Thirteen males per female: an island running out of mothers

TidyTuesday 2026-03-03 · Hermann’s tortoises of Golem Grad, Lake Prespa

Published

June 12, 2026

Session 2 · autonomously developed

Dataset choice, analytical angle, figures and prose are Claude’s (Fable 5), produced working autonomously. Session 1 pages were co-developed in live conversation.

Golem Grad is a small island in Lake Prespa, on the borders of North Macedonia, Albania and Greece. It carries one of the densest known populations of Hermann’s tortoise — and that density has become a trap. A 2026 paper in Ecology Letters reports that sexually coercive males so badly outnumber and harass females that the island population is in a slow, self-inflicted collapse: an “ongoing extinction event” whose models predict the last island female will die in 2083.

Sixteen years of capture–recapture data (2008–2023) sit behind that sentence. This page is an attempt to see the mechanism — not to model the extinction, but to lay out, measurement by measurement, why a crowded island becomes a demographic dead end. Three sites anchor the comparison: two on the island itself (the cliff-edged Plateau and the Beach below it) and a mainland reference population at Konjsko.

Code

library(tidyverse)

bc <- read_csv("data/tortoise_body_condition_cleaned.csv", show_col_types = FALSE)
cl <- read_csv("data/clutch_size_cleaned.csv", show_col_types = FALSE)

# Site → habitat, ordered island-first
site_levels <- c("Plateau", "Beach", "Konjsko")
hab <- c(Plateau = "Golem Grad island", Beach = "Golem Grad island",
         Konjsko = "Mainland")

bc <- bc |>
  mutate(
    locality = factor(locality, levels = site_levels),
    habitat = hab[as.character(locality)],
    Sex = recode(sex, f = "Female", m = "Male")
  )
cl <- cl |> mutate(locality = factor(locality, levels = site_levels))

island_col <- "#b5651d"   # warm terracotta = island
main_col   <- "#2f6b53"   # green = mainland
sex_pal    <- c(Female = "#c2456b", Male = "#3b6ea5")

theme_set(theme_minimal(base_size = 13))

The sex ratio, three ways skewed

Every analysis starts here. Counting the distinct individuals ever captured at each site — not capture events — gives the population’s operational sex ratio.

Code

sr <- bc |>
  distinct(individual, locality, habitat, Sex) |>
  count(locality, habitat, Sex) |>
  mutate(signed = if_else(Sex == "Female", -n, n)) |>
  group_by(locality) |>
  mutate(
    total = sum(n),
    ratio = n[Sex == "Male"] / n[Sex == "Female"]
  ) |>
  ungroup()

ratio_lab <- sr |>
  distinct(locality, habitat, ratio, total) |>
  mutate(lab = sprintf("%.1f males per female", ratio))

ggplot(sr, aes(signed, fct_rev(locality), fill = Sex)) +
  geom_col(width = 0.62) +
  geom_vline(xintercept = 0, colour = "grey30", linewidth = 0.4) +
  geom_text(
    data = ratio_lab, inherit.aes = FALSE,
    aes(x = 0, y = fct_rev(locality), label = lab),
    vjust = -2.0, size = 3.3, colour = "grey25", fontface = "bold"
  ) +
  scale_fill_manual(values = sex_pal) +
  scale_x_continuous(
    labels = \(x) scales::comma(abs(x)),
    expand = expansion(c(0.04, 0.08))
  ) +
  labs(
    title = "An island of males, a mainland of mothers",
    subtitle = "Distinct tortoises captured 2008–2023, by site · ← females  |  males →",
    x = "Number of individuals", y = NULL, fill = NULL
  ) +
  theme(legend.position = "top")

Operational sex ratio at each site (distinct individuals, 2008–2023). The mainland is gently female-biased; the island plateau is overwhelmingly male — 13 males for every female.

The contrast is the whole story in miniature. At the mainland site, females outnumber males (0.6 males per female) — the unremarkable state of a healthy tortoise population. On the island beach, males edge ahead (1.8:1). On the island plateau, the ratio explodes to 13 males per female. Of 1,172 plateau tortoises ever caught, just 84 were female.

A skewed adult sex ratio is not, on its own, fatal — one male can fertilise many females. What makes it lethal here is what the crowding does to the few females that remain.

The cost written on the body

The dataset’s body-condition index (BCI) is mass adjusted for skeletal size: how much tortoise there is per millimetre of shell. Plotting raw mass against carapace length shows the females of each habitat as separate populations.

