http://cps.sagepub.com/Comparative Political
Studieshttp://cps.sagepub.com/content/2/2/139The online version of
this article can be found at:DOI: 10.1177/001041406900200201 1969
2: 139 Comparative Political StudiesPhilip E. ConverseOf Time and
Partisan StabilityPublished by:http://www.sagepublications.com can
be found at: Comparative Political Studies Additional services and
information for http://cps.sagepub.com/cgi/alerts Email Alerts:
http://cps.sagepub.com/subscriptions Subscriptions:
http://www.sagepub.com/journalsReprints.nav Reprints:
http://www.sagepub.com/journalsPermissions.nav Permissions:
http://cps.sagepub.com/content/2/2/139.refs.html Citations: What is
This?- Jul 1, 1969 Version of Record>> at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from at
St Petersburg State University on November 4, 2013 cps.sagepub.com
Downloaded from at St Petersburg State University on November 4,
2013 cps.sagepub.com Downloaded from at St Petersburg State
University on November 4, 2013 cps.sagepub.com Downloaded from OF
TIME AND PARTISANSTABILITYPHILIP E. CONVERSEPHILIP E. CONVERSE is
Professor of Sociology and Political Science atthe University of
Michigan. His principle field of concentration is socialpsychology.
He isthe author of numerous articleson voting behaviorand political
alienment.IT IS A COMMONPLACE expectation that newly established
social andpolitical institutions somehow accumulate a deepening
stability withthe passage of time.In the very long run, of course,
once-establishedinstitutions may obsolesce, evolve into new forms,
or be more brusquelyoverturned. But the timing of these terminal
events, embedded as they willbe in a complex nexus of change, seems
almost impossible to forecast atlongrange. What is utterly
predictable, however, isthat threats to thesurvival of a new
institution will be very high in its infant stages. In thedegree
that itcan outlast these first precarious periods, it will
typicallyhave put forth roots of sufficient strength so that
challenges to its verybeing diminish in number and those that do
occur are warded off withincreasing ease.Itbecomes sanctified and
protected by &dquo;the weight ofhistorical
tradition.&dquo;Itisacrossthis sequence, from birth through
initialvicissitudes toestablished maturity, that currently
fashionableinterestsin &dquo;political*This paper was prepared
with the benefit of a Senior Post-Doctoral Fellowship/~o~ ~
~Vaf/o/M/~c:eMC~ FoM~Jaf/on.7~ cdd!~OMfof/:M H/ppo~, ~e at~Ac~from
theNational Science Foundation.In additiontothis support, the
authorgratefully acknowledges the indispensable aid of Mrs. Jean
Dotson and Mrs. Sarah(Barker) Horack in the preparation of the
empirical materials.[140]development&dquo; on one hand, and
&dquo;politicalstability&dquo; on the other,
clearlycoincide. New political forms are best nurtured in the haven
of politicalstability; once matured, they can act in turn as a
significant force towardthe maintenance of political
stability.Whilethese sequences arefamiliar
enough,systematicempiricaltreatment of the temporal processes
involved seems rare. Indeed, while thenatural history metaphor we
have used is both classic and compelling, itdoes not seem congenial
to quantification. On what scales, for example,are we to gauge the
&dquo;weight of historical tradition&dquo;?Yet it seems
evidentthat patterns of slowlyaccumulatingstability within new
social formsreflect temporal processes of habituation or
&dquo;socialization&dquo;on the part ofthe many factors
involved. In short, they are cases of social learning, and assuch
should be susceptible to quite rigorous specification.The main
purpose of this paper is to attempt such a specification, in
alimited though farfrom trivialcase.We are interested generally in
theadventand progressive stabilizationof democratic political
institutionsthat has represented perhaps the most noteworthy and
sweeping politicalchange over an ever broadening circle of nations
for the past century and ahalf.This &dquo;progress&dquo;
has, of course, occurred incrabwise fashion, withbackward
stepsnearly as frequent as forward ones at many times andplaces.
Yet these crabbed steps are part and parcel of our sense of
thevicissitudesof earlydevelopment;
wherenationshavemaintaineddemocratic forms for a sufficient period
of time, the implantation seemscomplete, and further reverses
nearly inconceivable.It would be self-defeating to try to treat the
development and stabilityof democratic forms in any general sense,
without further focus.
While&dquo;politicalstability&dquo;may be a unitary concept
at some abstract level, itsreferents and concrete indicators are
extremely multifarious. That there issomestatistical tendency for
symptoms of stability and instability toco-occur seems likely, and
without this there would be little meaningfuldiscussion of
stability in the abstract at all. Yet the correlation among anyfull
gamut of such symptoms ismild at best, and statements
aboutstability are likely toremain weak until more specific
components ofstability are singled out and their dynamics
understood.Historically, for example,incipient
democraticformshavebeenoverturnedbothfrom&dquo;above&dquo;and
from &dquo;underneath,&dquo; and thetwopatterns do not
always havemuch incommon. Revolutionary coupsagainst new democratic
systems have been executed by traditional eliteswithanarrowor
totally invisiblebaseof support inthebroaderpopulation. And
relatively fringe leadership, with minimal initial support[141]from
traditional elites, has succeeded in rallying sufficient mass
support toantidemocratic parties to derail a system, with the
Weimar Republic beingperhaps the prototypic case.The elite coup has
undoubtedly been the more frequent phenomenonof the two, in part
for the obvious reason that disaffectedelitesoftencontrol the kinds
of resources necessary to make seizure of governmentplausible.
