Ninety-Eight in the Hotel Room, Forty-Four in the Apartment: Scent Detection Dogs and the Gap Between Validation and Work

1. Introduction: the best-marketed technology in the trade

Scent detection dogs are sold on a number. That number comes from a laboratory, and a field evaluation exists.

1.1 Why this article exists

Because the first number is quoted constantly and the second almost never, and because the methodological reasons for the gap generalise well beyond dogs.

2. What the laboratory found

The validation work, which was good work and produced good results.

A 2008 university study reported that the dogs were 97.5 per cent accurate in distinguishing bed bugs from other household pests, 95 per cent accurate in distinguishing live bed bugs from bed bug debris, with a 3 per cent false-positive rate, and 98 per cent accurate in detecting six bed bug vials in a hotel room with no false-positive alerts.⁵

2.1 What each of those tested

The first tested discrimination between target and non-target insects. The second tested discrimination between a live infestation and the residue of a dead one, which is a harder and more useful distinction. The third tested location of hidden targets in a realistic room.

All three are well-designed questions, and the answers were strong. Nothing in this article disputes them.

2.2 The second result is the impressive one

Distinguishing live bugs from debris at 95 per cent is a harder task than it sounds. Cast skins and faeces carry the same chemistry as the insect that produced them, and a dog trained on the odour of an infestation has every reason to alert on its residue.

That the laboratory dogs made that distinction is the strongest evidence in the whole literature that the underlying discrimination is available. It also makes the §14 finding more striking, because in the field the residue is precisely what preceded the false alerts.

3. The researcher's own caveat

Which was published alongside the result and is rarely quoted with it.

One of the researchers stated: our study has shown that dogs can be trained to accurately locate live bed bugs and viable bed bug eggs; however, bed bug detecting canines and handlers are not all equal. The quality of canines depends on the efficiency of their training, handlers, and what the dogs are trained to do.⁴

3.1 The distinction being drawn

Dogs can be trained to do this. It does not follow that any particular dog has been.

A demonstration of what is achievable under good conditions is a capability claim. Marketing treats it as a performance claim about a specific team, and those are different statements about different things.

4. The gap nobody had measured

The stated motivation for the field study.

There are no data demonstrating the reliability of trained canines under natural field conditions.²

4.1 How long that was true

The technology entered commercial use in the early 2000s.⁵ The field evaluation was published in 2014.

So a detection service was sold for roughly a decade on laboratory validation alone. This journal made a similar observation about remote rodent monitoring, where deployment preceded independent evaluation, and the pattern is worth naming as a general feature of this trade.

4.2 Why nobody measured it sooner

Because the measurement is expensive and nobody in the market wanted the answer. It requires naturally infested homes with independently established infestation status, repeated visits, and the cooperation of commercial teams willing to be assessed.

The study records thanks to teams who donated their time and to residents who tolerated repeated visits by researchers.³ That is a considerable amount of goodwill to assemble for a result that was always likely to be unwelcome, and the teams who volunteered deserve some credit for it.

5. The field result

What was found when the question was asked.

Eleven canine detection teams were evaluated in naturally infested apartments. In three separate experiments, the mean detection rate was 44 per cent, with a range of 10 to 100, and the mean false-positive rate was 15 per cent, with a range of 0 to 57.¹

5.1 Reading the detection figure

Fewer than half of genuine infestations were found. In a setting where the purpose of the inspection is to establish whether an infestation exists, a negative result from such a team carries little information.

5.2 What a 44 per cent detection rate does to a clearance decision

Consider the use this service is most often sold for: checking a unit before a tenant moves in, or confirming that a treated apartment is clear.

If fewer than half of real infestations are detected, then an all-clear is consistent with an infestation being present more often than not. The inspection has not answered the question it was commissioned to answer, and the confident report it produces is worse than no inspection because it stops the search.

5.3 Why the false-positive figure is the cheaper error

An unnecessary treatment costs money and pesticide. A missed infestation costs a continuing infestation, further dispersal and, in multi-unit housing, the spread this journal described in the bed bug economics article.

