A population estimate can look reassuring because the same animals keep appearing in the records. But those familiar individuals may be the ones the observation method finds most easily. Animals rarely encountered can remain poorly represented even when the visible records are abundant.
The short answer is that unequal encounter probabilities can distort a simple capture–recapture estimate. When the same individuals are consistently easier to encounter on both occasions, the overlap can be larger than a representative-sampling argument assumes. A simple overlap formula can then make the population look too small. The direction depends on the observation process; it is not a universal correction rule.
This guide extends The Gold Standard of Zoology with original paper populations, probability calculations and changed-context checks. All animals, counts and probabilities below are fictional teaching constructions. Use cards or simulated records for the exercises. The article explains inference from encounter histories and does not give instructions for capturing, marking or handling wildlife.
eduKateSG · Zoology
Find your next learning step
Follow identifiable animals through the lists, then test what makes their overlap informative about abundance.
ROUTE 2 · CHAPTERS 6–11
Explain the mechanism
Derive the enrichment effect and test how its direction changes.
ROUTE 3 · CHAPTERS 12–18
Challenge the inference
Separate sampling variation, model limits and additional information.
ROUTE 4 · CHAPTERS 19–23
Practise and investigate
Change the population, inspect a report and check the target boundary.
Full chapter contents · Practice and transfer check · Zoology subject guide
Chapter contents
Build the model · Chapters 1–5
Explain the mechanism · Chapters 6–11
- 6. Explain why the overlap is enriched
- 7. Derive the general expected-count relationship
- 8. Distinguish conditional independence from pooled association
- 9. Compare time variation with individual variation
- 10. See why the direction is not universal
- 11. Examine a fixed first list with unequal second chances
Challenge the inference · Chapters 12–18
- 12. Keep sampling uncertainty separate from model failure
- 13. Understand what a small-sample adjustment cannot repair
- 14. Consider subgroup estimates without pretending the groups are known
- 15. Ask what a never-encountered group would change
- 16. Keep occupancy and abundance questions distinct
- 17. Preserve the information that helps distinguish explanations
- 18. See where spatial information can help
Practise and investigate · Chapters 19–23
Resolve and continue · Chapters 24–25
CHAPTER 1 OF 25 · Build the model
1. Start with individuals rather than photograph totals
Capture–recapture reasoning uses recognisable individuals observed on more than one occasion. The word capture can refer to an encounter in a suitable identification system; the inferential question is which individuals recur. A collection of photographs is not automatically a collection of distinct animals.
Imagine two lists of fictional animal identities. The first contains A, B, C and D. The second contains C, D, E and F. There are four individuals on each list, two in both lists and six distinct individuals observed altogether.
The six observed identities provide a direct count of what appeared at least once. A claim about additional unseen individuals requires a model of how encounters arise. That is a different step from removing duplicate records.
For the calculations below, assume identities are recorded correctly, each individual contributes at most one encounter indicator per occasion and the target population remains unchanged between occasions. These assumptions isolate the unequal-probability problem. Later chapters explain why identification and population boundaries still need separate checks in a real study.
Write n1 for the number of individuals encountered on the first occasion, n2 for the number on the second and m for the number encountered on both. A familiar simple estimate is N-hat = n1 × n2 / m, when m is positive.
Its intuitive argument compares fractions. If the first occasion identified n1 of the population’s N individuals, the identified fraction is n1/N. If the second occasion samples that population representatively with respect to first-occasion identification, its overlap fraction m/n2 should resemble n1/N.
Equating those fractions and rearranging gives the formula. The algebra is easy. The difficult part is whether the second list supplies the representative fraction that the explanation needs.
For example, n1 = 40, n2 = 50 and m = 10 produce an estimate of 200. This does not mean 200 individuals were observed. The observed union contains 40 + 50 − 10 = 80 distinct identities. The remaining difference comes from the model-based extrapolation. Label both numbers so a reader can see where observation ends and inference begins.
CHAPTER 3 OF 25 · Build the model
3. Understand what encounter probability describes
An individual’s encounter probability is its chance of appearing in a specified occasion’s record under the stated process. It belongs to the combination of individual, method, space, time and conditions. It is not simply an enduring personality score for the animal.
In a paper model, one card may have an encounter probability of 0.8 and another 0.2. Those numbers mean an 80% and 20% chance on the relevant occasion. They do not mean the first animal exists more strongly or represents four animals.