Indeed, whilewe know of no systematicchartings of suchdevelopments,
itis impressionistically clear why the consolidation of anewly
established democracy isslowand unsure with respect
toeliteattacks.In most such situations, thereis initially a
residuum of elitepersonnel, which either poorly understands the
constraints of democraticvalues, or which is downright hostile to
them. As these older cohorts dieout and are replaced by younger
generations of potential elites, socializedfrom the outset in the
fundamental &dquo;rulesof the game&dquo; characteristic
ofdemocratic values, the balance of ideologies shifts progressively
and thenew forms find increasingly firm footing.The mass side of
the picture is at least somewhat more perplexing, forthere isa good
deal of evidence that the more subtle and important
ofdemocraticvaluesneverhavemuch more thanan extremely
limitedabsorption. Democratic slogans of one sortor another may
have widepopular currency, but even in the United States where mass
exposure todemocratic forms is of the longest standing in the
world, there seems littlecapacity to translate these norms into
appropriate attitudes and behaviorswithin complex,
&dquo;reallife&dquo;situations (see, for example, Prothro
and Grigg,1960). From this point of view, elites must remain
&dquo;the keepers of theflame.&dquo; 1Yet itis hardly true
that mass populations exposed for generations toelectoral
participation failto differ significantly from those passed
intothissituation for the firsttime. There isevidence of a kind of
&dquo;settlingdown&dquo;or habituation to a competitive
party system, which occurs at amass level as a secular trend over
time, and which is quite analogous to theprogressive learning at an
elite level, although it may be oriented somewhatdifferently.While
there is undoubtedly a variety of indicators of this
habituation,the one most noteworthy in our eyes isthe progressive
&dquo;binding in&dquo;ofpopular loyalties to one or another
of the traditionally competing politicalparties. Where these
loyalties have not yet had time to develop, itseemslikely that
electoral support will have numerous capricious overtones, andthat
intimes of severe distress nontraditional and antidemocratic
partiesmay find ready support. When in an intermediate stage they
become at[142]least moderately developed, such probabilities may
decline quite visibly,although it may remain a frequent strategy of
political dissent to launch anew politicalparty in hopes of
striking sufficient popularsupport tobecomeviable.When these
loyalties becomestillfurther developed,however, even such a
strategy comes to seem forbidding in the extremeand is rarely
entertained for any purpose more serious than
short-termmaneuver.Thus we conceive such aggregate levels of
loyalty to existing parties in ademocratic system as an important
ingredient of democratic stability, andperhaps, for the mass level,
the most important ingredient, although anysuch claim must remain
speculative. And it is the temporal development ofthese loyalties,
fromthe point atwhichelectoral processes arefirstinstitutedina
polity to maturational equilibrium, which provides theprimary focus
for this paper.2 2POINTS OF DEPARTUREMore clearly than usual, our
inquiry was given its first impetus fromtwo superficially distinct
bodies of empirical fmdings which have been intheliteraturefor a
number of years. We can speak in a more efficientshorthand later if
we stop to review these findings briefly now.The Individual
LifeCycle and Strength of Partisanship. The generalfinding,long
since enunciated, that people tendtovotein quiteremarkable degree
forthe partysupportedby their parents intheprecedinggeneration, has
helped tostimulatea greatproliferation ofrecent studies on
political socialization.3 The hallmark of these studies isa focus
on the political learning that has already taken place in the
decadeand a half before the individual first joins the active
electorate.It is certainly true that the individual enters an
electorate as a &dquo;biasedcoin,&dquo;particularly with
respect to such simple matters as party inclination.Yet this fact
should not be allowed to obscure the equally important fact,vivid
for anyone working exclusively on adult cohorts from age 21 to
thelate years, that significant increments of political learning
are visible overalmost the whole course of the adult participation
in the electorate. Wewould intuit that this would necessarily be
the case where the politicallandscape changes in important ways
under the adults eyes. But it is alsotrue, and rather markedly,
forsuch a simplequantum as strength ofpartisanpreference, evenwhere
the parties haveconstituted a
&dquo;fixedfirmament&dquo;sincebirth. Indeed,
inonecorrelational analysis overa[143]number of our national
election surveys utilizing a battery of the mostobvious and
standard sociological variables (age, sex, income, education,group
membership, etc.), the strongestsingledemographic correlate ofparty
identification strength was consistently the age variable, even
thoughthe samples were restricted to adult citizens and thereby
failed to take intoaccount the slope of increasing identifications
known to occur betweenkindergarten and the age of 21.What thismeans
isthat vastly differentlevels of aggregate partisanattachmenttothe
traditional parties would characterizean electoraterestricted to
voters in their twenties from an electorate of retired people.It is
useful in this regard to think of qualitative responses
(&dquo;strong,&dquo; &dquo;notvery
strong&dquo;etc.) to party identification questions as locating
individualson a quantitative continuum of probabilities of
remaining loyal to a partyon any given vote. In a two-party system,
it is natural to conceive of such acontinuum as running from .00
(will vote for party A with certainty andcould not be seduced to
vote for party B even a single stray time) throughthe .5 midpoint
(no party &dquo;leaning&dquo;whatever) to 1.0 (will vote
for party Bwith certainty). Distributed across such a continuum,
the cohort of newvoters in any election displays a decidedly
bell-shaped curve, with modalpoint inthe midrange (.4-.6). Cohorts
of elderly votersshow inclearcontrasta U-shapedcurve, with only a
minimum of frequencies stillresiding in the midrange 4In short, the
partisan stability of voting behavior tends to accumulatequite
notably over the adult years of participation in the electoral
process.Itislittlewonder thatwithmonotonous regularity in country
aftercountry, major shifts in electoral outcomes are found to have
arisen verydisproportionately in the youngest cohorts of the
electorate. New votersare flexible/unstable, and much more
responsive to new events than areolder voting cohorts.In view of
the apparent theoretical importance of the
age-partisanshiprelation, we have probed its properties in greater
detail and reported themelsewhere.s Thus, for instance,
wehaveassuredourselvesthattherelationship is truly a life-cycle
phenomenon rather than one of differ-ential politicalgenerations
inthe United States (and recent data havecontinued tounderscore
this fact). More important for our subsequentdiscussion is another
knot of findings. Since sheer time (or passage of yearsin
chronological age) is obviously not in itself a causal agent, but
rather aconvenient means of marking off how much of some more
substantiveprocess can have been achieved, we were anxious to get a
little closer tothe process itself. We asked, for example, whether
a person who takes on a[144]new party loyalty relatively late in
life builds a loyalty at faster or slowerrates than a new entrant
into the electorate. Although case numbers werelimited, the results
were intriguing: length of psychological membership ina particular
party was the critical determining variable, and not age
itself.Furthermore, thecorrelationbetween age and strength of
partyidentification, with length of membership controlled, was
actuallynegative!The latterclusterof resultsis essentially what we
would be almostcertainto find, for example, were we to study
relative fluency in theEnglish language of foreign-born migrants to
the United States. We wouldexpect to find a primary correlation
between age and fluency which, uponmore detailed examination, would
be improved if &dquo;length of residence inthe United
States&dquo;were substituted as the time variable in place of
mereage; we would alsofind that, holding length of residence
constant, theage-fluency relation would be negative, with younger
migrants more fluentthan older ones. Rates of learning would
decline with the resistances andpsychological interferences of more
advanced age. These findings againunderscore admirably the simple
processes of learning through
exposure(&dquo;socialization&dquo;) which yield the
empirical patterns.McPhee etal.have proposed a model of mass voting
that contains alearning process of Polya type as a submodel to
generate the increasingaggregateprobabilities of party fidelity as
voting experience wears on.Although this model was conceived
independently of our own work and,forthat matter,
withouttheilluminationof ourdirectmeasures
ofunderlyingpartypredispositions(apart fromcurrentvote
intention),almost all aspects of the temporal performance of the
party identificationmeasure which have intrigued us inhere as
direct logical consequences oftheMcPhee model (see McPhee and
Glaser,eds., 1962: Chapters 5[McPhee and Smith] , 6 [McPhee and
Ferguson] ; and McPhee, 1963). Thedegree of empirical fitisa
jointcommentary on the brilliance of theMcPhee insights and the
relative simplicity of learning processes that seemto be
involved.The utility of theorizing about underlying processes that
produce theage-partisanshiprelationship (rather than accepting the
relation withoutfurther thought as some natural constant) becomes
clear as we ruminateabout plausiblelimiting or boundary
conditions.Let us imagine, forexample, anew democratic
systemholding itsfirst popular partisanelections. Obviously,
ifourearlier analyses of the &dquo;meaning&dquo; of
therelationship were correct, therewould beno ground to
expectanyage-partisanship relation at all, since every
participating cohort would have[145]had equal length of exposure
(brief) to the party system. Or again, imaginethat popular
elections have held over a period of 10 to 15 years. Then
anaccurate appraisal of the underlying processes would lead us to
expect thatwhile the age-partisanship relation would be quite
nonlinear, an averageslope acrossthe age continuum would be
negative, with older citizenstaking on party loyalites at slower
rates than younger ones. Thus the mostfamiliar positiverelationship
could only be expected to emerge clearlyaftersomemoreextended
history of partisan elections.Theoldergenerations would constitute
atleasta mild lag factor in this particularrate of
&dquo;politicaldevelopment.&dquo; We shall happily have a
chance to assessthese expectations empirically at a later point
below.Intergenerational Transmission of Partisanship. While it is
clear that thepartisan impulse strengthens across the individual
life cycle, if we are tosimulate the levels of partisan loyalty in
a moving population we must takeaccountof thefactthatthis impulse
must jump acrossthe synapsebetween parent and child as the
composition of the electorate undergoesits regular turnover.6 6Here
again, our current inquiry has been stimulated by a configurationof
data we published some years ago in a slightly different context,
whichwe reproduce in part as Table1. At the time, we were
interested in thedegree to which partisan transmission mechanisms
between the generationsTABLE 1CAN RESPONDENT RECALL ANY
PARTYIDENTIFICATION FOR FATHER?seemed comparable between France and
the United States. We were quiteastonishedatthesimilarities
displayedby theresults (Converse andDupeux, 1962).[146]Let us now
approach thistableina slightly different way. Let usassume that an
averaging of the values across the two countries-already
sosimilar-willdo no more than improve our estimates as to the
generictransmission processes involved. Let us interpret the
paternal recall item asasufficient indicator of whether or not
Generation Ihad an intensity ofattachment beyond a specified
threshold (i.e., thatderivedfrom thecutpoints oftheitemonwhich
thetableis based), andtaketherespondent as a representative of the
state of affairs in Generation II. Thenthe cell entries (roughly,
.8 and .5) are the conditional probabilities that
asecond-generation member will identify with a party, given the
partisansituationofhisfather.Andthesecell entries,along
withtheircomplements, can be taken as transition probabilities for
a Markov chaindescribing the partisan transmission process.If we
could assume that the process is describable as a first-order
chain,and thatthetransition probabilities remain constantover the
relevantlengthyperiods of time, thenotherdeductionscan be drawn.
