Plus ca change...: The New CPS Election Study Panel

Abstract

Between 1956 and 1960, the first long-term panel study of the American electorate was carried out at the University of Michigan. Among other findings from this original panel were sharp contrasts between the high individual-level stability of party identification and more labile individual preferences on major political issues of the day. Since 1960, several changes in the nature of the American electoral response have caught the attention of scholars, including an erosion of party loyalties on one hand and an increasing crystallization of issue attitudes on the other. Completion of a new panel segment, 1972-76, makes it possible to review the original 1956-60 findings in the light of these intervening changes. We discovered that the contrasts in individual-level continuity of party and issue positions remain nearly identical to those estimated for 1956-60. The theoretical significance of these counter-intuitive results is discussed.

Full text

Plus ca change...: The New CPS Election Study Panel
Author(s): Philip E. Converse and Gregory B. Markus
Source:
The American Political Science Review,
Vol. 73, No. 1 (Mar., 1979), pp. 32-49
Published by: American Political Science Association
Stable URL: http://www.jstor.org/stable/1954729
Accessed: 13/01/2009 23:42
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=apsa.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.
American Political Science Association is collaborating with JSTOR to digitize, preserve and extend access to
The American Political Science Review.
http://www.jstor.org

Plus
qa
change
.
.
.:
The New
CPS Election
Study Panel*
PHILIP E. CONVERSE
GREGORY B. MARKUS
University of Michigan
Between
1956
and
1960,
the
first long-term panel study
of
the
American electorate was carried
out at
the University of
Michigan. Among other
findings from this original
panel
were
sharp
contrasts between the high
individual-level stability
of party identification
and more labile
individual
preferences
on
major political issues of the
day. Since
1960,
several changes
in
the
nature
of
the
American
electoral response have
caught the attention of
scholars, including an
erosion
of party loyalties
on one
hand and an
increasing crystallization of
issue attitudes on the
other.
Completion of
a
new
panel segment,
1972-76, makes it possible to
review the original
1956-
60 findings
in
the light
of these intervening
changes. We discovered
that the contrasts
in
indiridual-level
continuity of
party and issue positions
remain nearly identical
to those estimated
for
1956-60. The theoretical
significance of these
counter-intuitive results is
discussed.
In some of
our
sister
sciences,
publication
of
frankly provisional
findings,
even
in
the most
polished
of
journals,
is
not
considered
ddclasse.
This is
particularly
true
where crucial
new data
bases
have
freshly
come
into
being.
Thus when
the first
rocks were returned from
the moon, a
haste
spate of
publications merely
recounted, in
the
languages
of
various
specialties,
what
the
rocks were like
in
simple, first-cut
terms.
To
be sure,
more
delicate
and
incisive
analyses
were
to be
supplied in
ensuing years.
However,
the
function
of such
early publication
for
the collective
process of
inquiry should be
transparent.
Even
the
first
crude
observations as
to
the
apparent age and structure of the
rocks,
or the
presence
of
water or
organic traces,
sufficed to
shoot
down
some
broad-gauge
astro-
physical
hunches as
to what
processes, on what
time
scale, had
actually molded the
moon's
surface.
And
correspondingly, these
primitive
observations drew
into
sharper focus
an
appro-
priate
research
agenda among
remaining tenable
hypotheses
for
the months
of
more probing
inquiry
which
lay
ahead. Thus
early publication
can
warn
the
community
in
advance
which
more
detailed
assays
of
the new
data
base are
worth,
and
not
worth, making.
Although
the
metaphor may
be
a little more
dramatic than current
political
science can
sustain,
it is
in
some such
spirit
that
this
article
is
offered. Analysts
of the sequence
of national
election studies
generated with
National Sci-
*The
research on which
this
report
is
based was
supported
by National
Science
Foundation Grant
SOC-7707537. We are
also grateful to
Jean Dotson,
Maria
Sanchez and
Peter Joftis for
their aid in data
preparation.
ence Foundation help by the Center for Politi-
cal Studies at
the
University of Michigan have
been frustrated
for
some
time at
the lack of
long-term panel linkages-the reinterviewing of
the same respondents-in successive replications
of the basic study design. The 1976 presidential
election marked the completion of the first
large-scale panel segment, stretching back to
1972, since
the
original four-year panel was
completed in 1960. These new panel data
constitute
a fresh
and
important
base
which
seems worth preliminary report.
Of
course the
study
of
electoral
change
has
scarcely
withered
away
in
the
interim
for
lack
of
full
longitudinal
studies. The
use of
less
expensive independent samples
at
biennial
elec-
tions has
greatly enriched
our
understanding
of
the
range
of
variation
in
American
electoral
processes since
the first electorate
panel
in
1960,
in
part because
the
intervening period
has
been
one
of
political change
of sufficient
magnitude
to
be plainly
visible without
the
extra sensor of a genuine panel design. How-
ever, it is exactly this marked change
that has
put an increasing premium on the advent of a
new round of panel data. Hence it will be
helpful background
if
we review, with a brevity
verging on oversimplification, some of the chief
findings from the original panel, as well as
revisions provoked by subsequent independent
samples on
into
the early 1970s.
The
Original
1956-60
Panel
and Later
Findings
Thanks
to
the
delinquency
of the
current
senior
author,
the
major analyses
of the
panel-
specific data
from
the
1956-58-60 reinterview
32

1979
The
New
CPS
Election
Study
Panel
33
sample were never
reported in a self-contained
monograph.
However,
some
of
the
more sur-
prising empirical
facts to emerge from those
analyses played
an
influential role
in
his essay
"The Nature of
Belief Systems
in
Mass Publics"
(Converse,
1964). Stripped
to its
essentials, one
brute fact which could
only
have arisen
from
long-term
longitudinal data was the discovery
that at the
individual level, the stability of
party identifications
in
the
1956-60 period
vastly
outstripped the stability of individual
positions on even
the most stable of the major
political issues of
the
period.
Such
stability
of
party identification had
been
suspected
for
some time
on
the
basis of
the
noteworthy
inertia
of
marginal distributions
for
the variable
in
independent cross-section
samples throughout
the
1950s.
In
fact, that
suspicion had been
central in
some of the major
theses of The
American Voter (1960). How-
ever, inertia
of marginals, or minimal net
change, is not
at all
incompatible with the
possibility of
high
rates of
gross change,
or
rapid individual-level
"turnover." Indeed, while
time series
data on
distributions
of
mass
opin-
ions
on major
political
issues
were
somewhat
more truncated because of the
flux
of salient
issues from
one
election
to the next,
where
repeated cross-section measurement did exist
for
political
issues,
the
marginals
were
often as
inert in
the 1950s
as
those
for party
identifica-
tion.
Therefore it was
particularly striking
to
discover
that
beneath
the
net
stability
of
party
identification
lay
a
very
marked
degree
of
gross,
or
individual-level
stability;
whereas
the inertia
or
net
stability
of
opinion
distributions
on
major political
issues concealed an
equally
surprising degree
of
individual-level
turnover
which
appeared
to
be almost
a
Brownian
motion. Let us
be
careful not to overstate
the
case. Party
identification was
not
perfectly
stable
for individuals
over the
1956-58-60
panel
observations,
in the
sense that
all
indivi-
duals
located themselves
in the
same
one of
seven
possible
categories
in
each
ensuing
mea-
surement. Substantial
numbers
changed
to an
adjacent
location
on the
continuum,
and
a
very
few appeared
to move long distances (5-7
"slots")
across
the
continuum.
Similarly,
it
was
not true that the individual-level
data
on
issue
positions
showed
no
stability
whatever.
For
one
thing,
there was
significant
variation
from issue
to
issue,
in
patterns
that seemed
decodable;
and
in
any event,
even the
issues with
the
greatest
individual-level turnover
showed
some
signifi-
cant
degree
of
continuity,
as indexed
by
the
correlation of each issue
with itself
over
time
departing
reliably
above zero.
Nonetheless, the contrasts
in stability be-
tween party and issue positions
remained abso-
lutely stark, and we would
find it hard to be
impressed with any theory
of the dynamics of
these political
attitudes which
failed
to
en-
compass such contrasts.
One implication
seemed to be that party
loyalties had a con-
siderable primacy
in
the
attitude systems in-
volved. In the 1950s, at least, issue
positions
were but
weakly aligned
with partisanship,
in
the minds of the voters, in
any event. Perceived
linkages between parties
and issues were, by
and large, vague and contradictory
as well. We
opined at the time (The
American Voter, pp.
179-83) that a major
cause of the confusion
might be that "actual" party
differentiation on
the major issues was at
low ebb in the mid-
fifties, a guess that later
proved to have some
merit. However,
in
the
modest degree that
partisanship
and
issue positions
did covary, the
contrasts in stability
from the panel data
suggested
that it was likely that these
issue
preferences
were more
often
brought
into
line
with
prior partisanship
than the reverse.
Once
again,
there is
no
need
to
exaggerate
the
case:
surely issue positions
could
affect partisanship
for
occasional individuals
at
any time,
and
at
some occasional junctures
might do so for
larger fractions of the
population. But
it
seemed
well-nigh impossible
to square
the
panel
contrasts in
party and
issue stability with
a
model
in
which the
preponderant
causal
flow
was,
in
any
intermediate
term,
from issue
position to partisanship,
at
least in
the late
1950s.
Cross-Section
Sample
Findings
from
the
1960s.
Although
no full-blown
panels
of the national
electorate were followed
in the 1960s, there
was
ample sign
of
major
trends
in these
matters
by
the
middle
years
of the decade. While
change
was
rapid
on
many fronts,
some of the
most
impressive
evidence involved
the
party-is-
sue nexus.
Burnham
(1968)
noted that
after a
lengthy
period in which the aggregate
distribution of
party
identification
had
remained
nearly
con-
stant,
the
first
signal
change began
in the
middle
1960s
with
a
rising proportion
of the
electorate refusing to report
identification with
either party, and insisting
on being classed
as
independents, thereby
swelling
the neutral mid-
dle
of the
party
identification
continuum. This
apparent
erosion
in
feelings
of
party loyalty
among
American
voters,
caught
at
an
incipient
stage by
the
first Burnham
accounts,
has since
proceeded majestically
in
the
same direction
for
close to
a
decade,
and
certainly
must rank
by