A real method may encounter individuals differently because of where they range, when they are available or how the observation system works. Which explanations apply needs biological and methodological evidence. The numerical model does not identify those causes merely by assigning different probabilities.
Keep population abundance and observation probability as separate unknowns. Many encounters can reflect more animals, easier detection, or both. Repeated identities help investigate that distinction, but their usefulness depends on the model connecting the observed histories to individuals never recorded.
Begin with a fictional population of 200 individuals. Every individual has probability 0.5 of encounter on each of two occasions. For this baseline, the two occasion indicators are independent for each individual.
The expected first count is 200 × 0.5 = 100. The expected second count is also 100. Each individual’s probability of appearing on both occasions is 0.5 × 0.5 = 0.25, so the expected overlap is 50.
Putting these expected counts into the simple ratio gives 100 × 100 / 50 = 200. The representative-fraction argument agrees with this deliberately uniform model.
This is a calculation using expected counts, not a guarantee for every random pair of lists. An actual draw can produce different counts and overlaps. Also, the ratio of expected counts is generally not the expected value of the random ratio estimator. We use it to expose the observation mechanism before adding sampling variation, rather than claim an exact unbiasedness result from three expectations.
Keep the true population at 200. Divide it into 100 high-encounter individuals and 100 low-encounter individuals. On each occasion, the high group’s probability is 0.8 and the low group’s is 0.2. Assume independence between occasions conditional on group.
The expected count on either occasion is 100 × 0.8 + 100 × 0.2 = 100. A dashboard showing only each occasion’s count would therefore resemble the uniform baseline.
The overlap changes. A high-group individual has probability 0.8² = 0.64 of appearing twice. A low-group individual has probability 0.2² = 0.04. The expected overlap is 64 + 4 = 68.
| Fictional group | Population size | Probability on each occasion | Expected first count | Expected overlap |
|---|---|---|---|---|
| High encounter | 100 | 0.8 | 80 | 64 |
| Low encounter | 100 | 0.2 | 20 | 4 |
| Combined | 200 | Population mean 0.5 | 100 | 68 |
Substituting those expected counts into the pooled ratio gives 100 × 100 / 68, approximately 147.1. The population did not shrink. The assumed representative overlap relationship failed because the same group was favoured on both occasions.
The first list is expected to contain 80 high-group and 20 low-group individuals. High-group individuals form half the population but four-fifths of the expected first list. The list overrepresents individuals with greater encounter probabilities.
The second list has the same expected composition. Individuals already represented after the first occasion are therefore more likely to appear again than a uniformly selected population member would be in this model.
The expected overlap fraction among the second list is 68/100 = 0.68, while the first list’s expected fraction of the whole population is 100/200 = 0.5. Substituting 0.68 for 0.5 in the representative-fraction argument makes the inferred denominator too small.
This is the mechanism worth explaining aloud. “Some individuals were hard to find” is a useful beginning, but it does not yet show why the overlap formula changes. Trace the chain: unequal probabilities change list composition; the changed composition increases recurring identities; the simple formula interprets that recurrence using a relationship the model does not satisfy.
CHAPTER 7 OF 25 · Explain the mechanism
7. Derive the general expected-count relationship
Suppose the fixed population has N individuals. Individual i has encounter probability pi on each occasion. Assume the two occasions are independent conditional on that individual’s pi, with pi unchanged between occasions.
The expected count on either occasion is Σpi. The expected overlap is Σpi², because each individual must appear on both occasions. The ratio formed from those expectations is (Σpi)² / Σpi².
Let μ be the population mean of the pi values and v their population variance, using division by N. Then Σpi = Nμ and Σpi² = N(μ² + v). The expected-count ratio becomes Nμ²/(μ² + v).
When μ is positive, this can be written N/(1 + v/μ²). If all probabilities are equal, v is zero and the expression equals N. If they vary, it is smaller. This algebra describes this particular stable-heterogeneity model; it is not a formula for correcting real data unless the needed probability information is independently justified. The unseen individuals’ probabilities are precisely part of what the encounter records do not reveal directly.
CHAPTER 8 OF 25 · Explain the mechanism
8. Distinguish conditional independence from pooled association
The two encounter indicators in the group example were independent after the group was specified. A high-group individual’s first result does not change its second-occasion probability of 0.8. The corresponding low-group probability remains 0.2.