Mostnotably, perhaps, any such process describable in these terms
will have a&dquo;fixed point,&dquo; or equilibrium level of
partisan attachments (in this case),toward which the system will
necessarily move. What is charming in thisspecial case is that the
fixed point of the Franco-American process matrixis between 71 %
and 72% finally &dquo;identifying with a party&dquo; as a
matter ofsimple arithmetic.As an empiricalmatter,
thefewcountrieswhosedemocratic voting histories are of sufficient
depth to imply arrivalata&dquo;mature equilibrium&dquo;
inthese regards, tendtoshow proportionsidentified at the relevant
threshold (Table 1) in the vicinity of 75% to 80%.Not a perfect
fit, but without question suggestively close.Having come this far
without disaster, a further heuristic step or two isobvious.We
might simulate an electorate launched de novo ina partysystem and
therebydisplaying themost impoverished levelsofidentification, and
apply the transition matrix to the population to askhow many cycles
(here, &dquo;generations&dquo;) it takes for the system to
mount tothe equilibriumpoint. Therearetwo problems with the
question inprecisely this form, but fortunately both are more
apparent than real andneed not rob us of an answer. The first
problem is that the process in thiscase does not
&dquo;arriveat&dquo;the equilibrium point in an exact
sense, but onlyapproaches it asymptotically. Thereforeamore
manageable questionconcerns the number of cycles necessary to bring
the process within veryclose range of equilibrium-so close, for
example, that an electoral sampleof standard sizecould not be
expected to discriminate the difference atany reasonable level of
confidence.[147]The second problem is truly illusory, but worth a
moments discussion.Since we must specify a very hard numerical
&dquo;startingpoint&dquo; for ourcycles, what do we choose
as the most plausible proportion of initial partyidentifiers (at
this threshold) as the system gets underway? 1%? 4%? 10%?Once
again, the mathematical properties of these curious tables can
rescueusevenfrom much hard thought. For we canbeassuredthat
anyreasonable startingpoint, from as large a proportion as 15% of
thepopulation initially identifying to as few as a tiny handful of
persons (orfor that matter, nobody!), would, with our transition
matrix, arrive in thegeneral vicinity of the fixed point at what is
to all intents and purposes&dquo;thesametime.&dquo;The
paths towardthe equilibrium levelwouldbedifferentintheirinitial
details, but would converge so rapidly in theirapproach as to be
essentially indistinguishable.The convergence on the fixed point,
given our transition matrix, wouldlikewisebe extremelyrapid. Two
cycles alone would bring thethirdgeneration to a levelof 65% to 66%
of the population identifying, thepercent difference representing
what would remain atthattime of adiscrepancy between a starting
point of zero or one of 15%. One morecycle and the differences
between the path from either starting point andthe ultimate
equilibrium level would be quite negligible.Therefore a provisional
answer to our question of developmental ratesmight well be that the
grandchildren of the initiating generation would, asanadult
population, show nearly &dquo;mature&dquo;or equilibrium
ratesofidentification, and certainly theirchildrena generation
later would bedisplaying full partisan maturity in this special
sense of the term.We label this answer provisional, for we do not
intend to stop here. Itmay be a perfectly good answer, but enough
unknowns remain that wehave been motivated to push farther. Among
these unknowns, a principaldoubtmustsurroundthe assumption of
constancy inthetransitionprobability matrix, and even the general
validity for any wide range ofcountries of the entries derived from
France and the United States in thatmatrix can scarcely be taken
for granted.Most important,perhaps, isourdesiretohaveamodel of
thisdevelopmental process which, though as unencumbered and
parsimoniousas possible, would still permit us to deal with fmer
slicesof time than&dquo;generations.&dquo; The little
Markov transmission process above, while usefulas a means of
illustrating the kind of ultimate product we seek, turns outto be
awkward in many ways as a tool for more detailed analysis.