We would rank the 44 per cent as the more serious of the two numbers for that reason, and flag the ranking as ours.

6. What the handlers believed

The detail that makes this study unusually instructive.

All handlers believed their dogs could detect infestations at a very high rate, of 95 per cent or greater.¹

6.1 The size of the discrepancy

Believed 95 per cent or better. Measured 44 per cent on average.

These are not dishonest operators. They are people working without feedback, which is the subject of §13, and their belief is what the absence of feedback produces.

It is worth sitting with how confident that belief was. Not most of the time, not usually, but 95 per cent or better, from every handler in the study. A uniform belief across independent practitioners is a sign that something structural is producing it rather than individual overconfidence.

6.2 Why this generalises

Any practitioner whose work is not independently checked will tend to believe it is going well, because the cases where it did not go well are precisely the cases that generate no signal.

This journal has made that argument about treatment outcomes, about inspection quality and about monitoring interpretation. Here it is measured, with a number attached to the gap.

7. The range is the story

The averages conceal something important.

Detection ranged from 10 to 100 per cent and false positives from 0 to 57 per cent.¹

7.1 What that spread means for a buyer

Some teams in that sample were excellent. At least one detected everything.

The problem is that a purchaser cannot tell which team they are hiring. The mean is not a description of the service anyone receives; it is a description of a lottery, and the variance is wider than the mean is low.

That also explains why operator testimony and published results can both be honest and still disagree. A handler who happens to be one of the good teams has accurate personal experience, and it does not generalise to the team a customer hires next.

7.2 The team experience factor

The teams evaluated had not worked together long, with some pairs working together for less than a year.⁵

That is a fair mitigating observation and it belongs here. A dog and handler who have worked together for five years may perform differently, and the study does not establish that they would not.

8. Against the proposed standard

A benchmark existed before the study, proposed for termite detection.

It is not unreasonable to expect a properly trained dog to meet a minimum standard with a positive indication rate of 90 per cent or greater and a false-positive rate of 10 per cent or less.³

Judged against it, only one out of 16 inspections met the 90 per cent detection standard, and only 5 out of 16 met the 10 per cent false-positive standard.³

8.1 The value of having a standard first

The benchmark was proposed in 2003, before this evaluation, by people working on a different target pest. It was not constructed afterwards to make a result look bad.

That sequence matters. A criterion set in advance by independent parties is the strongest form a performance standard can take, and this journal has complained elsewhere about its absence.

8.2 Whether the standard is reasonable

Ninety per cent detection with ten per cent false positives is demanding, and it is worth asking whether any working team could meet it.

One of the sixteen inspections did meet the detection bar, so it is achievable at least occasionally.³ And the authors who proposed it described it as not unreasonable for a properly trained dog, which is a claim about training rather than about biology.³

8.3 The comparison that should be made instead

The honest benchmark is not a standard at all but the alternative. A canine inspection is worth buying if it outperforms a careful visual inspection by a competent technician at a comparable cost.

We located no head-to-head field comparison of the two under equivalent conditions, which is a striking gap given that the entire commercial case is a comparison between them. Marketing supplies a visual-inspection figure of 30 to 50 per cent for early-stage infestations,⁶ but it appears beside the laboratory canine figure rather than beside the field one.

9. The correlation that changes the interpretation

The most methodologically interesting sentence in the study.

The false-positive rate was positively correlated with the detection rate.¹

9.1 What that means

Teams that found more infestations also alerted more often where there were none. The two error types moved together rather than in opposition.

That is the signature of a threshold rather than a skill. A team willing to alert on weaker evidence catches more real cases and more false ones; a conservative team misses more and errs less. Moving along that line is not improvement.

9.2 Why this matters for the marketing

A company quoting a high detection rate without a false-positive rate has quoted one end of a trade-off. The study's authors say so directly: it is necessary to consider both detection rate and false-positive rate to evaluate a team's effectiveness.²

9.3 The one thing it does not mean

It does not mean the better teams were worse. A team at 100 per cent detection with a high false-positive rate may be exactly what a client wants if the cost of a miss exceeds the cost of an unnecessary check.