Nevertheless, among randomly selected population members, first-occasion encounter is informative about second-occasion encounter. It makes high-group membership more likely. In the example, the joint encounter probability is 68/200 = 0.34, whereas multiplying the marginal probabilities gives 0.5 × 0.5 = 0.25.
There is therefore positive pooled association without a behavioural response to being encountered. This is a useful distinction when an explanation attributes every repeated identity to animals becoming more attracted to the method after the first encounter.
A behavioural response is another possible mechanism, but it is a different assumption. Identical-looking overlap patterns can have different explanations. State whether dependence arises through persistent differences among individuals, a change following an encounter, shared circumstances or some combination. The observed overlap alone does not identify which story is correct.
CHAPTER 9 OF 25 · Explain the mechanism
9. Compare time variation with individual variation
Unequal probabilities between occasions are not identical to unequal probabilities among individuals. Consider a different population of 200 in which everyone has probability 0.4 on the first occasion and 0.7 on the second, with occasion indicators independent.
The expected counts are 80 and 140. The expected overlap is 200 × 0.4 × 0.7 = 56. Their expected-count ratio is 80 × 140 / 56 = 200.
Changing the overall encounter rate did not produce the earlier mechanism because all individuals remained equally represented within each occasion. This does not establish that every real change of effort is harmless. It isolates one distinction in the mathematical model.
Compare the baseline, the stable two-group model and this time-varying model in separate rows. Ask which changes the average count, which changes overlap composition and which assumptions connect the two occasions. A precise explanation names the dimension of variation instead of treating the phrase “detection changed” as a complete diagnosis.
CHAPTER 10 OF 25 · Explain the mechanism
10. See why the direction is not universal
Now reverse which group is favoured. Of 200 individuals, 100 have probabilities 0.8 then 0.2, while the other 100 have probabilities 0.2 then 0.8. Assume independence between occasion indicators conditional on the group’s two probabilities.
The expected counts are again 100 and 100. The expected overlap is 100 × 0.8 × 0.2 + 100 × 0.2 × 0.8 = 32. The ratio formed from these expected counts is 312.5, above the true population of 200.
The second occasion now favours individuals that were harder to encounter on the first. The small overlap points in the opposite direction from the persistent-high-probability example.
This is why “heterogeneity always underestimates population size” is too broad. Underestimation was established for the specific unchanged-pi, conditionally independent model we derived. Changed probabilities, behavioural responses and other dependencies need their own analysis. Keep the useful result together with its conditions rather than memorising the direction while discarding the mechanism.
CHAPTER 11 OF 25 · Explain the mechanism
11. Examine a fixed first list with unequal second chances
There is another way to see the representative-fraction problem. Suppose the first list has already identified 50 of a closed population of 200. On the second occasion, each of those 50 has encounter probability 0.6, while each of the remaining 150 has probability 0.2.
The expected overlap is 50 × 0.6 = 30. The expected second-list count is 30 + 150 × 0.2 = 60. With the first list fixed, inserting those expected second-occasion counts gives 50 × 60 / 30 = 100.
The first list covers one-quarter of the actual population, but its members constitute half the expected second list. The second list is unrepresentative with respect to first-list membership.
This deliberately assigned model might represent several hypothetical mechanisms. It does not establish why the probability differs. Its teaching value is that the fraction mismatch is visible without requiring a random first-occasion calculation. Explain the sampling relationship first, then decide what biological evidence would be needed to connect that model to a real study.
CHAPTER 12 OF 25 · Challenge the inference
12. Keep sampling uncertainty separate from model failure
A correctly specified stochastic model still produces sampling uncertainty. Repeat a paper experiment with the same population and probabilities, and the lists, counts and overlaps vary. That is different from repeatedly applying a formula whose representative-sampling assumption is unsuitable.
In the stable two-group example, a large set of simulated replications would let a learner inspect the distribution of the actual ratio estimates, including occasions with very small overlap. Its average would not be obtained simply by dividing the expected counts we calculated earlier.
A useful classroom report can therefore have two parts. First derive the expected-count mechanism. Then describe the realised results from a clearly specified simulation. Keep the random-number procedure and number of replications visible if making numerical simulation claims.
No simulation results are claimed in this article. The cases are exact expectation calculations under supplied models. That distinction is more informative than attaching a confidence interval copied from a different method and presenting it as if it measured every uncertainty in the problem.
