Notes on Population Counting, Sampling, and Mark-Recapture

Population counting and sampling: humans, animals, and the environment

  • Enumeration versus estimation

    • For humans in the US, almost everyone is counted. Past methods included door-to-door data collection on Saturdays to reach people who might be traveling by train, air, or bus. Some people could be missed due to timing or location.

    • Now there is online data input via a website where households enter their information, which leads to more complete counting but still involves practical limits.

    • The key idea: some populations are counted exactly (almost everyone for humans in practice), while others are too large or mobile to count one-by-one, requiring estimates.

  • Counting non-human populations (e.g., animals) is not feasible for every individual

    • You cannot count every rabbit in a large area like Louisiana by counting each rabbit one by one.

    • Instead, you count in a representative sample area (a plot) and then extrapolate to the larger area:

    • Count the number of individuals in a sampled plot of known area, then scale to the total area.

    • Challenge: counting from one small area may not represent the whole region if there is spatial variation (some areas may have many rabbits, others few).

    • Therefore, you should count in multiple areas and average the results to improve the estimate.

    • Random sampling is crucial: pick several locations randomly rather than selecting a single location to avoid bias.

    • If you select a single section and it happens to be unusually high or low, the estimate will be biased.

    • Practical approach: randomly pick multiple locations, count in each, then average the results to estimate the overall population.

    • Limitation: you still cannot count everyone; sampling provides an estimate with some uncertainty.

  • A practical technique: mark-recapture (a.k.a. capture–mark–recapture) for mobile populations

    • When direct counting is impractical, you can estimate population size by tagging or marking a subset of individuals, releasing them, and then recapturing later.

    • Steps in the method:

    • First capture: catch n1 individuals and mark them (tag, GPS, color mark, etc.). Release them back into the population.

    • Second capture: after some time, catch n2 individuals. Among these, some m2 are marked (recaptured).

    • Estimate total population using the Lincoln–Petersen formula: N^=MCR\hat{N} = \frac{M C}{R} where

      • M = number initially marked (here n1),

      • C = size of the second sample (here n2),

      • R = number of marked individuals recaptured in the second sample (here m2).

    • Worked example from the transcript:

    • First catch: n1 = 200 (marked and released)

    • Second catch: n2 = 400

    • Recaptured marked individuals: m2 = 100

    • Estimate: N^=200⋅400100=800\hat{N} = \frac{200 \cdot 400}{100} = 800

    • Assumptions and caveats of mark-recapture:

    • The population is closed between the two captures (no births, deaths, immigration, or emigration affecting numbers).

    • Marks are not lost or overlooked, and they do not affect the chance of recapture.

    • Every individual has the same probability of being captured in each sample (equal catchability).

    • Marked and unmarked individuals mix back into the population without bias.

    • Real-world relevance: mark-recapture is commonly used in ecology to estimate wildlife populations (e.g., rabbits, birds, fish) when full counts are impractical.

  • Connection to real-world environmental context

    • Human activities (e.g., driving cars) contribute to environmental pollutants in exhaust (e.g., sulfur dioxide, carbon dioxide).

    • Carbon dioxide is a greenhouse gas contributing to the greenhouse effect and global climate regulation.

    • The Arctic and other global regions are part of the broader environmental system affected by these emissions.

    • Everyday practices outside professional counts also influence environmental data, such as energy sources and cooking fuels.

    • The transcript references smog in India and the use of cow dung as fuel, illustrating different regional practices that impact air quality and climate.

    • Visuals or diagrams (referred to in the talk) help connect local actions to global environmental patterns.

  • Key takeaways for data collection and estimation

    • For humans: census-like processes aim for near-complete enumeration; online and mail methods complement door-to-door efforts.

    • For non-human populations: exact counts are often infeasible; use sampling across multiple random locations to produce robust estimates.

    • Avoid single-location sampling, which can produce biased estimates; randomize location selection and average across samples.

    • Mark-recapture provides a practical estimation method for mobile or elusive populations, with an explicit formula and clear assumptions.

    • Always consider ecological and methodological assumptions when interpreting estimates (e.g., population closure, mark retention, equal catchability).

  • Formulas and numerical references (LaTeX)

    • Lincoln–Petersen estimator: N^=MCR\hat{N} = \frac{M C}{R}

    • Where $M$ = number marked in the first capture, $C$ = size of the second capture, $R$ = marked recaptured in the second capture.

    • Worked example: N^=200⋅400100=800\hat{N} = \frac{200 \cdot 400}{100} = 800

    • Density-based or area-based extrapolation (sampling over area):

    • If you sample $s$ plots with areas $Aj$ and counts $nj$ for each plot, density per plot: d<em>j=n</em>jAjd<em>j = \frac{n</em>j}{A_j}

    • Estimated density: d^=1s∑<em>j=1sd</em>j\hat{d} = \frac{1}{s} \sum<em>{j=1}^s d</em>j

    • Estimated total population: N^=d^⋅Atotal\hat{N} = \hat{d} \cdot A_{\text{total}}

    • If plots have equal area $Ap$ and the mean count per plot is nˉ\bar{n}, then: N^=nˉ⋅A</em>totalAp\hat{N} = \bar{n} \cdot \frac{A</em>{\text{total}}}{A_p}

  • Ethical, practical, and interpretive considerations

    • While estimates are valuable, they carry uncertainty and potential bias; report uncertainty where possible (confidence intervals, standard errors).

    • Census data implicates privacy and consent considerations; online collection must safeguard personal information.

    • Sampling design affects representativeness; strive for randomization and stratification when appropriate (e.g., stratify by habitat type or region).

    • Policy and resource decisions based on estimates should account for uncertainty and potential biases in data collection.