34
The American
Political Science
Review
Vol.
73
now as
change of
massive
magnitude. The
trend
has
nicely fit
Burnham's
broader
views as to
the
progressive
"decomposition"
of
the
traditional
party
system
in
the
United
States,
and has
become the
subject of a torrent of
literature
analyzing
details
of the
trend,
or
speculating
as
to its
implications for
partisan
realignment.
Where
issue
positions
of
the
public are
concerned,
Nie
(with
Anderson,
1974)
has
advanced
some
provocative
analyses
suggesting
that
soon after
1960, voter
policy
preferences
began
to
take
on
a
much firmer
"muscle
tone"
than was
characteristic of data from
the
1950s.
This
change
is
largely
indexed
by
a
marked
increase
in
the
cross-sectional
correlations
among issue
positions
in the
1960s, as
opposed
to
their feeble level
of
the
preceding
decade.
The reality of the
issue
trends charted
by
Nie
has
been
subject
to
somewhat
more
controversy
than
attends the
brute
empirical
facts
of
ero-
sion
in
party
loyalties,
since
some
(Bishop
et
al.,
1978;
Sullivan et
al.,
1978) have
suggested
that
the
changes
may be
more
an
artifact of
shifts in
question
format in
the
Michigan series
than
true
secular
change.
However this matter
may be
resolved,
the
Nie
thesis
has
strong
face
validity at
least,
since
increased
public
attention
to
policy
cleavages is
almost another
way
of
describing
the
onset of
political
turbulence
of
the
type
characterizing
American
politics
for
the
past
dozen years.
Moreover,
the
Nie
thesis
receives
supple-
mentary
support
from
changes
observed
in
linkages
between
partisanship,
issue
positions,
and
voting
decisions.
For
example,
Pomper
(1972)
pointed
out that
voter
perceptions
of
party
differences
on
policy
issues
showed
more
clarity
by
1964
than
had been true in
the
first
panel
period
of
the
late
1950s.
The
citizen
was
more
likely
to see
differences
between
the
parties
on
policy
matters, and
the
alignment
between
partisanship
and
issue
preferences
had
heightened as
well.
In the
most
exhaustive
recapitulation
of these
trends,
Nie,
Verba
and
Petrocik
(1
976)
have shown that
the
election-
by-election correlations
between
party
identifi-
cation and the
vote, as
compared with
issue-
vote
correlations,
changed
after
the
early
1960s:
the
apparent
weight
of
partisanship
in
the
voting
decision
had
undergone
significant
decline,
while
the
weight
of
policy
preferences
showed a
marked
advance.1
Again
the
magnitude
of
these
trends,
as well
as their
durability
into the
1970s,
remain
subject
to
scholarly
dispute
(see,
for
example,
lPresidential
voting
in
the
1976
election
seems
to
be
the
first
major
reversal
of
this
trend.
Margolis [ 1977]
on
Pomper). Similarly,
the
longer-range import of the differences
before
and
after the
early 1960s
is
subject
to
basic
disagreement. Some
see
these
changes
as secular
and permanent
or
even
accelerating
into
the
future. Others
see one or
the
other
political
period as the American norm, interrupted
by an
era of aberration. Still others see the
two
periods as local minima and maxima,
with a
putative normal lying somewhere between.
Nevertheless, some of the component
chang-
es are beyond dispute as empirical
facts,
and
taken
as a
gestalt,
they
have a
great
deal of
intuitive appeal.
The
parts
fit
together
admirab-
ly,
and
correspond nicely
with
impressions
most
of us
who lived
through
both
periods
would
have
harbored
in
any
event.
It
is
against
this
backdrop
that the advent
of
the new
1972-74-76 panel
version
of
the
Michigan
election series is
particularly welcome.
After
all,
the trends
catalogued
since
1960
have
all
been developed without the longitudinal
tracing
of
the
same
individuals over time.
This
is not to say the trends are spurious:
the
movements of marginal distributions and
drifts
in
correlations coefficients estimated
within
each slice
of
time
would
certainly
have
been
duplicated (apart
from
sampling
error and
variation due to population replacement)
if
panel studies
had
been mounted throughout
the
era since 1960.
However, one precious
class
of
informa-
tion-the continuity or individual-level
stability
of
the key variables-cannot be squeezed
from
the
independent-replication design.
In fact, it
cannot
even
be deduced from other attendant
slice-of-time facts
with
much
reliability,
as we
shall
shortly
see.
And given
the intervening
work on
electoral
change,
some
of
the first-cut
expectations as to how the new panel
data
might
look
are
extremely straightforward.
Ap-
parently party
is
down,
and issues
are
up.
The
main
question
to
be
asked
of
the
new panel
data
would
merely
seem
to
be
"by
how
much?"
We
shall
reveal
the
general
answer rapidly,
for
its
explication
can
easily occupy
us for the
rest
of
this article.
Where
party
and
issues
are
concerned,
and
with respect
to
those facets
of
change
which a
panel design
is
uniquely
equipped to
illuminate,
our first-cut
explora-
tions
show
that the
1972-76 panel
data are
very nearly
a
carbon
copy of
those
from
the
1956-60 panel!
That
is,
the
individual
stability
of
party identification,
relative
to that for
policy preferences, is just about as great
for the
latter
period as
for
the
former.
We shall
spend
most
of the
rest
of this article
trying
to
clarify just
what
those initial
results
do
and
do
not
mean. One
thing they
do not

1979
The
New CPS
Election Study Panel
35
mean
is that
no
change
has occurred
over
this
period.
A severe
handicap
in approaching
prob-
lems
of change
consists
of
the first
simplistic
assumption
that change
with respect
to a
given
variable
is
something
monolithic:
either
the
variable
changes
or it
doesn't,
in which
case
we
might
not
want
to call
it a
variable.
Such
a
statement
can have
merit
if
we
specify
it
more
closely,
as
we
might
by referring
exclusively
to
the
mean or
central tendency
of the
variable.
Then
it is
undoubtedly
true
that
either
the
mean
has
changed
or it has not.
But
more
generally
speaking,
change
is multifaceted,
even
if we
are addressing
the
same
variable
in
its
behavior
over
time.
These
multiple
facets
of
change,
even
with
respect
to
a
single
variable,
are recognized
in
more
formal
treatments
of
process.
Although
the
distinctions
are looser than
one
might
like,
processes
are
described
as
"strictly
stationary,"
only
"weakly
stationary,"
or "not
stationary
at
all,"
according
to whether all,
some
or none of
the
moments
of
the
univariate
distribution are
constant.
Thus
each
moment
is, unto
itself,
a
facet of change.
As
we shall
see momentarily,
there
are still
further
facets
of
change
with
respect
to
a
single
variable,
in addition to
the
moments,
and it
is one of these
which
shows
unexpected
constancy
or stationarity.
It is
exactly
because
impressive
changes
are
known
to have occurred
in the
party
and issue
variables
that
the
panel
discovery
of any
points
of
stationarity
is
particularly
exciting.
In
a
period
in
which
multiple
facets of
change
are
occurring,
the localization of some
points
of
stationarity
are
theoretically
pregnant
in
diag-
nosing
the
nature of the processes
underlying
the
total
manifold
of observations.
To
para-
phrase
Archimedes,
they
give
us
"'a place
to
stand"
from
which other
facets in
flux can
be
evaluated.
The
Case
of
Party
Identification
We
can
profitably
turn to
the
case
of party
identification
over
the comparative
1956-60
and
1972-76
time
segments
for concrete
exam-
ples
of the
varied
facets
of
change.
All too
frequently,
contradictions
appear
to
arise in the
treatments
of
change
carried
out
on the same
variables
by
different scholars,
simply
because
different
aspects
of change
are being
addressed,
and
for
quite
different
purposes.
This
potential
for
confusion
is particularly
virulent
as
we
approach
the
panel
feature of these
change
data,
so careful
description
is
important.
The
most
obvious
facet
of
change
in a
single
variable
based
on aggregated
observations
over
time
is the
displacement
of
the
mean
of the
distribution
from
one
reading
to the
next.
Several
authors dealing
with
the first panel
have
examined
changes
in
the
central
tendency
of
the
party
identification
variable
in
the
1956-60
period.
There
are
various
ways
such
change
can be
indexed
in
addition
to
the
obvious
integer
scoring
of
the
conventional
seven-point
scale
and
taking
an
actual
mean.
One
can
deal,
for
example,
with
the
percentage
Democratic
(or
Republican)
of the
total
sam-
ple.
or of
the
more restricted
set
of those
who
identify
with
one party
or another.
If, for
example,
we examine
the marginal
distributions
of
party
identification
in
the panel
years
in
terias of percentage
Democratic
(strong
or
weak)
of all
identifiers,
we find values
of about
60
percent
for
1956,
64
percent
for 1958,
and
61
percent
for
1960.
For 1972
the
value
is
63
percent,
for 1974,
65
percent,
and
for 1976,
63
percent.
These
are forms
of
change.
Such
statements
of
change
can
be
conceived
as
referring
to shifts
in
the
mean
or
central
tendency
of the
party
identification
distribu-
tion,
as
one
facet
of
change.
The
shifts are
customarily
small.
If we stick
to the
percentage
Democratic
of
all
identifiers,
the median
bien-
nial
displacement
over the
whole
election
study
series
runs
below
2
percent.
On the
other
hand,
unless
they
can be
thought
of
as
sampling
error,
these differences
are in
themselves
evidence
that
party
identification is
not
perfectly
stable
down
to
the
last voter.
Although
they
are
typically
small enough
to
lie
within
the tol-
erances
of plausible
sampling
error,
it has long
been
clear
however
that these
shifts
are
quite
systematic,
moving
nicely
in tune
with national
short-term
tides
favoring
one
party
or
the
other.
For
example,
the
three biennial
shifts
since 1952
which
exceed
3
percent
are all
associated
with landslide
presidential
elections.
More impressively
still,
first
differences
in
these
percentages
Democratic
among
identifiers
from
one
biennial
election
to
another
show a re-
sounding
correlation
of .75
with
first dif-
ferences
in the
Democratic
fraction
of
House
seats
after the
corresponding
elections since
1952. Clearly,
although
these
shifts
are
slight,
we
are not
dealing
with
mere
sampling
error
in
this
facet
of
party
identification
change.
In-
deed,
Brody
(1977)
in
working
with the
1956-58
and 1972-74
segments
of the
panel
studies,
focuses
a
great
deal of
his attention
on
these
shifts,
showing
within
the
panels
them-
selves
that
shifting
has
political
meaning
in
terms of other
attitudinal
dimensions.
How
important,
theoretically
and
practically
speaking,
depends
entirely
on the more
specific
class of
questions
one wishes
to
ask
about

36
The
American
Political Science
Review
Vol.
73
change.
Certainly
for
any
problems
of
short-
term
changes
in political
tides,
they
are a
crucial
indicator,
although
it must
be
recog-
nized that
a
disproportionate
share
of
these
swings
come
from
people
who do
not vote
in
any event.2
On the
other hand,
if one's
research
interests
concerning
change
are
longer-run
in their
per-
spective,
then
these
small
shifts
are
not of
gripping
interest.
This
is not
simply
because
they
are small,
but
also because
their
nature
as
short-term
oscillations
around
a mean
value
has
historically
been
quite
clear.
If the
proportion
Democratic
shifts
by
4 percent
between
two
biennial
elections,
as
it
did between
1956
and
1958,
this
is
not
likely to
be
an intriguing
harbinger
of
fundamental
partisan
change,
be-
cause
the
odds
are very
high
that
the shift
will
rapidly be
reversed.
It is as
though we
were
watching
reeds
growing
in a
shallow
pond.
Before an east
wind,
they
tilt to the
west.
Before
a west
wind
they
tilt
to
the east.
When
the
wind is
still,
they
stand
in
a
middling
position.
From
some
points
of
view the
motion
is
of interest,
since it
tells
us which
way
the
wind
is
blowing
at the moment.
From
some
other
points
of
view
it is
not of great
interest,
since we
know
the
reeds
are not
moving
in any
absolute
sense,
but
are firmly
anchored
in
a
fixed
position
by
their
roots.
Speaking
more
technically,
the
mean of the
party
identification
distribution
around
which
these
small oscillations
occur
has
been
almost
stationary
for the
past
25
years.
We
say
"almost"
because
there
appears
to
have
been
a
glacially
slow drift
of
the mean
in
a Democratic
direction
over
this
period,
probably
due
to
population
replacement.
Even
in
25
years,
the
total ground
covered
is
very
small.
This
secular
drift
is small
by comparison
with the
amplitude
of the oscillations
about
that
mean,
even
though
those
oscillations
are
in
turn
very
small
by
comparison
with the
swings
of
the
national
popular
vote.
It
is
important
to
recognize
these oscillations
for
what
they
are
in
the
panel
data,
to avoid
misinterpretations
of
the
changes
they generate.
Thus,
for example,
Brody (1977)
discusses
the
counter-intuitive
finding
that
in
both
the
1956-58
and
the
1972-74
segments
of
the
panel
studies,
Republican
identifiers
are
less
"directionally
stable"
(i.e.,
more
likely
to move
out
of
the
Republican
camp
into
at
least
independence
if not a Democratic
identifica-
2For a more
extended discussion
of the
significance
of these oscillations,
see Converse (1976),
pp.
126-30.
tion)
than
Democrats,
despite
the
fact
that
in
most
other
terms
Republicans
have
seemed
more
"anchored"
in their
political
behavior.
This perplexing
effect almost certainly
arises
because
both
the
1956-58
and the 1972-74
segments
bracket
major
shifts
in
short-term
partisan
fortunes
from
the Republicans
to
Democrats.
Thus
even
relatively
labile
respon-
dents
who
claimed to
be
Democrats
when
current
winds
were
favoring
the
Republicans
(1956
and 1972)
had
little
reason
to stop
being
Democrats
two
years
later
when
the winds
had
become
more
supportive.
Presumably
the
re-
verse
partisan
contrasts
would
occur
in tem-
poral
comparisons
where
the
wind
changes
were
simply
reversed.
While the
mean
of a
distribution
provides
one facet of potential
change
in
a variable,
it
scarcely
exhausts
the
subject
of
change
where
that
variable
is concerned. Aside
from
practical
problems
like
ceilings
and
floors on
measuring
instruments,
the
variance
of
the
distribution
can
move
independently
of the
mean. It
can
remain stationary
while
the
mean shifts
dra-
matically;
or
it
can
shrink
or
expand
dramati-
cally while
the mean is
stationary.
In fact, as
we
have
seen,
the
most striking
change
in
party
identification
since 1965 is
a shrinkage
of
the
variance
by
more
than
10
percent,
as more
and
more voters
cluster toward
the independent
middle
of
the
distribution.
This
is
again
true
change
in
party
identification,
and has
appro-
priately
been
treated
in
the
recent
literature as
having
major
theoretical
interest.
Yet
this
change
has
occurred
rapidly
while
the mean of
the distribution
has
been
essentially
constant.
There
is no need to
do
a
panel
study
to
monitor
change
in
either
the mean
or
the
variance of
a
distribution.
Successive
indepen-
dent
cross-section
samples
can
perform
this task
equally
well.
The
information
a
panel
study
uniquely
supplies
involves
the continuity
in
positioning
of the
same
individuals
over
time.
One
index of
such
continuity
is the
proportion
of the
panel
on
or
near
the
main
diagonal
of
the
cross-tabulation
of the
same variable
between
waves.
While
such
expressions
are
commonly
used,
they
are weak summarizations
not
only
because
of
the
arbitrary
nature
of the
category
width
that
constitutes the
"stable"
diagonal,
but also
because the
resulting proportions
are
strongly
influenced
by
the number
of
categories
into which
the variable is
partitioned,
so that
comparisons
across variables
of different
cate-
gory
"length"
become
nonsensical.
An obvious
improvement
is
to deal with
a summary expres-
sion
of correlation
or
regression.
We
have
approached
the
new
panel
data
on
party
identification
only
after
a consideration

1979
The New
CPS
Election
Study
Panel
37
of
changes
in
means
and variances for a
very
simple reason:
although
novices
are
constantly
confused by
the
fact, continuity
correlations
formed
on the same
variable
in
two
panel
waves
are as
independent
of
temporal
changes
in
the
mean and variance of
the distribution
as the
latter are
independent of
one
another. That
is
to
say,
a
continuity correlation
can
range
from
zero to
perfection whether both
mean and
variance are
constant,
or one or
both
shift
dramatically
between the two time
points.
This
is true by the
very
construction
of the
product-
moment
coefficient,
which
equates
or
nor-
malizes
the two
means
and
variances, thereby
partialling out
whatever
real
change
may
have
occurred
with
respect
to those
facets of the
variable.3
Since
continuity
correlations can
vary inde-
pendently of
changes in means
and variances,
they obviously
tap still another
facet of change.
It
is
important to
be
clear
just
what facet that
is.
Obviously, it has to do with
stability of
individual-level
positioning. What
is important
to
remember is that
it taps the
stability of
relative
individual
positions across a
population,
not absolute
ones.
While it is
thereby technically
quite possible
that
individual-level
continuity
correlations for
party
identification have
remained
at their high
levels
of the
late 1950s despite the
shrinking
variance
of
the
distribution, it
would not
intuitively
seem very likely.
After all, the
descriptors
commonly applied to the
surge in
independence
talk of the
"erosion" of these
identifications, and'the
"destabilization of par-
tisanship."
While such
terms are
entirely prop-
er, they
would certainly seem to
imply a major
unraveling of
individual-level
continuities in
party
preference,
even
in the
stability of rela-
tive
locations.
Moreover,
at a less
intuitive and
more technical
level,
if
error
variances remain
constant while
true variances
of
a variable
decline,
then the
magnitude of
correlations
involving the
variable will be
attenuated.4 Thus
obviously,
one can restore
some of this
informa-
tion
in
relatively compact
form
by
shifting
from
correlation coefficients
to raw or nonstandardized
regression
coefficients
plus intercept
terms. For some
purposes
this
is
exactly
what one should
do, although
such statements
operate
in a metric-bound
currency
which
can
impede
communication for those
unfamiliar
with the variable. For
simplicity
we
shall deal here
with
continuity
correlations.
4Note,
however,
that this
expectation
is
contingent
on
an
error variance
which is
gaining
in size relative to
the
true
variance.
In the
more
general
case,
the
continuity
correlation
is in
theory independent
of
the
variances of the
component
distributions.
the
natural shrinkage
of
observed
variance
of
party
identification
might
be expected
on these
grounds
alone to
dilute the
continuity
correla-
tions
correspondingly.
Some years
back we suggested
that
this
might
not turn
out empirically
to
be the case
with party identification
(Converse,
1975),
since
the
only
scrap
of a national panel con-
ducted during
the
1960s-a couple
hundred
stray
cases reinterviewed
between
1964
and
1966-showed
continuity
correlations
which
were virtually identical
to those
estimated
from
the 1956-60 panel,
despite
the fact that
this
panel
lapped over
into
the
period
in which
the
variance
of identification
reports had
begun to
shrink palpably.
Therefore
we
are
not
entirely
surprised
to
discover
that the
continuity
corre-
lations
calculated
for
party
identification
from
the new
1972-76
panel
are
just
about
as
high
as they
were
in
1956-60.
In absolute
terms
they
are a
hair
smaller,
but the
difference is too
slight
for statistical significance.
Probably
the
drop
is
in
some
degree
real,
as one would
expect
from the
truncation
of variance
in the
later
period.
However,
the
stunning
fact
is
that
the
decline
is
simply
vestigial:
if
one
could say
from
the
1956-60
panel
that individual-level
continuity
correlations
for
party identification
with
a
two-year
time
lapse
ran
in
the
low
.80s,
one
can
make
exactly
the
same statement for
the
1972-76 panel.
For reasons that should
be
apparent,
the
magnitudes
of
continuity
coefficients
are speci-
fic
to
the size
of the time
interval
elapsing
between
measurements. In
processes
of progres-
sive
(i.e.,
not
cyclical)
change,
they can
be
expected
to decline
as the
interval
becomes
more extended.
Indeed,
this property
emerges
clearly
in the
party identification
data. The
four-year
correlations
are
less than those
calcu-
lated
for a
two-year
interval; and the
latter
are
smaller than correlations formed
for
measure-
ments
separated
by only
a few weeks
in
time,
as
is
the case when
the party
identification
item
is
applied
to the same sample
in
both
the pre-
election
and
post-election
questionnaires.
Therefore
continuity correlations
in general
only
have
meaning
as
they
are tagged
with some
specific
time interval. If
we
stick
with the
two-year
correlations
in
order
to
encompass
comfortably
the
data
from
the
two-wave,
1964-66 mini-panel,
then we
find the follow-
ing
best
estimates
for
two-year
continuity
correlations;
5
5All
correlations
are
calculated
on
the
basis
of
party
identification
as
a
seven-point,
integer-scored
scale.
Here
as at
some
other
points
in this
article,

