Journal
of
Comparative
Psychology
2001. Vol.
115,
No. 4,
351-364
Copyright 2001
by
the
American
Psychological
Association,
Inc.
<5735-703fi»l/$5.00
DOI:
10.1037//0735-7036.115.4351
Early
Temperamental
Traits
in an
Octopus
(Octopus
bimaculoides)
David
L.
Sinn
and
Nancy
A.
Penin
Portland
State
University
Jennifer
A.
Mather
University
of
Lethbridge
Roland
C.
Anderson
The
Seattle
Aquarium
During
their
3rd
week
of
life,
73
Octopus
bimaculoides
were observed
to
test
whether
discrete
behaviors
could
be
grouped reliably
to
reflect dimensions
of
temperament. Frequencies
of
behaviors during Week
3
were subjected
to
principal-components
analysis (PCA), resulting
in 4
components (active
engagement,
arousal/readiness,
aggression,
and
avoidance/disinterest)
that explain
53% of the
variance.
Levels
of
temperamental
traits
were then evaluated
for 37
octopuses using composite
scores
at 3 time
points
across
the first 9
weeks
of
life. Profile analysis revealed significant change
for the
testing group
as a
whole
in
trait expression levels
from
Week
3 to
Week
6.
Results also suggest
a
significant
effect
of
relatedness
on
developing temperamental profiles
of
octopuses. Discussion focuses
on how
results apply
to the
life
history
of O.
bimaculoides
and
what temperament
can
reveal
about adaptive individuality
in a
protostome.
Individual differences
in the
behavioral tendencies
of
young
organisms
can be
described
as
differential expression
of
tempera-
mental traits.
By
definition, these traits
are
behavioral styles
(as
opposed
to
discrete
acts) that appear early
in
life,
are
partly based
on
innate
biological
processes,
and are the
precursor
to the
adult
personality
(Rothbart,
Ahadi,
&
Evans,
2000).
Temperamental
features
that
are
present
from
birth
are
shaped
and
modified
throughout early development
by
interaction
with
the
individual's
environment
(Braungart,
Plomin,
DeFries,
&
Fulker, 1992; Clarke,
1993). Thus, initial temperamental tendencies have repercussions
for
both early survival
and
later adult behavioral expression
(Clarke
&
Boinski,
1995).
Although there
is
debate
as to the
exact
proximate causes
of
these differences
in
humans
(McGue
&
Bou-
chard, 1998;
Plomin,
2000;
Turkheimer, 1998), there
is
general
David
L.
Sinn,
Department
of
Biology, Portland State University;
Nancy
A.
Penin,
Department
of
Psychology
and
Systems Science Doctor-
ate
Program,
Portland State University; Jennifer
A.
Mather, Department
of
Psychology, University
of
Lethbridge, Lethbridge, Alberta, Canada;
Ro-
land
C.
Anderson,
The
Seattle
Aquarium,
Seattle, Washington.
We are
grateful
to
Samuel Gosling
and
Jean
Boal
for
helpful
discussions
on
this project.
Peter
Edmunds
and
Rebecca
Habeeb provided constructive
criticism
on
early drafts
of
this article. David
L.
Sinn thanks Leonard
Simpson
for
laboratory
space
and
financial
support,
Carrie
Furrer
for
assistance
with
statistics,
Marwan
Adjaj
for
help with data collection,
Shane
Hockett
and
Judith
Briner
for
assistance
with octopus maintenance,
Chuck
Winkler
for
collection
of
octopuses,
and the
American
Malacologi-
cal
Society
and the
Hawaiian
Malacological
Society
for
partial
funding
of
this project.
This
work
was
conducted
by
David
L.
Sinn
as
partial
fulfill-
ment
of a
master's
degree
in
biology
at
Portland State University.
Correspondence concerning this article should
be
addressed
to
David
L.
Sinn,
who is now at the
University
of
Tasmania, School
of
Aquaculture,
Locked
Bag
1-370,
Launceston,
Tasmania
7250,
Australia. Electronic mail
may
be
sent
to
dsinn@utas.edu.au.
agreement
that
the
temperamental profile
of an
organism
is the
result
of
biological
factors (primarily
genes),
developmental con-
text,
and the
interaction
(or
covariance)
between these
two
(Gold-
smith
et
al.,
1987;
Kagan
&
Snidman,
1991; McGue
&
Bouchard,
1998).
