Capture-Mark-Recapture (CMR) is one of the most widely used statistical methods for estimating wildlife populations in ecological studies. It involves capturing a sample of animals, marking them in a way that allows for future identification, and then releasing them back into the wild. After some time, another sample is captured, and the number of marked individuals in this second sample helps researchers estimate the total population size using statistical models.

However, the reliability of CMR estimates depends on six key assumptions, which, if violated, can result in biased population estimates. In this article, we will discuss each of these assumptions in detail, their significance in statistical modeling, and the consequences of not adhering to them.

Overview of Capture-Mark-Recapture in Statistical Context

Before delving into the assumptions, it is important to understand why CMR is so commonly used in wildlife statistics. The method offers a non-invasive way to infer population sizes without needing to capture every individual. By marking a portion of the population and observing the proportion of those individuals recaptured later, researchers can apply statistical formulas to estimate the total population.

The simplest CMR model is the Lincoln-Petersen estimator, which assumes a closed population and equal catchability between marked and unmarked individuals. However, more sophisticated models have been developed to account for the complexities of real-world populations, including open populations, variation in catchability, and the possibility of mark loss.

The Lincoln-Petersen Estimator Formula:

The basic formula used in CMR is:

\[N = \frac{M \times C}{R}\]

Where:

  • $N$ is the estimated population size,
  • $M$ is the number of marked individuals in the first sample,
  • $C$ is the total number of individuals captured in the second sample,
  • $R$ is the number of recaptured marked individuals in the second sample.

This formula provides a starting point, but its accuracy depends on whether the six assumptions discussed below are fulfilled.

Assumption 1: Closed Population

In CMR studies, a closed population means that the population does not change in size between the marking and recapture phases. No individuals are allowed to enter or leave the population, and no births or deaths occur during the study period.

Statistical Implications:

A closed population is critical for ensuring that the marked-to-unmarked ratio stays consistent between the two sampling periods. If immigration or emigration occurs, the marked proportion of the population could decrease or increase, leading to over- or underestimation of the population size.

Models for Open Populations:

When a closed population cannot be assumed, researchers often turn to more advanced statistical models, such as the Jolly-Seber model, which accommodates changes in population size. These models include parameters for immigration, emigration, births, and deaths, but they require more extensive data and sampling to remain accurate.

Assumption 2: Equal Catchability

For CMR to yield reliable estimates, each individual in the population must have an equal probability of being captured. This assumption, known as equal catchability, ensures that the marked and unmarked individuals have the same chance of being recaptured in subsequent sampling events.

Statistical Bias in Unequal Catchability:

If marked individuals become more or less likely to be recaptured due to trap avoidance (trap-shyness) or attraction (trap-happiness), the estimated population size will be skewed. Trap-shy animals may avoid future capture, leading to overestimation of the population, while trap-happy animals may be recaptured disproportionately, leading to underestimation.

To account for unequal catchability, statisticians can employ heterogeneous capture models such as the Huggins model, which incorporates individual heterogeneity in capture probabilities.

Assumption 3: No Mark Loss

A critical assumption in CMR is that no marks are lost or overlooked between the marking and recapture events. If marks degrade, disappear, or are not identified correctly during recapture, marked individuals might be counted as unmarked, which would bias the estimate of population size.

Statistical Correction:

When mark loss is suspected or unavoidable, researchers may use multi-state models that account for the possibility of a transition from the marked to unmarked state over time. This requires a larger sample size and more complex statistical analysis but can help mitigate the effects of mark loss on the population estimate.

Assumption 4: Random Mixing of Marked Individuals

After the initial capture and marking, it is assumed that the marked individuals randomly mix back into the population. This ensures that the marked individuals are evenly distributed throughout the population, so the sample collected during recapture represents a random cross-section of both marked and unmarked individuals.

Impact of Non-Random Mixing:

If marked individuals remain in clusters or specific areas, the marked-to-unmarked ratio in the recapture sample may not reflect the true distribution of the population. This can lead to biased estimates, particularly in species with strong territorial behavior or social structures.

Statistical methods to address non-random mixing include spatially explicit capture-recapture (SECR) models, which account for the geographic location of captures and recaptures to produce more accurate estimates in cases where individuals do not mix randomly.

Assumption 5: Instantaneous Sampling

The assumption of instantaneous sampling means that the time between the marking and recapture events should be short enough to avoid any significant changes in the population. A longer interval increases the chances of violations of the closed population assumption, such as immigration, emigration, births, and deaths, as well as potential mark loss.

Temporal Considerations:

If the time interval between capture events is too long, the population may no longer remain effectively closed. In such cases, researchers can use robust design models, which allow for multiple sampling occasions within a study period, giving a more detailed understanding of the population dynamics.

Assumption 6: No Behavioral Change

Finally, it is assumed that the process of marking individuals does not affect their behavior in a way that influences their probability of recapture. If individuals become more wary of traps or more inclined to avoid humans after being marked, their likelihood of being recaptured decreases. Similarly, marked individuals could become more prone to recapture if the marking process makes them less fearful.

Behavioral Bias in Statistical Estimates:

When behavioral changes are suspected, models such as the Cormack-Jolly-Seber model can help account for variation in recapture probability due to changes in individual behavior. This model separates the capture probability from the survival probability, allowing for adjustments in cases where marked individuals are recaptured at different rates than unmarked ones.

