# Population Growth and Regulation

> In a closed population, growth is the per-capita rate times the population's size, and a population is regulated only when its net per-capita rate falls as density rises, from positive below a level to negative above it (negative density dependence, which in real populations may act after a lag). This lesson builds the exponential and logistic models from that rule, works a projection with real numbers, and then asks what field experiments on Yukon snowshoe hares and Bahamian reef fish, and long-term records of St Kilda sheep, say about what limits populations and how density dependence is detected. Afterwards you can project a population under either model, find its doubling time and its fastest-growing size, and name the assumptions a real population breaks.

Course: Ecology and Evolution (Environments and populations). Level: core university (third year). About 20 minutes.
Lesson page (link this one when citing): https://stemhop.com/lessons/population-growth/

Builds on: Demography and Survivorship · How to Test an Evolutionary Explanation

## What you'll learn

- Compare exponential growth (dN/dt = rN) with logistic growth (dN/dt = rN(1 - N/K)), giving the units of r and K, the assumptions of each model, r = b - d, lambda = e^r and the doubling time ln 2 / r.
- Explain that the logistic model assumes per-capita growth falls in a straight line with N, why total growth then peaks at K/2, and why K is an equilibrium rather than a ceiling.
- Explain why only negative density dependence (net per-capita growth falling as density rises, from positive below a level to negative above it) regulates a population, and distinguish density-dependent from density-independent effects using field experiments on snowshoe hares and reef fish and long-term records of Soay sheep, and explain why weather and density can interact.
- Make a numerical prediction from a growth model, compare it with data, and name the model assumptions that a field population breaks.

**The first film ran one rule with the per-capita rate held flat, then falling; this text adds the units, the doubling time, life above K, and what a real brake has to be.**

> **The one idea.** A population’s growth is its **per-capita rate** (births minus deaths per individual, per unit time) times its size: dN/dt = (b − d) × N. A population is **regulated** only when that net per-capita rate falls as **density** (individuals per unit area) rises, from positive below some level to negative above it, so the population returns to that level from either side. This is **negative density dependence**, and in real populations it may act after a lag.
>
> *Scope:* both models describe one closed population (no immigration or emigration) in a fixed area, so N and density rise together, with no age structure and continuous time. The exponential model holds b and d constant. The logistic holds r and K constant and assumes the net rate b − d falls in a straight line, from r at low density to zero at K, with crowding acting at once (no lag). These qualifiers hold for every model statement below.

## A fixed per-capita rate: exponential growth

Write N for the number of individuals and r for the **intrinsic rate of increase**, the per-capita rate when resources are not limiting. With a per-capita birth rate b and death rate d, r = b − d: if b = 0.5 per year and d = 0.3 per year, r = 0.2 per year. The whole population grows at dN/dt = rN, where dN/dt (said “d N d t”) is how fast N is changing right now. Calculus turns that into a projection (the card shows what it predicts, not the derivation):

> **Equation card: How many individuals an unchecked population will have after a given time.**
>
>     N(t) = N0 × e^rt
>
> In words: The population at time t is the starting population times e raised to the power r times t.
>
> - N(t): the number of individuals at time t (individuals)
> - N0: the number of individuals at the start, t = 0 (individuals)
> - e^rt: the growth factor so far: e, about 2.718, raised to the power r × t, where r is the intrinsic rate of increase (per day or per year) and t is the time elapsed in the same unit
>
> Sanity check: At t = 0, e^0 = 1, so N(0) = N0. If r = 0 the population stays at N0; if r is negative it shrinks toward zero. The units of r and t must match so that r × t has no unit: an r without its time unit means nothing.
> Try one number Start with 100 individuals growing at r = 0.2 per day, and look at day 5. r × t = 0.2 × 5 = 1, and e^1 = 2.718. N(5) = 100 × 2.718, about 272: more than one doubling (200), fewer than two (400).

Because r is a net rate (not a birth rate) acting at every instant, over one day the population multiplies by **lambda** = er: an r of 0.2 per day gives lambda = e0.2 = 1.22, a 22% rise per day, slightly more than 20% because newcomers start adding newcomers at once. Your [demography](https://stemhop.com/lessons/demography/) lesson shows how r and lambda come from a life table.

A constant r means a constant **doubling time**, ln 2 / r. Doubling means ert = 2; taking the natural log of both sides (ln undoes e; see [e and ln](https://stemhop.com/math-toolkit/#e-and-ln)) gives rt = ln 2 = 0.693, so t = 0.693 / r. With r = 0.2 per day the population doubles every 3.47 days, whether it holds 100 individuals or 100,000.