Although
temperament
as an
explanation
for
human behavioral
tendencies
has
been studied
for at
least
2,500
years (for
a
review,
see
Merenda,
1999), comparative animal studies within
the
past
25
years have begun
to map the
occurrence
of
similarly occurring
psychological traits across
a
wider range
of
vertebrate
taxa
(pri-
mates:
Stevenson-Hinde,
Stillwell-Bames,
&
Zunz,
1980a,
1980b;
Suomi,
Novak,
&
Well,
1996;
fishes:
Francis,
1990; Wilson,
Coleman,
Clark,
&
Biederman,
1993;
pigs:
Forkman,
Furuhaug,
&
Jensen,
1995; hyenas: Gosling, 1998;
for a
more comprehensive
review,
see
Gosling, 2001). Such comparative studies have
be-
come more common
as
researchers have begun
to
realize
the
merit
of
animal models
for
understanding
ecological
and
evolutionary
(i.e.,
selective
adaptations) aspects
of
temperament
and
personality
(Depue, 1995; Suomi,
1987).
The
results obtained
from
these
wider
ranging comparative studies
can
also give
us
valuable ref-
erence points
for
evaluating
the
development
of
cognitive
pro-
cesses
for
many taxa (including humans) along evolutionary time
scales (Gosling, 2001).
How
is
the
early individuality
of
organisms, which
is
expressed
through
their temperamental
tendencies,
important
in
understand-
ing
nonhuman
animal behavior? Unique expression levels
of
tem-
peramental
traits
can
provide behavioral material
for
natural
se-
lection (Clarke
&
Boinski, 1995)
by
conferring differential
goodness-of-fit
profiles
to
young organisms
in
reference
to
envi-
ronmental
conditions
that
they encounter
(Talwar,
Nitz,
Lemer,
&
Lerner,
1991). Variability
in
early temperamental traits among
individuals
may
also ensure
a
population's survival
in a
highly
variable,
unpredictable,
or
changing environment
(Katano,
1987;
Slater, 1981). Under
highly
unpredictable
and
fluctuating
condi-
351
352
SINN,
PERRIN,
MATHER,
AND
ANDERSON
tions,
genetic variability around behavioral traits would
be
con-
served through stabilizing
selection
(Plomin,
1981),
and
parents
could
"hedge
their
bets"
by
producing multiple
offspring
with
differing
levels
of
temperamental trait expression.
Maynard-Smith
and
Harper
(1988)
have illustrated three models
in
which stabiliz-
ing
selection could maintain
the
phenotypic
variability seen
in
aggressive traits, using songbirds
as
examples. Organism individ-
uality
could also
be
important
in
predator-prey interactions
be-
cause
an
individualized prey item would continually provide
a
protean search image
for
predators
(Sterrer,
1992). Instead
of
viewing
individual behavioral variation
in
their subjects
as
evolu-
tionary
"noise,"
researchers have
now
begun
to
examine
the
evo-
lutionary
and
ecological
implications
of
consistent individual dif-
ferences
in
behavioral styles
(Armitage,
1986;
Reale,
Gallant,
LeBlanc,
&
Festa-Bianchet, 2000; Wilson
et
al.,
1993).
Methods
for
Studying
Temperamental
Trait
Development
There
are a
number
of
methods that
can be
used
to
test
for
temperamental trait features
in
both human
and
nonhuman
ani-
mals.
One
method
is
purely exploratory
and
uses observational
data, preferably from several tests,
to
derive
a
number
of
temper-
ament dimensions that
are
thought
to
reflect underlying
processes
affecting
discrete
behaviors (e.g.,
Cattell
&
Peterson, 1959;
Stevenson-Hinde
et
al.,
1980a,
1980b;
Thomas
&
Chess, 1977).
It
is
worth noting that
an
exploratory method using three
different
behavioral
tests
was the
method chosen here. Another method
is
similar
to the
first
in
that
it is
exploratory
but
uses
human
response
data
acquired through questionnaires
to
describe subjects' behav-
iors
(e.g.,
Gosling, 1998). Questions
are
normally selected
by the
researcher
to
"pick
out" certain temperament dimensions
and
gen-
erally
are
answered
by
more
than
one
observer.
Yet a
third method
for
description
of
temperamental trait
features
is to
hypothesize
that
a
particular type
of
temperament already exists
for the
subject
organism
(e.g.,
Fox &
Henderson, 1999;
Maestripieri,
2000; Wil-
son
et
al.,
1993).
Behavioral tests
are
then performed
to
assess
whether
a
priori classifications
of
"type"
are
predictive
of
later
behavioral outcomes.
For
example, Wilson
et al.
(1993) classified
sunfishes
as
either
shy or
bold
on the
basis
of
individual tendencies
to
approach
a
novel stimulus
(an
unbaked
trap). Additional tests
then
were performed
to
assess whether these classifications could
be
used
to
predict behavioral responses
in
different
scenarios.
If
one of the first two
exploratory methods
is
chosen, then
the
behavioral dimensions that arise
from
their analyses should
be
assessed
for
reliability (the third method, although making theo-
retical assumptions regarding presence
or
absence
of
traits,
is
itself
a
test
of the
reliability
of
predictions).
By
reliability,
we
mean here
that temperamental
traits
should
not be
ephemeral
in
nature
but
should
be
strong indicators
of
underlying psychological processes.