What it means is that the two numbers have to be quoted together, and that comparing providers on detection alone will reliably select the team that alerts most readily rather than the team that discriminates best.

9.4 Where else this appears

This journal made the same argument about monitoring devices, where a trap catch is a joint product of population and trap attractiveness, and about diagnostic thresholds, where sensitivity and specificity trade against each other. A single accuracy number almost always hides one of a pair.

10. The finding that should worry everyone

Stated plainly by the authors and easy to read past.

The probability of a bed bug infestation being detected by trained canines was not associated with the level of bed bug infestations.¹

10.1 Why this is the most damaging result in the paper

A scent detector should find heavy infestations more reliably than light ones, because more insects produce more odour. That relationship is the mechanism the whole method rests on.

It was not observed. Detection did not improve with infestation level, which is difficult to reconcile with alerts being driven principally by target odour concentration.

There is a second implication that cuts against the usual sales case. The stated advantage of canine inspection is finding light, early infestations that visual inspection misses. If detection does not scale with infestation level, then the method is not specifically good at the light end; it is roughly as unreliable across the range.

10.2 What we are not saying

That the dogs cannot smell bed bugs. The laboratory work in §2 establishes that they can.

What this finding suggests is that under field conditions something other than target odour concentration was driving a substantial share of the outcomes. Sections 13 and 14 offer the candidates.

11. It was not an observer effect

The obvious objection, which the study anticipated.

Mean detection and false-positive rates in our study were similar regardless of whether or not detection firms were aware they were being observed.³

11.1 What that rules out

That the teams performed badly because they were nervous, or differently because they were being watched. Performance under observation matched performance when unaware.

It also rules out the reverse concern, that firms aware of observation performed unusually well and the true field figure is lower still.

11.2 Why that design choice matters

Evaluating a commercial service is hard precisely because the operators know they are being tested. Running both informed and uninformed conditions and finding no difference removes the most available objection to the result.

This journal has praised designs that rule out alternative explanations rather than merely supporting a preferred one, and this is another instance.

11b. The same team on two days

A finding that undermines the idea of a team having an accuracy at all.

Four canine detection teams evaluated on multiple days were inconsistent in their ability to detect bed bugs and exhibited significant variance in accuracy between inspections on different days.¹

11b.1 Why this matters more than the mean

Section 7 established that teams differ from one another. This establishes that a team differs from itself.

A quoted accuracy figure presumes a stable property of the team. If the same dog and handler produce significantly different results on two days in the same kind of setting, there is no stable figure to quote, and certification at a moment cannot represent performance on a Tuesday.

11b.2 A physiological reason

Part of it is mechanical. To cool the body dogs pant, and while panting they are unable to sniff, with an inverse relationship demonstrated between increased panting and detection.¹

That is a hard physical constraint rather than a training issue. A hot apartment, a long working day or a warm Winnipeg July is a condition under which detection capability is physically reduced, and no amount of handler skill changes it.

11b.3 The learning observation

In a controlled hide study, a dog exhibited learning behaviour over the course of four inspections in apartments.¹

A dog learning the pattern of a test rather than the odour is a known hazard in scent detection research, and it is part of why naturally infested settings with unknown status are a more demanding and more honest test than repeated planted hides. It is also a caution about the laboratory results in §2.

11c. What did not predict accuracy

The study tested the proxies a purchaser would actually use, and none of them worked.

There was no significant relationship between a team's detection rate and the length of time the handler and dog had been working together, nor between detection rate and whether the team was certified.¹

11c.1 The price signal ran backwards

A separate evaluation of seven canine detection firms, with each apartment inspected by three or four of them, found detection rates ranging from 11 to 83 per cent. The two companies that charged the most, at a thousand dollars, and had been in business more than three years had very low detection rates of 25 and 30 per cent, and false alerts were also high for those same two companies.⁹

11c.2 Why this is the most uncomfortable result in the paper

Tenure, certification and price are the three signals available to someone choosing a provider. In these evaluations none of them predicted performance and one of them ran in the wrong direction.

