<s>
A	O
funnel	B-Application
plot	I-Application
is	O
a	O
graph	O
designed	O
to	O
check	O
for	O
the	O
existence	O
of	O
publication	O
bias	O
;	O
funnel	B-Application
plots	I-Application
are	O
commonly	O
used	O
in	O
systematic	O
reviews	O
and	O
meta-analyses	O
.	O
</s>
<s>
In	O
the	O
absence	O
of	O
publication	O
bias	O
,	O
it	O
assumes	O
that	O
studies	O
with	O
high	O
precision	B-General_Concept
will	O
be	O
plotted	O
near	O
the	O
average	O
,	O
and	O
studies	O
with	O
low	O
precision	B-General_Concept
will	O
be	O
spread	O
evenly	O
on	O
both	O
sides	O
of	O
the	O
average	O
,	O
creating	O
a	O
roughly	O
funnel-shaped	O
distribution	O
.	O
</s>
<s>
A	O
funnel	B-Application
plot	I-Application
is	O
a	O
scatterplot	B-Application
of	O
treatment	O
effect	O
against	O
a	O
measure	O
of	O
study	O
precision	B-General_Concept
.	O
</s>
<s>
It	O
is	O
used	O
primarily	O
as	O
a	O
visual	O
aid	O
for	O
detecting	O
bias	O
or	O
systematic	O
heterogeneity	B-General_Concept
.	O
</s>
<s>
A	O
symmetric	O
inverted	O
funnel	B-Operating_System
shape	O
arises	O
from	O
a	O
‘	O
well-behaved	O
’	O
data	O
set	O
,	O
in	O
which	O
publication	O
bias	O
is	O
unlikely	O
.	O
</s>
<s>
An	O
asymmetric	O
funnel	B-Operating_System
indicates	O
a	O
relationship	O
between	O
treatment	O
effect	O
estimate	O
and	O
study	O
precision	B-General_Concept
.	O
</s>
<s>
This	O
suggests	O
the	O
possibility	O
of	O
either	O
publication	O
bias	O
or	O
a	O
systematic	O
difference	O
between	O
studies	O
of	O
higher	O
and	O
lower	O
precision	B-General_Concept
(	O
typically	O
‘	O
small	O
study	O
effects’	O
)	O
.	O
</s>
<s>
Whatever	O
the	O
cause	O
,	O
an	O
asymmetric	O
funnel	B-Application
plot	I-Application
leads	O
to	O
doubts	O
over	O
the	O
appropriateness	O
of	O
a	O
simple	O
meta-analysis	O
and	O
suggests	O
that	O
there	O
needs	O
to	O
be	O
investigation	O
of	O
possible	O
causes	O
.	O
</s>
<s>
A	O
variety	O
of	O
choices	O
of	O
measures	O
of	O
‘	O
study	O
precision’	O
is	O
available	O
,	O
including	O
total	O
sample	O
size	O
,	O
standard	B-General_Concept
error	I-General_Concept
of	O
the	O
treatment	O
effect	O
,	O
and	O
inverse	O
variance	O
of	O
the	O
treatment	O
effect	O
(	O
weight	B-Algorithm
)	O
.	O
</s>
<s>
Sterne	O
and	O
Egger	O
have	O
compared	O
these	O
with	O
others	O
,	O
and	O
conclude	O
that	O
the	O
standard	B-General_Concept
error	I-General_Concept
is	O
to	O
be	O
recommended	O
.	O
</s>
<s>
When	O
the	O
standard	B-General_Concept
error	I-General_Concept
is	O
used	O
,	O
straight	O
lines	O
may	O
be	O
drawn	O
to	O
define	O
a	O
region	O
within	O
which	O
95%	O
of	O
points	O
might	O
lie	O
in	O
the	O
absence	O
of	O
both	O
heterogeneity	B-General_Concept
and	O
publication	O
bias	O
.	O
</s>
<s>
In	O
common	O
with	O
confidence	O
interval	O
plots	O
,	O
funnel	B-Application
plots	I-Application
are	O
conventionally	O
drawn	O
with	O
the	O
treatment	O
effect	O
measure	O
on	O
the	O
horizontal	O
axis	O
,	O
so	O
that	O
study	O
precision	B-General_Concept
appears	O
on	O
the	O
vertical	O
axis	O
,	O
breaking	O
with	O
the	O
general	O
rule	O
.	O
</s>
<s>
Since	O
funnel	B-Application
plots	I-Application
are	O
principally	O
visual	O
aids	O
for	O
detecting	O
asymmetry	O
along	O
the	O
treatment	O
effect	O
axis	O
,	O
this	O
makes	O
them	O
considerably	O
easier	O
to	O
interpret	O
.	O
</s>
<s>
The	O
funnel	B-Application
plot	I-Application
is	O
not	O
without	O
problems	O
.	O
</s>
<s>
If	O
high-precision	O
studies	O
are	O
different	O
from	O
low-precision	O
studies	O
with	O
respect	O
to	O
effect	B-Application
size	I-Application
(	O
e.g.	O
,	O
due	O
to	O
different	O
populations	O
examined	O
)	O
a	O
funnel	B-Application
plot	I-Application
may	O
give	O
a	O
wrong	O
impression	O
of	O
publication	O
bias	O
.	O
</s>
<s>
The	O
appearance	O
of	O
the	O
funnel	B-Application
plot	I-Application
can	O
change	O
quite	O
dramatically	O
depending	O
on	O
the	O
scale	O
on	O
the	O
y-axis	O
whether	O
it	O
is	O
the	O
inverse	O
square	O
error	O
or	O
the	O
trial	O
size	O
.	O
</s>
<s>
Researchers	O
have	O
a	O
poor	O
ability	O
to	O
visually	O
discern	O
publication	O
bias	O
from	O
funnel	B-Application
plots	I-Application
.	O
</s>
