The Gold Standard of Zoology | Separating Cause-of-Death Proportions from Cause-Specific Mortality Risk

Begin with the population at risk, follow-up interval and outcome definition. A cause proportion among carcasses, a cumulative incidence in a cohort and a cause-specific hazard are not interchangeable labels for one percentage.

Begin with a measurement ledger. Name the quantity, its unit, the sample or observation that produced it, the time window and the transformation rule. Mark each input as observed, stipulated, estimated or assumed. This prevents a familiar label from importing a new denominator or a stronger claim. Read the ledger as a sentence before calculating; if the sentence is incomplete, the arithmetic is premature.

A fictional study follows 100 marked animals for one year. Twenty die from predation, ten from disease and seventy survive to the endpoint. Predation is 20/30=66.7% of recorded deaths and 20/100=20% of the starting cohort.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to state both results in full sentences and identify why neither denominator can be omitted State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Next chapter

Death from one cause prevents later death from another cause in the same follow-up. Competing outcomes therefore change the number still capable of experiencing the focal cause.

Keep the mechanism beside the number. Ask what physical, biological, statistical or strategic relationship connects the observation to the conclusion, and which conditions could break that link. A correct calculation can answer the wrong question when a changed setting alters the process. Draw one arrow for every inferential step, then challenge each arrow with a counterexample.

After ten early disease deaths, only ninety animals remain alive for later predation. A calculation that keeps all 100 in every instantaneous risk set ignores the event history even though 20/100 remains a valid crude endpoint fraction in this complete fixed-time example.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to move the disease deaths after the predation deaths and explain which time-based quantities can change State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Cause proportion describes composition among classified deaths. Cumulative incidence describes the probability of experiencing a cause by time t in the cohort under the study’s event and censoring structure.

Use an independent check that follows a different route. Reconstruct totals from components, compare a bound with a point estimate, inspect a distribution behind an average, or solve a changed case with the original definitions. Agreement does not prove every assumption, but disagreement locates a definition, unit or boundary that needs repair.

Site A has 10 predation and 0 other deaths among 100 animals, so predation is 100% of deaths but 10% of the cohort. Site B has 20 predation and 20 other deaths, so predation is 50% of deaths but 20% of the cohort.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to rank the sites under each measure and explain why the order can reverse State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

The 20/100 shortcut assumes the fictional cohort’s endpoint fates are completely observed and the target is a fixed-time crude risk. Censoring, staggered entry and unequal follow-up require time-to-event methods.

Match the claim to the evidence. A tidy decimal cannot repair selective observation, and one convenient run cannot establish every product, participant, signal or opponent. Report the result, the conditions that make it interpretable, and the next observation needed for the actual decision. Separate the numerical conclusion from the practical conclusion.

If 25 animals leave observation at six months with unknown later fates, 20/100 is an observed minimum fraction, not automatically the one-year cumulative incidence. The missing time cannot be repaired by dividing only by the 75 retained animals.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to bound the number of possible predation deaths if every censored animal is unresolved State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Age, release date, tagging date and calendar season can produce different risk clocks. The analysis must match the biological question and preserve delayed entry when animals join after the nominal origin.

Begin with a measurement ledger. Name the quantity, its unit, the sample or observation that produced it, the time window and the transformation rule. Mark each input as observed, stipulated, estimated or assumed. This prevents a familiar label from importing a new denominator or a stronger claim. Read the ledger as a sentence before calculating; if the sentence is incomplete, the arithmetic is premature.

One animal enters at birth and another at capture as an adult. Calling both ‘one year of survival’ can hide different age and exposure histories even if each contributes twelve observed months.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to redefine time as age and identify which records need left-truncation information State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

A cause label depends on carcass recovery, timing, condition, pathology and field evidence. Unknown and uncertain causes should remain visible instead of being forced into familiar categories.