Because
an
exploratory model
was
used here,
we
initially
asked
two
questions:
Do
octopuses display temperamental traits?
If so,
are the
trait dimensions reliable assessments
of
enduring
behav-
ioral styles?
A
common feature
of
human
temperament theory
has
been
the
emphasis
on the
continuity
of
temperamental characteristics
throughout
an
individual's life. Traits falling under
the
rubric
of
temperament
are
expected
to be
more stable
or
continuous relative
to
other personality variables
(A.
Buss, 1989; Strelau, 1989).
Relative
is
stressed
here because change
is an
essential
feature
of
development,
and
most
researchers
report some degree
of
both
stability
and
change
in
their results
(e.g.,
Carnicero,
Perez-Lopez,
Salinas,
&
Martinez-Fuentes,
2000;
Stevenson-Hinde
et
al.,
1980a,
1980b;
Suomi
et
al.,
1996).
Indeed, developmental studies
of
temperament
present
a
paradox
because
change
is an
essential
aspect
of any
developmental
function,
and yet
some element
of
stability
or
continuity
is
also
necessary
for the
maintenance
of
individual
distinctiveness
(Sackett,
Sameroff,
Cairns,
&
Suomi,
1981). Thus, some researchers have interpreted
findings of low
stability
as
indicating problems
in the
concept
or
measurement
of
temperament
(Hubert,
Wachs,
Peters-Martin,
&
Candour,
1982),
whereas
others have suggested that lack
of
stability reflects devel-
opmental
changes
in the
expression
of
temperament itself (Riese,
1987). Simply put, consistency
in
developmental features
or at-
tributes
is not
easily definable
and is
highly complex. Caspi
(1998)
described
five
different
kinds
of
stability that
can be
classified
within
two
major
types
of
continuity:
homotypic
and
heterotypic
(Kagan,
1969).
The first
type
of
continuity,
homotypic,
refers
to
the
stability
of
similar behaviors
or
phenotypic attributes over
time.
For
example,
the
same level
of
shy—bold
tendencies
to
novelty
within
individuals
or a
group observed
in
infancy
is
expressed
in
later childhood.
In
contrast, heterotypic continuity
refers
to the
continuity
of a
genotype
or
underlying trait despite
changes
in its
phenotypic expression. Therefore, heterotypic con-
tinuity
allows
for
differing
levels
of
response along temperamental
trait
dimensions even though
the
underlying dimension itself
in-
fluencing
a
response does
not
change. Considering these features
of
temperament trait development then,
we
asked
a
third question
regarding octopus temperament:
If
octopuses
do
indeed display
temperamental
features throughout their early
life,
how
might
we
characterize temperamental trait continuity
and/or
change through
time?
Using Octopuses
as
Models
for
Psychological
Trait Testing
The
divergence
of the
protostome
and
deuterostome lineages
(Wray,
Levinton,
&
Shapiro, 1996)
can
provide
a
useful
reference
point
for
evolutionary
psychologists
when making comparative
assessments. Understanding
the
individuality
of
representative
taxa
from
both
major
lineages
can
provide
useful
information
regarding convergent
and
divergent
processes
associated
with
the
evolutionary
development
of
temperamental traits. Currently, there
is
a
paucity
of
knowledge
of
individuality
in
invertebrates com-
pared
with that
in the
vertebrate
deuterostomes,
despite
a
relatively
larger number
of
species
and a
high
degree
of
behavioral diversity
in
the
former
(Barnes, 1987). Several arthropod studies
(bees:
Pflumm
&
Wilhelm,
1982;
ants:
Bonavita-Cougourdan
&
Morel,
1988; Retana
&
Cerda, 1991)
and one
octopus study (Mather
&
Anderson,
1993) have begun
to
delineate possible trait dimensions
of
individuality
in
invertebrates. Mather
and
Anderson
(1993)
provided
the first
psychological trait testing
in the
octopuses
and
outlined
three broad personality traits
in
adult
O.
rubescens:
ac-
tivity,
reactivity,
and
avoidance.
In
their study, Mather
and
Ander-
son
observed sexually mature
O.
rubescens
of
unknown
age in
behavioral tests
for 2
weeks
and
obtained results using
a
single
analysis
of
summed frequencies
of
behaviors.
In the
current study,
we
attempted
to
further
understand
the
individuality
of
these
invertebrates
by
documenting temperamental features
in an
octo-
TEMPERAMENTAL
TRAITS
IN
OCTOPUSES
353
pus.
We
intentionally
chose
to use the
construct
descriptor
tem-
perament
in our
study
as
opposed
to
personality.
Although
there
is
confusion
over
this
distinction
in
both
human
and
nonhuman
animal
literature
(Gosling,
2001),
our use of the
term
temperament
was
based
on our
observations
of
octopus
behaviors
that
occurred
early
in
their
life.