CHAPTER 13 OF 25 · Challenge the inference
13. Understand what a small-sample adjustment cannot repair
A common adjusted two-list expression is the Chapman estimate: ((n1 + 1)(n2 + 1)/(m + 1)) − 1. It addresses properties of the estimator within its relevant sampling model; adding one does not make an unsuitable encounter process representative.
Using the earlier illustrative counts n1 = 100, n2 = 100 and m = 68 gives (101 × 101 / 69) − 1, approximately 146.84. The adjustment leaves the result close to the pooled unadjusted value of 147.06, far below the stipulated population of 200.
Those counts are being reused as a numerical diagnostic, not claimed to be a random study outcome or a proof about the adjusted estimator’s average bias. The lesson is that a change to the arithmetic cannot automatically repair persistent differences in whose identities enter the lists.
When a learner suggests a correction, ask which problem it addresses. Small-sample behaviour, identification error, population change and heterogeneity are different problems. A method can help one while leaving another unresolved. The formula’s name is less useful than understanding the observation assumptions it requires.
CHAPTER 14 OF 25 · Challenge the inference
14. Consider subgroup estimates without pretending the groups are known
In the two-group teaching population, the analyst was given group membership and the true probabilities. If expected counts are analysed separately, the high group gives 80 × 80 / 64 = 100 and the low group gives 20 × 20 / 4 = 100.
Adding gives the stipulated total of 200. This demonstrates how pooling groups with different persistent probabilities created the earlier expected-count discrepancy.
It is not a general fieldwork recipe. An unseen individual has no observed record from which its group can automatically be assigned. Even recorded groups can contain further heterogeneity, and small subgroups may have little overlap. A convenient observed label does not establish equal probabilities inside that label.
Use stratification only when its statistical and biological conditions are justified for the real question. In the classroom, keep the supplied complete population table visible so students understand which information the exercise grants that an analyst may not actually possess. Otherwise the apparent solution quietly uses the unknown answer as an input.
CHAPTER 15 OF 25 · Challenge the inference
15. Ask what a never-encountered group would change
Suppose a record system can encounter one set of individuals but has zero probability of encountering another set during the study. Adding more individuals to that completely unobservable set changes the total population while leaving every possible observed encounter history unchanged.
This exact zero-probability construction makes a nonidentifiability problem easy to see: the record alone cannot distinguish those totals. Individuals with very small rather than zero probabilities create a less absolute but still serious practical information problem over a limited observation period.
William Link's 2003 study, indexed by USGS, examines population-size inference when heterogeneous detection models can imply different abundance conclusions from very similar observations. That research supports taking model assumptions seriously even when datasets are large.
The paper-card construction here is narrower and original. It does not prove that every abundance problem is impossible. It shows why more precision about the observed part of a population does not necessarily supply information about a part the method never reaches. Define the target and accessible population before deciding what the records identify.
CHAPTER 16 OF 25 · Challenge the inference
16. Keep occupancy and abundance questions distinct
The existing all-zero survey-history guide asks how repeated non-detections affect a claim about presence. The question here concerns how recurring identifiable individuals inform population size.
Both require an observation model, but their outputs differ. A site can be occupied by one animal or by many. Detecting a species establishes neither an exact count nor the identities needed for an individual capture–recapture analysis.
For a changed task, replace the two individual lists with two checkboxes saying the species was detected at a site. The overlap of positive checkboxes is no longer the overlap of recognisable individual animals. The numerical estimator cannot be transferred merely because both datasets contain zeros and ones.
Ask what a row represents: individual, site, visit, photograph or event. Then identify what the model estimates. This small check prevents an advanced-sounding analysis from beginning with the wrong observational unit. Follow the all-zero guide for the presence question and return here when individual encounter histories and abundance are the actual task.
CHAPTER 17 OF 25 · Challenge the inference
17. Preserve the information that helps distinguish explanations
A useful encounter-history table keeps an individual’s identity and its observation indicator for every defined occasion. It also preserves the occasion definitions and relevant metadata needed to interpret those indicators.
Do not replace histories with only the total number of encounters per animal if the timing matters to the proposed model. An individual seen early and an individual seen late can have the same total while providing different information about temporal conditions.
Identification uncertainty also needs a visible route. Incorrectly merging two identities can inflate overlap; splitting one identity can reduce it. Those errors can imitate or oppose the heterogeneity mechanisms in our examples. Retain reviewed determinations and the evidence supporting them rather than silently altering identities to improve model fit.
