In brouwern/avesdemazan: What the Package Does (One Line, Title Case)

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>", 
  warning = F
)

Introduction

Model community abundance to test functioning of glmmTMB and determine if there is an overall decline in total captures over time.

library(avesdemazan)

# Model fitting
library(lme4)       # glmer() for glmm
library(glmmTMB)    # glmm with AR-1  
# library(blme)       # bglmer (used?)

# # Model evaluation
# library(afex)       # functionall_fit()
# library(multcomp)   
# library(emmeans)
# library(bbmle)      # aictab
# library(arm)        # se.fit()

# Result visualization
library(ggplot2)    

#  markdown etc
library(knitr)

Model TOTAL individuals captures

Load data. All individuals captures within a session totaled up, ignoring species.

data(ecuador_tot)

The data is 86 by 6

dim(ecuador_tot)

The data frame looks like this:

head(ecuador_tot, 10)

summary(ecuador_tot)

NOTE: This shows total captures of species that are included in the GLMM, NOT total captures of ALL species. Species that did not have enough captures to meet the inclusion critera are not included, and unbanded individuals are not included.

Split data by session and plot by year. This is NOT how the data are modeled because time is set up as a continuous variable.

ggplot(data = ecuador_tot,
        aes(y = N/tot_net_hours*1000,
            x = year,
            color = session)) +
  geom_point() +
  facet_wrap(~Location) + 
  geom_smooth(method = lm,se = F) + 
  theme_bw()

Ignore sessions and plot by time_cts.

ggplot(data = ecuador_tot,
        aes(y = N/tot_net_hours*1000,
            x = time_cts)) +
  geom_point() +
  facet_wrap(~Location) + 
  geom_smooth(method = lm,se = F) + 
  theme_bw()

Models of community abundance change

GLM model of community abundance change

Basic GLM for model exploration.

glm_fit <- glm(N ~ 1 + time_cts +   # continuous time 
                     Location +     # Location
                     time_cts*Location +  # 
                     offset(log(tot_net_hours)),
                   family = poisson,
                   data = ecuador_tot)

Time is significant, but with no time_cts*:Location interaction.

summary(glm_fit)

Basic GLMM with lme4::glmer()

Poisson GLMM; no autocorrelation, poisson errors.

glmer() Modeling with fixed effects for time*location

glmm_fit01 <- lme4::glmer(N ~ 1 + time_cts +   # Null
                          Location +
                          Location*time_cts +
                     #(1|session) +
                     #(1|Location) +
                     (1|year:session) + # all captures w/in a session-year
                     (1|year:session:Location) + # all caps w/in a session-year-loc  
                     #(0+time_cts|Location) + # rand slope for location
                     offset(log(tot_net_hours)),
                   family = poisson,
                   data = ecuador_tot)

summary(glmm_fit01)

glmer() Modeling with ranefs for time*location

Not significant

glmm_fit02 <- lme4::glmer(N ~ 1 + time_cts +   # Null
                          #Location +
                          #Location*time_cts +
                     #(1|session) +
                     #(1|Location) +
                     (1|year:session) + # all captures w/in a session-year
                     (1|year:session:Location) + # all caps w/in a session-year-loc  
                     (0+time_cts|Location) + # rand slope for location
                     offset(log(tot_net_hours)),
                   family = poisson,
                   data = ecuador_tot)

summary(glmm_fit02)

# fixef(glmm_fit)[2]
# slopes_loc <- fixef(glmm_fit)[2]+ranef(glmm_fit)$Location
# 
# CI.up
# CI.lo <- fixef(glmm_fit)[2]-ranef(glmm_fit)$Location
# 
# 
# +se.ranef(glmm_fit)$Location*2
# fixef(glmm_fit)[2]-se.ranef(glmm_fit)$Location*2

Poisson glmmTMB without ar1

library(splines)
glmmTMB_poisson <- glmmTMB(N ~ 1 + time_cts +   # Null
                          Location +
                          Location*time_cts +
                     (1|year:session) + # all captures w/in a session-year
                     (1|year:session:Location) + # all caps w/in a session-year-loc  
                     (0+time_cts|Location) + # rand slope for location
                     offset(log(tot_net_hours)),
                   family = poisson,
                   data = ecuador_tot)

summary(glmmTMB_poisson)