modelling of Common Myna impact in Canberra

Subject: modelling of Common Myna impact in Canberra
From: Andrew Taylor <>
Date: Sun, 19 Aug 2012 14:33:01 +1000
A recently published paper [1] claimed to have established that
Common Mynas had harmful effects on Canberra populations of 11 bird species -
the first time this has been demonstrated. 

I posted some quick comments on birding-aus but as these claims have
been widely publicized [2] and they'll be used  to justify myna control
programs, I've written a more detailed explanation of why I think the
paper's claims are unsupported and the paper provides no useful evidence
of myna impacts.

The paper [1] describes a model built on  COG Garden Survey Data.
The authors divide Canberra into 4 regions and convert the survey results
into biannual population estimates for 20 species of birds over the 29
years the survey has run.

The external model variables are essentially population, dwellings,
extent of 5 vegetation types, mynas and year.

The model attempts to explain the changes in 20 bird populations over
1981-2009 in the 4 regions using these model variables.

The demographics of Canberra myna populations, as estimated in the
paper vary between regions but overall there is a peak 5-15 years ago
followed by a decline.  In two regions the decline is recent and in 2009
myna populations are still near 50% of their peak.  In the other two
regions the peak is earlier and by 2009 myna populations are below 20%
of their peak.

The effects mynas are suspected to cause are largely density-dependent
(proportional to the number of mynas) so the obvious model input to
estimate myna impact is myna density.  The model instead uses a proxy
variable, years since myna establishment.  This yields a variable which
linearly increases from 1991, 1993, 1981 & 1989 in the respective regions.

The paper doesn't compare this variable to myna density but its clear
there will be a large disparity.  For example from eyeballing the
graph, the region 3 1986 myna density looks to be about double the 2009
myna density, but the proxy variable will be 5 in 1986 and 28 in 2009.
In other words 2009 myna density is about half 1986 but the modelled
myna impact is a factor of over 5 times larger.

The justification in the paper for this  choice of variable makes no
numerical or ecological sense to me.  This variable choice is crucial
because all the claims in the paper are based solely on attributing
changes in the COG data to this myna variable.

This reduction of changes in myna density to a simple linear increase
raises concern about correlations with variables not included in the
model. Its not uncommon for variables to show roughly linear increase
or decrease for a period of time. For example there conceivably might
be a bias in the COG data related to observer recruitment, and if it
increased or decreased roughly linearly and roughly over the same time
period as myna establishment, part of its effects might erroneously be
attributed to mynas.

Its important to consider what variables are included in the model.

One obvious omission is rainfall.  Its  easy to obtain. It has been linked
to bird abundance and it shows large variation through the time period.
For example, on a quick look at BOM data for the the Botanic Gardens,
Sep-Nov rainfall appears to vary by a factor of 6 between 1981-2009.
Rainfall may be less important in a suburban environment its still seems
some of the variation in some species will be explained by rainfall.

Another omission is the effects of birds other than mynas. For example,
Pied Currawong, Red Wattlebirds and Noisy Miners have all been linked
to negative effects on other bird species in suburban environments, but
the model excludes this possibility.  Particularly problematic is the
omission of Noisy Miners because their Canberra population is roughly
constant until 2000 then shows a large roughly linear increase [3].

As Noisy Miners aren't included the model its impossible to exclude their
effects being erroneously attributed to mynas.  There is good reason to
suspect there might be such effects because a number of researchers have
noted such effects elsewhere and there has been such a large increase
in Canberra's Noisy Miner populations.  There is also good reason to
suspect such effects could be erroneously attributed to mynas in the
model because the growth in  Noisy Miner population follows a pattern
that correlates with the mynas proxy variable.

In other words the concern is if that if the authors had taken the same
data and same approach to modelling Noisy Miner impact they would have
found similar impacts but attributed them to Noisy Miners.

There are other species whose omission as model variables is
problematic. For example, cockatoo numbers triple through the time period.
Their density and the variation in their density is similar in magnitude
to mynas.  While the authors suggest mynas affect other cavity nesters
by competing for nest sites, their model assumes cockatoos do not affect
other cavity nesters by competing for nest sites.  This asymmetry is
hard to justify.

So, for example, it is impossible to exclude the negative impact on
Galahs the model attributes to mynas being actually caused by cockatoos.

Its suspected that mynas have a negative impact on cavity nesters.
However the model's outputs for cavity nesters are equivocal: 4 species
supposedly negatively affected and 3 positively affected.

3 of the 4 species supposedly negatively affected species populations
grew through the time period, for example cockatoo numbers tripled,
but the model output indicate mynas slowed this growth.

The only cavity nester which showed a large decline, starling, supposedly
benefited from mynas - an unexpected result. The authors suggest this
large decrease in starling numbers is hiding the negative effect from
mynas.  This is plausible. The model attributes the majority of this
decline to the linear pseudo-variable year. If this large  unexplained
effect is non-linear then it could  mask a negative effect from starlings.

Eyeballing the starling & myna density graphs raises the suspicion that
if the model had used myna density as a variable they might instead have
found the expected negative effect.

This problematic presence of large unexplained effects also occurs for
other species. For example, a large part of the increase in cockatoos
is unexplained by the model, and is also attributed to the linear
pseudo-variable year.  If this large effectively unexplained effect is
non-linear then part of it could be being erroneously attributed to the
myna variable.

Note when the model produced an unexpected positive impact from mynas
(starlings) the model output was rejected, but when the  model produced
an unexpected negative impact (cockatoos) the model output was accepted,
and  conveyed via press release to Australia's major media outlets.

Another concern is the extent to which the model is opaque.  The PLOS
paper presents only a table of coefficients.  We aren't given the values
of the model variables, e.g. as graphs, to which these coefficients apply.
The authors discuss only the signs of the coefficients and only for the
myna variable.  Some of the coefficients look to have implausible values
but the information needed to interpret them is absent.

It easy to criticise models - all models are wrong, some are useful -
and even given the lovely 29 years of COG data its no small task to
tease apart competing effects - but this model fails to do so and fails
to support the claims made.

Personally I think is likely given the high density they reached mynas
did affect at least a few Canberra bird species, but this modelling leaves
me none the wiser as to which species or the magnitude of these effects.

On the other hand 2 recent observational papers [4,5] (both paywalled)
on Sydney & Newcastle mynas I found while writing this did suggest to me
I (and others) might be over-estimating myna impacts.


[3] Canberra Bird Notes 37 (2) June 2012 Noisy Miners in the COG Garden Bird 
Survey Martin Butterfield

<Prev in Thread] Current Thread [Next in Thread>

The University of NSW School of Computer and Engineering takes no responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the Canberra Ornithologists Group mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the list contact David McDonald, list manager, phone (02) 6231 8904 or email . If you can not contact David McDonald e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU