How Science Works | Why Removing One Species Can Reshape an Ecosystem

“Species P is important” leaves almost everything unspecified. Important for which outcome: the number of resident species, vegetation cover, decomposition, nutrient retention or another function? Over which area, season and length of time? Compared with what level of P’s abundance?

A testable version might be: reducing P in this community causes a larger decline in non-focal species richness than occurs in comparable communities where P remains. “Non-focal” excludes P itself. That exclusion matters because removing one species trivially subtracts it from a total count; the scientific question concerns effects on the rest of the community.

The proposed mechanism could be that P consumes a strong competitor, M, allowing other residents to persist. This predicts more than a final richness change. M should increase or expand its occupation of space, and changes in other residents should be consistent with that expansion.

Other mechanisms would make different predictions. P might alter prey behaviour without greatly changing prey numbers. It might transport nutrients or modify habitat. Naming the mechanism determines which intermediate observations are informative.

Finally, a large effect is not automatically a keystone effect. Removing a dominant habitat-forming organism can have major consequences because it supplies much of the physical structure or biomass. To argue for disproportionate influence, the effect must be interpreted alongside the focal species’ representation and the chosen comparison. Keystone status is an evidence-based description of a relationship within a setting, not a permanent badge attached to a species name.

Contents · Next chapter

Robert Paine’s 1966 paper described an intertidal comparison at Mukkaw Bay, Washington. Pisaster sea stars were absent from an experimental shoreline area while an adjacent control remained. Barnacles initially occupied much of the available space, followed by increasing occupation by mussels and other space competitors. The reported resident-species count declined from fifteen to eight while the control’s appearance remained comparatively unchanged. Paine linked the losses to competition for space and indirect effects, rather than the predator eating every affected species. Original paper, especially pages 70–71.

The paper also stated that the changing area had not reached equilibrium and that microfaunal changes were unmeasured. The focal shoreline comparison was not a modern many-site randomized experiment. These limits narrow the inference without erasing the importance of observing an intervention and a plausible mechanism together.

The reasoning advance is transferable. A description of two naturally different shores leaves open many causes. Changing one ecological component while following a comparison area provides a more directed test. Recording the intermediate changes then helps distinguish a proposed causal pathway from a mere before-and-after coincidence.

The next chapters use a separate invented dataset to make that logic quantitative. Its replicated sites, counts and cover values are teaching constructions, not reconstructed Paine data. Keeping those two evidence types separate prevents a worked example from quietly acquiring the authority of a historical experiment.

Contents · Previous chapter · Next chapter

Imagine six independent study sites in a hypothetical research dataset. Three are assigned to a predator-reduction condition, and three to an unchanged comparison condition. Measurements occur before and after the intervention using the same survey effort. Assume, for the exercise, that assignment and measurement are suitable and that sites do not influence one another.

At the three comparison sites, non-focal resident-species richness changes from 14 to 13, from 15 to 14 and from 13 to 12. At the three reduction sites, it changes from 14 to 8, from 16 to 9 and from 12 to 7.

These are site-level values. Each site could contain many photographs or sample squares, but those subsamples would help estimate that site’s condition rather than turn one intervention into many independent interventions. The number of independently assigned sites remains central to the design.

A before–after comparison within a reduction site removes its fixed starting difference from other sites. A comparison with unchanged sites then asks how much additional change occurred beyond the background trend. This is the logic of a before–after–control–impact comparison, often called BACI.

The controls are not assumed to remain perfectly constant. In fact, every comparison site loses one recorded species in this exercise. Weather, regional recruitment or seasonal change could affect all sites. Their purpose is to inform what might have happened in the reduction sites had the predator remained, not to provide an artificially motionless picture of nature.

Contents · Previous chapter · Next chapter

The comparison-site mean starts at (14 + 15 + 13)/3 = 14 species and ends at (13 + 14 + 12)/3 = 13. Its mean change is −1 species.

The predator-reduction mean also starts at 14 species: (14 + 16 + 12)/3. It ends at (8 + 9 + 7)/3 = 8. Its mean change is −6 species.

The difference between changes is therefore −6 − (−1) = −5 species. Under the exercise’s causal assumptions, the estimated additional richness loss associated with predator reduction is five species per site on average. Reporting the raw six-species decline alone would attribute the shared background decline to the intervention as well.