## A falling per-capita rate: the logistic model

The logistic model writes the first film’s tilted line as a brake on the per-capita rate:

> **Equation card: How fast a population grows when crowding slows growth as it nears a limit.**
>
>     dN/dt = r × N × (1 − N/K)
>
> In words: The growth rate is r times N times one minus N over K.
>
> - dN/dt: the change in population size per unit time (individuals per day)
> - r: the intrinsic rate of increase, the per-capita rate when the population is tiny (per day)
> - N: the number of individuals now (individuals)
> - K: the carrying capacity: the population size at which the model’s growth stops, in the same units as N (individuals)
>
> Sanity check: The bracket, 1 − N/K, is the brake. When N is tiny it is close to 1 and growth is nearly exponential. At N = K it is 0 and growth stops. Above K it is negative and the population falls.
> Try one number Take r = 0.2 per day and K = 1,000, as in the worked problem below, with N = 250. N/K = 0.25, so the brake is 1 − 0.25 = 0.75. dN/dt = 0.2 × 250 × 0.75 = 37.5 individuals per day.

Divide both sides by N for the per-capita rate:

> **(1/N) dN/dt = r(1 − N/K)**

Against N this is a straight line falling from r through zero at K, the **carrying capacity** (real lines are often curved; Going deeper). The numbers behind the first film’s hump hold one surprise:

| N | per-capita rate r(1 − N/K), per day | total growth dN/dt, per day |
| --- | --- | --- |
| near 0 | 0.20 (the low-density limit) | near 0 |
| 100 | 0.18 | 18 |
| 250 | 0.15 | 37.5 |
| 500 (K/2) | 0.10 | **50 (the maximum)** |
| 750 | 0.05 | 37.5 |
| 900 | 0.02 | 18 |
| 1,000 (K) | 0 | 0 |

At N = 100 and N = 900 the population has the same instantaneous growth rate, dN/dt = 18 individuals per day.

**Predict.** Immigrants push this population to 1,200, above K. What does the model say happens next?

> **Check your prediction.** It falls back. At 1,200 the per-capita rate is 0.2 × (1 − 1.2) = −0.04 per day, so at that moment the population is shrinking at 48 a day, and more slowly as it nears 1,000. K is a **stable equilibrium**, a level the population returns to from either side, not a ceiling. From below, the continuous model rises smoothly to K and never overshoots: a one-variable continuous model with instant feedback cannot.

**Worked problem.** Suppose a population starts at N0 = 100 individuals, with r = 0.2 per day and K = 1,000 individuals, all held constant. These numbers are illustrative, not measured for any species.

1. *Exponential projection at day 10.* N(10) = 100 x e0.2 x 10 = 100 x e2 = 100 x 7.389 = 738.9, about 739 individuals.
2. *Logistic projection at day 10.* The solution of the logistic equation (you can use it without deriving it) is N(t) = K / (1 + (K/N0 - 1) e−rt). Here K/N0 - 1 = 1,000/100 - 1 = 9, and rt = 0.2 x 10 = 2, so N(10) = 1,000 / (1 + 9 x e−2). With e−2 = 0.1353, N(10) = 1,000 / (1 + 9 x 0.1353) = 1,000 / 2.218 = 450.9, about 451 individuals.
3. *Initial logistic rate.* dN/dt = 0.2 x 100 x (1 - 100/1,000) = 0.2 x 100 x 0.9 = 18 individuals per day, compared with 20 per day under the exponential model.
4. *Time to K/2.* Set N(t) = 500: 1,000 / (1 + 9 e−0.2t) = 500, so 1 + 9 e−0.2t = 2 and e−0.2t = 1/9. Taking the natural log of both sides, t = ln 9 / 0.2 = 2.197 / 0.2 = 10.99 days.
5. *Peak total growth.* At N = 500, dN/dt = 0.2 x 500 x 0.5 = 50 individuals per day, which is rK/4 = 0.2 x 1,000 / 4.
6. *Check the answers against each other.* At day 10 the logistic population (451) is just short of K/2, consistent with reaching 500 at day 10.99. By day 20 the exponential model predicts 5,460 individuals (100 x e4) and the logistic about 858 (1,000 / (1 + 9 x e−4)), about one-sixth as many.

*Interactive on the lesson page:* Exponential and logistic growth, side by side

Now name what a field population breaks. Real populations have **age structure**: juveniles do not breed, so a population of mostly young animals grows more slowly than its N suggests. Feedback has **time lags**: food eaten this winter limits births next spring. And **K changes** with season, weather and food supply. These three can make a real population overshoot or cycle around K, though none always does; so can a model that steps a season at a time, if growth is strong enough (Going deeper).