The&dquo;children&dquo; represented by the second
generation output of the transitionmatrix are truly
&dquo;childrenof all ages.&dquo; There isno immediate way
of[148]folding in our relatively rich information on the
accumulation of loyaltiesforthe individual over the life cycle. And
finally, for many interestingspecial cases-for example, the polity
that establishes a democratic system,loses itfor a period of time
and then resumes it-such a gross model wouldyield no estimates
atall. Clearly a more differentiated cohort model
isdesirable.CONSTRUCTING A MODEL OF AGGREGATE PARTISAN
DEVELOPMENTTherefore we turn from our two empirical points of
departure to themain task of this paper: the construction of a
model to represent the mostcritical processes known to affect this
developmental sequence, and to doso on a base of annual age
cohorts, or measurement in years rather than ingenerations. While
there is some premium on keeping the model as simpleas possible,
the following would seem indispensable for inclusion as
criticalprocesses:(1) a learning process: the basic increments in
partisan loyalties shownby theindividualover hislife cycle as a
direct function of hiscontinuing experience with the party
system;(2) a resistance phenomenon, representing the declining ease
of learn-ing as a function of the absolute age at which the
individual com-mences his experience within the system;(3) a
transmission process,capturing whatever vicissitudes may sur-round
the transfer of partisan feeling from one generation to thenext;
and(4) a forgetting process,handling therateof decay in retention
ofpartisanloyaltiessubsequent to anysuspension ofdemocraticprocess
eliminating the mass relevance of the traditional parties.Of these
subprocesses, itshould be self-evident that (1) and (3) are byfar
the primary elements, and (2) and (4) quite secondary. Process (4)
isnot needed in the model at allsave for the rare cases of
suspension ofdemocratic process, although it might have one other
minor role to benoted later.Process (2) comes into play more often,
as itis relevant notonly in cases where democratic process is
suspended, but also for cohortsreceiving
thevotefranchiseforthefirsttimeatadvanced ages.Nevertheless, itisno
more than a minor &dquo;modifier&dquo;of process (1),
andwill be treated as such explicitly.Itis apparent that for any
finished piece of work itwould be highlydesirable as well to have
some body of data that might serve for empiricalverification.In
effect, we needed estimates of the strength of
partisan[149]loyalitesby age cohorts as finely differentiated as
possible, in as manycountrieswithmaximal variability in history of
democratic process.Despite the increasingaccessibility of
cross-national poll data in recentyears, these specifications were
harder to meet than might appear. Theprincipaldifficulty wasthatthe
majority ofthese polls neitherconceptualize nor attempt to measure
an underlying strength of partisanloyalty. Instead, the
&dquo;party location&dquo; variable is most typically
elicited bya question like, &dquo;If you were voting right now,
what party would youchoose?&dquo;While this item undoubtedly
fits client needs admirably, the datayielded simply lack the
conceptual significance of questions tapping theintensity of more
enduring party loyalties, and their empirical propertiesare clearly
not the same. Moreover, even where a more general (i.e.,
lessmoment-specific) partisanship question is found, strength of
attachment israrely probed as well. And for reasons we have not
entirely sorted out, wehaveseen questions which, on their surface,
seem equivalent toourstandard items,yet havesomewhatdifferent
wordings and show thecritical aging effects only more faintly.At
about the time we had conceived what we wished to do and
werebecoming discouraged forlackof empirical reference material,
GabrielAlinond and Sidney Verba kindly made the data from their
ambitious&dquo;five-nation study&dquo; availabletothe
community (almond and Verba,1963). The fivecountries sampled
presented anexcellent portfolio ofvariationsrelevantforour needs.
There were two nations (the UnitedStatesandGreat Britain) with
lengthy and uninterrupted historiesofpopular voting. Two other
nations (Germany and Italy)provided casesof sharpinterruption and
then resumption of democratic process, withsome further differences
in time depth of widespread suffrage before theinterruption
occurred. The fifthnation (Mexico) presented a case of arelatively
new democracy,particularly wherefemale participation innational
electoral politics was concerned.The items concerning partisanship,
as comparably translated as possibleacross these countries, were
worded in a fashion that departed somewhatfrom the question
sequence standard on our prior work, yet which seemedto approach
the same spirit:Now wewouldliketofindout something about
yourpartypreference and how you vote. Are you currently a member of
anypolitical party or organization?(IFNO)... Do you consider
yourself a supporter of anyparticular political party?(IF NO)
Towards which political party do you lean? [Almondand Verba,
1963:531].~[ 150]These items are capable of yielding four
gradations of party attachment:membership, support, leaning only,
or &dquo;independent&dquo;(&dquo;none&dquo; in
answerto the final probe). However, allof our subsequent treatment
will be interms of a three-point scale combining
&dquo;members&dquo;and &dquo;supporters&dquo;
forthe highest levelofattachment.Thereasonforthis discarding
ofinformationis simple: theconditionsfor party membership vary
fromcountry to country asa matter of sociostructural arrangements
quitedistinct from our &dquo;learning processes&dquo;
(whether or not, for example, unionmembership is equivalent to
membership inalabor party, asitisinBritain); moreover, the
membership question was not even posed in theUnited States, so that
data comparability would have been infringed uponhad we tried to
maintain the distinction. In sum, our primary
dependentvariableisbasedon an integerscoring(0,1,2) of
thethreelevelsofattachment mentioned..Our firstconcern was whether
the index of partisanship based in thisfashion on the Almond-Verba
data would show at least the gross contours,countryby country and
age cohort by age cohort, thatour modelelements would predict.l In
view of the notable differences in electoraldevelopment
acrossthefive countries, these expectations were quitediverse. Yet
as Figure 1 suggests, the first glimpse of the Almond-Verbadata
suggested a splendid correspondence. The overall means were
highestforthe two countries with the longest uninterrupted
electoral
traditions,andwithinbothofthesecountriesthefamiliarcurvesof
life-cycleincrementsin partisanship were reasonably well
represented., I TheGerman and Italian curves, preceeding at the
expected lower level, showedtantalizingly-matched
&dquo;ann6escreuses&dquo;in exactly the age cohortsonewould
have expected to have been most retarded by the fascist
intrusions.And the curve for the newest democracy, Mexico, showed
the negativeslope that we had again expected to find at such a
stage of development(see discussion above).Vastlyencouraged,
itseemed well worth the labor to setasidetheAlmond-Verba dataand
work out in high detail atthe level of annualcohorts actual values
for the independent variables based on the votinghistory
ofeachofthesefivecountriesthatwere specifiedby thedevelopmental
model, and which could be expected to predict levels ofpartisanship
as one or another function of time.Stage I: Personal partisanship
experience. This rubric isintended tocoverthefirsttwoofour
specifications, orthesheer temporalaccumulation of electoral
experience on the part of the individual himself,[151]as modified
where relevant by the age-resistance term. It will be apparentto
the reader that any such estimates must immediately take account
ofthe fact that for a majority of the age cohorts of Figure 1,
there is strongand systematic variationin expectation by sex within
each cohort, forfemale suffrage has been instituted within the
lifetimeof atleastsomemembers of allfive national samples, and
indeed very recently in Mexico.Therefore allcalculations of
independent variables for annual cohortsproceeded on a sex-specific
basis. When the annual-cohort computationsFigure 1.STRENGTH OF
PARTISANSHIP, BY AGE AND COUNTRY(ALMOND-VERBA)[152]are reaggregated
into the grosser Almond-Verba age categories for tests ofpredictive
value ,1 2 this means that 70 discrete values of each
independentvariable (5 countries, 2 sexes, 7 age cohorts) were
computed, in contrast tothe 35 observations (5 by 7) presented in
Figure 1.13Since the initial process component supposed that
partisanship wouldincreaseasadirectfunction of length of personal
experience inthesystem, a first predictor established for the
annual cohorts was purely andsimply the number of years that each
cohort had been eligible to vote incompetitive democratic elections
up to the time of the data collection in1959. This variable was
denoted Ye or &dquo;years eligible.&dquo; In addition to
theobvious counting operation, the variable took into account such
things asthe year(s) in which female suffrage was effected, for the
female cohorts.Italsoexcluded in itsaccumulation for German and
Italian cohorts theyears during which competitive partisan process
was suspended by fascistregimes. IntheGerman case, this suspension
was consideredtohaveoccurred after the 1933 elections; in the
Italian case it was dated from thefascistelectorallawof May,1928,
thatabolishedthe relatively newuniversal suffrage and limited
voting by those male taxpayers still eligibleto an acceptance or
rejection of candidates proposed by the Fascist GrandCouncil.