That is the credence goods problem this journal set out separately, appearing here with measurements attached. The buyer has no usable quality signal, and the signals they are offered are uncorrelated with the thing they are trying to buy.

11c.3 What the trade literature concluded

That variability in accuracy among canine scent firms is great, that additional studies are necessary to understand the factors influencing accuracy, and that guidance is needed for maintaining a quality canine scent programme.⁹

That was published in 2011. We have not found evidence the gap has been closed.

12. Why the field differs

The authors' explanation, which is about information rather than about ability.

An obvious difference between performance under controlled and field conditions is that errors by dogs and handlers are identifiable and correctable in a controlled setting, while the natural field setting is very complex with odors from many different sources where errors can easily occur, go unidentified and thus remain uncorrected, reinforcing the incorrect behavior. This creates challenges in the ongoing training and evaluation of a team's performance.²

13. The feedback problem

The core of it, and the part that generalises furthest.

In training, the trainer knows where the bed bugs are. In an apartment, nobody does. That is the reason the inspection is happening.

13.1 The consequence for learning

A dog that alerts incorrectly in training is corrected. A dog that alerts incorrectly in an apartment is rewarded, because the handler has no way to know it was wrong and the alert is treated as a find.

So the working environment does not merely fail to improve the team. It actively trains error, because an unverifiable alert is indistinguishable from a correct one and gets the same reinforcement.

13.2 Why performance would be expected to drift

Under that regime, the longer a team works without independent verification, the further from its certified performance it can drift, and in no particular direction that anyone can observe.

That is our inference from the authors' account rather than a measured finding, and it is testable: it predicts that verification programmes should preserve accuracy and their absence should degrade it.

14. The cueing mechanism

The specific pathway the authors identified for some of the false alerts.

False-positive alerts occurred in some of the apartments where old evidence of fecal spotting, carcasses, and exuvia were readily visible and recognized by handlers during the inspection. This may have led to unintentional cues to the dog by the handler.²

14.1 What is being described

The handler sees evidence of a past infestation, forms a belief that bugs are present, and transmits that belief through posture, pace or attention without intending to. The dog responds to the handler rather than to the odour.

The word unintentional is doing important work. Nobody in this account is cheating.

That is also what makes it hard to fix. A handler who knew they were cueing could stop; a handler who does not cannot, and the standard remedy in other fields is to blind the handler to the information, which is impossible when the information is a visible stain on a mattress.

14.2 Why old evidence is the worst possible trigger

Fecal spotting and cast skins persist after an infestation has been eliminated. An apartment treated successfully last year still carries them.

So the cue that most reliably produces an alert is the cue least diagnostic of a current infestation, which is close to a worst case for a technology whose selling point is finding what visual inspection cannot.

15. The wider canine literature

This is not confined to bed bugs.

A trade account reports a study in which researchers found that detection dogs falsely alerted their handlers to the presence of drugs or explosives over 200 times, in a building where the researchers never brought any contraband items or explosives, with false positive alerts by canine teams in every room they inspected.⁸

15.1 The status of that report

We have it through a commercial pest control company's blog rather than from the primary paper.⁸ We have not verified the design, the number of teams or the conditions, and a reader should treat the specific figures accordingly.

We include it because it is consistent with the mechanism in §14, and because the bed bug field study itself references work on detection of wild brown tree snakes as a comparable case.²

16. The termite comparison

The only other pest detection evaluation the study cites.

The only other study investigating the accuracy of trained dogs for termites reported a mean detection rate of 81 per cent and false-positive rate of 28 per cent in a laboratory setting.³

16.1 What that comparison shows

The bed bug field result of 44 per cent detection and 15 per cent false positives is described by the authors as more in line with that termite laboratory figure.³

Which means the termite laboratory result was itself well short of the standard proposed for termite work five years later. The 28 per cent false-positive rate in a controlled setting is nearly three times the proposed ceiling.