Personality
measures,
on the
other
hand,
are
normally
used
to
describe
later
adult
behavioral
styles,
which
have
been
modified
through
learning
(Mather
&
Anderson,
1993).
O.
bimaculoides
(Pickford
&
McConnaughy,
1949)
was
chosen
as
our
subject
organism
because
it can be
reared
in the
laboratory
from
birth,
with
the
young
immediately
adopting
a
benthic
lifestyle
after
hatching
(Forsythe
&
Hanlon,
1988b).
This
is in
comparison
with
small-egged
Octopus
species
(spp.),
which
are
significantly
more
difficult
to
rear
from
birth
in
laboratory
settings
because
of
their
planktonic
young
(Boyle,
1987).
Currently,
most
observa-
tions
of
young
O.
bimaculoides
have
come
from
laboratory
studies
(Forsythe
&
Hanlon,
1988a;
Sinn,
2000);
only
one
study
has
reported
limited
in
situ
observations
(Lang,
1997).
Eggs
are
nor-
mally
3 to 4 mm in
length,
teardrop-shaped,
and
laid
in
groups
of
150 to 500
that
are
then
brooded
by the
females
until
hatching
(Forsythe
&
Hanlon,
1988a;
Sinn,
2000).
At
hatching,
the
young
octopuses
are
thought
to
disperse
by
crawling
or
swimming
(Lang,
1997).
O.
bimaculoides
are
found from
central
California
to mid
Baja,
Mexico,
where
they
inhabit
rocky
reefs,
kelp
forests,
and
mudflats
(Lang,
1997).
Little
is
known
concerning
the
juvenile
life
history
of
this
or any
other
Octopus spp.
because
of the
difficulty
of
observing
young
in the
wild
(for
an
exception,
however,
see
Ambrose,
1988).
In
summary,
the
aims
of
this
study
were
as
follows:
(a) to
document
whether
early
behaviors
in a
protostome,
O.
bimacu-
loides,
could
be
factored
reliably
into
categories
reflecting
dimen-
sions
of
temperament,
(b) to
assess
whether
early
temperamental
development
in an
invertebrate
could
be
described
according
to
vertebrate
temperamental
trait
theory
(i.e.,
developmental
continu-
ity
and
change),
and (c) to
assess
the
possible
effects
of
relatedness
on
octopus
temperamental
trait
development.
The
last
aim of our
study
is
consistent
with
temperamental
trait
theory
that
considers
temperamental
trait
levels
to be
influenced
by
biological,
or
innate,
components
(Goldsmith
et
al.,
1987).
We end by
discussing
pos-
sible
correlates
of
temperament
with
regard
to the
life
history
of O.
bimaculoides
and the
relevance
of
using
protostome
representa-
tives
as
comparative
data
points
for
vertebrate
psychological
processes.
Method
Subjects
Female
Octopus
bimaculoides
(n
=
8)
with
eggs were collected
in the
spring
of
1999
from the
wild
in the
Long Beach,
California,
area,
shipped
to
Portland, Oregon,
and
maintained separately
in
holding tubs
(76
L; 1.0 X 0.3 X 0.3
m)
until eggs were hatched. Holding tubs were
maintained
as
part
of a
single in-line closed system (3,028.33
L) and
were
filled
with artificial seawater (Instant Ocean brand mixed
with
deionized
water)
that
was
maintained
at 18
°C
with
salinities varying between
34
and
36
ppt.
An
in-line chiller tank
with
a
2,839
L/hr
pump
was
used
to
maintain
temperature
and
water circulation.
Individual
holding tubs con-
tained
crushed oyster shell substrate, plastic
sea
grass blades,
and
other
shelter (i.e., clay pots, small
pieces
of PVC
pipe,
and
rocks).
The
system
received
overhead fluorescent
lighting
in
addition
to
natural, indirect
sunlight
from
large adjacent windows.
The
day-night
cycle
of the fluores-
cent
lights
was
kept approximately
the
same
as the day
length
of
Portland,
OR,
during
all
phases
of
experimentation.
Because
females were
wild-
caught,
the
fathers)
of
each
individual
egg/juvenile
octopus
was
unknown.
Multiple
male
O.
bimaculoides
are
known
to
mate with
the
same female
in
the
lab
(Forsythe
&
Hanlon, 1988b),
and it is
possible
that more
than
two
parental
genotypes contributed
to the
genotype
of
individual octopuses
in
this
study.
Because
of
their small size
(< 6 mm
mantle length)
and
high mortality,
juvenile
octopuses were
not
tested
until
14
days
of
age.
At 14
days
post-hatching,
octopuses were removed
from the
holding tubs, weighed
and
measured,
and
housed separately
in
testing containers, thus allowing
for
identification
by
individual
and by
brood.
The
mean
wet
weight
and
mantle
length
for
octopuses
(n = 62) at 14
days
old was 1.1 g
(SD
=
0.45)
and
6.5 mm (SD =
0.90), respectively. Once placed individually into
testing
containers, octopuses were
not
moved
for the
remainder
of the
study.