Otherwise its accumulation as a simple &dquo;sumof
years&dquo; was verystraightforward,although
itdidneedtotakeintoaccountaswelloccasional minor variants by
country in the age of initial eligibility to vote.There was no
attempt in forming the Ye variable to express the factthatnotall
portions of an age cohort might indeed have been legallyeligible to
vote. Since cohorts were sex-specific, we were able to reflect
themost important singlesuffrage variation. Nonetheless,
particularly in theearly historiesofsomeof thefive countries,
therewere noteworthyfurther limitations on male eligibility
springing from property criteria andthe like. Furthermore, under
occasional circumstances, the gap betweenformal eligibility and
rates of voting turnout was so large within cohortsthat one could
not conceive of ignoring it. Thus, for example, in an earlyItalian
period the hostility ofthechurch to popular
electionswassufficiently great, and the influence itexerted over
its faithful sufficientlyintense, that many Italians
formallyeligible tovoterefrainedfromparticipating in the electoral
system.Therefore the Ye variable was considered as nothing more
than a firstapproximation, and a second variable was developed as a
modification ofit, denoted Y~ (&dquo;yearsvoted&dquo;).
Forthese purposes, theannualunitaccumulationof votingexperienceby a
generallyeligible cohortwas[153]reduced by acoefficient (between
zeroand one) based on levelsofvoting turnoutestimatedforthatcohort
atthenext preceding majornational election. Thus, to take a very
extreme hypothetical example forpurposes of illustration, letus
suppose that parameter information sug-gested that only half of the
males in a cohort were represented on therollsof eligible voters,
and thatforother reasons only half again ofthose eligible were
actually participating in the voting system. Then
theannualaccumulationof votingexperience forthecohortwould
bereduced from one unitto .25,reflecting the putativeparticipation
ofonly a quarter of the cohort. Needless to say, the modifications
in theYe variable introduced in this fashion were almost nowhere as
extremeasthis. Nonetheless, there seemed little question butthatthe
Y vvariablemore accurately reflectedthe accumulation of personal
votingexperience in an age cohort than the Ye variable.l4The second
modification of the personal-experience term arises fromtheneed to
express the age-resistance phenomenon. Itisrelevant onlyfor those
cohorts which, due to extensions of the franchise during
theiradulthood, became eligible to participate in partisan
electoral process ata greater orlessertimeafterthestandard age
ofinitial eligibility(generally, 21 years of age). While our
operations up to this point haveinvolved simplecountings of
relatively &dquo;hard&dquo; quanta, like years orfractions
thereof, to arrive at &dquo;external&dquo;indicators of
voting experience,we arenow faced with proposing some function of
time (or length ofexperience) which isa more dynamic property of a
learning process. Inprinciple, we had evidencenot only of the
existence of the phenome-non, but also an empirical basis
independent of the Almond-Verba datathemselves for estimating the
way inwhich speed of partisan learningwas retarded by latenessof
initial exposure tosuch experience. Wemight have been helped in the
details of this estimation had it been easytoarriveatsome
equivalences between specific values of the Almond-Verba
partisanship measure and &dquo;comparable&dquo; values of
the partisanshipmeasure on which many of our independent analyses
had been effected.Severaltactics employed toestablishsuch
correspondences gave, how-ever,sufficiently anomalous results that
we were obliged to proceed insomewhat rougher fashion,although it
may bestressedthatthe pro-posed function remains based on data
independent of the Almond-Verbamaterials.LetA represent
theindividuals chronologicalage,conventionallymeasured in years,
atwhich he isfirst permitted the voting franchise.[154]Then a is
defineda =A -211 (1)and the degree of resistancetonew learning (R)
isdefined simply asR = a;0 < a < 50 (2)100-aThus learning
resistanceisconceivedto be nilwithin a cohort thatbegins voting
participation at age 21, and growing but still quite trivial ifthe
cohort isexcluded from voting untiladecade orsolater.But itmounts
with increasing speed, and is complete if the cohort is not
giventhe franchise until age 70 or later.The independent variable
that will be taken to summarize the overallimpact (Ip) of personal
voting experience is, therefore:Ip - (1-R) Y (3)It may be
illuminating tonote that across our set of 70 observations,the
modification introduced by the coefficient (1-R) on the values of Y
yissmall indeed, Ip and Yy being correlated at.99. This is so
because 42cohorts of the 70 began voting
&dquo;ontime&dquo;and hence the values of thetwo
variablesforthesecohortsareidentical. Moreover, the valueof(1-R)
slips below .9 in only 10 of the 70 cohorts, these largely
centeringamong older Mexican and Italian women.Therefore
theresistancemodification may seem scarcely worth theeffort,
although itis clear that in estimations restricted to relatively
newdemocratic systems such a modification would play a stronger
role. Andeveninthecaseat hand, weshallsee shortly
thattheresistancemodification is of more than passing importance. 1
5At this terminal point in the discussion of the Ip cluster of
variables,it may be of interest to skip out of our chronological
narration momen-tarily to report thatthezero-order
linearcorrelationbetween Y y andthe mean strengths of partisanship
for the 70 Almond-Verba cohorts is awelcome .519.The same
correlation for the more refined Ip predictormounts to .558.Stage
77:Inherited Partisan Experience. With afirst predictor
com-missioned to express the accumulation of personal voting
experience, weshallnow turntothe problem of intergenerational
transmission.Aglance at Figure 1issufficientto suggest thata very
substantial pro-[155]portion of the overall variance (total sum of
squares) in partisanship inthese data, atleastatthecohort level,
istied up in between-countryvariance.Now the between-country
variancein voting history forthecurrentadult generation is palpably
lessthan thatwhich would havecharacterizedthe preceding adult
generation inthecaseof thesefivecountries. Thus, while some
relevant between-country variance does getcaptured in the Ip
variable, itisclear that further gains in explanationcould be
achieved by a reflection of the larger between-country differ-ences
in political experience of the parental generation.Itisobvious
thatour model, to be atall manageable, must make anumber of
simplifyingassumptions about thesheer mechanics of thetransmission
process. Itwould greatly facilitate our task, for example, ifwe
were ableto pretend that transmission occurs relatively
instantane-ously, and ataconstantmoment in the joint life cycle of
parent andchild.Thisof course isunrealisticin every direction. If
common sensehad not alreadyapprised us of the fact, recent
literature would makeclear that the relevant political
socialization iswell distributed over thetime the child is growing
to adulthood within his family. Similarly, thevariationin ages
atwhich parents beartheirchildrenis substantial,particularly
forfathers.There isinadditionsome variation, althoughless, in the
age at which a youth departs definitively from his family
oforigin,therebydiluting the possibility of political
influencefromparents. Thus since our model must hinge on a fairly
precise synchroni-zationof socializationevents with
historicaleventsinannual cohorts,therealisticsourcesof temporal
variationintheseaffairsarenumerou s.16Therefore we shall proceed
with a highly simplified version of suchevents, intendedtobeno more
thana rough approximation of theexpected values, or joint means, of
these temporally dispersed matters.Ourfirst simplification isto
hinge thewhole political socializationprocess on the father, for
daughters as well as sons.This decision flies inthefaceof so many
recent and entirely credible findings that it mayseem downright
perverse. In point of fact, it is predicated on a sequenceof
analyses with allof our American data, which is far too intricate
andtedious torecount here.The point isnot that anything is wrong
withthe current socialization findings, but rather that the