Keep the mechanism beside the number. Ask what physical, biological, statistical or strategic relationship connects the observation to the conclusion, and which conditions could break that link. A correct calculation can answer the wrong question when a changed setting alters the process. Draw one arrow for every inferential step, then challenge each arrow with a counterexample.

Of 30 deaths, 18 are verified predation, two are probable predation, seven are verified disease and three are unknown. A transparent table can report verified, expanded and bounded interpretations rather than silently calling all ambiguous cases predation.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to compute the verified minimum and the maximum if every unknown case were predation while keeping probable cases separate State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Removing unknown deaths changes the denominator and can bias cause composition if diagnosis success differs by cause. Report the classification process and sensitivity of conclusions to unresolved cases.

Use an independent check that follows a different route. Reconstruct totals from components, compare a bound with a point estimate, inspect a distribution behind an average, or solve a changed case with the original definitions. Agreement does not prove every assumption, but disagreement locates a definition, unit or boundary that needs repair.

If three unknown deaths are excluded, verified predation is 18/25=72% of classified deaths. Among all 28 known deaths plus unknowns, verified predation accounts for 18/28=64.3% of all recorded deaths. Each sentence needs its denominator.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to move two unknowns into disease and recalculate without changing the cohort total State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Causes that leave recoverable remains can be overrepresented. Telemetry failure, scavenging and access affect which deaths become diagnosable, so the recorded cause mixture can differ from the true mortality process.

Match the claim to the evidence. A tidy decimal cannot repair selective observation, and one convenient run cannot establish every product, participant, signal or opponent. Report the result, the conditions that make it interpretable, and the next observation needed for the actual decision. Separate the numerical conclusion from the practical conclusion.

Suppose 90% of predation deaths but 40% of disease deaths yield recoverable evidence. Observed carcass proportions can exaggerate predation even when underlying counts are equal. Without supported recovery probabilities, do not apply an invented correction.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to supply hypothetical recovery probabilities and show how a sensitivity calculation differs from a validated estimate State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

A cause-specific hazard is an instantaneous event rate among animals currently alive and at risk. It is not the same as the fraction of the original cohort dead from that cause by a later time.

Begin with a measurement ledger. Name the quantity, its unit, the sample or observation that produced it, the time window and the transformation rule. Mark each input as observed, stipulated, estimated or assumed. This prevents a familiar label from importing a new denominator or a stronger claim. Read the ledger as a sentence before calculating; if the sentence is incomplete, the arithmetic is premature.

Two periods can have equal ten-percent predation incidence but different timing. Ten deaths in the first week and ten spread across a year produce different hazard patterns despite equal endpoint counts.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to split the year into intervals and update the risk set after each event State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Cumulative incidence for a cause depends on both its event process and survival from all causes. Treating competing deaths as ordinary independent censoring can overstate absolute risk.

Keep the mechanism beside the number. Ask what physical, biological, statistical or strategic relationship connects the observation to the conclusion, and which conditions could break that link. A correct calculation can answer the wrong question when a changed setting alters the process. Draw one arrow for every inferential step, then challenge each arrow with a counterexample.

A severe disease wave removes many animals before the predation season. Predation’s cause-specific hazard among remaining survivors could stay constant while its one-year cumulative incidence falls because fewer animals remain available.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to hold the predation hazard conceptually fixed while increasing early disease and predict the direction of cumulative incidence State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Events per animal-time summarize how quickly events occur over observed risk time. A rate can aid comparison, but it is not automatically a probability and cannot be converted to risk without a model.

Use an independent check that follows a different route. Reconstruct totals from components, compare a bound with a point estimate, inspect a distribution behind an average, or solve a changed case with the original definitions. Agreement does not prove every assumption, but disagreement locates a definition, unit or boundary that needs repair.