Testing
containers were brown
plastic
cylindrical plant
potholders
approximately
10 cm in
diameter
and 14 cm
deep. Containers were opaque
on
all
sides
and had
opaque lids, which were
removed to
permit
access
to
octopuses
during
testing.
The
last
12 mm of a
polystyrene
13-
X
100-mm
culture
tube
was
provided
as a den for the
hatchlings,
with
the top
third
of
the
tube
clear
to aid
observation
and the rest
painted
with
black
epoxy
paint.
Test containers were
floated in
662-L water tables separate
from the
holding
tubs. Water depth
in
testing containers
was
maintained
at 10 cm,
and
containers were
continuously
illuminated
with
indirect
red
light
(25
W)
to
allow
for
accurate
visual
observation during nighttime hours.
Before
octopuses were
14
days old, they were
fed a
variety
of
foods
including
littorinid
snails,
mysid
shrimp, limpets,
amphipods,
mole crabs,
and
appropriately sized (crab carapace
width
less
than
one
half
the
mantle
length
of the
octopus) live shore crabs.
After
Day 14,
octopuses were
not
fed
on
testing days except during
the
last phase
of
experimentation;
on
days
when
testing
did not
occur, octopuses were presented with
food
items (live
shore
crabs, mysid shrimp,
and
littorinid snails)
ad
libitum. Human inter-
action
with
experimental octopuses
was
minimized during testing. How-
ever, interaction
with
test octopuses outside
of
testing situations
was
inevitable
because containers were cleaned
and
octopuses were
fed
during
nontesting
periods.
Additional
developmental histories
of
octopuses before
Day 14 and
during
testing were
as
follows.
After
hatching, some nontested octopuses
were
removed from
holding
tubs
to
minimize
densities
of
octopuses
in
these
tubs.
This resulted
in the
numbers
of
octopuses
in the
holding tubs
before
Day 14 to
range
from
27 to 45,
which
are
most likely exaggerated
densities compared
with
natural
conditions. Holding
tub
availability, hatch-
ling
die-off,
and the
need
for
identifiable testing animals were
the
neces-
sary
parameters
in the
decision
to
maintain these densities
in
holding tubs.
Previous
to 14
days
of
life, octopuses interacted
with
siblings only.
After
Day
14,
all
tested octopuses were
isolated,
as
mentioned previously.
Previous
to Day 14,
holding
tubs were checked
at
least once daily
to
remove
waste material,
and no
major
differences
in
degree
or
type
of
interactions
were noted between broods. Octopuses were
not
weighed
or
measured
during testing
to
minimize contact.
At the end of the
testing
period, octopuses were maintained
until
their death, which occurred before
sexual
maturity
at
approximately
5
months
of
age. Preliminary dissections
of
young
animals were
not
conclusive
regarding
sex,
so sex is not
reported
here.
Test
Procedure
Octopuses were observed
in a
three-part
testing
series
on
each testing
day.
Tests
began
within
30
min
of
sunset,
because
preliminary observations
indicated
octopuses were crepuscular
in
activity.
All
direct observations
of
octopus
behavior were performed
by one
person (David
L.
Sinn). Objective
recordings
of
behavior
from
more than
one
observer
are
ideal
(Feaver,
Mendl,
&
Bateson,
1986),
but
single
observer
results
are
common
in
354
SINN,
PERRIN,
MATHER,
AND
ANDERSON
studies
of
this type
(e.g.,
wolves,
MacDonald,
1983; octopuses, Mather
&
Anderson, 1993;
fishes,
Budaev, 1997; primates,
Maestripieri,
2000).
Observations were recorded
by
noting
the
occurrence
of a
behavior
verbally
into
a
microcassette
recorder.
At the end of
each testing
night,
all
results were transcribed
from
this
voice
tape into
a
computer. Although
octopuses
across
known broods were tracked individually throughout
the
study
period,
analyses
were
not run
until after
all
data were collected.
Because
of
this, results
of
types
of
temperament dimensions
as
well
as the
discrete
behaviors that would consequently load
on
each dimension were
unknown
at the time
observations were made. This
was
done
to
minimize
the
possibility that
an
unconscious bias
on the
part
of the
observer would
influence
results. Testing methods were similar
to
those used
by
Mather
and
Anderson
(1993)
and
were designed
to
represent naturalistic circum-
stances.
The first
test,
termed alert, consisted
of the
experimenter
lifting
the
opaque
lid to the
testing compartment
and
leaning over
the top of the
container. Octopus behaviors were recorded
at the
start
of
visual contact
for
30
s. The
second
test,
termed threat, occurred directly
after
alert,
and
consisted
of the
experimenter touching
the
octopus with
a
4-mm
diameter
test-tube brush. Behaviors were recorded
for 30 s
beginning with
the
touch
of the
test-tube brush.