additional varianceto be explained by taking account of
permutations in parental identifi-cationsandsexofchildren (while of
significance in themselves) isdwarfed by theextreme complications
thatwould be entailedinthismodel. The cost-benefit answer was
simply negative.[156]Once we arefocused on the fathers, then, we
assume the followingpattern generically:(1) the child is born when
the father is 30 years old;(2) therelevant political
socializationoccurs when thechildis age15, and the father age
45.Given this idealized (or skeletal) description of the timing of
sociali-zation events, the question becomes one of what itisthat
the fathertransmitsatthemoment designated. Here theanswercan be
morerealistic. Clearly he transmits something of his own
accumulated partisanvotingexperience. And furthermore, we have
already designed a fairlythoughtful measure of the partisan voting
experience accumulated by anindividualat any point inhislife cycle,
as a function of the relevanthistorical sequences he (and his
cohort) has experienced. Therefore whatismore natural than to
suppose that the father transmits some quantumthat is a simple
function of whatever his own Ip may be at age 45?Hence we have two
chores: to say somewhat more precisely what the&dquo;simple
function&dquo;of Ip should be, and to cycle backward a
generationinorder to compute Ip values appropriate forthe specific
cohorts offathers of citizens in the Almond-Verba samples. The
latter problem is aroutine matter requiring no further comment. The
former can occasionmore labor.The Ip of an individual can be
conceived as (an indicatorof) asortof
partisan&dquo;push&dquo;deriving from electoral
participation. Inthis light, the question can be formulated in
terms of how much of such&dquo;push&dquo;(hopefully in
terms of years or elections worth of experience) apartisan father
(orhome) gives his children, relativetochildrenofnonpartisan
background. This formulation led to a detailed examinationof the
relativerates of ascent of partisanship curves in the limited set
ofour American data enjoying the relevant information, according to
thepartisan or nonpartisan nature of the family background. Here as
earlier,we were anxious to develop our specifications as fully as
possible fromother data, preserving the application to the
Almond-Verba materials asan independent test of the model. Some
conclusions did emerge from this comparative examination oflearning
ratesin partisan and nonpartisan homes. For example, itwasquite
clearthatthe partisanship of childrenfrom nonpartisan homes(even
excluding chronicnonvoters among the children)
neverdoes&dquo;catch up&dquo; in intensity tothe mounting
partisanship of adults origi-nating in partisan environments. This
was a welcome fact, as we realized[157]that we had unwittingly been
building such a property into our model.But the key question-how
much voting experience the individual withno partisan push from his
family would have to accumulate to matchtheinitial partisanship
ofthenormal amalgam ofindividualsfrompartisan,semipartisan, and
nonpartisan homescharacteristicofa&dquo;mature&dquo;
democracy-received a lessdefiniteanswer due to small casenumbers
and irregular functionsinthetest group atcritical
points.Nonetheless, the data did suggest thatthe
&dquo;push&dquo;expressed in years ofvoting experience
should lie between 14 and 25 years.We were prepared to take the
answer from this analysis as a conserva-tive one, for the simple
reason that while an individuals home may havebeen without partisan
flavor, in the United States there would be manyassumptions of some
partisan identificationintheculturalairoutsidethe home. Therefore
itwas assumed that any individual growing up insuch an atmosphere
should starthis electoral experience with at least aslightly
greater partisan push than an individual in a country where
thecultural atmosphere(as wellasthe parentalsituation) made
fewerpresumptions of partisanship. And these assumptions would in
turn leadus tofavorforour model an estimatelike25 years, ratherthan
theinner limitof 14. Yet this seemed somewhat extraordinary: at age
45,thefathersown Ye in any country could only be 25 (21 to 45);
andhis Y~ or Ip somewhat less.Hence the data seemed to imply that
all ofhis partisan push was transferred into initial partisan
momentum for hischild.This unexpected result might be resolved of
course by taking a thirdgeneration intoaccount. Thus the father
would represent not only hisown electoral experience up to age 45,
but in addition whatever initialmomentum hehadreceivedinturnfrom
hisown parents. Then histransmission to his child would more easily
be depicted as the expectedfractional value only of his own
experience.Atthis point, however, we decidedtocommission the
experienceterm for the father as an explicit variable, and
letconventional estima-tion procedures seektheir optimal weight in
the test against empiricaldata. Therefore we proceeded with the
definitionIr = The value of Ip as computed at age 45 for the
annualcohort 30 years ahead of (older than) ego,where If may be
thought of as &dquo;impact, father.&dquo;[158]The
establishment of If values across our 70 cohorts needs
littlenewexplanation. Itcanbeseenthat only 35 independent
valueswereneeded, as If was considered to be the same for both male
and female(new-generation) cohorts of the same age. The main
configuration thatrequiredany new decisionrulesaroseinconnection
with thefascistsuspensions of democratic process in Germany and
Italy. What shouldwe imagine tobethefateof thetransmission
processduring theseperiods?For the firsttime we found ourselves
without a shred of informationapart
fromtheAlmond-Verbadatathemselveson which tobaseajudgment. At the
outset we considered turning the parental contributionbrusquely
tozeroforthefullfascist period ineach country. Onreflection,
however, this seemed needlessly unrealistic. Informal evidencewould
suggest thatdemocratic expectations arenot instantaneouslydropped
under such conditions:even after some passage of time at
leastasmall minority of the population clings covertly to party
loyalties ofthe prefascistperiod. Moreover, setting
thesocializationtransmissionimmediately tozero would make fora
discontinuity inthe temporalfunctions that belies other outside
knowledge. Thus, for example, somesocializationwould already have
occurredinthecohorts lessthan 15years of age, even before
democratic process was suspended.Therefore,
itseemedmostreasonableto posit some more gradualdecomposition of
the socialization effects in the period after suspension.In
particular, we decided to assume that there would be a 20%
annualdecrement inthe strength of the partisanimpetus
availablefortrans-mission as of the suspension. In other words, if
the father under 45 (andhence toarriveatthe socializingyear during
the fascist period) has aquantum to transmit of If at time t (the
year of suspension), then whathe actually transmits in the nth year
of the fascist period is assumed tobeIf,, t+n - (.8)n If tt (4)We
emphasize that this function isan ad hoc estimate of these
effects.At the same time, itaffects only about a dozen of the 70
cohorts, andinfluences predictions evenformostof theseina very
minor wayrelativetothealternative hypothesis ofasocializationthat
simplyterminates during such a change of regime.The Empirical
Test.Whilethemodelatthis point hasnot yetincorporated
alloftheinitial specifications, itisnot premature to[159]examine
itsfit. The one element still missing is a representation of
someforgetting(decay of Ip) that logically would be expected
forcohortsexperiencing normal socialization and entry into a
democratic electorate,butwhose participation was then suspended
fora lengthy period oftime.The cohorts in question herearethe older
German citizens andItalianmales. The Ip values for these cohorts
have been mildly limitedby the fact, already reflectedin the model,
that these people failedtoaccumulate partisan experience atnormal
rates during youth to middleage. Butnodecrementhasbeen postulated:
themodel as itstandsportrays them to be resuming the suffrage after
the fascist period at thesame levelsof experiencethey had built up
by the time the intrusionfirstoccurred. Clearly some provision for
partialforgetting ofthisexperience during the hiatus is
necessary.However, we have no independent basis for estimating