38
The American
Political
Science Review
Vol.
73
The
1956-60 period
= .835
The 1964-66
mini-panel
=
.836
The 1972-76 period
= .813
There
are several
ways
of
processing
these
raw continuity
coefficients
into more
refined
components.
One of the
easiest
partitions,
dependent
on having
at least
three waves of
measurement,
is to break
the raw coefficients
into (1) a
reliability component,
reflecting
the
complement
of
the
proportion
of error
vari-
ance; and
(2)
a
"true"
stability coefficient,
reflecting
the
degree to
which the measurement
at
two points
in
time
would be correlated
if
there were
not
intrusion
of unreliability.
Such
procedures
have
been
described by
Heise
(1969), and
refined a
further step by Wiley
and
Wiley (1970).
These
manipulations
rest, as
always, on
a
series of assumptions
about
the
character
of the data and
the attendant
error
structures.
One of the less pleasing
sets of
assumptions-that
the measurement
reliability
(Heise)
or
the
absolute error
variances (Wiley
and Wiley)
are
constant
from wave to
wave in a
given panel
sequence-can
be relaxed in
the case
of the
1956-60
panel,
where
instead
of three
waves
of
measurement
of party
identification
we
can
profit
from five.6 While it is
regrettable
that
party
identification
was measured
only
three times
in
the new panel
(1972
pre-,
1974,
and 1976
pre-),
there is sufficiently
striking
consistency
between the
two panels to
enable
us to
use
the
additional
information
afforded
by
the first
panel
to reassure ourselves
concern-
precise numbers
of cases are
difficult
to
provide,
since
the estimates
are more
often
than not
"synthetic,"
resting on
multiple bases.
Thus,
for
example,
the
value
of
.835 is
based on an
averaging
of
four different
estimates available
in the
1956-60 panel
for
the
correlation
after about a two-year
interval.
While all
of
these
values
are
very
similar in
appearance,
each is
based on a rather
different
number
of
cases.
61n
addition
to
the
full-blown
fourth wave
present
on
ICPSR
tapes
due
to a
full-sample application
of the
party
identification
item in the
post-election
as
well as
the
pre-election
study
for
1960,
a
stray
set
of
165
cases
were re-asked
the
party
identification
question
in
the
1956 post-election
study
as
well.
Estimates
from
this
mini-sample
are
of
course
less stable
than their
counterparts
in
the
four
other
waves.
However,
where
comparisons
are
possible,
they
are
so
gracefully
consistent
with
what
can
be learned from the
pre-
and
post-1960
waves that
we
do not
have much
hesitation
in
using
them
to
yield
the additional
degrees
of
freedom
provided
by
a fifth
wave.
Naturally,
in
all
relevant
calculations,
we take
account
of the
fact
that
the
points
of
observation
are
very unequally
spaced
in
real time.
ing
various
assumptions
which
must
be
made
for
the
later
batch
of
panel
data.
Also,
of
course,
we
can
proceed
no
further
with
the
mere
two
waves
offered
by the
1964-66
mini-panel,
although
the data
presented
above
also
give
some
assurance
that
we
are
dealing
with
process
parameters
which
have
a
con-
siderable
stability
in
the
intermediate
term.
There
are other
assumptions
embedded
in
calculations
of
the
Heise
or
Wiley
and
Wiley
type
that
are
not
entirely
well
met
by the
data.
Thus,
for
example,
there is reason
to
believe
that both
the
measurement
reliability
and
the
true stability
of
the party
identification
item
is
less
than
nicely
homogeneous
over
the
popula-
tion.
In
later,
more
complicated
assays
of these
data,
we
shall
take
such
characteristics
into
fuller
account.
Nonetheless,
as
first-cut
approxi-
mations,
it
is worth
seeing
what
can
provisional-
ly be
estimated
concerning
the
relative
propor-
tions
of
change
due
to
unreliability,
as opposed
to
"true
change,"
in
these
longitudinal
measure-
ments.
Where
they
can be
compared
most
fully
(as
in the
five-wave
panel),
calculations
of
the
Heise
and
Wiley
and
Wiley
type
give
virtually
identical
estimates
with these
data.
Both
calcu-
lations
agree
that
the main
source
of
the
slight
decline
in overall
continuity
between
the
two
large
panels
is
a
slight
increase
in
unreliability,
exactly
as one
would
expect
if
the
main
change
in
the
data
were
a
shrinkage
in
the
true
variance.7
In
both
calculations,
the
decline
in
reliability
is roughly
from
(a
central
tendency
of
about)
.88
in
the first panel
to
about .84
in
the second.8 Once
this
slight
decline
in
reliabili-
ty
is
set
aside,
as
it
is
when
we
calculate
the
true stability
coefficients,
we
find
comparisons
of
the
following
sort
as
our
best
estimates
for
the
two-year
interval:
1956-
1972-
58-60
74-76
Two-year
stability
(Heise
style)
.951 .972
Two-year
stability
(Wiley
stule)
.958
.972
The
homogeneity
of
these
estimates,
both
between calculation
methods
and
more
espe-
7Actually,
our
detailed calculations
show not only
no
shrinkage
in
the error
variance,
but
if
anything
a
slight
absolute
increase.
Thus the decrease
in
reliability
(or
the
proportion
of
true
variance
to total
variance)
has a double source.
8Throughout
this article
we are
providing
mere
"bottom-line"
estimates that
conceal more
detailed
consideration
of
angularities
in the
data.
These will
be
dealt
with more
explicitly
in
subsequent
technical

1979
The
New CPS
Election
Study Panel
39
cially across two periods in which
the nature of
party identification has seemed
so different,
when other facets of
change
in the
variable are
considered, is very striking. There
are a variety
of ways of expressing the
implications of these
findings. We shall postpone some
of the broader
theoretical implications for our
conclusions.
For
now,
we might briefly
consider what the
above data imply for the
long-term "staying
power" of party identifications
at the indivi-
dual
level.
If
we merely knew
that
the
raw two-year
continuity correlation
for
party
identification
was
about .82
(as
a
period-free
average),
then
we
might
be
tempted
to
conclude
that
over
a
period of 16 (24) years, we could
only expect
a
continuity
of
.45 (or .824),
a value
which
would
not be
entirely impressive.
However,
if
we
trust
the
above
estimates
as to the
propor-
tion of observed
change
which is mere
unreli-
ability
of
measurement,
then
we
would expect
the
16-year
observed
(i.e.,
error-attenuated)
stability
coefficient to
look
something
like
.73
(or .86
[.964],
where the
.86
is the
period-free
reliability estimate) instead
of
.45. This
is
a
vastly different picture of the
long-term indi-
vidual-level stability
of
the
variable, especially
as it is appropriate to square these
two values,
yielding
a
contrast
in
shared
temporal
covari-
ance
of
.53 instead
of
.26.
And we
feel
considerably less
cavalier
about
cycling hypo-
thetically
over
longer periods of
time
in
this
way,
once we
have at
least
some shred of
assurance,
which
the
above data
give,
that
the
true
stability
of
the
party
identification
mea-
sure
is
remarkably
close to
stationary
across
two
political periods
in
which,
in
other
senses,
it
has been
subject
to
major change.
However,
even
this
assumption
is
subject
to
challenge.
After
all,
the
decline
in
reported
strength
of
partisanship
has
proceeded
secularly
from 1965 to 1975. While
our
panel
observa-
tions
in
1964-66
and
1972-76 bracket
the
beginning
and end of
this
fascinating period,
they
leave
uncovered a
crucial
segment
of
time
from 1966 to 1972. Perhaps it
was
in
this
six-year
interval
that
continuity
correlations on
party identification might have
registered a
distinct
sag.
Here
again
we
can
import
a
crucial
datum
which
suggests
that the
stationarity of these
articles.
One
such
angularity being
momentarily ig-
nored
in
this
statement, although
dealt with
in
our
calculations,
is the
presence
of
some systematic
progression
of
the
unreliability
term across waves of
the same
panel.
continuity
correlations
has been
complete over
the past
two
decades.
In
1973
Jennings
and
Niemi (1977)
reinterviewed
the parents
of their
1965 graduating
high
school students,
who
had
originally
been interviewed
in 1965.
In princi-
ple this
is a 1965-73
panel, covering
exactly
the years
that
we
are
missing
in our
national
samples.
The
difficulty
is
that
these parents
cannot be
seen as a proper
sample
of the adult
electorate,
even though they may
be a perfectly
honest national sample
of a more restricted
population.
What the
Jennings-Niemi
respondents
have
in
common
is having
been parents
of high
school seniors
in 1965. This means
that
they
are
a
very homogeneous
age cohort,
relative to
the electorate as
a
whole. The
group lacks
newer
and
younger
voters completely
and,
given their
association
with children
not drop-
ping out
of high
school,
they
should show
higher
than
normal social
status.
Now
it can
be
calculated that
if
such
respondents
were really
representative
of the
full
electorate, and
if our
panel continuity
coefficients were
truly
sta-
tionary
in
the
long
run,
then
over
an
eight-year
span
these
respondents
should
show
a
continui-
ty
correlation for party
identification
of .76.
However,
the
age
and class bias of
the
Jen-
nings-Niemi
respondents
would
lead
one to
expect
a
slightly higher
figure. It
turns out that
the observed
figure
for the
party
identifications
of the
Jennings-Niemi
parents
over the
1965-1973
interval
is
.79.
This
value
is
either
right
on
target or,
if
anything,
a shade high:
certainly
an
hypothesis
which
maintains that
continuity sagged between
our
panel
points
of
1966
and 1972
finds no comfort
whatever
in
these
data.
All of
the
available
evidence,
then,
suggests
that there
has
been
no change
in the
individual-level
continuity
of
party
identifica-
tion
over the past
20 years, despite
an incon-
testable
change
in the
likelihood
of reported
partisanship.
The
Continuity
of
Policy
Preferences
in the
'50s
and the
'70s
A
comparative
examination
of
the individu-
al-level
stability
of
preferences
on political
issues
poses
somewhat
different
analysis prob-
lems,
chiefly
because
while
the two
major
parties
have
long been
constants
of
the political
scene,
the
objects
of
issue
orientation
are
in
perpetual
evolution.
The issue positions
which
seemed
central
to
political
controversy
in
the
early 1970s
were
somewhat
different
from
those
central
in the late
1950s.
Moreover,
the

40
The
American
Political
Science
Review
Vol.
73
election studies
have been
for some time
measuring issue
positions
in a different
format
than was used
in
the late 1950s, and there
are
reasons to
expect that these
pure methods
changes should
affect some
of the process
parameters which
interest us (Bishop
et al. and
Sullivan
et
al.,
1978).
Hence
strict
comparisons
of
issue
continui-
ties
are
impossible.
Nevertheless,
there
are
some
matches
between
issue
items from the
panel
studies of
the
two periods
which
are
tighter
than
others,
so that some
rough comparisons
are
possible.
Moreover,
the
continuity correla-
tions
for
issue
positions
in the
original
panel
fell
within a
range
which was vastly
different from
the values
for
party identification,
and there-
fore it is of interest
to see how
even the newer
issues
are behaving
in
these terms
in the more
recent panel.
We shall begin with
a few of the
closer matches
and then summarize
the general
trends
over
all
of
the
new
panel
issues as
a set.
One issue
item in the later
panel
was
worded
identically
to the original 1956-58-60
study.
This is
a
foreign policy question,
in
agree-dis-
agree
form,
intended to
tap
isolationism:
"This country
would be better
off if we just
stayed
home
and did not concern
ourselves
with problems
in other parts of the
world."
Despite
the
identical
wording
of
questions,
there were
minor variations
in
format.
Most
notably,
the item responses
were
handled
di-
chotomously
in
the new
panel,
whereas
they
had been
spread
out
into four main
points
("strong
agree"
to
"strong
disagree")
in all
waves of
the
original panel.
Also
in
the last two
waves of the
first panel, a "filter"
was added to
assess
whether
respondents
felt
they
had
an
opinion
on
the
matter.
Both
of these format
differences would seem
likely,
by
conventional
wisdom
at
least,
to
dull
the
correlations
sur-
rounding
the
more recent
application
of the
isolationism item,
other things
equal.
Unfortunately,
this
foreign
policy
item was
asked
only
in
1972
and
1976
in
the second
panel, so
we
can
only
make
direct
comparisons
with the
four-year (I195 6-60)
continuity
corre-
lation from the first panel. Since
the same
limitation applies
to
the
only
other plausible
foreign policy
item
match,
we shall
take
that
item in the
same breath,
even though item
wording is
much less
parallel.
Both
panels
had
an item
on
foreign
aid, although
with the
following
variation
in
wordings:
(1956-60)
"The
United States
should give
economic help
to the poorer countries
of the
world even if
those countries can't
pay for it."
(1972-76)
"The
United States should
give
help
to
foreign
countries even
if
they
don't stand
for
the
same
things
we
do."
The
same
differences
in
response
categories
pertain
for these
foreign
aid items
as for the
isolationism item.
The
four-year
continuity
correlations for
these
foreign
policy items are as
follows:
1956-60
1972-76
r
(N)
r (N)
U.S. Stay
Home
.347
(1086)
.309 (1113)
Foreign
Aid .292
(1009)
.264
(1039)
The
similarities
between the
two
panels are
quite
impressive.
And
if
there is any
merit in
our a
priori
expectation that
format
differences
would tend to
attenuate the
more recent
panel
results
by some
small
amount,
it
would be hard
to
argue
that
these
continuity
estimates are
any
less
stationary than
those
for party
identitica-
tion,
although
of course
they
fall at
a
totally
different
level of
correlation.
In
the
original
panel,
foreign
policy prefer-
ences
tended to be
among
those issue
items of
lesser
individual-level
stability.
We
shall see that
the
case is
exactly
the same in
the
new
panel.
Domestic
issues
in the old
panel
fell
into
two
substantive
domains: social welfare
policy
and
civil
rights
policy. Both
domains
are
repre-
sented with
three-wave
issue
items
in
the new
panel,
although
the
item
format is
always
different,
and the
wording
shows
greater
or
lesser
variation.
The nearest
thing
to
a
match
among
the
social welfare
items
involves
questions
about
government
guarantees of
employment.
The
old
agree-disagree
item
has been
changed to
a
forced-choice
format, with
two
contrasting
substantive
alternatives.
Moreover,
in
this
case-in
contrast
to
the
foreign
policy items-
response
alternatives for
the
new
panel
are
more
numerous than
for
the
old,
with
a
7-point
rating
scale
being
used to
replace
the
four-cate-
gory
agree-disagree
form.
Thus, by
the
same
reasoning
as
used
before,
we might
expect
slightly
higher
correlations with
the
new
panel
than
with the
old.
Despite
such
differences
in
form,
the
intent of the two
items is
surely
identical:
(1956-60)
"'The
government
in
Washington
ought
to
see
to
it
that
everybody
who
wants to
work
can find a job.'
Now
would you have an
opinion
on
this
or
not?
(IF
YES)
Do
you
think
the
government
should do
this?"
(1972-76)
"Some
people
feel
that
the
govern-
ment in
Washington
should see
to
it that
every
person
has a
job and
a
good
standard
of
living.
Suppose
that these
people
are
at one
end
of this