It is also worth noting how thin this literature is. Two laboratory evaluations across two pest species and one field evaluation is the whole evidence base for a service that has been sold commercially for more than two decades.

16.2 The honest reading

Laboratory results in this field are not uniformly high. One laboratory study produced 98 per cent; another produced 81 per cent detection with 28 per cent false positives.

Quoting the better laboratory result as though it were the laboratory result is a selection that happens before the field question even arises.

17. How the result is marketed around

What is currently being said, quoted so the contrast is visible.

One commercial page states that published peer-reviewed studies consistently demonstrate that properly trained and certified bed bug detection dogs achieve 95 to 98 per cent accuracy rates, that this far exceeds traditional visual inspection methods, which typically catch only 30 to 50 per cent of early-stage infestations, and that false negatives are rare with properly certified and trained teams.⁶

17.1 The word doing the work

Properly trained and certified. Every claim is conditioned on it, which makes the claim unfalsifiable in the same way §18 describes: any team that performs badly was, by definition, not properly trained.

17.2 The consistently claim

Published studies do not consistently demonstrate 95 to 98 per cent. One laboratory study did. Another laboratory study reported 81 per cent detection with 28 per cent false positives for a different pest, and the field study reported 44 per cent.¹³

18. The unfalsifiable defence

The sentence that makes the whole claim unfalsifiable, quoted in full because it deserves attention.

False positives, where the dog alerts but no bugs are found visually, can occur occasionally. Interestingly, these false positives often mean bed bugs ARE actually present, but hidden so deeply that even careful visual inspection cannot locate them.⁶

18.1 Why this cannot be argued with

If an alert with visible bugs is a success, and an alert without visible bugs is also a success because the bugs must be hidden, then no alert can ever be wrong.

A claim that cannot be wrong cannot be evidence of anything. That is not a rhetorical point; it is the reason the field study had to measure against known infestation status rather than against whether bugs were subsequently found.

18.2 The part of it that is true

Deeply hidden bugs do exist, and visual inspection does miss things. That is the whole reason to use a dog.

The problem is not that the explanation is impossible. It is that it is applied to every failure without a mechanism for distinguishing the cases where it holds from the cases where it does not.

18.3 What would make it testable

Follow-up. An alert in an apartment with no visible bugs is a prediction, and the prediction can be checked by monitoring or by whether an infestation emerges.

We located no evidence that such follow-up is routinely performed, and we would treat a company that does it as making a materially stronger claim than one that does not.

18.4 The incentive problem underneath

Follow-up is the one procedure that could generate the feedback §13 identifies as missing, and it is also the procedure most likely to show that an alert was wrong.

A provider who treats every alert as correct has no operational reason to build it, and a provider whose inspections are followed by their own treatment work has a reason not to. We are describing an incentive rather than alleging conduct, and the remedy is that the party checking should not be the party selling the treatment.

19. The certification question

The word that all the marketing rests on.

One source notes minimal certification oversight, with many certified dogs having come through wildly different certification processes, as they are not standardized across the industry, and lists among the risks that poorly trained dogs or handlers can produce false positives, leading to unnecessary extermination costs.⁷

19.1 Why an unstandardised certification is worse than none

No certification at least leaves a purchaser aware they are taking a risk. A certificate with no common standard behind it transfers confidence without transferring information.

That is not only a theoretical objection. Section 11c reports that certification showed no significant relationship with detection rate in the field evaluation.¹ The credential and the performance were uncorrelated.

This journal has made the same point about product claims and about inspection reports. A credential is only as useful as the consistency of what it attests to.

19.2 What would fix it

Not more certificates. Publishing the test: what the team was asked to find, under what conditions, how often it alerted correctly, and how often it alerted on nothing.

Those four numbers would let a purchaser compare providers on the same basis, and their absence is why a laboratory figure from 2008 is still doing the selling.