Twenty predation deaths over 80 animal-years give 0.25 deaths per animal-year in a stipulated ledger. Writing 25% would incorrectly turn a rate into a one-year probability without assumptions about hazard and competing events.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to halve follow-up time while keeping events fixed and explain why the rate changes State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

The existing lesson on bounding survival from verified fates and unresolved follow-up records keeps unknown fates visible. Here, the added question is how known and uncertain deaths are divided by cause.

Match the claim to the evidence. A tidy decimal cannot repair selective observation, and one convenient run cannot establish every product, participant, signal or opponent. Report the result, the conditions that make it interpretable, and the next observation needed for the actual decision. Separate the numerical conclusion from the practical conclusion.

A cohort can have a well-bounded all-cause survival fraction yet a poorly resolved cause distribution because several carcasses lack diagnoses. Survival evidence and cause evidence need separate completeness checks.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to hold all-cause deaths fixed while reallocating unknown causes and identify which conclusions remain stable State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

An absent animal is not a diagnosed death. Tag failure, permanent emigration and missed detection can resemble mortality in encounter data, as the tag-retention lesson explains.

Begin with a measurement ledger. Name the quantity, its unit, the sample or observation that produced it, the time window and the transformation rule. Mark each input as observed, stipulated, estimated or assumed. This prevents a familiar label from importing a new denominator or a stronger claim. Read the ledger as a sentence before calculating; if the sentence is incomplete, the arithmetic is premature.

If eight animals disappear and only two carcasses are found, assigning all eight to a cause creates events the study did not observe. The mortality ledger should retain alive, dead by cause and unresolved states.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to add a later live resighting and show which earlier inference must be withdrawn State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Raw death counts cannot compare mortality risk when cohort sizes differ. Preserve the same follow-up definition before dividing by starting or dynamically at-risk populations.

Keep the mechanism beside the number. Ask what physical, biological, statistical or strategic relationship connects the observation to the conclusion, and which conditions could break that link. A correct calculation can answer the wrong question when a changed setting alters the process. Draw one arrow for every inferential step, then challenge each arrow with a counterexample.

Site C records 15 predation deaths among 300 animals; Site D records 8 among 80. Their crude risks are 5% and 10%, so the smaller raw count belongs to the higher cohort fraction.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to double Site D’s cohort with no new deaths and recalculate without changing cause proportion among its deaths State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

A six-month cohort and a two-year cohort do not become comparable by sharing a percentage label. Use aligned endpoints, survival methods or clearly named rates supported by the design.

Use an independent check that follows a different route. Reconstruct totals from components, compare a bound with a point estimate, inspect a distribution behind an average, or solve a changed case with the original definitions. Agreement does not prove every assumption, but disagreement locates a definition, unit or boundary that needs repair.

Ten predation deaths among 100 animals over six months and ten among 100 over two years both yield a ten-percent crude observed fraction, but exposure opportunities differ.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to truncate the two-year record at six months and identify which data are needed State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

A treatment can change other-cause mortality and thereby alter the focal cumulative incidence even if the focal biological mechanism is unchanged. Compare all outcomes and estimands.

Match the claim to the evidence. A tidy decimal cannot repair selective observation, and one convenient run cannot establish every product, participant, signal or opponent. Report the result, the conditions that make it interpretable, and the next observation needed for the actual decision. Separate the numerical conclusion from the practical conclusion.

An intervention prevents disease deaths, leaving more animals alive during predator exposure. More later predation deaths need not mean the intervention caused predation; it can reflect altered competing survival.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to draw the event timeline and state which causal claim needs randomization and fuller modelling State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Imperfect cause assignment can move events between categories. Blind review, standard necropsy criteria and sensitivity analyses help without turning uncertain diagnoses into certainty.

Begin with a measurement ledger. Name the quantity, its unit, the sample or observation that produced it, the time window and the transformation rule. Mark each input as observed, stipulated, estimated or assumed. This prevents a familiar label from importing a new denominator or a stronger claim. Read the ledger as a sentence before calculating; if the sentence is incomplete, the arithmetic is premature.