The same arithmetic can be written as a counterfactual prediction. Starting from a reduction-site mean of 14, apply the comparison trend of −1 to predict a no-reduction final mean of 13. The observed final mean is 8, which lies five species below that reference.

Diagram to sketch: put before and after on the horizontal axis and mean non-focal richness on the vertical axis. Both groups begin at 14. Draw the comparison line down to 13 and the reduction line down to 8. At the final time, mark the five-species gap. Label the reference as an estimated counterfactual, not an observed second history of the treated sites.

The calculation provides an effect estimate, not a confidence interval or a guarantee of causality. Six invented site records do not justify a claim of statistical certainty. Real analysis would examine site variation, sampling uncertainty, baseline trends and the actual assignment mechanism.

Contents · Previous chapter · Next chapter

Suppose M, a space-occupying competitor, is the proposed intermediate link. In the same invented dataset, its cover at comparison sites changes from 30%, 35% and 25% to 35%, 40% and 30%. Mean cover rises from 30% to 35%, a five-percentage-point increase.

At predator-reduction sites, cover changes from 30%, 40% and 20% to 80%, 85% and 75%. Mean cover rises from 30% to 80%, an increase of fifty percentage points. The difference between changes is 50 − 5 = 45 percentage points.

The larger expansion of M is consistent with the proposed pathway: predator reduction → increased occupation by M → less available space for other residents. It strengthens the explanation beyond a richness count alone. Nevertheless, both M and richness might also respond to an unmeasured treatment-associated change.

To evaluate the pathway more closely, researchers could combine timing, feeding observations, recruitment records and independently informative contrasts that distinguish M’s effect from other consequences of predator loss. Merely putting M into a statistical model does not automatically establish that it mediates the whole effect.

Pay attention to units. Cover changing from 30% to 80% is a fifty-percentage-point increase, not a fifty-per-cent increase. Relative to its starting value, it is an increase of about 166.7%. Neither wording describes the richness effect, which is measured in species per site.

A clear conclusion would say that the richness and cover contrasts support the predicted sequence under the design’s assumptions. It would not claim that a cover graph has proved every arrow or that no other pathway contributed.

Contents · Previous chapter · Next chapter

The central counterfactual assumption is that, without predator reduction, the treated sites would have experienced a comparable background change. Multiple pre-intervention observations can help evaluate whether their earlier trajectories were similar. They cannot guarantee that an unobserved future disturbance would affect both groups equally.

If a storm strikes only the reduction sites, the difference between changes combines predator effects with storm effects. Random assignment helps balance unknown differences on average, but small samples can still be imbalanced and later events can complicate interpretation.

Spillover is another concern. If predators or prey move between treatment and comparison sites, one site’s intervention can change another site’s outcome. The neat separation assumed in the exercise then fails. Ecologists must choose spatial units appropriate to the organisms and account for connections rather than treating nearby patches as automatically independent.

The measurement process can also generate an apparent loss. In a separate toy detection model, suppose twenty species truly remain present throughout. If each has a detection probability of 0.8 during a survey, the expected observed richness is 20 × 0.8 = 16. If detection probability falls to 0.5, the expectation becomes 10, despite no true species loss.

These equal detection probabilities are invented for explanation; real species vary. The example shows why altered visibility or sampling conditions can mimic a six-species decline. Repeated comparable surveys and detection-aware analyses address this problem. A larger spreadsheet does not repair a measurement process that changes systematically with the treatment.

Contents · Previous chapter · Next chapter

A predator making up a small fraction of community biomass could have a large effect if it restrains a competitively dominant species. But “small fraction” needs a denominator: which organisms, spatial boundary, season and biomass measure are included? Counts and biomass can tell different stories when body sizes vary greatly.

Consider two hypothetical focal species, each associated with the same standardized effect on a community outcome after comparable reductions. One initially contributes 2% of the defined community biomass and the other 40%. The first has the more disproportionate effect relative to that biomass measure. This comparison still says nothing by itself about which species is more valuable or which management action is appropriate.

Raw ratios can become unstable when the abundance estimate is tiny or uncertain. Effect estimates can also depend on how far abundance was reduced: losing the final few individuals may cross a threshold that a modest initial reduction does not. A single “importance score” should not conceal those conditions.

