count regression models: implications for estimating standing dead tree abundance",
abstract = "The presence and abundance of standing dead trees (SDTs) in forests is typically characterized by an excess number of zeros and high variation. The variability inherent in SDT data naturally leads to the assessment of novel quantitative methods to represent SDT populations and their role in contributing to forest ecosystem structure. This analysis assessed the performance of count regression methods fit with Bayesian mixed-effects models that estimate SDTs (all dead trees with a diameter at breast height ≥ 12.7cm found on plots 0.07-ha in size) on over 17,000 forest inventory plots across the US Lake States (Michigan, Minnesota, and Wisconsin). Random effects models that used the independent variables basal area, mean annual temperature, and plot ownership (i.e., publicly- or privately-owned) as fixed effects and forest type as a random effect outperformed a method which used fixed-effects, alone. Standard and zero-inflated negative binomial models were effective in accounting for overdispersion present in the data (variance/mean ratio for SDT counts was 72.6), whereas Poisson count models were not. Random effects were calibrated on a new population of SDTs. Employing informative prior distributions from the developed model led to improved estimates of SDT abundance by reducing root mean square error by seven percent. This analysis highlights an approach that uses existing models for representing population averages while calibrating their local random effects with new data to arrive at improved estimates of SDTs across the region.",
keywords = "Biomass, Carbon, Coarse woody debris, Forest inventory, Markov chain Monte Carlo (MCMC), Zero-inflation",
author = "Russell, {Matthew B}",
year = "2015",
month = "1",
day = "1",
doi = "10.1007/s10651-014-0290-7",
language = "English (US)",
volume = "22",
pages = "145--160",
journal = "Environmental and Ecological Statistics",
issn = "1352-8505",
publisher = "Springer Netherlands",
number = "1",

}

TY - JOUR

T1 - Influence of prior distributions and random effects on count regression models

T2 - implications for estimating standing dead tree abundance

AU - Russell, Matthew B

PY - 2015/1/1

Y1 - 2015/1/1

N2 - The presence and abundance of standing dead trees (SDTs) in forests is typically characterized by an excess number of zeros and high variation. The variability inherent in SDT data naturally leads to the assessment of novel quantitative methods to represent SDT populations and their role in contributing to forest ecosystem structure. This analysis assessed the performance of count regression methods fit with Bayesian mixed-effects models that estimate SDTs (all dead trees with a diameter at breast height ≥ 12.7cm found on plots 0.07-ha in size) on over 17,000 forest inventory plots across the US Lake States (Michigan, Minnesota, and Wisconsin). Random effects models that used the independent variables basal area, mean annual temperature, and plot ownership (i.e., publicly- or privately-owned) as fixed effects and forest type as a random effect outperformed a method which used fixed-effects, alone. Standard and zero-inflated negative binomial models were effective in accounting for overdispersion present in the data (variance/mean ratio for SDT counts was 72.6), whereas Poisson count models were not. Random effects were calibrated on a new population of SDTs. Employing informative prior distributions from the developed model led to improved estimates of SDT abundance by reducing root mean square error by seven percent. This analysis highlights an approach that uses existing models for representing population averages while calibrating their local random effects with new data to arrive at improved estimates of SDTs across the region.

AB - The presence and abundance of standing dead trees (SDTs) in forests is typically characterized by an excess number of zeros and high variation. The variability inherent in SDT data naturally leads to the assessment of novel quantitative methods to represent SDT populations and their role in contributing to forest ecosystem structure. This analysis assessed the performance of count regression methods fit with Bayesian mixed-effects models that estimate SDTs (all dead trees with a diameter at breast height ≥ 12.7cm found on plots 0.07-ha in size) on over 17,000 forest inventory plots across the US Lake States (Michigan, Minnesota, and Wisconsin). Random effects models that used the independent variables basal area, mean annual temperature, and plot ownership (i.e., publicly- or privately-owned) as fixed effects and forest type as a random effect outperformed a method which used fixed-effects, alone. Standard and zero-inflated negative binomial models were effective in accounting for overdispersion present in the data (variance/mean ratio for SDT counts was 72.6), whereas Poisson count models were not. Random effects were calibrated on a new population of SDTs. Employing informative prior distributions from the developed model led to improved estimates of SDT abundance by reducing root mean square error by seven percent. This analysis highlights an approach that uses existing models for representing population averages while calibrating their local random effects with new data to arrive at improved estimates of SDTs across the region.

KW - Biomass

KW - Carbon

KW - Coarse woody debris

KW - Forest inventory

KW - Markov chain Monte Carlo (MCMC)

KW - Zero-inflation

UR - http://www.scopus.com/inward/record.url?scp=84939417645&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84939417645&partnerID=8YFLogxK

U2 - 10.1007/s10651-014-0290-7

DO - 10.1007/s10651-014-0290-7

M3 - Article

VL - 22

SP - 145

EP - 160

JO - Environmental and Ecological Statistics

JF - Environmental and Ecological Statistics

SN - 1352-8505

IS - 1

ER -