Consequences of Assumption Violations

Violations of any of the six assumptions can introduce biases into population estimates derived from CMR studies. In most cases, these biases lead to either overestimation or underestimation of the population size, which can significantly impact wildlife management decisions, conservation efforts, and our understanding of species’ ecological roles.

Addressing Violations:

Several advanced statistical models are available to help address potential violations of the CMR assumptions:

  • Jolly-Seber models for open populations,
  • Huggins models for unequal catchability,
  • Multi-state models for mark loss,
  • SECR models for non-random mixing,
  • Robust design models for longer time intervals between sampling events,
  • Cormack-Jolly-Seber models for behavioral changes.

These models add complexity to the statistical analysis, but they allow researchers to adjust for the realities of working with wildlife populations in dynamic environments.

Conclusion

Capture-Mark-Recapture is a robust and widely used method in wildlife population estimation, but its reliability depends heavily on fulfilling six key assumptions: closed population, equal catchability, no mark loss, random mixing, instantaneous sampling, and no behavioral change. Understanding and addressing potential violations of these assumptions is crucial for obtaining accurate and unbiased population estimates.

When these assumptions cannot be fully met, advanced statistical models offer alternative approaches that account for various violations, making CMR an adaptable and reliable method even in challenging field conditions. By carefully designing CMR studies and employing appropriate statistical techniques, researchers can ensure that their population estimates are as accurate and informative as possible.

Appendix: Advanced Statistical Models in Capture-Mark-Recapture (CMR)

The accuracy of population estimates in Capture-Mark-Recapture (CMR) studies can be influenced by violations of the method’s core assumptions. To address these issues, several advanced statistical models have been developed. Below is an overview of key models used in CMR studies to correct for assumption violations.

Jolly-Seber Models for Open Populations

The Jolly-Seber model extends the basic CMR framework to populations that are open, meaning that individuals can enter or leave the population between sampling periods. This model accounts for immigration, emigration, births, and deaths, providing estimates of population size, survival rates, and recruitment rates.

  • Use Case: When the assumption of a closed population cannot be met.
  • Advantages: Allows for population changes between sampling events.
  • Limitations: Requires more complex data collection and a higher number of recapture events.

Huggins Models for Unequal Catchability

The Huggins model addresses the issue of heterogeneous capture probabilities, where individuals in a population do not all have the same chance of being captured. It allows capture probabilities to vary by individual and uses a conditional likelihood approach that eliminates the need to estimate population size directly in the initial phase.

  • Use Case: When individuals have varying capture probabilities due to behavioral or environmental factors.
  • Advantages: Accounts for individual heterogeneity without inflating the population size estimate.
  • Limitations: Requires detailed covariate data on factors influencing capture probabilities.

Multi-State Models for Mark Loss

In situations where marks can be lost or not detected, multi-state models offer a way to handle transitions between states (marked and unmarked). These models account for the probability that an individual’s mark may degrade or go unnoticed during recapture, which could otherwise bias the population estimate.

  • Use Case: When marks degrade over time or when detection of marks is imperfect.
  • Advantages: Provides robust estimates even when some marks are lost or misidentified.
  • Limitations: More complex to analyze and requires careful tracking of marking conditions.

Spatially Explicit Capture-Recapture (SECR) Models for Non-Random Mixing

SECR models account for non-random mixing of marked individuals by incorporating spatial data into the capture-recapture analysis. These models consider the locations where individuals are captured, enabling more accurate population density estimates and addressing issues where animals do not mix randomly after release.

  • Use Case: For populations where individuals have strong territorial behaviors or limited movement.
  • Advantages: Provides spatially explicit density estimates and addresses spatial heterogeneity in capture probability.
  • Limitations: Requires spatial data on capture locations, adding complexity to fieldwork and analysis.

Robust Design Models for Longer Time Intervals

The robust design model is used when there is a need for longer intervals between capture events, but the assumption of a closed population must still be approximated. This model combines both open and closed population components within a single study design, allowing for multiple sampling occasions while accounting for population changes over time.

  • Use Case: When multiple sampling events occur over a long period, but a closed population model is desired during shorter intervals.
  • Advantages: Combines the strengths of closed and open population models for more flexible study designs.
  • Limitations: Requires more frequent sampling and careful study design to differentiate between short-term and long-term population dynamics.

Cormack-Jolly-Seber Models for Behavioral Changes

The Cormack-Jolly-Seber model is an extension of the Jolly-Seber model that separates recapture probability from survival probability. This model is particularly useful when marked individuals exhibit behavioral changes after being captured and marked, such as becoming trap-shy or trap-happy.

  • Use Case: When individuals are suspected to alter their behavior after being marked.
  • Advantages: Allows researchers to estimate survival and recapture probabilities independently, correcting for behavioral biases.
  • Limitations: Requires detailed data on individual capture histories and behavioral trends.

Summary of Model Applications

Model Purpose Key Assumption Addressed
Jolly-Seber Open populations Closed population
Huggins Unequal catchability Equal catchability
Multi-State Mark loss No mark loss
SECR Non-random mixing Random mixing
Robust Design Long time intervals between events Instantaneous sampling and closed population
Cormack-Jolly-Seber Behavioral changes after marking No behavioral change

These models provide flexible solutions to real-world complications in CMR studies, allowing for more accurate population estimates when key assumptions are violated. By incorporating these models into their analyses, researchers can correct for common biases and improve the reliability of their estimates in dynamic wildlife populations.

You may also enjoy