If two of twenty predation labels could plausibly be disease, the verified predation count might be eighteen. Report 18 verified plus 2 uncertain rather than choosing the result that best supports a narrative.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to swap the direction of misclassification and show how the cause fraction responds State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Include original cohort, entry dates, follow-up, all outcome categories, unknown fates, diagnosed deaths, cause definitions and the exact denominator for every percentage. Tables should let readers reconstruct totals.

Keep the mechanism beside the number. Ask what physical, biological, statistical or strategic relationship connects the observation to the conclusion, and which conditions could break that link. A correct calculation can answer the wrong question when a changed setting alters the process. Draw one arrow for every inferential step, then challenge each arrow with a counterexample.

A row reading ‘predation 66.7%’ is incomplete. ’20 of 30 diagnosed deaths; 20 of 100 enrolled animals by one year under complete fictional follow-up’ makes both summaries auditable.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to remove the follow-up interval and explain which comparison becomes unsafe State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Most recorded deaths does not mean most animals died from that cause; a rate is not a probability; censoring is not survival; and a competing death is not missing data. Restore event, time and risk-set definitions.

Use an independent check that follows a different route. Reconstruct totals from components, compare a bound with a point estimate, inspect a distribution behind an average, or solve a changed case with the original definitions. Agreement does not prove every assumption, but disagreement locates a definition, unit or boundary that needs repair.

When a headline says two-thirds of animals died from predation, check whether it actually means two-thirds of the subset of diagnosed deaths. Rewrite it before interpreting population impact.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to create a headline for the 20/30 result that cannot be mistaken for 66.7% cohort mortality State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

Check cohort construction, time origin, entry, tracking, carcass recovery, diagnosis, unknowns, censoring, competing events, units and analysis method. Reproduce simple totals before fitting a model.

Match the claim to the evidence. A tidy decimal cannot repair selective observation, and one convenient run cannot establish every product, participant, signal or opponent. Report the result, the conditions that make it interpretable, and the next observation needed for the actual decision. Separate the numerical conclusion from the practical conclusion.

The ledger must satisfy 20 predation + 10 disease + 70 alive = 100 in the complete toy example. A mismatch exposes missing categories before a polished percentage conceals them.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to add five censored animals and redesign the mutually exclusive endpoint table State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

A descriptive cohort can estimate defined outcomes under its sampling and follow-up assumptions. It does not by itself identify why causes differ between sites or predict a new population.

Begin with a measurement ledger. Name the quantity, its unit, the sample or observation that produced it, the time window and the transformation rule. Mark each input as observed, stipulated, estimated or assumed. This prevents a familiar label from importing a new denominator or a stronger claim. Read the ledger as a sentence before calculating; if the sentence is incomplete, the arithmetic is premature.

Habitat A’s higher predation cumulative incidence could reflect predator density, animal behaviour, monitoring, age structure or competing deaths. A mechanism claim needs evidence that separates those explanations.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to name one additional measurement for each of two plausible mechanisms State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Next chapter

The USGS review of cause-specific mortality with competing risks explains why wildlife mortality requires estimand-aware methods. Return to the Zoology root and its survival, tag and encounter branches for the wider observation system.

Keep the mechanism beside the number. Ask what physical, biological, statistical or strategic relationship connects the observation to the conclusion, and which conditions could break that link. A correct calculation can answer the wrong question when a changed setting alters the process. Draw one arrow for every inferential step, then challenge each arrow with a counterexample.

The transferable habit is to make every percentage carry its population, event set and time window. The same repair works for machine failures, medical outcomes and customer attrition with competing reasons.

Before moving on, record the observation that would distinguish the two most plausible explanations. If no feasible observation exists, narrow the claim instead of pretending the ambiguity has disappeared.

A changed-context transfer check is to apply the denominator audit to equipment failures split by cause and identify the competing events State what would change the result and what the example cannot establish. That limit belongs inside the answer.

Contents · Previous chapter · Return to Zoology