20. What the critics inside the trade say

Notably, the sharpest criticism we found came from a pest control company.

That operator writes that dogs are simply dogs, that their ability to detect scent is excellent but that does not mean they will consistently provide dependable results, that some dogs get fatigued, some get distracted, some get confused by inconsistent training or handler error, and that certified is a meaningless term in my estimation because once that dog leaves the training area and the certifying instructors it will be subject to the handlers and caretakers who often do not have the same level of skill in managing the dog's performance.⁸

20.1 Why that is worth quoting

It is an operator's description of the §13 mechanism in plain language, arrived at from experience rather than from the paper.

The observation that a certificate attaches to a moment rather than to a career is the sharpest version of the drift argument, and it identifies the right remedy: periodic re-testing against known targets, by someone other than the team.

20.2 The commercial motive to be aware of

A company that does not offer canine inspection has an interest in criticising it, exactly as a company that does has an interest in the 98 per cent figure. We have quoted both and the study sits between them.

Another source warns that some pest control companies will use the dogs as a way to showcase that your bed bug issue is worse than it actually is.⁷ That is an allegation rather than a finding, and we record it as such.

21. What we would conclude

The capability is real. Dogs can be trained to locate live bed bugs and viable eggs.⁴

A specific team's performance is unknown to the buyer. The field range was 10 to 100 per cent.¹

Ask for the false-positive rate too. The two move together and one alone is meaningless.¹²

A negative result should not be treated as clearance. Not at a measured mean detection of 44 per cent.¹

An alert should be treated as a reason to look, not as a finding. Fifteen per cent mean false positives, up to 57.¹

Visible old evidence makes an alert less informative, not more. It is the documented cueing trigger.²

Ask what verification the team receives. Feedback is the variable the study identifies.²

Do not rely on tenure, certification or price. None predicted accuracy, and price ran backwards.¹⁹

Consider the conditions on the day. Panting blocks sniffing, and the same team varies significantly between days.¹

22. Limitations and open questions

One field study. Eleven teams, sixteen inspections, one region, published in 2014.¹ It is the only field evaluation we located and one study is one study.

The teams were relatively inexperienced. Some pairs had worked together under a year, which is a genuine limitation on generalising the result.⁵

Training and certification may have improved since. Eleven years have passed and we located no more recent field evaluation, which is itself the point but also means the current position is unmeasured.

The laboratory figures reach us through trade sources. We have not read the 2008 paper directly, and the numbers come from secondary accounts.⁴⁵

The drugs and explosives study is unverified. Stated in §15.1.

The seven-firm evaluation is a trade publication. It is a field evaluation reported in a trade technical magazine rather than a peer-reviewed paper.⁹

Four sources are commercial. Two promote canine inspection, one criticises it, and one is general consumer content; all have positions.⁴⁶⁷⁸

Sections 9.1, 13.1, 13.2, 14.2, 17.1, 18.1 and 19.1 are our reasoning. The threshold interpretation, the error-reinforcement argument, the drift prediction, the old-evidence point, the unfalsifiability readings and the certification argument are ours rather than sourced positions.

Our commercial position. APC Exterminators does not operate a detection dog. That gives us an interest in the critical reading, and a reader should weigh this article knowing that. We have tried to meet it by quoting the laboratory results in full, recording the mitigating factors in §7.2 and §22, and stating in §21 that the capability is real.