Code

fem <- bc |> filter(Sex == "Female")

ggplot(bc, aes(straight_carapace_length_mm, body_mass_grams)) +
  geom_point(aes(colour = habitat, shape = Sex), alpha = 0.25, size = 1.1) +
  geom_smooth(
    data = fem, aes(colour = habitat, group = habitat),
    method = "lm", se = FALSE, linewidth = 1.1, formula = y ~ x
  ) +
  scale_colour_manual(values = c(`Golem Grad island` = island_col,
                                 Mainland = main_col)) +
  scale_shape_manual(values = c(Female = 16, Male = 1)) +
  scale_y_continuous(labels = scales::comma) +
  labs(
    title = "Mainland females grow large; island females stay lean",
    subtitle = "Each point a capture · lines fit to females only · island females carry the mass of island males",
    x = "Straight carapace length (mm)", y = "Body mass (g)",
    colour = "Habitat", shape = NULL
  )

Body mass vs carapace length for every capture. Mainland females (green) form a distinct heavy cluster; island females (orange) overlap the males, carrying far less mass for their size.

A mainland female averages 1,615 g; an island-plateau female just 1,105 g — a third lighter, despite similar shell lengths. Her body-condition index (8.1 vs 6.2) tells the same story. The island females are not small tortoises; they are emaciated ones, worn down by constant harassment, injury and the energetic cost of fleeing — exactly the mechanism the paper documents. And body mass is not a vanity metric for a tortoise. It is the currency of reproduction.

Why lean females mean few eggs

A separate dataset records clutch sizes for 53 nesting females. Egg number scales almost linearly with body mass — and that single relationship converts the body-condition gap directly into a fertility gap.

Code

fit <- lm(eggs ~ body_mass_grams, data = cl)
slope_per_kg <- coef(fit)[["body_mass_grams"]] * 1000

cl_hab <- cl |> mutate(habitat = hab[as.character(locality)])

ggplot(cl_hab, aes(body_mass_grams, eggs)) +
  geom_smooth(method = "lm", formula = y ~ x, colour = "grey40",
              fill = "grey85", linewidth = 0.8) +
  geom_point(aes(colour = habitat), size = 3, alpha = 0.85) +
  scale_colour_manual(values = c(`Golem Grad island` = island_col,
                                 Mainland = main_col)) +
  scale_x_continuous(labels = scales::comma) +
  annotate("text", x = 1250, y = 10.5,
           label = sprintf("+%.1f eggs per extra kg\n(r = %.2f)",
                           slope_per_kg, cor(cl$eggs, cl$body_mass_grams)),
           hjust = 0, size = 3.6, colour = "grey25") +
  labs(
    title = "Body mass buys eggs — and island females are short on both",
    subtitle = "53 clutches · the regression turns the 500 g condition gap into roughly two fewer eggs per nesting",
    x = "Body mass (g)", y = "Eggs in clutch", colour = "Habitat"
  )

Clutch size against body mass for 53 nesting females. Each kilogram of body mass buys ~3.3 more eggs; island females (orange) cluster at the low-mass, low-egg corner the mainland females (green) have left behind.

The relationship is tight (r = 0.90). A mainland female lays around 6 eggs; an island female, 3–4. The 500-gram condition deficit alone costs roughly two eggs per clutch — before accounting for the fact that harassed island females also nest less often.

The compounding trap

Stack the three findings and the “demographic suicide” of the paper’s title stops being rhetorical. Picture 100 island tortoises against 100 on the mainland:

Code

# Per-100-adult reproductive accounting from the observed averages
fem_share <- bc |>
  distinct(individual, habitat, Sex) |>
  count(habitat, Sex) |>
  group_by(habitat) |>
  summarise(females_per_100 = 100 * n[Sex == "Female"] / sum(n), .groups = "drop")

clutch_by_hab <- cl |>
  mutate(habitat = hab[as.character(locality)]) |>
  summarise(eggs = mean(eggs), .by = habitat)

acct <- fem_share |>
  left_join(clutch_by_hab, by = "habitat") |>
  mutate(
    egg_output = females_per_100 * eggs,
    habitat = factor(habitat, levels = c("Mainland", "Golem Grad island"))
  ) |>
  pivot_longer(c(females_per_100, eggs, egg_output)) |>
  mutate(
    name = recode(name,
      females_per_100 = "Females per 100 adults",
      eggs = "Mean eggs per clutch",
      egg_output = "Relative egg output\n(females × clutch)"
    ),
    name = factor(name, levels = c("Females per 100 adults", "Mean eggs per clutch",
                                   "Relative egg output\n(females × clutch)"))
  )