## Tilting the line: what counts as a brake

A **limiting factor** holds numbers or growth down: if adding food raises numbers compared with untouched control plots, food was limiting there. A **density-dependent** factor has a per-capita effect that changes with density; it tilts the line. A **density-independent** factor shifts the per-capita rate by the same amount at every density.

Only one direction of tilt regulates: births falling or deaths rising with density, so that net per-capita growth is positive below a level, zero at it and negative above it. The logistic meets this whenever r and K are positive, with the level at K; real feedback can arrive after a lag (the time lags above). A tilt the other way, at very low density, pushes a population away (an Allee effect, Going deeper). Neither label comes with the agent: whether a predator or a frost is density dependent is a hypothesis, tested on per-capita rates.

**Predict.** Add to the worked population a hazard that removes individuals at a per-capita rate of 0.05 per day at every density: a removal rate of 5 a day at N = 100, 45 a day at N = 900. Does it regulate the population?

> **Check your prediction.** No. It kills more in total only because a fixed fraction of a bigger number is bigger. The per-capita line drops by 0.05 at every density with its slope unchanged, so it crosses zero earlier: where 0.2(1 − N/K) = 0.05, at N = 750. It lowers the level; the tilt that creates a level still comes from crowding.

So a brake shows only in per-capita rates measured across densities, which is what the second film goes looking for.

**Film transcript: Catching the Brake**

### Film transcript: Catching the Brake

The first film tilted the per-capita line by assumption. In a real forest or on a real reef, how would you catch that line tilting? Start in the Yukon, near Kluane Lake. Snowshoe hares there rise and fall in a cycle of roughly a decade.

Charles Krebs and colleagues worked on blocks of forest. They added food, fenced out mammal predators, did both, or did neither. During the cycle's peak and decline, fencing out mammal predators raised hare density, and so did food. Doing both raised it far more than either alone. So food and predators both limited the hares. But were individuals doing worse at high density?

Counts of hares cannot say. For that you need a per-capita rate, measured at several densities. Mark Hixon and Mark Carr did this on small patch reefs in the Bahamas. A per-capita measure counts outcomes per individual, not heads. Here it is the fraction of young fish that died during the study on each reef. They stocked newly settled blue chromis, a small reef fish, at different densities. Some reefs got few young fish, some got many. They also crossed two kinds of predator.

Residents live on the reef. Transients swim in from open water and move between reefs. Then they plotted the fraction of young fish that died against starting density. A fraction dying is a per-capita measure, so it is teal. With both predator groups present, the line rose steeply. The more young fish that settled on a reef, the larger the share that died. That rising line is density-dependent mortality. Predators caused it on these reefs, at least when both kinds hunted there together.

Now take away one of the two predator groups, either one. Does the teal line still rise as density goes up? With only one group, no rise was detected. But one of those tests had low statistical power, too little data to catch a real slope reliably. Not detected is not the same as absent. Notice what made this a test of density dependence. Not more deaths in total, but a larger fraction of the fish dying. The hares and the chromis answer two different questions. Adding food and removing predators showed what limited the hares. The chromis showed density dependence, within one generation and across reefs. A brake is negative density dependence.

As density rises, deaths rise or births fall, so net per-capita growth falls. You find it only by measuring per-capita rates.

From the film

- Adding food and fencing out mammal predators raised snowshoe hare numbers, which shows both were limiting, but counts alone do not show density dependence.
- On Bahamian patch reefs, the fraction of young blue chromis that died rose with starting density when both resident and transient predators were present, measured across reefs within one generation.
- With only one predator group, no rise was detected, and one of those tests had low statistical power: not detected is not the same as absent.
- A brake makes net per-capita growth fall as density rises, so it is judged by per-capita measures, such as the fraction that die, not by the total number killed.

## What regulates real populations

The second film skipped two limits. At Kluane each predator treatment had one fenced block, and the fences kept out lynx and coyotes but not birds of prey, so exclusion was partial.

*Figure (AI-generated illustration).* Food and predator removal showed what limited hare numbers, but each predator treatment was one block, and the fences let birds of prey through. Image: four equal squares of boreal forest in a two-by-two grid, each with the same two snowshoe hares. One is untouched, one has a feeding station, one is fenced with a lynx and a coyote outside, and one, outlined in gold, is both fenced and fed. A bird of prey flies over a fence. A small inset shows a fence stopping a lynx but not a bird.

The reef-fish slope, measured across reefs in one generation of young blue chromis, is density dependence in space. Regulation is a claim about return through time, and showing it needs evidence through time: long records, repeated density manipulations, or a tested demographic model built from measured per-capita rates.