effects of thiskind, although itwould be quite clearfrom other
psychological litera-turewhat generalfamily of functionistobe
prescribed.Up tothispoint we have been quite successful (with one
minor exception: specifi-cationof the decay of transmission effects
when democratic process isinterrupted, treated just above) in
basing our estimates of functions andparameters on dataother than
the Almond-Verba materials, preservingthe status of the latter as a
fresh and independent test. Lacking furtherinformation,however,
itseems appropriate toconsider atestof thenearly complete model at
this point. Thus we will leave the introductionof a forgetting
orretention function to a frankly a posteriori stage ofthe
analysis.We have already noted above that the Ip variable taken
alone showedat this point a simple correlation of .56 with the mean
values of partyidentification strength acrossthe70sex-and
country-specificagecohorts.The If variable reflecting
transmissionof parentalexperienceshows only a very mild correlation
(.17) with the Ip variable over thecohorts, so that collinearity
problems are minimal. As expected also, theadditionof the If
variableas a predictorgreatly increases the fit (forthis sample of
nations): the multiple correlation from If and Ip to
meanpartystrengthby cohortisatthis point a satisfying .886.The
co-efficientofdetermination (R2 ) is .785, or nearly four-fifthsof
thecohort variance in partisan strength accounted for.The
&dquo;Forgetting&dquo; Function. The residuals from the
regression analysisabovecouldbeexaminedtodeterminewhether they
reflectedthe[160]incomplete state of the model at this point. In
particular, the absence ofthe forgetting functionshould have given
a systematic overestimate ofparty identification strength within
the six oldest German cohorts (threeoldest, subdivided by sex) and
the three oldest grades of Italian males.This anticipated effect
was indeed present, and quite clearly so. Eight ofthe nine
criticalcohorts showed negative residuals (actual
identificationsweaker than predicted), while the ninth was barely
positive. The effectwas clearest in the German cohorts: all six
fell among the nine residualsmost extreme in the predicted
direction.A forgetting functionwas thenaddedtothe model,
determinedpartiallyby theoreticalconsiderations and partially by
the behavior oftheresiduals.That is, itwas assumed that a retention
curve (&dquo;percentsavings&dquo;) would show a negative
but deceleratingslope(negative firstderivative,positive second
derivative) asa functionof the passage oftime unexposed to partisan
democratic process. It was also assumed, onsomewhat lesscertain
grounds, thatthe rateof loss would be slightlyfaster for the
cohorts that were youngest atthe time of the suspensionofdemocratic
process, relativetothosewho were already
intheirthirtiesorfortiesatthetime of suspension. Within
thesetheoreticalconstraints on the likely character of the
forgetting function, the actualchoice of parameters was
deliberately made so as to minimize the size ofthe residual values.
1 7Afterthe forgetting modifications were incorporated and the
modelas originally envisionedwas completed, the multiple
correlationhadrisen from .89 to .91.The Socialization of
Disfranchised Females. A second examination ofresidualsatthis point
revealed a strong pattern suggesting that we hadmade one rather
egregious error in the initial organization of the model.Itwill be
recalled that both sonsand daughters cohorts were creditedequally
with reception of whatever partisan experience their fathers hadto
offer during theirsocialization. While this represented an
intentionalsimplification and one which for most purposes would be
adequate, itconstitutes an affront to common sense with respect to
those cohorts ofdaughters for whom, at the time of socialization,
politics remains solidlya male world and female suffrage isstill
decades away.The point was brought home tous forcefullyby the fact
that thestrongest negative residuals (the model overestimating
partisanship) nowclustered among the older cohorts of Italian and
Mexican women. Thesewomen had the least personal voting experience
of any of our cohorts,[161]havingonly recently
receivedthefranchise.Their personalexperience(I ) was being dimmed
further, interms of the model, by high age-resistance.Thuswherethe
Ip variablealonewas concerned,onlyvestigial levels of partisanship
were being predicted for them. The over-estimatewas arisingentirely
from thecontribution of the If variable(partisanexperience
&dquo;inherited&dquo;from father), imputed to be as
strongforthese long-to-be-disfranchised females asfortheir voting
brothers.This whimsical implication of thea priori model cried out
for adjust-ment.Inthe original model, itwillbe recalled,every
cohort was taggedwith an age-resistance value (R). For allcohorts
entering the electorateautomatically upon reaching adulthood, this
value was zero. It ascendedto increasinglysignificant levelsthe
laterinlifeacohort was
enfran-chised.Thereforeitseemedareasonablemodificationto apply
thecoefficient (I-R)referring to egos own cohortnot only to
egosimmediate partisan experience, butalsoto apply itto the
&dquo;inherited&dquo;experience communicablefrom egos
father.Such a correction wouldhere affect only those female cohorts
becoming enfranchised after adult-hood had been attained. In
effect, this correction would impute a fairamount of
&dquo;anticipatory socialization&dquo;:those women being
socialized ata time when their own enfranchisement was imminent
would receive thesame, or nearly the same,partisanpush
fromthefatherastheirbrothers. The more remote in time thatfemale
suffrage becomes, how-ever, the less political impact such a
socializing father would transmit toa daughter.After this further
post hoc modification was introduced in the valuescomputed for the
fivenations of the Almond-Verba study, the multiplecorrelation
coefficient ascended to .925, with nearly six-sevenths of
thevariance in cohort means accounted for. Table 2 summarizes the
outputof the regression analyses at the two main stages of
testing.The fit might have been further improved in several simple
ways hadtheeffortseemedwarranted.Asone example, thereis
-somemildsuggestion thatthelinear regression
modelontheseuntransformedvariables isnot optimal. Indeed, semi-log
and double-log versions of theregressions wereexamined.A log
transformationofthe dependentvariable (commonly indicated in
problems of time and growth) did givealmostassuccessfula
fitastheuntransformed version, and itseemslikely
thatsomeintermediatetransformationwouldhave actuallyimproved the
magnitude of the correlations. Or again, there remain some[162]few
small patterns in the residuals that suggest reasonable further
adjust-ments of themodel. However, inview of thefactthatthese
wouldinvolve the importation of concepts alien to the learning base
on whichthe model rests, and in view of the real danger of
overfitting the modeltothis particular batch of data, we have
pursued none of these othercorrections. The level of fitneeds no
apologies as itstands.TABLE 2FIRST AND SECOND REGRESSION
ANALYSESNOTE:The dependent variable, Ps, referstotheindex of
strength of partisanidentificationcalculated asdescribed inthe text
from the Almond-VerbaFive-Nation Study. The regression coefficients
are not normalized.Some Caveats.Inviewof what may seem a remarkably
closefitbetween the model predictions and the Almond-Verba data, a
number ofqualifying observations are in order.First, whilethe
multiple correlationsareindeed pressing theouterlimitof what might
be expected, given inevitable sampling and measure-ment error,
itshouldbe remembered that the sheer magnitude of thecorrelationsis
substantially enhanced by thefactthatwe have beenmaking somewhat
aggregated, ratherthan individual-level,predictions.That is, we
have been attempting to predict the mean values of partisanstrength
registered by each of the 70 cohort groupings. Within every oneof
these cohorts, of course, thereisfurtherindividual-level
variationaround the mean, and thisvariationhas been systematically
excludedfromconsiderationinour predictions. To the degree
thatthemodelpivots on experience thatcohort members share in
common,l8 itfailstoaccount forwithin-cohortvariation.If the data
were disaggregated,bringing thisfurthersourceofvariationinto view,