1979
The
New CPS Election
Study Panel 41
scale-at point
number
1.
Others think the
government should
just
let
each
person
get
ahead
on his own.
Suppose these
people
are at
the other end-at
point
number
7.
And,
of
course,
some
other
people have
opinions
some-
where
in
between.
Where
would
you place
yourself on
this scale,
or
haven't you
thought
much
about
this?"
For
these
items, asked
in
all
three
major
waves
of
both
panels,
the
average
two-year
continuity
correlation in
each
study is:
1956-60
1972-76
Government Job
Guarantee
.457
.493
Again,
we
may
well
be struck
by the
similarity
in
values
between the
two
periods,
especially
by
comparison
with the
much
more
stable
party
identification
items,
or the
visibly
less
stable
foreign
policy
questions. Or
if
we
want to
pay
attention
to
the
small
differences in
absolute
values
between
the two
estimates, the
value
for
the
new
panel
is in
fact
slightly
higher,
as
format
changes
would have led us
to believe.
In civil
rights,
an
item on
school
desegrega-
tion occurs in
both
panels,
although
it
is
another
instance
in
which
the
question
was
only
asked
twice
(1972
and
1976)
in the
second
panel. The latter
item is a
forced-choice
version
of the
earlier
form,
with
those
respon-
dents
claiming
no
opinion
being
filtered off in
both
instances.
(1956-60)
"The
government in
Washington
should
stay
out
of the
question
whether white
and colored
children
go to the
same
school."
(1972-76)
"Do
you
think
the
government
in
Washington
should see
to
it
that
white
and
black
children
go
to the same
schools,
or
stay
out of this area as
it is
not its business."
In
general
we
would
expect
the
forced-choice
version to
produce
slightly
more
discriminating
measurement and
hence
higher
correlations;
yet
the item in
the second
panel
was
only
scored as
a
dichotomy,
instead of the
four-point
agree-
disagree
scale
used
originally.
Therefore
there
are no
very
clear
expectations
as
to
pure
methods
differences
in
the
continuity
correla-
tions.
The
actual
four-year
correlations
are,
however,
very similar:
1956
-60
1972-76
r
(N)
r
(N)
School
Desegregation .397
(1059)
.410
(714)
By
this time
it
will
not
be
surprising
if
the
reader
is
beginning
to
find
the
parallels
between
all
of these
panel
results, despite
a decade
and
a
half of
intervening political
change, as
eerie
as
these parallels have
seemed
to
us.
We
have
by
now
exhausted all of the
issues
for
which
very
close
matches
in
wording
and
intent, if
not in format or in
letter,
are
available.9
While the similarities in
individual-
level
stability for these well-matched
items
are
striking,
we
should not conclude that
the
issue
materials from the new
panel
more
generally
look just
like those from the
earlier data set.
This
is true
chiefly
because
there are new
issues
present
in the
1972-76 panel which are
not
inheritances
from
the
past; and
these items
sometimes
have
a
different cast
in continuity
terms.
A
good
example
is
provided
by
the civil
rights issue
concerning school
busing to pro-
mote
racial
integration. This issue
was not even
in the air at
the time of the
earlier panel, so
that
its absence
from those
interviews was not
at
all a
matter
of editorial
oversight, but rather
a
token of
a truly different
environment of
policy debate. The
issue is also
of particular
interest because
casual observation
would
sug-
gest that
this issue, probably
more than any
other
in the
civil rights area or, for
that matter,
the
total
arena
of policy debate,
has had an
inflammatory or polarizing
quality that has
bitten
deep
into
even
relatively
inattentive
segments of
the public. In other
words, if as we
have
earlier
suggested, it is true that
the general
levels
of
these continuity
correlations reflect
the
degree
to
which
items tap into
more or less
thoroughly
crystallized attitude
structures
(where
the
limiting
case of
no
crystallization
whatever
is the
"non-attitude"), we
would
expect
rather substantial
continuity correla-
tions for
the
busing
items. This
expectation is
sustained.
The
best estimate of the
two-year
continuity correlation
is .575, a
value still
falling very
well
below
the
corresponding value
for
party
identification, but nonetheless
signifi-
9There
is
one
other
pairing,
also
in
civil
rights,
where
a
somewhat
rougher
match
might
be
made. In
the
first
panel,
a
question inspired
by
the FEPC
legislation
of
the
time
asked
about the
appropriateness
of
the
government
attempting
to
ensure
that
Negroes
get
"fair treatment in
jobs
and
housing."
In
the
second
panel, respondents
were
asked
whether
the
govern-
ment should make
"every possible
effort to
improve
the social and economic
position
of blacks
and
other
minority
groups."
These are
divergent
questions,
both
in
substance
and
in
group
of reference,
although
there
is some obvious
kinship
between
them. The
first item
was
in
four-point
agree-disagree form;
the
second
was
in
forced-choice, seven-point
scale form.
The
average
two-year
continuity
correlations
in the first instance
was
.46
8;
in
the
second,
a
somewhat
higher
.535.

42
The American
Political
Science Review
Vol.
73
cantly higher
than that
registered
for
any issue
items
in the
whole
1956-60
panel.
There
are at least
two
other issues
new
to
the 1972-76
panel
which
show
even higher
continuity
values.
These items
are
not
only
new: together
they
seem
to stand
for a
co-
herent class
of issue
which
was not
even
of
much salience
in
the 1950s.
One
is a 7-point
scale
where the extremes involve
the
legaliza-
tion
of
marijuana at
one end,
and the
setting
of
higher
penalties
for
its use
at the
other.
The
other is
an item on
abortion,
in
which
the
respondent
is asked
to choose
between
four
levels of
lenience
in availability,
essentially
bounded by "under
no circumstance"
at
one
end
and "at
any need"
on the
other.
Neither
of
these
items
was asked
in
the middle
wave
in
1974,
so
that
only
estimates
of the
four-year
continuity
correlations
are
available.
As
we
have
pointed
out,
coefficients
over
this time
lapse
tend
to run
lower than
their
two-year
counterparts.
Thus,
for example,
if
the school
busing issue
runs
about .574
for
a two-year
lapse,
it stands
at
.535 for
a four-year
lapse.
The
only
two
issues which
exceed it
are
marijuana
and
abortion,
with four-year
coeffi-
cients of .640
and .617,
respectively.
These
are
high
values,
and
they
seem
clearly
associated
with a class of
moral issues
of
the
kind
brought
to salience
by
the
counterculture
confrontations
of the preceding
decade.
Indeed,
the
only
other
specific
issue item
with
a
four-year
continuity
correlation
over
.50,
in
addition
to
marijuana,
abortion
and
busing,
is
an item
on whether
women should
have
an
equal
role
with
men
in
business
and govern-
ment,
or should keep
their
place
in the
home
(.519).
Again,
this
is
an
item
which
pits
the
cutting edge
of new
mores
against
an
array of
traditional
values.
For those
who
have worked
in other
politi-
cal cultures, it
is not entirely
surprising
that
such moral
issues should
have
a
deeper
reso-
nance among
those not
normally
attentive to
much
political
controversy
than
is
true for
policy
debates
of most
other
kinds.
Perhaps
these
uncommon
magnitudes
represent
some-
thing
other
than a
sharper
crystallization
of
attitudes
across
a
larger
fraction of the
elec-
torate
than is true of other
issues,
but for the
time
being
such would seem
to be
the most
plausible
diagnosis.
There
is one other
item,
not
referring
to a
specific
policy
issue
or
for that matter
any
"new"
issue
at
all,
but
relevant
to a
large
range
of
issues,
which also
shows
a
four-year
continui-
ty
correlation
over
.50,
although
the
high
value
occurs
under
a
special
configuration
of
circum-
stances
important
to
make
clear. This
item is
the
ideological
item
involving
self-placement
on
a 7-point
liberal-conservative
scale,
with
a
four-
year
continuity
correlation
of
.564.
The
special
circumstances
arise
because
the correlation
is
being
figured
on
a
much more
limited
fraction
of
the population
than
is true
of
any
of
the
other
specific
issues.
This high
attrition
occurs
because
such
a continuity
correlation
can
only
be
computed
for those
respondents
who
choose
some
content
response
at
both
years
involved
in
the
correlation.
On
items
like
the
abortion
question,
almost
all respondents
choose
a
con-
tent
alternative,
so
that the
continuity
correla-
tion
refers
to
a quasi-totality
of
the
electorate.
On
the liberal-conservative
item,
however,
over
one-third
of
the
population
must
be
set
aside
at
the outset
as
missing
data,
either
because
they
say
they
do
not
know
their position
or
because
they accept
a "no
opinion"
filter.
It
may
seem
surprising
that
only
as
few as
35
percent
or so
fail
to
give
content
locations
on
two
successive
administrations
of this
ideo-
logical
scale,
and
hence
do
not
figure
in the
continuity
correlations,
when
for
many
years
items
asking
people
what
the
labels
"liberal"
and
"conservative"
mean
have shown
some
40
percent
or so
who
do
not
know.
The
implica-
tion
is
either that some
people
who would
not
profess
to
know
what
political
difference
is
intended
are
nonetheless
choosing
a
location
on
the
continuum,
or that
the
40 percent
value has
changed,
although
it had
not
changed
even
after
we were
deep
into the
turmoil
of
the
1960s.
Unfortunately,
the item
tapping
the
meaning
of
the
terms was
never
asked
in any
of
the
five
waves
of
the
new
panel.
However,
work
in
Western
Europe
has
made
it
clear that people
who
locate
themselves
on
an
analogous
left-
right
scale
despite
confession
that they
do not
understand
what the terms
mean
politically,
tend
in
large
numbers
to
place
themselves
at the
exact
midpoint
of
the
continuum
offered (e.g.,
Converse
and
Pierce,
1970).
This
is
an
easy
solution
when
one
is
unsure of the significance
of
the extremities.
And
indeed,
the
midpoint
of
the
United
States
scale
is
very
well
populated
also.
In
fact,
in the turnover
cross-tabulations
of
the liberal-conservative
scale,
the
cell
which
marks
the
exact
center on both
administrations
tends
to be
two
to three
times
as
populous
as
any
other
cell
in
the matrix.
Hence
the
special
circumstances
which
ap-
pear
to sustain
the
high
liberal-conservative
continuity
correlation
seem
to be that
better
than one-third
of the
population
is
removed at
the
outset;
and
among
the
remaining
two-
thirds,
substantial
numbers
express
their
uncer-
tainty
by
persistently
locating
themselves
at
dead center,
thereby
not
adding
to the
con-