The
third
test,
feeding,
took place
30 to 60
min
after
the
threat
test.
Feeding
was a
live
food
presentation (live
shore
crabs),
and
behaviors
in
this last test were recorded
for 10 min or
until
capture
of the
crab. Behaviors
in
feeding
tests
were recorded beginning with crab place-
ment
in the
testing container.
Octopuses
received the
three-part
testing series twice weekly during
their
3rd
(Days
16 and
19),
6th
(Days
37 and
40),
and 9th
(Days
58 and 61)
week
of
life.
A
total
of 73 O.
bimaculoides
were tested during
the 3rd
week
of
life,
37
were tested during
the 6th
week
of
life,
and 37
were tested
during
the 9th
week
of
life.
The
attrition
of
octopuses
was due to
die-off
or
escape
of
those that then could
not be
positively identified. Data
from
the two
testing periods
in a
given week were combined
to
create Week
3,
Week
6,
and
Week
9 frequency
scores
for
each
behavior.
A
given behavior
was
recorded
as
occurring more than once within
a
test
only
if
there
was at
least
a
5-s
break between instances
of mat
behavior.
This
5-s
rule
was
imposed
before
experimentation.
Data
Analyses:
Aim 1
PCA
and
reliability
and
stability measurements
on PCA
data were used
to
address
the
first
aim of our
study:
Can
early behaviors
in an
octopus
be
reliably
factored
into categories reflecting dimensions
of
temperament?
PCA.
Results obtained during
the 3rd
week
of
testing only were used
to
define
possible
temperamental
trait dimensions. Temperamental traits
normally
are
used
to
describe human infants
less
than
2
years
old
(Roth-
bart,
1986; Thomas
&
Chess, 1977),
and 1
week
in an
octopus's
lifespan
(assuming
a
lifespan
of
18
months, likely
for O.
bimaculoides;
Forsythe
&
Hanlon,
1988a)
would correspond roughly
to 1
year
in a
human (assuming
a
72-year lifespan).
Furthermore,
within
this short
lifespan,
octopus devel-
opment
(physical, physiological,
and
behavioral)
is
rapid.
The 3rd
week
was
our
earliest
time
point,
and a
week
was the time
span that
was
chosen
to
best represent
the
rapid behavioral development
in our
octopuses while
not
obscuring developmental
processes.
We
observed
19
behaviors during
the 3rd
week
of
testing,
and
15
were
chosen
for
analysis (Table
1)
because
they
met the
criteria
of
contributing
at
least
5% of the
total behavioral
repertoire
observed during Week
3
testing.
The
summed
frequencies of
these
15
behaviors were then subjected
to
exploratory factor analysis (PCA model;
Tabachnick
&
Rdell,
1996)
with
direct
oblimin
rotation using SPSS
for
Windows
(1997;
n =
73).
Oblique
rotation
was
chosen
on a
priori grounds that temperamental
dimensions
were
most likely
related, and not
orthogonal
(or
uncorrelated)
to
one
another. Orthogonal rotation
was
performed
as
well,
and the
results
matched
those obtained
by
oblique rotation. Because this
was an
explor-
Table
1
Definitions
and
Frequencies
of 15
Behaviors
Recorded
for
Octopus
bimaculoides
(n = 73) in
Tests
During
Their
First
9
Weeks
of
Life
Behavior
M
SD
Operational definition
No reaction
Shrink
Crawl
Head movement
Respiratory change
Touch stimulus*
Pupil
change
Papillae change
Posture change
Color change
Alert
posture
Jet or
swim
Arm
probe
Grab
stimulus"
Pull
stimulus'"
1.22
1.55
2.79
0.66
0.53
0.84
0.89
0.25
1.34
2.38
0.18
0.37
0.59
0.79
0.26
1.00
0.85
2.97
0.90
1.04
1.07
0.89
0.79
1.52
2.15
0.42
0.89
0.88
0.71
0.53
No
visible reaction
was
noted.
A
specific type
of
posture change:
The
arms
and
lower part
of the
body
remained
stationary while
the
head
and
mantle increased
the
distance
from
the
stimulus.
Octopus
moved along
the
bottom
or
side
of the
testing compartment using
the
arms
as its
means
of
locomotion.
Octopus made vertical
(up and
down), horizontal (side
to
side),
or
vertical/horizontal
(combination)
movement
of the
head only.
Octopus
at
least doubled
its
ventilation rate
and
then maintained this elevated rate
for at
least
a
5-s
period.
Octopus
initiated contact
with
the
stimulus
with
one arm
only.
Octopus's pupils were enlarged.
Octopus raised skin surface
in
papillae.
Octopus's head
and
body changed position while maintaining
a fixed
point
in
space;
this behavior
was
not
directional
in
regard
to the
stimulus.
Octopus's overall skin color pattern
was
changed.