Sensitive location information belongs within an appropriate access arrangement. The zoology root provides the broader responsible-recording route. The aim is enough traceability for qualified review, with the statistical unit and observation process intact. A polished abundance number without its encounter histories and assumptions is harder to assess than a modest result whose evidence remains understandable.
CHAPTER 18 OF 25 · Challenge the inference
18. See where spatial information can help
Individuals may differ in encounter probability because their activity areas relate differently to observation locations. A model that includes where encounters occur can ask a different and more informative question than a pooled list-overlap calculation.
The USGS publication page for oSCR and spatial capture–recapture describes a framework connecting spatial encounter histories with population and landscape processes. It is an onward research route, not a claim that installing software resolves every source of heterogeneity.
For a paper exercise, draw three observation locations and two fictional activity centres. Ask why an individual near several locations may have different encounter opportunities from an individual near the edge. Do not calculate density until a defined spatial model, region and relevant assumptions have been supplied.
This helps learners see why additional kinds of information can matter more than accumulating the same summary count. The appropriate specialist method depends on the target, design and data. A complicated model fitted to an unsuitable record does not automatically provide a stronger biological conclusion.
CHAPTER 19 OF 25 · Practise and investigate
19. Test a changed population without copying the first numbers
Consider 300 fictional individuals: 100 with probability 0.6 and 200 with probability 0.3 on each of two occasions. Assume the probabilities remain unchanged and the occasions are independent conditional on individual probability.
The expected count on either occasion is 100 × 0.6 + 200 × 0.3 = 120. The expected overlap is 100 × 0.36 + 200 × 0.09 = 54. The ratio of expected counts is 120 × 120 / 54, approximately 266.67.
Now replace the two groups by 300 individuals each having probability 0.4. The expected occasion count is still 120, but the expected overlap becomes 300 × 0.16 = 48. The ratio becomes 300.
The average encounter probability is identical across the two models. Persistent individual variation changes the overlap. Explain that difference before computing either ratio. If the learner can only say the first formula produced a smaller number, return to the enriched-list explanation. Understanding the mechanism lets the result travel to a changed population and a changed group balance.
CHAPTER 20 OF 25 · Practise and investigate
20. Diagnose a fictional report in a useful order
An imaginary report says: “We recorded 100 individuals twice, with 68 recurring identities, so exactly 147 animals live in the area.” Begin by clarifying whether each occasion contains 100 distinct identities and whether 68 is their intersection. Ambiguous wording should be repaired before analysing the estimator.
Next separate the observed union, 100 + 100 − 68 = 132, from the inferred total. Then ask which closed-population and encounter assumptions justify the chosen estimate. The word exactly is unsupported for a model-based estimate from stochastic observations.
Finally consider which evidence could challenge the model: individual histories, observation locations, identity review, occasion conditions and biological information relevant to availability. These are questions for examining a report, not a new wildlife-intervention plan.
A useful rewrite would identify the estimator, report its uncertainty under the stated model and explain sensitivity to persistent encounter differences. If those checks have not been performed, say so. The aim is a conclusion readers can assess and improve, rather than replacing one unjustified exact number with another equally confident number chosen from a classroom example.
CHAPTER 21 OF 25 · Practise and investigate
21. Understand what known probability weighting would require
There is a useful theoretical contrast with the overlap formula. On one occasion, suppose every individual’s encounter probability pi is known, positive and appropriate to that occasion. Let Ii equal one if individual i appears and zero otherwise. Then E[Ii/pi] = pi/pi = 1.
Summing Ii/pi over the encountered individuals therefore has expectation N under those supplied inclusion probabilities. This is an inverse-probability counting argument. The proof follows by adding each individual’s expected contribution; it does not require every individual to have the same probability.
For the fictional high and low groups, expected encounters are 80 and 20 with probabilities 0.8 and 0.2. The expected weighted total is 80/0.8 + 20/0.2 = 200. A recorded low-probability identity receives a larger weight because fewer such individuals are expected to appear.
The important word is known. Assigning an estimated encounter probability to an observed identity is not automatically equivalent to knowing the true inclusion probability. Estimation uncertainty, misspecification and the lack of observed information about unseen groups remain relevant. Zero inclusion probability also prevents this calculation from covering the entire target population.
Do not calculate 1/pi by substituting an animal’s fraction of recorded visits and assume the proof still applies unchanged. The probability in the argument is the relevant inclusion probability under a defined sampling design or justified model, not any convenient empirical rate. The theoretical contrast identifies valuable information; it does not conjure that information from a pair of lists.