*Figure (AI-generated illustration).* With both predator groups present, the fraction of young chromis that died rose with density, measured across reefs in one generation, not through time. Image: four equal patch reefs in a two-by-two grid with the same four blue chromis each: no predators, a resident predator in a crevice, a transient predator swimming in from open water, and both, outlined in gold. An inset compares a reef with fewer settled chromis and one with more.

Weather and density can also act together. Soay sheep on Hirta, in the St Kilda archipelago off Scotland, crash in some winters. Tim Coulson and colleagues reported in 2001 that lambs, adult males and adult females respond differently to density and to winter weather, so two populations of equal size under identical weather can follow different paths. The population’s makeup, which the logistic leaves out, changed the outcome.

## Checking a model against data

A model earns trust by predicting populations it was not fitted to (Drake’s water fleas, Going deeper). Pearl and Reed’s 1920 logistic fitted United States census counts well, but a close fit to points you already have does not show that the mechanism in the equation is at work.

Short records are a trap. Ecologists use **time series**, counts of the same population year after year, and ask whether a high count is followed by a low per-capita growth rate the next year.

**Predict.** Take many such series and test each one alone: the pooled picture is little evidence of a fall in per-capita growth after high counts. Are most of those populations unregulated?

> **Check your prediction.** Not necessarily. Loïc Thibaut and Sean Connolly reanalysed series from the Global Population Dynamics Database, a public collection of such records, and reported in 2020 that individual series gave little evidence of density dependence, while the series analysed together gave strong evidence. That points to low **statistical power** in short single series (too little data for a test to detect an effect that is there), not absent regulation.

The next two lessons, on [competition](https://stemhop.com/lessons/competition/) and [predation](https://stemhop.com/lessons/predation/), build their models directly on the logistic.

### Design problem

Take the Kluane result from the second film, a crossed design (each treatment alone and both together): during the peak and decline, hare densities were about 2 times control with mammalian predators excluded, 3 times with food added, and 11 times with both. Interpret the result. What does it show, what does it not show, and what would you do next?

> **One way to do it.** Start with a null expectation: if food and predation acted independently and their effects multiplied, the combined treatment should give about 2 x 3 = 6 times control density. If their gains above control simply added, it would be smaller still: 1 (baseline) + 1 (predator-exclusion gain) + 2 (food gain) = 4 times. The observed 11 times exceeds both, which suggests that the two factors interact, as the authors concluded. With one fenced block per predator treatment, the interaction is suggested, not statistically tested. One reading is that extra food helps more when fewer hares are killed, and predator removal helps more when hares are well fed. Both factors limited hare density in this experiment, but that does not show that food shortage drives the cyclic decline. Krebs’s later reviews conclude that predation is the key cause, including stress from predator chases that lowers reproduction.
>
> What it does not show: these density ratios are not r or K, and the experiment is not a test of the logistic model. The blocks were few, in one region over one run of years, so do not extend the result to all hare populations or all cycles.
>
> Next: replicate the combined and single treatments on more blocks, ideally in another region, and measure per-capita survival and reproduction against density in each treatment. That turns “density was higher” into the per-capita evidence that regulation requires.

## Going deeper

### Two equilibria

With r and K positive, N = 0 is also an equilibrium of the logistic, but unstable: any population above zero grows away from it, while K is stable.

### Real per-capita lines are often curved

One analysis of 1,780 animal time series (Sibly and colleagues, Science, 2005) found that per-capita growth typically fell fast at low density and then flattened, rather than falling in the logistic’s straight line. Such curves are hard to estimate from noisy counts, and sampling error in the counts can itself create spurious density dependence in a regression of growth on density.

### Allee effects

At very low density the per-capita rate can instead rise as density rises, so the line slopes up before it slopes down. This is an **Allee effect** (positive density dependence), and the logistic leaves it out.

### Chaos from the simplest model

Robert May showed in Nature in 1976 that a discrete-time logistic, with growth in seasonal steps so that this year’s crowding acts on next year’s number, can settle, oscillate, or fluctuate chaotically depending on r, with nothing random in the equation. Keep the continuous and discrete models apart: erratic counts need not mean an erratic environment.

### Drake’s Daphnia test

John Drake’s 2005 analysis followed 281 laboratory populations of *Daphnia magna* for 104 days under food supplies of differing variability. He fitted models to some populations and tested their extinction predictions on others. Models that let demographic chance (random variation in which individuals happen to be born or die) vary with density predicted the extinctions accurately or with only slight bias; models that assumed demographic chance was the same at every density were severely biased and underpredicted them. That result belongs to those microcosms.

---

From STEMhop (https://stemhop.com), free lessons by Christopher Deppmann. Lesson text © the authors. Practice questions: https://stemhop.com/lessons/population-growth/