Octopus raised eyes
and the
mantle
was
held
at a 45°
angle downward
from
vertical.
The
body
was
held
fixed,
and the
arms were tucked close
to the
body
and
used
to
raise
the
head/eyes
further
in the
vertical plane.
Octopus moved
by jet
propulsion,
with
no
contact
to the
bottom
or
sides
of the
container.
Octopus
moved
one or
more arms laterally, maintaining contact with
the
container,
but not
touching
any
stimuli.
Octopus
used more than
one arm to
contact
the
stimulus.
Octopus,
after
touching
or
grabbing, maintained hold
with
suckers
and
attempted
to
shorten
arm(s).
Note.
Means
and
standard deviations
are
given
for
behaviors observed
in
Week
3
testing only because
it is
these data that contributed
to the
principal-components analysis
and
resulting temperamental dimensions. Operational definitions were held constant throughout
the
testing
period.
All
behaviors were displayed
at all
three
time
points
and in all of the
three test situations
(alert,
threat,
or
feeding)
unless
otherwise noted.
*
Occurred only during threat
and
feeding.
b
Occurred
in
threat test only.
TEMPERAMENTAL TRAITS
IN
OCTOPUSES
355
atory
factor analysis,
no a
priori assumptions were made regarding
the
number
of
components
to
retain.
A
decision regarding
the
number
of
components
was
based
on the
following criteria:
(a) a
scree
test
(Cattell,
1966),
(b)
eigenvalues
> 1
rule
(Kim
&
Mueller, 1978),
(c) a
parallel
analysis
(Montanelli
&
Humphreys, 1976),
and (d)
interpretability
of the
factors
themselves (Zwick
&
Velicer, 1986).
For
interpretation, behaviors
with
loadings
of at
least
± .4
were considered
to
contribute
to the
meaning
of
a
component. Acceptable
levels
of
loadings
can be as low as ± .32 in
some
cases
(Tabachnick
&
Fidel],
1996).
Reliability
and
stability
measurements.
A
number
of
methods were
used
to
assess
the
reliability
and
stability
of PCA
results obtained
from
Week
3
data.
First,
interfactor
correlations obtained during oblique rotation
were used
to
assess
divergent stability
of
Week
3
results (Campbell
&
Fiske,
1959).
Divergent stability
is a
measure
of how
well
the
components
obtained
at
Week
3
were predictors
of
independent
processes,
and not two
measurements
of
essentially
the
same underlying temperament trait. Sec-
ond,
the
similarity
of
pattern matrices obtained through
differing
methods
of
rotation (oblique
vs.
orthogonal)
was
used
to
assess
convergent validity
of
the
factors
at
Week
3
(e.g.,
Budaev, 1997; King
&
Figueredo,
1997).
Convergent validity measures
a
component's strength
to
predict itself using
different
methods
of
prediction. Third,
communalities
of
variables
and
magnitude
of
variable loadings obtained during
PCA on
Week
3
data were
used
to
assess
the
strength
of the
relationships
of
variables
in the
Week
3
pattern
matrix.
An
additional analysis
was
used
to
address
the
longitudinal
reliability of
measurements obtained using Week
3
data.
1
A
comparison between
the
Week
3
pattern matrix
(n = 73) and
pattern matrices obtained through
PCA
of
Week
6 (n = 37) and
Week
9
data
(n = 37) was
used
to
assess
whether
the
patterns
of
behaviors
on
components obtained
at
Week
3
were consis-
tently
found
among behaviors across
the
entire study period.
Although
confirmatory
factor analysis
is the
preferred analytical tool
to
compare
pattern
matrices between
time
points,
the
smaller sample
sizes
at
Weeks
6
and
9
prevented
the use of
this analysis (our model
would
have resulted
in
a
15-variabIe,
37-subject
model
at
each
time
point;
confirmatory
factor
analysis
requires
a
variable-to-subject
ratio minimum
of
1:10
to
1:15).
Instead, component loadings were
first
normalized using Fisher's
r-to-z
ratio
(Snedecor
&
Cochran,
1980)
and
entered into
a new 15 x 12
(Variables
X
Components
at
each week) data set. Pearson correlations
(two-tailed) were then computed
(«
= 15)
using
the
normalized component
loadings
across
the
four
components
and
three
time
points (Weeks
3,6,
and
9).
Convergent stability
of the
components
throughout
the
study period
was
assessed
by
high correlations between
the
same dimensions
at
different
time
points;
divergent stability
was
assessed
through
relatively lower
correlations between
different
dimensions
at
different
time
points.
of
which loaded positively,
and
Component
4 had two
positively loading
behaviors
and one
negative. Composite
scores
were created
for
Weeks
3,
6, and 9. For
example,
to
create
a
composite score
for
Component
3 for an
octopus
at
Week
6
based
on the
Week
3
pattern matrix,
we
added (positive
loading)
the
observed
number
of times mat
individual octopus performed
grab
stimulus, pull stimulus,
and
posture change (behaviors that
loaded
highly
on
Component
3)
during that
week's
testing,
and we
subtracted
(negative
loading)
the
number
of times
that
individual
octopus
was ob-
served
to jet
during that week's testing. Thus, each octopus
had a set of 12
scores
(4
temperament dimension
scores
at
each
of
three
time
points).
