<s>
In	O
robust	O
statistics	O
,	O
Peirce	B-Language
's	I-Language
criterion	I-Language
is	O
a	O
rule	O
for	O
eliminating	O
outliers	O
from	O
data	B-General_Concept
sets	I-General_Concept
,	O
which	O
was	O
devised	O
by	O
Benjamin	O
Peirce	O
.	O
</s>
<s>
In	O
data	B-General_Concept
sets	I-General_Concept
containing	O
real-numbered	O
measurements	O
,	O
the	O
suspected	O
outliers	O
are	O
the	O
measured	O
values	O
that	O
appear	O
to	O
lie	O
outside	O
the	O
cluster	O
of	O
most	O
of	O
the	O
other	O
data	O
values	O
.	O
</s>
<s>
First	O
,	O
the	O
statistician	O
may	O
remove	O
the	O
suspected	O
outliers	O
from	O
the	O
data	B-General_Concept
set	I-General_Concept
and	O
then	O
use	O
the	O
arithmetic	O
mean	O
to	O
estimate	O
the	O
location	O
parameter	O
.	O
</s>
<s>
Peirce	B-Language
's	I-Language
criterion	I-Language
is	O
a	O
statistical	O
procedure	O
for	O
eliminating	O
outliers	O
.	O
</s>
<s>
The	O
test	O
,	O
based	O
on	O
a	O
likelihood	B-General_Concept
ratio	I-General_Concept
type	O
of	O
argument	O
,	O
had	O
the	O
distinction	O
of	O
producing	O
an	O
international	O
debate	O
on	O
the	O
wisdom	O
of	O
such	O
actions	O
(	O
Anscombe	O
,	O
1960	O
,	O
Rider	O
,	O
1933	O
,	O
Stigler	O
,	O
1973a	O
)	O
.	O
"	O
</s>
<s>
Peirce	B-Language
's	I-Language
criterion	I-Language
is	O
derived	O
from	O
a	O
statistical	O
analysis	O
of	O
the	O
Gaussian	O
distribution	O
.	O
</s>
<s>
Peirce	B-Language
's	I-Language
criterion	I-Language
was	O
used	O
for	O
decades	O
at	O
the	O
United	O
States	O
Coast	O
Survey	O
.	O
</s>
<s>
Peirce	B-Language
's	I-Language
criterion	I-Language
was	O
discussed	O
in	O
William	O
Chauvenet	O
's	O
book	O
.	O
</s>
<s>
An	O
application	O
for	O
Peirce	B-Language
's	I-Language
criterion	I-Language
is	O
removing	O
poor	O
data	O
points	O
from	O
observation	O
pairs	O
in	O
order	O
to	O
perform	O
a	O
regression	O
between	O
the	O
two	O
observations	O
(	O
e.g.	O
,	O
a	O
linear	O
regression	O
)	O
.	O
</s>
<s>
Peirce	B-Language
's	I-Language
criterion	I-Language
does	O
not	O
depend	O
on	O
observation	O
data	O
(	O
only	O
characteristics	O
of	O
the	O
observation	O
data	O
)	O
,	O
therefore	O
making	O
it	O
a	O
highly	O
repeatable	O
process	O
that	O
can	O
be	O
calculated	O
independently	O
of	O
other	O
processes	O
.	O
</s>
<s>
This	O
feature	O
makes	O
Peirce	B-Language
's	I-Language
criterion	I-Language
for	O
identifying	O
outliers	O
ideal	O
in	O
computer	O
applications	O
because	O
it	O
can	O
be	O
written	O
as	O
a	O
call	O
function	O
.	O
</s>
<s>
A	O
.	O
Gould	O
attempted	O
to	O
make	O
Peirce	B-Language
's	I-Language
criterion	I-Language
easier	O
to	O
apply	O
by	O
creating	O
tables	O
of	O
values	O
representing	O
values	O
from	O
Peirce	O
's	O
equations	O
.	O
</s>
<s>
A	O
disconnect	O
still	O
exists	O
between	O
Gould	O
's	O
algorithm	O
and	O
the	O
practical	O
application	O
of	O
Peirce	B-Language
's	I-Language
criterion	I-Language
.	O
</s>
<s>
In	O
2003	O
,	O
S	O
.	O
M	O
.	O
Ross	O
(	O
University	O
of	O
New	O
Haven	O
)	O
re-presented	O
Gould	O
's	O
algorithm	O
(	O
now	O
called	O
"	O
Peirce	O
's	O
method	O
"	O
)	O
with	O
a	O
new	O
example	O
data	B-General_Concept
set	I-General_Concept
and	O
work-through	O
of	O
the	O
algorithm	O
.	O
</s>
<s>
This	O
methodology	O
still	O
relies	O
on	O
using	O
look-up	O
tables	O
,	O
which	O
have	O
been	O
updated	O
in	O
this	O
work	O
(	O
Peirce	B-Language
's	I-Language
criterion	I-Language
table	O
)	O
.	O
</s>
<s>
In	O
2012	O
,	O
C	O
.	O
Dardis	O
released	O
the	O
R	O
package	O
"	O
Peirce	O
"	O
with	O
various	O
methodologies	O
(	O
Peirce	B-Language
's	I-Language
criterion	I-Language
and	O
the	O
Chauvenet	O
method	O
)	O
with	O
comparisons	O
of	O
outlier	O
removals	O
.	O
</s>
<s>
In	O
order	O
to	O
use	O
Peirce	B-Language
's	I-Language
criterion	I-Language
,	O
one	O
must	O
first	O
understand	O
the	O
input	O
and	O
return	O
values	O
.	O
</s>
<s>
Because	O
Peirce	B-Language
's	I-Language
criterion	I-Language
does	O
not	O
take	O
observations	O
,	O
fitting	O
parameters	O
,	O
or	O
residual	O
errors	O
as	O
an	O
input	O
,	O
the	O
output	O
must	O
be	O
re-associated	O
with	O
the	O
data	O
.	O
</s>
