Marsh or Willow Tit (Europe, determining) (1 Viewer)

[TomSmith]

but this was not a thorough study and it had not been verified on a large sample so far. Nobody seemed to use it! However, it is being verified this winter in UK (and on skins from other European races) and is working at 98% accuracy (on hundreds of birds). This is for a paper to be submitted later this year. So it seems highly reliable.

Has this paper been published? Was just wondering - dug up this thread while trying to ID a marsh-willow tit seen yesterday.

 

[Flemingo]

A Testing of the bill thing (unpublished as yet) apparently shows it to be 98% accurate. That beats all other known criteria (including tail feather lengths).

Distinguishing features that are less than 100% accurate are of little concern if both species are almost equally abundant. Unfortunately than can become significant if one species is much rarer.

Lets presume Willow tits are only 10% as abundant as marsh tits in a given area.
Lets look at a 1000 ids, 900 are really Marsh and 100 really Willow.
If something is 98% accurate then out of the 900 Marsh 18 will be misidentified as Willow and only 2 Willow will be misidentified as Marsh.
So we have out of the 1000 ids 98 real Willows and 18 false Willows so that roughly 1 in 6 birds identified as Willow tits will actually be Marsh tits!

If Willow tits are only 1% as abundant then 2 in 3 Willow tits would actually be Marsh tits.

If Willow tits are absent then 100% of the 18 'Willow tits' will be Marsh.

If the relative populations are nearer 50:50 then the misidentification is irrelevant. Its not to do with the accuracy (specificity and sensitivity) of a feature which is constant. It is all to do with the positive predictive value (how likely the bird really is the species you think it is if the feature is present) which is heavily influenced by relative abundance.

 

[Ken Hall]

My experience of Marsh and Willow Tits amounts to one Marsh, identification clinched by call. Collins states that Marsh Tits may be seen at birdtables in winter, but not so for Willow Tits. How reliable is this? But Collins also states that Crested Tits do not visit birdtables, which I know to be incorrect. That's nearly always where I see them! Any comments welcome.

 

[bitterntwisted]

Collins states that Marsh Tits may be seen at birdtables in winter, but not so for Willow Tits.

It's plain wrong, and should be removed from the new edition. Both use feeders and tables regularly.

Graham

 

[Jos Stratford]

Collins states that Marsh Tits may be seen at birdtables in winter, but not so for Willow Tits. How reliable is this? But Collins also states that Crested Tits do not visit birdtables, which I know to be incorrect.

Indeed Collins is incorrect, I get all three species at my feeders, Marsh Tits throughout the year and in greatest abundance, Willow Tits regularly in winter and occasionally in summer and Crested Tits only in winter.

 

[Ken Hall]

Indeed Collins is incorrect, I get all three species at my feeders, Marsh Tits throughout the year and in greatest abundance, Willow Tits regularly in winter and occasionally in summer and Crested Tits only in winter.

thanks for that. I had my suspicions already but it's nice to have it confirmed.

 

[Jos Stratford]

Distinguishing features that are less than 100% accurate are of little concern if both species are almost equally abundant. Unfortunately than can become significant if one species is much rarer.

Lets presume Willow tits are only 10% as abundant as marsh tits in a given area.
Lets look at a 1000 ids, 900 are really Marsh and 100 really Willow.
If something is 98% accurate then out of the 900 Marsh 18 will be misidentified as Willow and only 2 Willow will be misidentified as Marsh.
So we have out of the 1000 ids 98 real Willows and 18 false Willows so that roughly 1 in 6 birds identified as Willow tits will actually be Marsh tits!

If Willow tits are only 1% as abundant then 2 in 3 Willow tits would actually be Marsh tits.

If Willow tits are absent then 100% of the 18 'Willow tits' will be Marsh.

If the relative populations are nearer 50:50 then the misidentification is irrelevant. Its not to do with the accuracy (specificity and sensitivity) of a feature which is constant. It is all to do with the positive predictive value (how likely the bird really is the species you think it is if the feature is present) which is heavily influenced by relative abundance.