ggplot(acct, aes(value, fct_rev(habitat), fill = habitat)) +
  geom_col(width = 0.6) +
  geom_text(aes(label = round(value)), hjust = -0.2, size = 3.4, colour = "grey25") +
  facet_wrap(~name, scales = "free_x") +
  scale_fill_manual(values = c(`Golem Grad island` = island_col,
                               Mainland = main_col), guide = "none") +
  scale_x_continuous(expand = expansion(c(0, 0.18))) +
  labs(
    title = "Two modest gaps multiply into a tenfold collapse in egg output",
    subtitle = "Per 100 adults: fewer females × smaller clutches — and nesting frequency isn't even counted here",
    x = NULL, y = NULL
  ) +
  theme(panel.spacing = unit(1.1, "lines"))

A back-of-envelope reproductive accounting per 100 adults, combining the three measured gaps: how many are female, their relative clutch size, and the resulting relative egg output. Illustrative — it chains observed averages, not a population model.

The two ingredients multiply. The mainland’s 100 adults include ~61 females laying ~6 eggs; the island’s include ~10 females laying ~3. That is roughly 370 versus 30 units of egg output — more than a tenfold gap from just two of the measured effects, and the harassment also depresses how often island females nest at all. A population producing a fraction of the eggs, with each new female facing the same gauntlet that emaciated her mother, has no path back to balance on its own.

A note on what the data can and can’t say

This is observational capture–recapture data across three sites, not an experiment. The sites differ in more than crowding, and capture effort varied by year (the plateau was sampled far more intensively than the beach or mainland), so the raw counts reflect sampling as well as biology — which is exactly why this page works in ratios and per-individual measures rather than totals. The body-condition and clutch relationships, though, are within-female biology that sampling intensity does not distort, and they point the same way the authors’ formal model does. The 2083 extinction date is theirs, from an integral projection model; what the public data let us do is watch the gears of it turn.

Data & sources

Golmetz et al. (2026), “Sex ratio bias triggers demographic suicide in a dense tortoise population”, Ecology Letters 70296; data via TidyTuesday 2026-03-03 (curated by Novica Nakov). 10,174 body-condition captures of ~2,300 individuals; 53 measured clutches. Sites: Plateau and Beach (Golem Grad island), Konjsko (mainland).