23. Conclusion

Dogs located planted vials in a laboratory hotel room at 98 per cent with no false positives, and the researcher who ran that work said in the same breath that canines and handlers are not all equal.⁴⁵ Six years later, eleven teams evaluated in naturally infested apartments averaged 44 per cent detection and 15 per cent false positives, while every handler believed their dog performed at 95 per cent or better.¹ One inspection in sixteen met a detection standard that had been proposed independently a decade earlier.³

Two results carry most of the weight. The false-positive rate rose with the detection rate, which means the two are a trade-off rather than a pair of independent virtues and a single accuracy number describes neither.¹ And detection probability was unrelated to infestation level, which is hard to square with alerts being driven mainly by how much target odour was in the room.¹

The explanation the authors give is not about noses. In training, someone knows the answer and errors get corrected. In an occupied apartment nobody knows the answer, which is why the inspection was booked, so a wrong alert is rewarded exactly like a right one. The technology was validated in the one setting where it can be checked and sold for use in the setting where it cannot, and the claim that a false positive is really a deeply hidden true positive is the mechanism by which that arrangement keeps working.

References

1. Cooper, R., Wang, C. and Singh, N. (2014). Accuracy of Trained Canines for Detecting Bed Bugs (Hemiptera: Cimicidae). Journal of Economic Entomology, 107(6), 2171. PubMed record 26470083. Principal source. Used for the study design evaluating the accuracy of 11 canine detection teams in naturally infested apartments; for the finding that all handlers believed their dogs could detect infestations at a very high rate of 95 per cent or greater; for the results that in three separate experiments the mean detection rate was 44 per cent with a range of 10 to 100 and the mean false-positive rate was 15 per cent with a range of 0 to 57; for the finding that the false-positive rate was positively correlated with the detection rate; and for the finding that the probability of a bed bug infestation being detected by trained canines was not associated with the level of bed bug infestations. https://pubmed.ncbi.nlm.nih.gov/26470083/
2. Cooper, R., Wang, C. and Singh, N. (2014), full text. Journal of Economic Entomology, 107(6), 2171. Used for the statement that there are no data demonstrating the reliability of trained canines under natural field conditions; for the authors' position that it is necessary to consider both detection rate and false-positive rate to evaluate a team's effectiveness; for their explanation that errors by dogs and handlers are identifiable and correctable in a controlled setting while the natural field setting is very complex with odours from many different sources where errors can easily occur, go unidentified and thus remain uncorrected, reinforcing the incorrect behaviour, creating challenges in ongoing training and evaluation; for their observation that false-positive alerts occurred in some apartments where old evidence of fecal spotting, carcasses and exuvia were readily visible and recognised by handlers during the inspection, which may have led to unintentional cues to the dog by the handler; and for their reference to comparable work on detection of wild brown tree snakes. https://academic.oup.com/jee/article/107/6/2171/796348
3. Cooper, R., Wang, C. and Singh, N. (2014), author manuscript hosted by the Rutgers University Department of Entomology. New Jersey Experiment Station publication. Used for the standard proposed by Brooks and colleagues (2003) for termite detection, that it is not unreasonable to expect a properly trained dog to meet a minimum standard with a positive indication rate of 90 per cent or greater and a false-positive rate of 10 per cent or less; for the report that the only other study investigating accuracy of trained dogs for termites, Lewis and colleagues (1997), reported a mean detection rate of 81 per cent and false-positive rate of 28 per cent in a laboratory setting, and that the observed bed bug field figures were more in line with that result; for the statement that mean detection and false-positive rates were similar regardless of whether detection firms were aware they were being observed; and for the assessment that when judged on a 90 per cent detection rate only one out of 16 inspections met the proposed standard, and when judged on a 10 per cent false-positive rate only 5 out of 16 met it. https://entomology.rutgers.edu/personnel/changlu-wang/docs/Cooper2014AccuracyCanines.pdf
4. Wang, C. Bed bug monitor detection tools and techniques. Pest Control Technology, 2011. Trade technical publication reporting a field evaluation. Used for the June to July 2011 evaluation of seven canine detection firms, with each apartment inspected by three or four firms and detection rates ranging from 11 to 83 per cent; for the observation that the two companies charging the most, at a thousand dollars, and having been in business more than three years had very low detection rates of 25 and 30 per cent with high false alert rates; and for the conclusions that variability in accuracy among canine scent firms is great, that additional studies are necessary to understand the factors influencing accuracy, and that guidance is needed for maintaining a quality canine scent programme. https://www.pctonline.com/article/pct0811-bed-bug-monitor-detection-tools-techniques/