You may be interested in this paper

The probability applies to each bird individually, not each group of birds. In any event, the paper includes probabilities for each species on its own (e.g. if we're talking populations then out of every 100 Willow, less than 4 would be wrongly i.d.'d).

 

[Flemingo]

You may be interested in this paper

The probability applies to each bird individually, not each group of birds. In any event, the paper includes probabilities for each species on its own (e.g. if we're talking populations then out of every 100 Willow, less than 4 would be wrongly i.d.'d).

Its not a case of how many willow tits are misidentified its a case of how many birds identified as wilow tits are actually willow tits. Its the difference between sensitivity, specificity of a test and positive /negative predictive values. The former are independant of relative population numbers but the later are sensitive to it. I have altered my calculations to fit in with the figures given in the paper quoted.

Taking a 1000 bird census in which we know 10% are Willow tits (you have to presume a given level to do the calculation) then using the figures in the paper approximately 12% of birds identified as Willow tits will actually be Marsh but less than 1% of those identified as Marsh will be Willow.

In more marginal areas of Willow tit range were we can suggest only 1% of the Willow/Marsh tits are really Willow then the figures would be 90% identified as Willow tits are really Marsh and .06% identified as Marsh are really Willow.

this presumes only a single criteria is used.

Obviously if there were only Marsh tits ALL those identified as Willow tits using this criteria ( 13 per 1000 individuals) would be Marsh.


Never rely on a single criteria to seperate two species where one is much rarer than the other unless its 100%!

Any mathematicians out there feel free to put me right :t:

 

[Indiana]

And what is this?

 

[bitterntwisted]

You may be interested in this paper

The probability applies to each bird individually, not each group of birds. In any event, the paper includes probabilities for each species on its own (e.g. if we're talking populations then out of every 100 Willow, less than 4 would be wrongly i.d.'d).

Try it this way, Jos.

A pharmaceutical company devises a new test for a rare illness, which affects only around 6000 people worldwide. The test is very accurate - they say it's 99.999% accurate (i.e. it produces only 0.001% false positives)
You take the test and are shocked to find that you test positive. What is the actual probability that you have the illness... <<scroll>>





... The probability you actually have the illness, despite the test's accuracy, is only 9%. The rarity of the condition means the test must be even more accurate to be a useful diagnostic tool.

The same problem arises with biometric security, and does apply to the Willow/Marsh problem in the UK in particular where one or other species are very rare in many locations. The actual diagnostic value of the KN formula depends on the relative abundance of the species in the location, not just on the false positive and false negative rates.

Graham

 

[Jos Stratford]

Try it this way, Jos.

A pharmaceutical company devises a new test for a rare illness, which affects only around 6000 people worldwide. The test is very accurate - they say it's 99.999% accurate (i.e. it produces only 0.001% false positives)
You take the test and are shocked to find that you test positive. What is the actual probability that you have the illness... <<scroll>>



... The probability you actually have the illness, despite the test's accuracy, is only 9%. The rarity of the condition means the test must be even more accurate to be a useful diagnostic tool.

The same problem arises with biometric security, and does apply to the Willow/Marsh problem in the UK in particular where one or other species are very rare in many locations. The actual diagnostic value of the KN formula depends on the relative abundance of the species in the location, not just on the false positive and false negative rates.

Graham

Way beyond me Lucky both species are common out here.

Besides, I'm crap with taking medical tests, so in all likelyhood would be dead before I ever actually tested to find out if I was ill. It seems I may have TB at present - had an idea for over 6 weeks and only today I have begun to check it out (and only then 'cos I was pressurised to :-O )

 

[bitterntwisted]

And what is this?

Clear-cut cheek divide and pale spot at bill base make it a Marsh Tit (depending on the relative abundance of Marsh and Willow at the location)

Graham

 

[marklhawkes]

Has this paper been published? Was just wondering - dug up this thread while trying to ID a marsh-willow tit seen yesterday.

Hi Tom,

Where was your tit? If seen in Cambridgeshire, then it will almost certainly have been a Marsh Tit (Willow Tit is quite proabaly gone as a breeding species in the county currently).