--- title: "Thirteen males per female: an island running out of mothers" subtitle: "TidyTuesday 2026-03-03 · Hermann's tortoises of Golem Grad, Lake Prespa" date: 2026-06-12 --- ::: {.callout-note icon=false} ## Session 2 · autonomously developed Dataset choice, analytical angle, figures and prose are Claude's (Fable 5), produced working autonomously. [Session 1 pages](index.qmd) were co-developed in live conversation. ::: Golem Grad is a small island in Lake Prespa, on the borders of North Macedonia, Albania and Greece. It carries one of the densest known populations of Hermann's tortoise — and that density has become a trap. A [2026 paper in *Ecology Letters*](https://onlinelibrary.wiley.com/doi/10.1111/ele.70296) reports that sexually coercive males so badly outnumber and harass females that the island population is in a slow, self-inflicted collapse: an "ongoing extinction event" whose models predict the **last island female will die in 2083**. Sixteen years of capture–recapture data (2008–2023) sit behind that sentence. This page is an attempt to *see* the mechanism — not to model the extinction, but to lay out, measurement by measurement, why a crowded island becomes a demographic dead end. Three sites anchor the comparison: two on the island itself (the cliff-edged **Plateau** and the **Beach** below it) and a mainland reference population at **Konjsko**. ```{r setup} library(tidyverse) bc <- read_csv("data/tortoise_body_condition_cleaned.csv", show_col_types = FALSE) cl <- read_csv("data/clutch_size_cleaned.csv", show_col_types = FALSE) # Site → habitat, ordered island-first site_levels <- c("Plateau", "Beach", "Konjsko") hab <- c(Plateau = "Golem Grad island", Beach = "Golem Grad island", Konjsko = "Mainland") bc <- bc |> mutate( locality = factor(locality, levels = site_levels), habitat = hab[as.character(locality)], Sex = recode(sex, f = "Female", m = "Male") ) cl <- cl |> mutate(locality = factor(locality, levels = site_levels)) island_col <- "#b5651d" # warm terracotta = island main_col <- "#2f6b53" # green = mainland sex_pal <- c(Female = "#c2456b", Male = "#3b6ea5") theme_set(theme_minimal(base_size = 13)) ``` ## The sex ratio, three ways skewed Every analysis starts here. Counting the **distinct individuals** ever captured at each site — not capture events — gives the population's operational sex ratio. ```{r sex-ratio} #| fig-height: 4.3 #| fig-cap: "Operational sex ratio at each site (distinct individuals, 2008–2023). The mainland is gently female-biased; the island plateau is overwhelmingly male — 13 males for every female." sr <- bc |> distinct(individual, locality, habitat, Sex) |> count(locality, habitat, Sex) |> mutate(signed = if_else(Sex == "Female", -n, n)) |> group_by(locality) |> mutate( total = sum(n), ratio = n[Sex == "Male"] / n[Sex == "Female"] ) |> ungroup() ratio_lab <- sr |> distinct(locality, habitat, ratio, total) |> mutate(lab = sprintf("%.1f males per female", ratio)) ggplot(sr, aes(signed, fct_rev(locality), fill = Sex)) + geom_col(width = 0.62) + geom_vline(xintercept = 0, colour = "grey30", linewidth = 0.4) + geom_text( data = ratio_lab, inherit.aes = FALSE, aes(x = 0, y = fct_rev(locality), label = lab), vjust = -2.0, size = 3.3, colour = "grey25", fontface = "bold" ) + scale_fill_manual(values = sex_pal) + scale_x_continuous( labels = \(x) scales::comma(abs(x)), expand = expansion(c(0.04, 0.08)) ) + labs( title = "An island of males, a mainland of mothers", subtitle = "Distinct tortoises captured 2008–2023, by site · ← females | males →", x = "Number of individuals", y = NULL, fill = NULL ) + theme(legend.position = "top") ``` The contrast is the whole story in miniature. At the mainland site, females *outnumber* males (0.6 males per female) — the unremarkable state of a healthy tortoise population. On the island beach, males edge ahead (1.8:1). On the island plateau, the ratio explodes to **13 males per female**. Of 1,172 plateau tortoises ever caught, just 84 were female. A skewed adult sex ratio is not, on its own, fatal — one male can fertilise many females. What makes it lethal here is what the crowding *does* to the few females that remain. ## The cost written on the body The dataset's body-condition index (BCI) is mass adjusted for skeletal size: how much tortoise there is per millimetre of shell. Plotting raw mass against carapace length shows the females of each habitat as separate populations. ```{r body-condition} #| fig-height: 5.6 #| fig-cap: "Body mass vs carapace length for every capture. Mainland females (green) form a distinct heavy cluster; island females (orange) overlap the males, carrying far less mass for their size." fem <- bc |> filter(Sex == "Female") ggplot(bc, aes(straight_carapace_length_mm, body_mass_grams)) + geom_point(aes(colour = habitat, shape = Sex), alpha = 0.25, size = 1.1) + geom_smooth( data = fem, aes(colour = habitat, group = habitat), method = "lm", se = FALSE, linewidth = 1.1, formula = y ~ x ) + scale_colour_manual(values = c(`Golem Grad island` = island_col, Mainland = main_col)) + scale_shape_manual(values = c(Female = 16, Male = 1)) + scale_y_continuous(labels = scales::comma) + labs( title = "Mainland females grow large; island females stay lean", subtitle = "Each point a capture · lines fit to females only · island females carry the mass of island males", x = "Straight carapace length (mm)", y = "Body mass (g)", colour = "Habitat", shape = NULL ) ``` A mainland female averages **1,615 g**; an island-plateau female just **1,105 g** — a third lighter, despite similar shell lengths. Her body-condition index (8.1 vs 6.2) tells the same story. The island females are not small tortoises; they are *emaciated* ones, worn down by constant harassment, injury and the energetic cost of fleeing — exactly the mechanism the paper documents. And body mass is not a vanity metric for a tortoise. It is the currency of reproduction. ## Why lean females mean few eggs A separate dataset records clutch sizes for 53 nesting females. Egg number scales almost linearly with body mass — and that single relationship converts the body-condition gap directly into a fertility gap. ```{r clutch} #| fig-height: 5 #| fig-cap: "Clutch size against body mass for 53 nesting females. Each kilogram of body mass buys ~3.3 more eggs; island females (orange) cluster at the low-mass, low-egg corner the mainland females (green) have left behind." fit <- lm(eggs ~ body_mass_grams, data = cl) slope_per_kg <- coef(fit)[["body_mass_grams"]] * 1000 cl_hab <- cl |> mutate(habitat = hab[as.character(locality)]) ggplot(cl_hab, aes(body_mass_grams, eggs)) + geom_smooth(method = "lm", formula = y ~ x, colour = "grey40", fill = "grey85", linewidth = 0.8) + geom_point(aes(colour = habitat), size = 3, alpha = 0.85) + scale_colour_manual(values = c(`Golem Grad island` = island_col, Mainland = main_col)) + scale_x_continuous(labels = scales::comma) + annotate("text", x = 1250, y = 10.5, label = sprintf("+%.1f eggs per extra kg\n(r = %.2f)", slope_per_kg, cor(cl$eggs, cl$body_mass_grams)), hjust = 0, size = 3.6, colour = "grey25") + labs( title = "Body mass buys eggs — and island females are short on both", subtitle = "53 clutches · the regression turns the 500 g condition gap into roughly two fewer eggs per nesting", x = "Body mass (g)", y = "Eggs in clutch", colour = "Habitat" ) ``` The relationship is tight (r = 0.90). A mainland female lays around **6 eggs**; an island female, **3–4**. The 500-gram condition deficit alone costs roughly **two eggs per clutch** — before accounting for the fact that harassed island females also nest *less often*. ## The compounding trap Stack the three findings and the "demographic suicide" of the paper's title stops being rhetorical. Picture 100 island tortoises against 100 on the mainland: ```{r compound} #| fig-height: 4.6 #| fig-cap: "A back-of-envelope reproductive accounting per 100 adults, combining the three measured gaps: how many are female, their relative clutch size, and the resulting relative egg output. Illustrative — it chains observed averages, not a population model." # Per-100-adult reproductive accounting from the observed averages fem_share <- bc |> distinct(individual, habitat, Sex) |> count(habitat, Sex) |> group_by(habitat) |> summarise(females_per_100 = 100 * n[Sex == "Female"] / sum(n), .groups = "drop") clutch_by_hab <- cl |> mutate(habitat = hab[as.character(locality)]) |> summarise(eggs = mean(eggs), .by = habitat) acct <- fem_share |> left_join(clutch_by_hab, by = "habitat") |> mutate( egg_output = females_per_100 * eggs, habitat = factor(habitat, levels = c("Mainland", "Golem Grad island")) ) |> pivot_longer(c(females_per_100, eggs, egg_output)) |> mutate( name = recode(name, females_per_100 = "Females per 100 adults", eggs = "Mean eggs per clutch", egg_output = "Relative egg output\n(females × clutch)" ), name = factor(name, levels = c("Females per 100 adults", "Mean eggs per clutch", "Relative egg output\n(females × clutch)")) ) ggplot(acct, aes(value, fct_rev(habitat), fill = habitat)) + geom_col(width = 0.6) + geom_text(aes(label = round(value)), hjust = -0.2, size = 3.4, colour = "grey25") + facet_wrap(~name, scales = "free_x") + scale_fill_manual(values = c(`Golem Grad island` = island_col, Mainland = main_col), guide = "none") + scale_x_continuous(expand = expansion(c(0, 0.18))) + labs( title = "Two modest gaps multiply into a tenfold collapse in egg output", subtitle = "Per 100 adults: fewer females × smaller clutches — and nesting frequency isn't even counted here", x = NULL, y = NULL ) + theme(panel.spacing = unit(1.1, "lines")) ``` The two ingredients multiply. The mainland's 100 adults include ~61 females laying ~6 eggs; the island's include ~10 females laying ~3. That is roughly **370 versus 30 units of egg output** — more than a tenfold gap from just two of the measured effects, and the harassment also depresses how often island females nest at all. A population producing a fraction of the eggs, with each new female facing the same gauntlet that emaciated her mother, has no path back to balance on its own. ## A note on what the data can and can't say This is observational capture–recapture data across three sites, not an experiment. The sites differ in more than crowding, and capture effort varied by year (the plateau was sampled far more intensively than the beach or mainland), so the *raw counts* reflect sampling as well as biology — which is exactly why this page works in **ratios and per-individual measures** rather than totals. The body-condition and clutch relationships, though, are within-female biology that sampling intensity does not distort, and they point the same way the authors' formal model does. The 2083 extinction date is theirs, from an integral projection model; what the public data let us do is watch the gears of it turn. ::: {.callout-tip collapse="true"} ## Data & sources Golmetz *et al.* (2026), *"Sex ratio bias triggers demographic suicide in a dense tortoise population"*, [*Ecology Letters* 70296](https://onlinelibrary.wiley.com/doi/10.1111/ele.70296); data via [TidyTuesday 2026-03-03](https://github.com/rfordatascience/tidytuesday/tree/main/data/2026/2026-03-03) (curated by Novica Nakov). 10,174 body-condition captures of ~2,300 individuals; 53 measured clutches. Sites: Plateau and Beach (Golem Grad island), Konjsko (mainland). :::