Why Rescuers May Search the Less Likely Place
U.S. Air National Guard photo by Staff Sgt. Edward Eagerton. Public domain. The appearance of U.S. Department of War (DoW) visual information does not imply or constitute DoW endorsement.
Curated and edited by John Pile Jr.
A missing person is more likely to be in the woods than in the open field. Before deciding where to search first, the team needs another estimate: how likely are they to spot the person in each place?
Search-and-rescue planning separates two questions: where the person probably is, and how likely a particular search is to find them if they are there. The Coast Guard calls their combination the probability of success.1
These examples use hypothetical numbers and simplified conditions, with comparable time or effort for each search.
A stationary person is somewhere in two areas, and either search takes one hour:
| Area | Chance person is there | Chance of detection if there | Chance this search finds them |
|---|---|---|---|
| A: dense woods | 60% | 20% | 12% |
| B: open field | 40% | 80% | 32% |
The arithmetic is multiplication: 60% × 20% = 12%. The woods are the likeliest location, but searching the field offers the greater chance of finding the person in the next hour.
Both things must happen for the search to succeed: the person must be there, and the search must detect them. A promising location can be a poor next search when the second step is difficult.
The same calculation appears in wildlife surveys, astronomical observations, and inspections. It also offers a practical way to think about searching historical records or investigating a software failure.
A marriage record in the less likely state
Suppose a couple lived in a town on the Colorado–Wyoming border. Other evidence suggests a 75% chance their marriage was registered in Colorado and a 25% chance it was registered in Wyoming.
The records service has digitized just 20% of the relevant Colorado marriage records, compared with 90% of Wyoming’s.
| Online collection | Chance marriage was registered there | Chance relevant record is included | Chance this search finds the record |
|---|---|---|---|
| Colorado | 75% | 20% | 15% |
| Wyoming | 25% | 90% | 22.5% |
Wyoming is the better first online search. Colorado remains the more likely place of registration.
That distinction also explains why an unsuccessful Colorado search settles little. Most of the relevant records were unavailable through that service. Searching another website with the same underlying collection may leave the same gap.
Finding any frogs
An ecologist looking for evidence that a frog species is present faces a similar problem. Suitable habitat raises the chance that frogs live there. Survey conditions affect whether they will be detected.
Researchers explicitly estimate these separately as occupancy and detection probability. A foundational study applied this approach to frog surveys at 32 wetlands in Maryland.2
Imagine two wetlands and the same amount of survey time at each:
| Wetland | Chance the species is present | Chance survey detects it if present | Chance this survey finds it |
|---|---|---|---|
| A: more promising habitat, poor survey conditions | 70% | 20% | 14% |
| B: less promising habitat, good survey conditions | 40% | 80% | 32% |
If the goal is to establish whether the species lives in the area, B is the better next visit. The ecologist needs evidence of any frogs of this species. Finding them at B could answer the question, making a search of A unnecessary.
Choosing a patch of sky
After a gravitational-wave alert from a neutron-star merger, astronomers can receive a probability map showing where the source might be. Telescopes then search for light associated with the event.
Location probability is only part of the problem. The source might be too faint for a particular exposure. Published observation-planning methods combine location information with models of detectability and allocate limited telescope time accordingly.3
Suppose one telescope field has a 25% chance of containing the source, but an exposure would detect it only 40% of the time if present. That observation has a 10% chance of success.
A second field has only a 15% chance of containing it, but a 90% chance of detection if present. Its chance of success is 13.5%.
The second field is the better next search. Spending longer on the first could improve detection, but would leave less time to search elsewhere.
Inspecting for a crack that is already there
Engineering inspections have their own detection probabilities. NASA uses probability-of-detection studies to evaluate how reliably inspection methods reveal flaws of specified sizes.4
Suppose a component has a 30% chance of containing a crack. An inspection has a 60% chance of revealing it if present. The chance of finding it is 18%: 30% × 60%.
A second component has a 15% chance of containing the same kind of crack, with a 90% chance of detection. Its chance of a discovery is 13.5%.
This time, the more likely location also offers the better chance of discovery. The first component is the better place to look. Accounting for detection difficulty can change the order, but it does not always reverse it.
A passed inspection may still have missed a crack that was already there. “We inspected it” tells us more when we know how likely that inspection was to reveal the crack.
Similar mechanisms exist in healthcare. The probability that a condition is present combines with a test’s sensitivity, its chance of detecting the condition when present. This is one reason a negative result must be interpreted in light of both the suspected condition and the test used.5
What did “we checked that” accomplish?
Return to the woods. Suppose the team searches there and finds nothing. Does the open field now become the likelier location?
Even after the unsuccessful search, the woods remain more likely.
To see why without a formula, imagine 100 otherwise identical situations. In 60, the person is in the woods. The search finds 20% of those people: 12. It misses the other 48. In the remaining 40 situations, the person is in the field, so the woods search also finds nothing.
That leaves 88 unsuccessful searches. In 48 of them, the person is still in the woods. After the negative result, the chance they are there is about 55%: 48 out of 88.
The search reduced the estimate from 60% to roughly 55%. It supplied some evidence, but its low detection probability limited how much could be learned from finding nothing.
The same question belongs in an investigation. A software team may have checked a suspected subsystem using a test that rarely creates the conditions needed to expose the fault. A family historian may have checked a collection missing most of the relevant records. Both searches happened. Neither necessarily supplied much evidence against the suspected location.
“We already checked that” becomes more useful when followed by a question: If it had been there, how likely were we to find it?
Join the discussion