CHAPTER 22 OF 25 · Practise and investigate
22. Plan a paper investigation that distinguishes the mechanisms
Create a complete population table on paper with 200 labelled rows. Give half the rows probability 0.8 on each occasion and half probability 0.2. In another table assign everyone probability 0.5. These tables define the two models before any lists are generated.
First calculate expected list sizes and overlaps analytically, as earlier chapters did. Then, if using a computer simulation, generate independent occasion indicators conditional on each row’s assigned probabilities. Preserve the complete hidden table separately from the encounter-only table that a hypothetical analyst receives.
The separation matters. The teacher can evaluate an estimate against the known population. The analyst cannot simply read the hidden group labels for individuals never encountered. Students should state which information belongs to each viewpoint before proposing a method.
Add a third model with probabilities 0.8 then 0.2 for one group and the reverse for the other. Predict the direction of the expected-count overlap effect before generating any records. This checks whether students understand persistent versus opposing encounter opportunities, rather than merely associate unequal probabilities with one memorised bias direction.
If the simulation is run, report its actual settings and realised output. If it is not run, retain the exercise as a proposed investigation and do not invent a histogram or a claimed average. The analytic answers already establish a substantive comparison. Additional numerical work is useful when it answers a new question, such as how variable the realised estimator becomes when overlap is small.
CHAPTER 23 OF 25 · Practise and investigate
23. Check the population boundary before interpreting overlap
The closed-population examples keep the same individuals available across the two occasions. If individuals enter, leave, are born or die between occasions, changes in list membership can have another explanation. A small overlap might then reflect a changing target population rather than the opposing encounter probabilities in our constructed model.
For a paper comparison, keep the first list fixed and replace twenty members of the underlying population before constructing the second list. The second occasion can no longer encounter those departed individuals. Applying the unchanged-population argument without acknowledging the replacement gives the equation a job its assumptions do not support.
This is a distinct check from whether the study area’s boundary was drawn clearly. A geographic boundary can remain fixed while the individuals using the area change. Conversely, a specified set of individuals may be followed through changing locations. Define which population the question concerns and how availability relates to that definition.
An abundance report should explain the relevant time and spatial target in ordinary language. That gives a reader a concrete basis for assessing the closure assumption before considering a more advanced model of individual differences.
CHAPTER 24 OF 25 · Resolve and continue
24. Questions about unequal encounter probabilities
Do more recaptures always mean a smaller population?
Holding the two list sizes fixed, the simple formula decreases as overlap increases. Whether that is a sound biological inference depends on how the overlap was generated. Persistent encounter differences can increase overlap without any reduction in true abundance.
Can we multiply the estimate by the factor from the example?
No. The two-group factor follows from probabilities supplied for a fictional complete population. It is not an empirically established adjustment for a different dataset. Choosing a correction requires evidence and an appropriate inferential method.
What if the overlap is zero?
The unadjusted ratio is undefined. Zero overlap does not establish an infinite population. It signals limited information for that calculation and calls for model-specific assessment of the design and data, including identification and whether the two occasions access the same population.
Is every capture–recapture estimate unreliable?
No. These methods can answer important questions when designs, records and models fit the target. The lesson is to examine the encounter process and sensitivity, rather than accept or reject an entire family of methods from one formula.
CHAPTER 25 OF 25 · Resolve and continue
25. Continue through animals, observation and inference
For the biological foundation and responsible recording, return to Zoology. For an introductory mathematical route, read Biodiversity Surveys, Species Richness and Capture–Recapture. It introduces the basic overlap estimator; this guide develops persistent individual differences and their effect on that reasoning. For the related presence question, use Interpreting All-Zero Survey Histories with Imperfect Detection. Their common lesson is to connect a claim with the observation process that could support it, while retaining each question’s different statistical unit.
For specialist research, continue into Link's study of heterogeneous detection and nonidentifiability and the spatial capture–recapture framework described with oSCR. These sources establish the research context. The constructed populations, algebraic comparisons and report-repair exercise above are independent teaching examples, not results from those studies.
The wider learning route is The Gold Standard of Education: Learning That Transfers into Life within How X Works. Change the group sizes, keep the average encounter probability fixed and predict the overlap before calculating. Then explain which assumptions made that prediction possible. That is a concrete step from remembering an overlap formula to understanding what its answer depends on.