Related terms identify different mechanisms. A foundation species helps define habitat or community structure, often through its abundance or physical presence. An ecosystem engineer changes the physical environment in ways that affect other organisms. A dominant species is highly represented according to a stated measure. These descriptions can overlap, but none is interchangeable with demonstrated disproportionate influence.

Importantly, the original richness exercise lacks a predator biomass or abundance estimate. It can illustrate a substantial community effect and its causal analysis. By itself, it cannot complete the relative-abundance argument needed for a strong keystone classification. Recognizing that missing measurement is part of doing the science.

Contents · Previous chapter · Next chapter

An interaction’s consequences depend on the surrounding community. Alternative predators may compensate, prey may have refuges, resources may limit growth, or competitors may redirect the response. These possibilities are testable explanations for variation, not excuses to ignore an inconvenient result.

A 2025 study by Langendorf and colleagues compared two thirty-year kelp-community datasets associated with sea-otter recolonization, off Vancouver Island and around San Nicolas Island. The classic otter–urchin–kelp cascade was stronger in the former setting and muted in the latter. Their dynamic community models linked the contrast to changing interaction strengths and indirect pathways. This was long-term observational evidence combined with modelling, not a randomized otter-removal experiment. Original PNAS paper.

Another 2025 study, by Yeeles and colleagues, experimentally suppressed dominant ant species in Australian communities. Remaining and newly arriving species compensated, and the measured multifunctional performance increased rather than collapsing. The authors also identified potential sensitivity to further species loss. The bounded lesson is that abundance decline need not translate directly into loss of every measured function; it is not a general recommendation to remove dominant species. Original study in the authors’ university repository.

These findings point to a better question than “Is this species always a keystone?” Ask which interactions make its effect strong here, which alternatives weaken that effect elsewhere, and whether those conditions change through time. Community context is part of the mechanism being explained.

Contents · Previous chapter · Next chapter

A species can disappear through emigration, selective extraction, failed recruitment or mass mortality. These routes do not necessarily leave the same ecological surroundings. Mortality can add decomposing material even as it removes living consumers; emigration does not leave the same local resource pulse.

Tye and colleagues’ 2024 freshwater food-web experiment explicitly distinguished predator loss from the resource input associated with predator decomposition. The resulting dynamics differed from removal or resource addition alone. Their findings warn that a loss of living predators and the material consequences of their death can act together, changing the expected response. Original research abstract.

Timing also changes what an endpoint means. Immediately after a predator decline, prey behaviour may change before prey abundance does. Later, recruitment, competition and habitat modification may become more important. A survey at one date can capture only one stage of that sequence.

Restoring a predator does not guarantee that the community retraces its earlier path. Some residents may no longer be nearby to recolonize, habitat may have changed, or the returned predator may face different prey and competitors. A reversal test can be informative, but imperfect reversal is not automatically proof that the original causal explanation was wrong.

For the fictional sites, imagine richness remains low even after M’s cover decreases. That would motivate questions about recruitment, dispersal and time lags rather than an immediate declaration of permanent collapse. Recovery must be defined by explicit outcomes and observed over biologically relevant times, with ethical and authorized research decisions left to qualified investigators.

Contents · Previous chapter · Next chapter

For the paper exercise, a defensible conclusion is: the predator-reduction sites lost five more non-focal species per site on average than predicted from the comparison-site trend, while the proposed space competitor increased by an additional forty-five percentage points of cover. The pattern supports the predicted indirect pathway if the assignment, counterfactual and detection assumptions hold.

Then state what remains unresolved. The numbers do not establish a confidence interval, identify every intermediate cause or show disproportionate influence relative to predator biomass. They do not predict the response of a different shore, season or community. Those are additional questions requiring additional evidence.

A useful reading checklist has five questions. What exactly changed? Which outcome was measured? What estimates the missing no-change history? Which observations support the proposed pathway? Which plausible alternative could still explain the contrast? Answering them turns an appealing ecological story into an auditable explanation.

For safe classroom practice, use published data, photographs taken without disturbance or the invented records here. Compare assumptions and redraw causal arrows. No organism needs to be removed for you to learn why an intervention can reveal connections that a species list hides.

The central lesson is that ecological importance emerges through relationships, and relationships require testing. A small part of a community can exert a large influence, but its effect belongs to a particular network, place and time. Continue with Ecology, use the Food Webs guide for network foundations, or return to the How Science Works master guide.

Contents · Previous chapter · Continue to Ecology