1979
The New
CPS
Election
Study
Panel
43
tinuity
correlation,
but
not detracting
much
from
it either.
It is
likely
that
for those
with
truly
crystallized
positions
on the
liberal-con-
servative
continuum,
the individual-level
stabili-
ty
of position
is
very high,
even
as
high or
higher
than
party
identification.
But this
frac-
tion
of the
population
is likely
to be
quite
limited.
In our discussion
we
have
introduced
a
number
of issues
or
meta-issues
which
were
not
present
in the
original
panel,
but are
new
entrants
in the
1972-76
period.
All
of these
items
discussed
show
higher
levels
of continuity
than
the
most
stable
issues
in the
original
panel.
This
might
lead
to
a
conclusion
that levels
of
continuity
in issue positions
are up
quite
notably,
if not
because
of
heightened
continu-
ity
from
issues
that can
be
specifically
matched
across the
panels,
at
least
because
issues
newly
arisen
in
the
real world are of
a
kind
that
tap
more sharply
crystallized
opinion.
One
can
defend
a
very
modest
version
of this
argument
if one
limits
attention
to the
two or
three
moral
issues.
However,
a number
of other
"new
issues"
in the
panel
have not
been
discussed,
and these
generally
show
levels
of
continuity
that
are
utterly
redolent
of
issues
in
the
1956-60
period.
Thus,
for
example,
two
items
which
have overtones
of
the
"law
and
order"
themes
of the
past
decade-one
dealing
with
repression
of
urban
unrest,
another
dealing
with
protection
of
the
rights
of
accused persons
as opposed
to stopping
crime
at all
costs-show
two-year
continuity
coefficients
of
about
.36
and .45
respectively,
values
which are
very
middling
relative
to the
range
"normal"
for
domestic
issues
late
in the
1950s.
An
item
addressed
to the
possibility
of
a
more
or less
progressive
taxation
rate
shows
a
four-year
value
of
about
.31,
and the
figure
is
only .28
for
an
item
asking
whether the
federal
govern-
ment
has become
too
powerful.
The latter
figures
are lower
than
any
parallel
figures
to
be
found for
domestic issues
in the
original
panel.
To
summarize,
then,
we
have found that
where specific
issues
can
be
directly
matched,
continuity
values seem
amazingly
stationary
across
the two
panels.
One
type
of issue
not
salient
in
political
debate
in the late 1950s-
moral
or counterculture
issues-do
show
signally
higher
continuity
values
than
any
other
specific
issues
in
either
panel.
Once
beyond
these moral
issues,
however,
even
new
issues
do
not
in
general
show
any
higher
levels
of
continuity
than the
old.
A
great
deal of work of
a
more
probing
sort
can be
done
with
these
issue materials.
For
example,
more
intensive
investigations
of
the
apparent
substantial
error
structures
in these
responses
are possible, particularly
as some
few
of the
new panel
issues
have
been applied
on
four waves
instead
of merely
three.
Such
empirical
wedges
are
important
to arrive
at a
proper
diagnosis
as to
what these
error
struc-
tures
really mean.
Achen
(1975)
has concluded
that
the low
levels of
continuity
marking
issue
responses
occur
simply
because
the items
are
unreliable,
probably
because
of
poor
wording.
It was
the amazingly
low
"test-retest"
reliabili-
ty
of the
1956-58-60
panel
issues
that
touched
off our
earlier
investigations
as
to why
this
degree of
unreliability
occurred,
and
led to
our
conclusion
that the
low apparent
reliabili-
ties
were
not
simply
a function
of the
items
themselves in
a
shoddy-wording
sense,
but
rather
were a
joint function
of
the substance
of
the items and
the variable
degree
of
attitude
crystallization
which
more and
less attentive
respondents
brought
to
the particular
substance
(Converse,
1964,
1970).
We
still
feel our interpretation
is preferable
to Achen's
revision.
Among
other
things,
Achen's
explanation
does
not
give
us much
help
in
understanding
why
high political
elites
show
higher
continuity
coefficients
and
hence
ap-
parently
higher
reliabilities
than does the mass
public,
even when asked to respond to
the
very
same
simplistic
issue
items,
although
this
dif-
ference
is the
most
direct
corollary
of our
original
interpretation.
Nor
does the Achen
interpretation
prepare
us to
expect
that the
same team
of
question-writers
who
manage
to
word
questions
quite
effectively regarding
per-
sonal
morals
cannot
find
comparably
effective
wording
when
asking
about
foreign
policy
debates.
Again,
we
need
only imagine
that
disagreements
about
personal
morals
grip
more
people
more
vitally
than do
debates
over
foreign
affairs
in order to fold
the
new
findings
nicely
into
our
original
interpretive
structure.
The
important
point
is
that the new
panel
data
base offers
a
challenging
field for
in-
vestigating
some
of
these differences
more
thoroughly.10
For the
moment,
however,
it
seems
fair
to conclude
that the
individual-level
relative
stability
of issue
positions
in the Ameri-
I0We
postpone
consideration
of
the
detailed parti-
tioning
of
the
issue items into
reliability
and
stability
components
until a subsequent
technical
paper,
since
a
proper
treatment
requires
taking
account
of
differing
response
variances
in
different
population
segments.
These
complications
are too elaborate
for
space
available
here.
However,
it
may
be of interest
to
note
that
a
simple partitioning
of these
issue
items, ignoring
the
complications,
suggests
that the
unreliability
com-
ponent
of
responses greatly
outruns
a
true
change
component,
just
as was true
for the
original panel.

44
The
American
Political
Science
Review
Vol.
73
can public
has changed
only
slightly between
the two
panels,
and that no
significant
change
whatever
can
be discerned
where
the same
issue
substance
can
be
matched
across the
two eras.
Evaluations
of Political
Figures
Of
the
original
Survey
Research Center
triad
of
parties,
candidates
and
issues,
all empirical
work
of a
panel
type
has focused
to date on
parties
and
issues,
for
the simple
reason that
the
1956-60
period
bridged
two
presidential
elec-
tions with no
continuity of
candidates.
The
same
lack of
continuity
marks
the 1972-76
interval.
However,
the new panel
does
include
multiple
evaluations,
on
a
thermometer
scale,
of the most prominent
partisan
political
figures
in the land,
even if not
running
for president.
It
is
naturally of interest
to ask
how stable
these
personality
evaluations
are, by
comparison
with
the
high
stability of
party identification
and the
relatively
low
stability
of personal
positions
on
political
issues. Therefore we shall
close our
review
of
some
of
the
first
panel
findings
with
a
brief consideration
of these raw
continuity
results.
Ratings
of Hubert
Humphrey,
Ted
Kennedy,
Scoop
Jackson,
Richard
Nixon
and
George
Wallace were drawn
in
1972,
1974
and
1976.
In
addition,
a
rating
of
George
McGovern was
applied
in
1972
and
1976.
Over
all
instances,
the
100-point
thermometer
scale was
explained
and
then
names were offered seriatim
as
stimu-
lus
words
for
evaluation.
The
continuity
corre-
lations
available
for these
political
figures
are
summarized
in
Table 1. A
number of
provision-
al
observations
about this
array
are
fairly
obvious.
Table
1.
Continuity
Correlations in
the
1972-76
Panel
Six Major
Political
Figures
Continuity
Correlations
Average,
2-Yr. 4-Yr.
Ests.
Ests.
Edward
"Ted"
Kennedy
.722
.671
George
Wallace
.683
.609
Hubert
Humphrey
.592 .531
Richard
Nixon
.585
.461
George
McGovern
-
.554
Henry
"Scoop"
Jackson .445
.343
Source:
1972-74-76
panel study
of
the
national
'elec-
torate carried
out by
the
Center for Political
Studies,
University
of
Michigan.
One
is that
the levels
of
continuity
correla-
tions
across
six of the most
prominent
political
figures
of
the
era
covered
by the
new panel
show
a
very
wide range of
variation.
While
it
is
not
perfectly
certain
in
what currency
such
comparisons
should
be made,
we
find it
reason-
able
to
square
the above correlational
values
and
speak
in
terms
of the
proportions
of
temporal
covariance involved
in the
various
evaluations.
In this
currency,
evaluations
of
Kennedy appear
some
21/2
to 4 times
more
stable
than
evaluations of Jackson.
Given
such
variation,
it
is tempting
to
try to
explain its
sources.
One
hypothesis
might
be
that
the
stimulus words
"Henry
'Scoop'
Jackson"
are
more
poorly
worded
as question
items
than the
stimulus
words
"Edward
'Ted' Kennedy,"
and
that
hence
the
Jackson
item is
simply
less
reliable. A plausible
rival
hypothesis,
and
one
that we considerably
prefer,
is
that more
American
voters
bring
more
meaning
to
the
Kennedy
name
than to the
Jackson
name, have
more
crystallized
attitudes
toward
it as
a
stimulus,
and
hence
respond
more stably
to it.
The
apparent
reliability
is
not, by
this construc-
tion,
a
function of the
measuring
instrument
taken
in
a
vacuum,
but rather
a
product
of the
interaction between item content
and the
vari-
able
degree
of
meaning
this content
has
for
various
citizens.
However,
a
second observation
helps
to
keep
us
from considering
these
continuity
correla-
tions
too
woodenly
as
though they
were
test-
retest reliability
coefficients.
Although
it may
not
dazzle the
unaided
eye,
the
pattern
of
continuity
coefficients for evaluations
of
Rich-
ard Nixon
are
quite
unique
relative
to
all
other
contents of
the
new
panel discussed
to date.
We
know from
the panel
marginals
(as
if
we needed
them!)
that true
change
in
evaluation
of
Richard
Nixon
between
1972
and
1974
or
1976
was
of massive
proportions.
In
1971,
according to
the Gallup
Poll, the
American
people
considered
Richard
Nixon
the most
admirable
person
in the entire
world.
His
thermometer
ratings
in
the
1972
wave
of
the
panel
confirm this
level
of
adoration,
since
he
received
a
mean
rating
of
65-66
in
that
year,
relative to a
rating
of 52
or 53
for popular
figures
like Kennedy
or Humphrey.
By
1974
he
had
tumbled
into
the
high 30's,
and
by 1976
to
31.
Thus there
is
a
major injection
of
real
change
in
the Nixon
case,
although
as we have
seen,
it
may
or
may
not
have been
a
type
of
change
that
affects the
stability of
relative
individual
positions.
That
is,
it remains
possible
that
people
who
were
wildly
enthusiastic
about
Nixon
in
1972
merely
were
reduced
to luke-

