# Detecting density dependence in R
# https://tidyecology.com/posts/detecting-density-dependence/
#
# The R code of this Tidy Ecology tutorial, chunk by chunk, in the order of the post.
# Run it top to bottom in a fresh R session. Code the post shows but does not run
# is commented out. The prose, and the numbers it quotes, are in the post itself.
# Post last modified: 2026-10-01

## ---- setup ----
suppressPackageStartupMessages({library(ggplot2); library(dplyr)})
ink<-"#16241d"; body<-"#2c3a31"; forest<-"#275139"; gold<-"#c9b458"
paper<-"#f5f4ee"; line<-"#dad9ca"; faint<-"#5d6b61"; aban<-"#b5534e"
theme_te <- function(){
  theme_minimal(base_size=12)+
  theme(plot.background=element_rect(fill=paper,colour=NA),
        panel.background=element_rect(fill=paper,colour=NA),
        panel.grid.minor=element_blank(),
        panel.grid.major=element_line(colour=line,linewidth=.3),
        axis.title=element_text(colour=ink), axis.text=element_text(colour=faint),
        plot.title=element_text(colour=ink,size=12,face="bold"),
        strip.text=element_text(colour=ink,face="bold"),
        legend.position="bottom", legend.title=element_text(colour=ink),
        legend.text=element_text(colour=body))
}

## ---- helpers ----
fit_ar1 <- function(x){
  n <- length(x); x0 <- x[-n]; x1 <- x[-1]; m <- n-1; xb <- mean(x0)
  Sxx <- sum((x0-xb)^2)
  b <- sum((x0-xb)*(x1-mean(x1)))/Sxx
  a <- mean(x1) - b*xb
  r <- x1 - (a + b*x0); s2 <- sum(r^2)/m
  se_b <- sqrt((sum(r^2)/(m-2))/Sxx)
  d <- x1 - x0; a0 <- mean(d); s20 <- mean((d-a0)^2)
  LR <- m*(log(s20) - log(s2))       # 2*(logLik_full - logLik_null), constants cancel
  list(a=a,b=b,s2=s2,se_b=se_b,m=m,LR=LR,a0=a0,s20=s20,grow=d,x0=x0)
}
sim_gompertz <- function(T,a,b,s,x1){ x<-numeric(T); x[1]<-x1
  for(t in 1:(T-1)) x[t+1] <- a + b*x[t] + rnorm(1,0,s); x }

## ---- simulate ----
Tlen<-35; Nstar<-140; Xstar<-log(Nstar); b_dd<-0.60; s_dd<-0.20; a_dd<-(1-b_dd)*Xstar

set.seed(932)
Xdd <- sim_gompertz(Tlen, a_dd, b_dd, s_dd, x1=log(70))   # density dependent

set.seed(3075)
Xrw <- sim_gompertz(Tlen, 0, 1, 0.20, x1=Xstar)          # random walk (b = 1)

## ---- fits ----
fdd <- fit_ar1(Xdd)
frw <- fit_ar1(Xrw)
naive_p <- function(f) pt((f$b-1)/f$se_b, df=f$m-2)      # one-sided t test of b < 1

round(c(dd_b=fdd$b, dd_se=fdd$se_b, dd_Nstar=exp(fdd$a/(1-fdd$b)),
        rw_b=frw$b, rw_se=frw$se_b), 3)
round(c(dd_naive_p=naive_p(fdd), rw_naive_p=naive_p(frw)), 4)

## ---- fig-growth ----
gd <- rbind(
  data.frame(x=fdd$x0, r=fdd$grow, series="Density dependent (b<1)"),
  data.frame(x=frw$x0, r=frw$grow, series="Random walk (b=1)"))
ggplot(gd, aes(x, r, colour=series))+
  geom_hline(yintercept=0, colour=faint, linewidth=.3)+
  geom_point(size=1.9, alpha=.85)+
  geom_smooth(method="lm", se=FALSE, linewidth=.9, formula=y~x)+
  scale_colour_manual(values=c("Density dependent (b<1)"=forest,"Random walk (b=1)"=gold))+
  labs(x="log abundance in year t", y="per-capita growth  log(N[t+1]/N[t])",
       colour=NULL, title="Per-capita growth against density")+
  theme_te()

## ---- bias-sim ----
set.seed(2468); M<-3000; bh<-numeric(M); rej<-logical(M)
for(i in 1:M){
  Xr <- sim_gompertz(Tlen, 0, 1, 0.20, x1=Xstar)
  f <- fit_ar1(Xr); bh[i] <- f$b; rej[i] <- naive_p(f) < 0.05
}
round(c(mean_bhat=mean(bh), bias=1-mean(bh), median=median(bh), naive_reject=mean(rej)), 3)

## ---- fig-bias ----
dfb <- data.frame(b=bh)
ggplot(dfb, aes(b))+
  geom_histogram(bins=45, fill=forest, colour=paper, linewidth=.15)+
  geom_vline(xintercept=1, colour=aban, linewidth=.8)+
  geom_vline(xintercept=mean(bh), colour=gold, linewidth=.8, linetype="22")+
  annotate("text", x=1, y=Inf, label=" true b = 1", hjust=0, vjust=1.6, colour=aban, size=3.3)+
  annotate("text", x=mean(bh), y=Inf, label="mean b_hat ", hjust=1, vjust=1.6, colour="#8a7a1e", size=3.3)+
  labs(x="OLS slope b_hat (density-dependence coefficient)", y="count",
       title="OLS slope is biased below 1 under a pure random walk")+
  theme_te()

## ---- pblr ----
pblr <- function(x, B=1999){
  f <- fit_ar1(x); T <- length(x); LRstar <- numeric(B)
  for(bb in 1:B){
    xs <- numeric(T); xs[1] <- x[1]
    e <- rnorm(T-1, f$a0, sqrt(f$s20))        # simulate under H0: b = 1, drift a0
    for(t in 1:(T-1)) xs[t+1] <- xs[t] + e[t]
    LRstar[bb] <- fit_ar1(xs)$LR
  }
  list(p=(1+sum(LRstar>=f$LR))/(B+1), LRobs=f$LR)
}
set.seed(9321);  pdd <- pblr(Xdd, B=1999)     # density-dependent series
set.seed(30751); prw <- pblr(Xrw, B=1999)     # random-walk series
round(c(dd_LR=pdd$LRobs, dd_pblr=pdd$p, rw_LR=prw$LRobs, rw_pblr=prw$p), 4)

## ---- calib ----
set.seed(1357); M2<-500; B2<-299; prej<-logical(M2)
for(i in 1:M2){
  Xr <- sim_gompertz(Tlen, 0, 1, 0.20, x1=Xstar)
  f <- fit_ar1(Xr); LRo <- f$LR; LRs <- numeric(B2)
  for(bb in 1:B2){
    xs <- numeric(Tlen); xs[1] <- Xr[1]
    e <- rnorm(Tlen-1, f$a0, sqrt(f$s20))
    for(t in 1:(Tlen-1)) xs[t+1] <- xs[t] + e[t]
    LRs[bb] <- fit_ar1(xs)$LR
  }
  prej[i] <- (1+sum(LRs>=LRo))/(B2+1) < 0.05
}
round(mean(prej), 3)