5. The ABCs of Bed Bug Sniffing Dogs: Accuracy, Benefits, and Costs. Commercial pest control company article. Trade source with a commercial interest, cited as attributed material reporting a university study. Used for the report that in a realistic hotel room setting the dogs showed a 98 per cent accuracy rate in locating six hidden vials of bed bugs with no false positives; for the caution that extrapolating those accuracy rates to all bed bug dogs may be somewhat misleading; and for the quoted statement by one of the researchers that the study has shown dogs can be trained to accurately locate live bed bugs and viable bed bug eggs, however bed bug detecting canines and handlers are not all equal, and the quality of canines depends on the efficiency of their training, their handlers, and what the dogs are trained to do. https://mandmpestcontrol.com/the-abcs-of-bed-bug-sniffing-dogs-accuracy-benefits-and-costs/
6. How Accurate Are Bed Bug Dogs. Commercial canine inspection service article. Trade source with a direct commercial interest, cited as attributed material. Used for the report of a 2008 University of Florida study finding dogs 97.5 per cent accurate in distinguishing bed bugs from other household pests, 95 per cent accurate in distinguishing live bed bugs from bed bug debris with a 3 per cent false-positive rate, and 98 per cent accurate in detecting six bed bug vials in a hotel room with no false-positive alerts; for the statement that extermination companies started using bed bug dogs during the early 2000s; and for the observation that in the 2014 field study the canine teams had not worked together long, with some pairs working together for less than a year. https://discreetbedbuginspections.com/how-accurate-are-bed-bug-dogs/
7. Bed Bug Sniffing Dogs: Cost, Accuracy and What to Expect. Commercial canine inspection service article. Trade source with a direct commercial interest, cited here as an example of the marketing claims this paper examines. Used for the statements that published peer-reviewed studies consistently demonstrate that properly trained and certified bed bug detection dogs achieve 95 to 98 per cent accuracy rates; that this level of accuracy far exceeds traditional visual inspection methods which typically catch only 30 to 50 per cent of early-stage infestations; that accuracy depends heavily on training quality, handler experience and regular certification testing; that false positives, where the dog alerts but no bugs are found visually, can occur occasionally but often mean bed bugs are actually present though hidden so deeply that even careful visual inspection cannot locate them; and that false negatives are rare with properly certified and trained teams. https://www.custombedbug.com/blog/everything-you-wanted-to-know-about-bed-bug-sniffing-dogs/
8. Bed Bug Detection Dogs: Fact or Fiction? Consumer information site. Non-peer-reviewed source. Used for the statements that poorly trained dogs or handlers can produce false positives leading to unnecessary extermination costs; that there is minimal certification oversight, with many dogs certified to detect bed bugs having come through wildly different certification processes which are not standardised across the industry; that factors such as distractions, competing scents or the size of the infestation can influence performance and potentially cause false negatives; and for the allegation that some pest control companies will use the dogs as a way to showcase that a bed bug issue is worse than it actually is by misrepresenting the process or the results. https://www.webvet.com/bed-bug-detection-dogs/
9. How Accurate are K9 Bed Bug Inspections? Commercial pest control company article. Trade source with a commercial interest in the opposite direction from references 5 and 6, cited as attributed material. Used for the reported study in which researchers found that detection canines falsely alerted their handlers to the presence of drugs or explosives over 200 times in a building into which the researchers had never brought any contraband items or explosives, with false positive alerts by canine teams in every room inspected; and for the author's assessment that dogs are simply dogs, that their ability to detect scent is excellent but does not mean they will consistently provide dependable results, that some dogs get fatigued, distracted or confused by inconsistent training or handler error, and that certified is a meaningless term because once a dog leaves the training area and the certifying instructors it is subject to handlers and caretakers who often do not have the same level of skill in managing the dog's performance. https://spearheadpestcontrol.com/how-accurate-are-k9-bed-bug-inspections-2/