1979
The
New
CPS Election Study
Panel
45
warm
feelings
about him by
1976;
and
those
just
mildly
warm
were
reduced
to
cool
feelings,
etc.,
such
that
there
was
little
leap-frogging
of
individual
positions.
What
is peculiar
about
the
Nixon
pattern,
however,
is
that
a
good
deal
of
true
change
(in the
latter sense)
was
touched
off
by
the
Watergate
drama.
If
we
make
the
relevant
Heise
calculations
for the
Nixon
case,
we
find that
quite uniquely
relative
to
all other
panel
materials,
true
change
competes
on
nearly
equal
terms
with
unreliability
in
explaining
why
the raw
continuity
coefficients
are
as
low
as
they
are.
A
final observation
concerning
Table
1
returns
us
to the
main
themes
of this
essay.
We
introduced
the
panel
data
on
personages
by
way
of asking
where
"candidates"-broadly
construed-fit
in the hierarchy
of
continuities
defined
up
to now
by parties
and issues, given
prior
panel
material.
If
we are
correct
in
our
interpretation
of
the
reasons
why
Kennedy
evaluations
are so
much
more
stable
than
those
for
Jackson,
then
it
should
be clear
that
we
could generate
variations
in
these
continuity
coefficients
for
political
leaders
almost
at
will,
by playing
with
the known
prominence
of
such
figures.
In the limiting
case,
we
could
probably
generate
a continuity
coefficient
approaching
zero
with
a
stimulus
like "Senator
Zablocki,"
although
it
is
reasonable
to imagine
that
a
persistent
"ethnic
vote" might
keep
such coeffi-
cients
from actually
touching
zero.
Nevertheless,
while
it would
be
a
chimera
to
imagine
that some
restricted
range
of continui-
ty
coefficients
could
reliably
encompass
any
political
figures
whatever
(by
comparison
with
parties
or
issues),
it
is
reasonable
to
limit our
attention
to
the
few
truly
prominent
(in
a
mass-public
sense) political
personages
in
the
land,
just
as
the
authors
of the
voting
surveys
try
to restrict
their issue
measurements
to
the
most salient
policy
cleavages
of
any particular
period.
And
if we
make such
a limitation,
then
the
data in
Table
1,
coupled
with
those
cited
earlier
for
parties
and
issues,
might
strongly
suggest
that
in
the
hierarchy
of continuity
that
these
panel
data
help
to
establish,
evaluations
of
major
political
leaders
might
properly
be
wedged
in between
parties
as the
most
stable
objects
of
orientation
in terms of relative voter
positioning,
and
issue
positions
as the
least.
Thus
major
political
personages
in
any given
epoch
may
serve
as
important
cynosures
of
attention
for
the
electorate,
and as
significant
cues
for
other more
substantive
forms
of
political
evaluation.
Summary
and
Implications
Figure
1
provides
a
handy
graphic
summary
of
the continuity
correlation
information
that
we
have
reviewed.
In
a
nutshell,
this
summary
says
that
apart
from
the
emergence
of
some
political
cleavages
having
to
do
with traditional
moral
values
in
the
later
period,
there
has
been
scarcely
any change
in
the
comparative
continu-
ity
of
party
and
issue
positioning
between
the
two
eras,
despite
manifold
reasons
to
expect
not
only
change,
but
change
of major
propor-
tions.
The individual-level
continuity
of
party
attachments
continues
to
dwarf
that
for
issue
positioning
in
the
1972-76
period,
as
it did
in
1956-60.
And affection
and
dislike
for
the
most
prominent
political
leaders
of
the
epoch
seem
to show
levels
of
continuity
which-again
absent
the
moral
issues-clearly
lie
between
parties
and
issues.
The display
in Figure
1
is
rich
with
implica-
tions,
some
few
of
which
we
shall
develop
here.
First,
however,
let
us
consider
an
apparent
implication
or
two
that
should
not
be
drawn,
particularly
with
regard
to
the
surprising
infor-
mation
concerning
the
continuities
in
party
identification.
We
have
emphasized
before
that
these
continuities
do not mean
that
evidence
of
weakening
partisanship,
as
registered
by
change
in the party
identification
marginals
since
1965,
is
somehow
imaginary
or
meaningless.
It
should
be kept
clear that
Figure
1 does
not
even
address
this
facet of change
in
any
direct
way,
even
though
it
is
very
tempting
to
confuse
the
two.
Thus,
for example,
one
improper
implication
to be
drawn
from
these
data
is that
since
the
individual-level
continuity
for
panel
respon-
dents
remains
high
while
partisanship
is
weak-
ening,
it
must
mean
that
the
weakening
is
all
attributable
to
new
entrants
into
the
electorate
who
do
not
figure
in
a
long-term
panel
table.
Hence
it would
seem
that
one
major
point
to
these
data is
that the
weakening
can be
traced
entirely
to
population
replacement
processes.
However,
such
a
conclusion
again
involves
a
subtle
confusion
between
the two
facets
of
change.
Clearly
generational
replacement
has
made
a
major
contribution
to
the
weakening
of
aggregate
partisanship,
especially
in
the
later
portions
of
the
past
ten
years.
But
with
equal
clarity,
persons
already
in
the
electorate
before
the
decline
began
have
also made
further
important
contributions
to
it,
and
have contin-
ued to
do
so
all the
way
to 1975
(Converse,
1976).
Such
persons
do
figure
in our
panel
tables,
and
indeed
they
show
a
weakening
of
partisanship
even within
those
tables:
over
all
segments,
including 1964-66,
1965-73
and

