CSI and cadaver dogs - some facts and statistics.

[Alice Purjorick]

probably best just to accept it's all over your head then

I like to learn at the feet of undoubted masters. I know you and OB are; you both told me yourselves.

[G-Unit]

Doesn't it only work if the dogs get one out of every 5 alerts wrong? Where does that assumption come from?

[OxfordBloo]

Doesn't it only work if the dogs get one out of every 5 alerts wrong? Where does that assumption come from?

It is the average of the published peer reviewed blinded academic investigations finding on accuracy- between 90% and 60%.

[G-Unit]

It is the average of the published peer reviewed blinded academic investigations finding on accuracy- between 90% and 60%.

do you have any links to the investigations please?

[OxfordBloo]

do you have any links to the investigations please?

They have been cited repeatedly and are available on line. Not one gives 100% accuracy.

If you feel they are other faultless, please find a reference.

So long as the probability is less than 100% the parallel probability will reduce by the product of the two uncertainties.

[G-Unit]

They have been cited repeatedly and are available on line. Not one gives 100% accuracy.

If you feel they are other faultless, please find a reference.

So long as the probability is less than 100% the parallel probability will reduce by the product of the two uncertainties.

Please could you provide links to the studies you are quoting? Thank you.

[OxfordBloo]

Please could you provide links to the studies you are quoting? Thank you.

Google is your friend.

My point is simply that two dog alerts make the probability less than single alerts.

[slartibartfast]

Bumped for slartibartfast to share his knowledge of statistical analysis by addressing the mathematics rather tan trying to shoot the messenger.

Sorry, had a busy day in the real world, will address tomorrow.

[G-Unit]

Google is your friend.

My point is simply that two dog alerts make the probability less than single alerts.

 You say dog alerts are false 4 out of 5 times. I simply asked on what you based that statement. If you can't give the source upon which you are basing your assertion then your assertion can't be verified and your point is null and void.

[OxfordBloo]

You say dog alerts are false 4 out of 5 times. I simply asked on what you based that statement. If you can't give the source upon which you are basing your assertion then your assertion can't be verified and your point is null and void.

I said IF they are correct 4 out of 5 times. All probabilities are less than one, so when multiplied as in be example the probability drops. This is my point.

I have wasted two much time quoting from the research only to have people nitpick. If people want to suggest that dogs are 100% perfect, then all they need to do is produce blinded peer reviewed academic papers proving that; in eight years I have never seen one.

But despite that the Maths holds.

[G-Unit]

I said IF they are correct 4 out of 5 times. All probabilities are less than one, so when multiplied as in be example the probability drops. This is my point.

I have wasted two much time quoting from the research only to have people nitpick. If people want to suggest that dogs are 100% perfect, then all they need to do is produce blinded peer reviewed academic papers proving that; in eight years I have never seen one.

But despite that the Maths holds.

Sorry, you're quite correct, you said they're right 4 out of 5 times. I just want to know where you got that figure?It's not nitpicking to ask someone to provide backup for their assertions is it?

[OxfordBloo]

Sorry, you're quite correct, you said they're right 4 out of 5 times. I just want to know where you got that figure?It's not nitpicking to ask someone to provide backup for their assertions is it?

Do you not understand conditional. IF!

My point is not that the dogs are 80% accurate, but IF they are, their accuracy is reduced by the requirement for two parallel alerts.

Most papers suggest accuracies between 60% and 90%.

Whatever the actual accuracies, they are reduced by the product of their individual uncertainties

90% becomes 81%
80% 64%
70% 49%
60% 36%

[Alice Purjorick]

Do you not understand conditional. IF!

My point is not that the dogs are 80% accurate, but IF they are, their accuracy is reduced by the requirement for two parallel alerts.

Most papers suggest accuracies between 60% and 90%.

Whatever the actual accuracies, they are reduced by the product of their individual uncertainties

90% becomes 81%
80% 64%
70% 49%
60% 36%

I understand squaring numbers I did that for 11+. and I understand basic probability all though it is unclear why you consider the parallel events have a dependent relationship.
You say however "most papers suggest" but you are unable or unwilling to show where or what those papers are. Saying Google is your friend don't cut it. You could be kind enough to tell us which papers your opinion is reliant on. It is a simple enough request.

[G-Unit]

Do you not understand conditional. IF!

My point is not that the dogs are 80% accurate, but IF they are, their accuracy is reduced by the requirement for two parallel alerts.

Most papers suggest accuracies between 60% and 90%.

Whatever the actual accuracies, they are reduced by the product of their individual uncertainties

90% becomes 81%
80% 64%
70% 49%
60% 36%

So your original figures upon which you are basing your calculations are figures you have decided are probably about right? It's all based on an assumption? Change the assumption and you change the results.

[OxfordBloo]

So your original figures upon which you are basing your calculations are figures you have decided are probably about right? It's all based on an assumption? Change the assumption and you change the results.

No. My contention simply is that a joint Eddie Keela alert has a higher uncertainty than a single alert whatever the actual vale of certainty for a single alert.

80% is a fair average of the published papers, but that is not my point.

The point is that the uncertainty increases because of the parallel nature of the test.