The
composite
scores
for the
four
temperament dimensions
at
each
of
the
three
time
points were subjected
to
profile analysis,
a
type
of repeated-
measures
multivariate
analysis
of
variance,
to
assess
parallelism (Tabach-
nick
&
Fidell,
1996).
Parallelism tests
the
hypothesis that there
is no
difference
in the
pattern
of
expression
of the
four
temperament dimensions
at
each
different
time
point.
This
analysis
was
considered
the
developmen-
tal
function
analysis because
it
described
the
normative pattern (Chalmers,
1987;
Wohlwill,
1973)
of all the
octopuses
in the
study.
Of
particular
interest
in
these tests
is the
interaction term between
time and
temperament
dimension
on
composite
score.
A
significant
interaction means that
the
pattern
of the
four
temperament dimensions
relative to one
another changes
over
time. A
significant
main
effect
in
this analysis
of
either
time or
temperament
dimension indicates that
the
composite
scores
for the
entire
sample
changed across
time or
dimensions.
Simple
and
deviation contrasts
were used
to
examine pairings
of time
points
(e.g.,
Week
3 vs.
Week
6,
etc.)
and
temperament dimensions (Component
1 at
Week
3 vs.
Compo-
nent
1 at
Week
6,
etc.)
within
any
overall significant main effects,
and
graphical
representations were used
to
help interpret significant
interactions.
Canonical
correlation.
The
second method used
to
address
the
char-
acterization
of
temperamental trait development
across
time was
canonical
correlation
(Tabachnick
&
Fidell,
1996).
Whereas profile analysis analyzed
expression levels
of
traits
for the
study group
as a
whole, canonical
correlation
was
used
to
understand
the
expression
of
traits
across
time at
the
level
of the
individual octopus. This analysis
is a
multivariate regres-
sion
analysis
and
asks whether
an
octopus's
four
composite
scores
at a
certain
time
point
are
predictive
of the
same
scores
at a
later
time
point.
Canonical
correlations were computed only
for
those
octopuses
that
were
present
throughout
the
testing period
(n =
37). Correlations were
run
between
the
four
composite
scores
at
Week
3 and the
four
composite
scores
at
Week
6 and
between
the
four
composite
scores
at
Week
3 and the
four
composite scores
at
Week
9.
Data
Analyses:
Aim 2
Profile
analysis
and
canonical correlation were used
to
address
the
second
aim of our
study:
Can
temperamental trait development
in a
protostome
be
characterized according
to
current developmental themes
normally
used
in
vertebrate temperamental trait theory?
Profile
analysis.
This
was the
first
analysis used
to
characterize
the
stability
of
expression
levels
of
octopus temperamental traits through
time.
Composite
scores
were computed
for
each individual octopus that
had
data
available
for the
entire testing period
(n = 37) on the
dimensions that arose
from
the
Week
3 PCA
(e.g.,
Stevenson-Hinde
et
al.,
1980b).
These
scores
were created
for
each temperament dimension
by
summing
the
behaviors
that loaded most heavily
(> ± .4) on
each component.
Variables
that
met
this
criterion
and had a
negative factor loading were given
a
weight
of
—
1
in
the
composite score equation,
and
variables meeting
this
criterion
and
having
a
positive factor loading were assigned
a
weight
of + 1. For
Component
1,
five
behaviors satisfied this criterion
and
loaded positively.
Component
2
contained
five
behaviors
as
well,
with
three loading posi-
tively
and
two
negatively. Component
3
consisted
of
four
behaviors, three
Data
Analysis:
Aim 3
A
second profile analysis
was
conducted
to
address
the
third
aim of our
study:
Are
temperamental trait expression levels
of
octopuses influenced
by
their
relatedness
to one
another? Individual
octopuses
were grouped
by
brood,
and
three broods
(X, n =
11;
Y, n = 19; I, n = 5)
were used
to
assess differences
in
parallelism
of
composite
score
patterns
across
the
three broods
at the
three
different
time
points.
The
unequal number
of
broods
tested
reflects
testing container availability,
and not
densities
in
holding
tubs
or
total number
of
individuals present
in
each brood. Once
again,
simple
and
deviation
contrasts
were used
to
examine
the
pairings
of
broods,
time
points,
and
temperament dimensions within significant main
effects,
with graphical
representations to aid in
interpretation
of
significant
interactions.
1
We
gratefully
acknowledge
an
anonymous
reviewer for
this suggestion
for
longitudinal
analysis.