46
The
American
Political
Science
Review
Vol.
73
PART I
ES
LEADERS
ISSUES
1.00
j
I
I
1.00
0.90
0.90
0.80
0.80
0.70
0*70
(1)
~~~~~~~~~~~~0
*M
07
z
<
J
0.60
M
0.60
_j
_j
~~~~0
gm
UJQ
C
30
0.50
o.)
0
0.50
l
l
l
.f.
D
0.50
>_
a
0
D<;
D|
~~-
~~~?
~0
D *D
zD
I!.4|l| (lc '?
0
C~~~~~~~~~~
z
%-I
F
0.30
3
0.30
*D
0.20
0.20
M
=Morol
Issues
C
Civil
Rights
0.10
D
Domestic
-0.10
Issues
F
Foreign
issues
c
I I
I
I
o
(00
k(
k1%,
(00 ~
(O
(0
A
_0
0)
Source:
1972-74-76
panel
study
of the
national electorate
carried
out
by
the Center
for
Political
Studies,
University
of
Michigan.
*Where
items can be
plausibly
matched
between
panel
segments,
the
respective
data
points
are linked
by
a
trend
line.
**The
whole
figure
is
expressed
in the
currency
of
two-year
continuity
correlations. Where
only
four-year
correlations are
available, plausible
two-year
interpolations
have been made
to
permit
inclusion
of
a
maximal
number of
observations.
Figure
1.
Continuities
in
Public
Response
to
Parties,
Leaders and
Issues*

1979
The New CPS
Election
Study Panel
47
1972-76,
later
observations
on
these
people
show lesser variance in partisanship than
earlier
ones, although continuity
levels remain
intact.
Thus
the decline
in
reported
strength
of
party identification over the past decade
re-
mains an important but virtually independent
fact. Moreover, there is evidence that
defection
rates for
given
levels
of identification
strength
have
been higher
in recent
years than
they were
in
the 1950s. Both changes attest to
a weaken-
ing of partisanship
in
the electorate.
Both are
thoroughly documented,
and
both
have had
real
effects on the operation
of the American
political system.
For
many analytic purposes
and
prognostica-
tions,
the
weakening
of
partisanship can
be more
important information than the near-constancy
of
continuity
correlations summarized in Figure
1.
For
other
purposes, however,
the
reverse
is
true, and
it remains
for us to suggest
what some
of the
genuine implications of Figure
1
would
seem to be.
Some
of these
implications
are
relatively
technical or
methodological. Thus for
example,
the kind of
cycling
of coefficients over
time
periods that we carried out with respect
to
party identification on page 39
makes
sense
only
if the facet
of
change
indexed
by
the
continuity coefficients
is
essentially
stationary,
as
we
have
found
it
to be. More generally,
the
virtual
lack
of
period dependency
of
the
kind
of
estimates we
have
generated
from the
panel
can provide a potent assist for a
variety of
structural
modeling
ventures
aimed
at
sorting
out
in a
more
satisfactory way
the
causal
interplays
between
evaluations
of
parties,
issues
and candidates. To
be
carried out
effectively,
such
enterprises presume knowledge
of things
like
measurement
reliability. Since
such
infor-
mation is
rarely available, investigators
often
plug in quite fanciful guesses in
order to
proceed. The panel data
in
themselves
provide
better-grounded
estimates.
And
the fact that
these
estimates seem
quite
invulnerable to
changes
in
period-even
when one
would
least
dare
to
expect
them
to
be,
as in
the
case of
partisanship, 1964-76-can
be an important
liberating
fact for
the
conduct of
further
research.
The
findings
have
provisional
implications
for research methods which
are
broader still.
For some
time,
the
cost-benefit
ledger
for
panel
designs,
as
opposed
to less
expensive
indepen-
dent
replications,
has been
subject
to
some
dispute.
One
can
learn
a
good
deal
about
important
facets of
change
without
making
the
extra investment
in
a multi-wave panel.
How-
ever,
other facets of
change,
such as those
central
to this
report,
can
only be
observed
with
the
help of
a
panel. One swallow does not
a summer make, and it would obviously be
premature to conclude that the stationarity of
continuity coefficients which marks these poli-
tical data would emerge for panel studies in all
periods with any subject matter. On the other
hand, these continuities seem so pervasive
across a variety of types of variables in the
election panels that the hint of universality is
hard to miss. And if there is some generality to
the
findings, then the premium that might be
placed on doing panels persistently over a series
like the election studies is at least somewhat
reduced, although occasional panel links would
remain
important.
The
findings also
have
strong bearing on
various substantive questions. For example, one
of
the points of most vigorous theoretical
ferment in the past decade in electoral politics
has
involved the
nature
of partisan realign-
ments. The subject has captured uncommon
scholarly
interest not
merely
because
of its
obvious importance
in
any theory
of
elections,
but also because of the
seeming
imminence
of
such a
realignment
after
1964
or so.
Given
past
periodicities, the time was ripe. Much more to
the
point,
several
key ingredients
associated
with
past realignments
had
begun
to
emerge.
Very crudely put,
the most
important
of
these
ingredients was the outbreak of uncom-
monly galvanizing political turmoil, of the sort
which
certainly
marked the
late
1960s
and
early 1970s. More finely drawn,
we
have the
excellent effort of Sundquist (1973) to recon-
struct
the internal dynamics of a partisan
realignment,
with stress
upon
crucial
emergence
of
important cross-cutting cleavages
which
the
leaderships of the extant traditional parties
bound to older
cleavage lines,
fail
to articulate.
Again,
such
ingredients
were
present, par
excel-
lence, by the
late
1960s. That these new
cleavages
were
gripping for
common voters
is
attested to
at least
indirectly by
the
high
continuity
values associated
in the
new
panel
with items
like busing
and the
life-style
or
moral issues. That these
new
issues
posed
cleavages which were truly cross-cutting (i.e.,
relatively orthogonal to or
uncorrelated with
prior
lines of
party alignment) is clearly
demon-
strated
by Miller and Levitin (1976, p. 82),
who
show the lack
of
much association
between
voter persuasions
on
the
newer issues
and their
party
identifications.
One must insist
that the
new issues be
cross-cutting
ones to set the
stage
for a
realign-
ment,
since otherwise the
emergence
of new
issues more or
less
aligned
with
old
ones would
merely
reinforce
existing party
differences.
In
the
late
1960s
and the
early 1970s,
some

48
The
American
Political
Science
Review
Vol.
73
Democrats
were
affronted
by
counterculture
trends while
others were
not; likewise,
some
Republicans
approved
while others
disap-
proved.
What is important
about these dif-
ferential
within-party
affinities is that if
potent
enough,
they should
produce changes in parti-
sanship
of
a
leap-frogging
kind, disturbing
the
relative
ordering of
individuals on the continu-
um of
partisan
loyalties.
However,
it is
exactly
this
possibility
of
leap-frogging
that our con-
tinuity
correlations
are
admirably designed
to
register,
however
insensitive
they may be
to
other
forms of
change.
And they give us
the
message
that
the
leap-frogging type
of
change
in
partisanship
was
scarcely more
prevalent
over
the whole of
this
period
than it had
been
in
1956-60.
Hence while
the
emergence
of cross-
cutting
cleavages seems
to have been real
enough,
it also seems clear
that they did
not
produce the
relevant
changes
in
partisanship in
this
period,
even
of
the
compensating kind
which
leaves
the
net
balance
of
partisanship
in
the
aggregate
unaffected.
Another
set
of
implications
flows
from
the
panel
data
patterns of the
issue
items. Our
original
theorizing led
us to imagine that
constraint
among belief
elements measured at
a
point
in
time
and the
stability
of those
beliefs
for individuals
across time
would
tend to
covary,
although no
empirical
test
of
this
assumption
between periods of
time has
been
possible
until the new
panel.
In
effect,
if
Nie's
(1974) demonstration
of increased constraint
in
the
recent
period
has
substantive
merit,
then
we
would
be
obliged to expect
the stability of
these
types
of items
to
advance
in the
new
panel
as well.
The fact
that
no
such
increase
is
observed
poses
an
interesting problem. One
possibility
is
that
the
original
assumption,
which
sees
varia-
tions
in
attitude
crystallization
as
producing
variations both
in
constraint
and in
stability,
is
simply
wrong
and
needs to be
rethought.
Another
possibility which,
as
we
have
seen,
has
already
been
raised
(Bishop
et
al., 1978; Sulli-
van
et
al., 1978) is that the
Nie displays are
largely a
product of changes
in question
format,
rather than true
increases
in
constraint.
At
first
glance,
the
accrediting
of such
challenges might
seem to solve
the
problem
by removing
the
weight
of the
expectation
that
issue stabilities
should
have
advanced in the
new
panel.
How-
ever,
even if
the Nie
results
were in fact
largely
due
to
question
format,
one
might
still
expect
comparable variations
in
stability
on
the same
base. Now
we
saw as we
passed through
the
issue
items that
small
changes
in
continuity
values
did appear to follow
in
expected ways
along
lines of
shifts
in
question
format.
How-
ever,
these latter
tests
were
few and
at points
equivocal,
so
that
not a great
deal
should
be
made
of
them. But
in
any
event,
even these
changes
were
of very
limited
magnitude.
Thus
some fundamental
problems
remain
which
chal-
lenge
further
research,
at least
some
of which
can
be carried
out
with other
probings
of the
new
panel data
set.
A
final cluster
of
implications
which is
of
great
theoretical
importance
flows
from the
hierarchy
of continuities
across
parties,
leaders
and
issues suggested
in
Figure
1, and
is relevant
for
controversies
over
the
prevalence
of issue
voting. It is
not hard
to imagine
that
attitudes
which show higher
individual-level
stability
have
causal
primacy
relative
to less
stable
attitudes.
If this simple
surmise
is
meaningful,
then it would seem that
where
political
evalua-
tions
are concerned,
a
party
-*
leader
-*
issue
flow
may be
dominant,
even across
periods
as
disparate
as those
before and
after
1960.
It
must be
kept
in
mind that
the
new panel
data
are
merely suggestive
in
this
regard,
rather
than
definitive.
Even the gross
discrepancies
in
continuity
between
partisanship
and
issue posi-
tioning
cannot guarantee
that party
identifica-
tion is
nearly
always
causally primary
when
party
feelings
are
discovered
to be aligned
with
issue preferences.
Among
other
things,
we must
remember
that
there are
only
two
major par-
ties,
whereas the number
of
political
issues
which
may
exercise one
or another voter
is
legion.
Thus
it would always
be possible,
if
indeed
a
bit
strained,
to
imagine
that
the
typical
voter
might
be
viscerally gripped
by
one
issue out of
some
30
in the
active
universe
in
some
period,
with
very
uncrystallized
attitudes
on the
other
29.
If
party
positions
on such
crucial
but idiosyncratic
issues remained
con-
stant
over
long periods
of
time,
these
policy
preferences
might
totally
dictate
party
identifi-
cations,
at
the very
same time that
measure-
ment over a dozen or so of
the
30
potential
issues
might
show
very high
aggregate
instabili-
ty,
since
only
one
voter
in 30
would,
on the
average,
be
gripped
by
each.
While
such a
model remains
conceivable,
it
is
obviously
not
the most
plausible
reading
of
current data.
The
huge
discrepancies
in
indivi-
dual-level
continuities between
party
loyalties
on
one
hand
and
issue
positions
on
the
other,
particularly
in view of
what now seems
to be
their
long-term
constancy,
argues
for
an
over-
whelming primacy
of
the
party
term
when
party-issue
congruence
does
occur.
However,
further causal
modeling
work with the
new
panel,
profiting
from some of
the kinds of
estimates presented
in this
report,
can mold
more
incisive
conclusions
from what must

1979
The
New
CPS
Election
Study
Panel
49
remain only suggestive ones at this point.
The additional fact, also made clear by the
new panel data, that reactions to political
leaders display levels of continuity intermediary
between partisanship and policy preferences,
raises a number of further interesting possibili-
ties. Such a data configuration may, for exam-
ple,
suggest a significant margin for policy
leadership available to major political figures,
whereby admirers
are
quite susceptible to influ-
ence leading them to adopt policy positions
more
congruent
with
those
espoused by
their
heroes. Or again, the greater continuity
in
evaluation of leaders as compared with issue
positions may simply
reflect the fact that
party
attachments
anchor both
leader assessments
and
issue
positions,
but
do
so more
effectively
in
the case of leaders because the common
voter
can
maintain much more firm and unequi-
vocal cognitive
links
between
parties
and their
most
prominent leaders than between parties
and positions on various issues.
This is
obviously
not
the place to begin
sorting
out
such
possibilities. However,
the
first-cut results from the new CPS panel are
striking
indeed.
They already
rule
out some
constructions
which
might
be
placed upon
the
electoral history of the past decade, and begin
to
draw
into focus
a
number of
new research
questions
which,
with more
refined "milling"
of
the
panel data,
we
may
well
be
able to
answer
effectively.
Bibliography
Achen, Christopher
H1. (1975).
"Mass Political
Atti-
tudes and
the
Survey
Response."
American
Politi-
cal
Scien
ce Revie w
69:1 218
-3
1.
Bishop,
George F.,
Alfred J. Tuchfarber,
and Robert
W. Oldendick (1978).
"Change
in the
Structure
of
American Political
Attitudes:
The Nagging Ques-
tion of Question
Wording."
American Journal
of
Political
Science
22:250-69.
Brody, Richard
A.
(1977).
"Stability
and Change
in
Party
Identification:
Presidential
to Off-years."
Paper
presented at
the annual
meeting of
the
American
Political
Science Association,
Washing-
ton,
D.C.
Burnham,
Walter
Dean
(1968).
"American Voting
Behavior
and the
1964 Election."
Midwest Journal
of
Political
Scietice
12:1 -40.
Campbell,
Angus, Philip
E. Converse,
Warren E. Miller,
and
Donald
F.
Stokes (1960).
The American
Voter.
New
York: Wiley.
Converse,
Philip E. (1964).
"The
Nature of
Belief
Systems
in
Mass
Publics."
In David
E.
Apter (ed.),
Ideology
and Discontent.
Glencoe:
Free
Press, pp.
206-61.
(1970).
"Attitudes
and Non-Attitudes:
Contin-
uation
of a
Dialogue."
In
Edward R. Tufte
(ed.),
The Quantitative
Analysis
of
Social
Problems.
Reading, Mass.:
Addison-Wesley,
pp.
168-89.
(1975).
"Public
Opinion
and
Voting
Be-
havior."
In Fred
I.
Greenstein and Nelson
W.
Polsby (ed.),
Handbook of Political
Science, Vol.
4. Reading,
Mass.: Addison-Wesley.
_-
(1976). The Dynamics of Party
Support.
Beverly Hills:
Sage.
and
Roy
Pierce
(1970).
"Basic
Cleavages
in
French Politics
and the Disorders
of May and June,
1968."
Paper
presented at
the
Seventh World
Congress of
Sociology, Varna,
Bulgaria.
Heise, David
R. (1969). "Separating
Reliability and
Stability
in Test-Retest Correlation."
American
Sociological
Review
34:93-101.
Jennings, M. Kent
and Richard G.
Niemi (1977). "The
Persistence
of Political Orientations:
An Overtime
Analysis
of Two
Generations."
British Journal of
Political Science
8:333-63.
Margolis, Michael
(1977). "From
Confusion to Confu-
sion: Issues
and
the American
Voter
(1956-1972)."
American Political
Science
Review
71:31-43.
Miller, Warren
E.
and Teresa
E. Levitin (1976).
Leadership
and
Change. Cambridge,
Mass.: Win-
throp.
Nie, Norman
H.
with Kristi
Andersen (1974). "Mass
Belief
Systems
Revisited:
Political Change and
Attitude Structure."
Journal
of
Politics
36:541-91.
Pomper, Gerald
M. (1972). "From
Confusion to
Clarity:
Issues
and American
Voters, 1956-1968."
American Political
Science
Review 66:415-28.
Sullivan,
John
L., James
E.
Piereson,
and
George
E.
Marcus
(1978).
"Ideological
Constraint
in
the Mass
Public: A
Methodological Critique
and Some New
Findings."
American Journal
of Political Science
22:233 -49.
Sundquist,
James
(1973). Dynamics
of
the
Party
System:
Alignment
and
Realignment of
Political
Parties in
the United States.
Washington, D.C.:
Brookings.
Wiley,
David E. and
James A.
Wiley (1970).
"The
Estimation
of Measurement Error
in
Panel
Data."
American
Sociological
Review.
35:112-17.