thecorrelationswould be considerably closerto ranges familiarfrom
other individual-[163]levelworkwith samplesurvey materials.For
comparison wehaveconsidered the results in this disaggregated form;
then the final multiplecorrelation recedes from .93 to .46.At the
same time we may emphasize that there is no reason to
avoidprediction to aggregated cohort values, provided the
difference in mean-ing of statistics reflectinggoodness of fitis
kept in mind. In point offact, for the study of secular trends in
social and political change wheretimeand population turnoverseem
important mechanisms, thetreat-ment of cohorts as units of analysis
may be very much indichted.Insofar as our cohort-level model
purports to deal in what seem to berelatively basic and inexorable
processes, it would make sense at anotherstage of analysis
toaddress the explanation of individual-level (within-cohort)
variations in partisan strength with the effects of these
processesfirstextracted.In other words, the measurement of
individual partisanstrength would be recast as residuals from the
cohort-level predictions.Successin accounting
fortheseresidualindividual-levelvalues wouldthen be suggestive with
respect to situations that inhibit or stimulate therate of growth
of such loyalties.This observation leads toasecond caveat.We do
takefor grantedthat &dquo;otherfactors&dquo;can inhibit or
stimulate the development of partisanloyalties atan individual
level, and it logically follows that such factors,if differentially
present across a range of polities, would lead to
charac-teristicdifferencesinthe speed withwhich aggregateloyalties
coulddevelop
andeveninthe&dquo;saturationlevel&dquo;towardwhich
particularpolities tend. Indeed, ifotherwork is valid, France may
provide aninstance in which communication peculiarities (among
other things) haveexertedsome inhibitions on the development of
partisanship (Converseand Dupeux, 1962). Conversely, if politicized
mass organizations tend tostimulatethe development of
partisanloyalties asis frequentlysup-posed, thena societypervadedby
such organizations toan unusualdegreemight be expected to display
unusual speed in the mobilizationof aggregate loyalties.The point
of our second caveat, then, should be clear. If we are usingthe
current model to extract some sense of a general
&dquo;timetable&dquo;for themobilization of aggregate
partisanship, we arenot in the least imaginingthatourestimates
constitute&dquo;lawsof development,&dquo;particularly
withrespect to any specificparameters. At the very best, the
general time-tabletobe presented below must be seen as a
&dquo;mainline&dquo;estimate ofthese effects, around which
considerablesocietalvariation may
occur;eventhis&dquo;mainline&dquo;is only more orless
adequate according tothe[164]degree towhich our haphazard sample of
fivenations is representativeof the larger universe of cases. 1 9A
third caveat has to do with our interpretation of determinants
ofthese processes. In view of the specific
&dquo;habit/attitude&dquo; being investi-gated here
(partisanship), we have ineffect operationalized the vagueentity
often described as &dquo;the weight of historical
tradition&dquo; largely interms of the transmission of parental
experience. Nonetheless, the readershould be aware that any of a
wide range of variables correlated withthe sheer length of time
these democratic systems have been under waywould produce
substantialcorrelations with the levelsof partisanshipobserved in
the Almond-Verba data. Itis even possible that
methodolog-icalartifacts have contributed to the result, since
certain known defectsof the Almond-Verba data become more intensein
the countries thatareless developed andhence
are&dquo;newer&dquo;democracies.Some of themost blatantof
thesedefects we have been safein ignoring, since themodeofdata
analysis cancelsouttheireffects.2 0 Others, however,would be of
concern.For example, it might be argued that Italy andMexicoshow
low levelsof partisanship because partypreference isajealously
guarded secret in more peasant cultures, particularly when thequery
comes froma strange interviewer.There is undoubtedly someelementof
such a phenomenon in these data. Nevertheless, work withevasions
and refusals to indicate partisanship in other similar settings
hasconvinced us thatthesetend tocenterin segments of the
population(women, the poorly educated, the geographically remote)
that are leastinvolved and informed about national politics, and
are likely to concealignorance far more often than fervent
feelings. Any such syndrome, ofcourse, fitsthecurrent model of
socializationor learning very admi-rably.In any event, we do not
feel apologetic about the interpretation ourmodel casts on the
data, in part because of the obvious logic of thematter, andin part
because of thesuccessof severalrather irregularpredictions (in
particular, those surrounding the cases of fascist suspen-sion).
Furthermore, expectations hinged on the model have been sus-tained
in a wider array of contexts than can be explored here.21 Yet
itremains useful to stress that the success of the predictions in
itself is noguarantee that our choice of mechanisms is the most
appropriate.A finalcaveat has to do with the structure of the model
itself. Thereare quite anumber of free parameters that require
estimation here, andestimatescould varyconsiderablyaccording to the
order in which theestimation proceeds. We have attempted to cope
with these potential[165]indeterminacies by starting with
thoseeffectsmost abundantly docu-mented from independent sources
and, considering these fixed, haveproceeded to make other estimates
on the basis of residual, unexplainedvariation.Thisseemsareasonable
strategy underthe circumstances,although itis obviously notstrict
proofagainst theintrinsic logicalambiguities.2 2It may be
appropriate to point out, however, that most of the ad
hoccomplications thatarisesurroundthe two unusual instancesof
fascistsuspension of democratic process. Without the further terms
necessaryto express theoutcomeof componentprocessesduringperiods
ofinterruption, themodelwouldrestona veryparsimonious
setofestimates and indeed is straightforward to the point of almost
embarrass-ing simplicity.CONCLUSIONLipset and Rokkan have recently
observed that much contemporane-ous electoral analysis has
proceeded without attention to the factthatquite radical
differences exist in the democratic age of the systems
beingdiscussed. They go on to observe that most of the substantive
alterna-tivesoverwhich (European)parties still compete inthe1960s
were&dquo;frozen&dquo; by cleavages salientatthose
historical points when widespreadsuffrage andhencemass parties were
instituted (Lipset and Rokkan,1967:
&dquo;Cleavagestructures,partysystems andvoter alignments:
anintroduction&dquo;).We have attempted here to assesssome of
the obvious regularities intheratesatwhichthesedemocratic systems
&dquo;freeze&dquo;ina slightlydifferent but entirely
complementary sense. That is, the fixing of partyalternatives has
tended to occur rather suddenly, as Lipset and Rokkanimply, and
thisas an outcome of elite competition at acritical point.Mass
loyalties only follow apace, and itisthis pace that we have
beeninterested in estimating.By way of summary, we present in
Figure 2 an idealized chart of theprogressive freezing of a system
as a function of time, derived immedi-ately from the model used
above. Itis important to stress, in additionto earlier caveats,
that this isan &dquo;idealized&dquo; picture rather than in
anysense a typical one.Inthe typical polity, the fullmass public is
onlyenfranchised piece by piece (e.g.,propertied males, then all
males, thenall adults), with decades often intervening between
stages. Thus Figure 2should be taken to refer either to an
electorate that is granted
universalzO_Haaupee4ca-ia1=0t-)ui-imu.0(AI=aHi>m2:&dquo;CO)
z4C*PGC2aHa0Uicc ocoati.0II-0ccC9N0zLL[167]adult suffrage atthe
outset, ortoa subpopulation thatissobe-stowed.2 3 Theadditionof
positions for specificpolities at particulartimes
totheidealizedcurveisthus partially inappropriate; these addi-tions
are meant only to be suggestive, and take into account the
actualstagesby which these politiesexpanded their suffrage.They
arethusrelatedto the main curve only by virtue of putative
(model-predicted)levels of partisanship at the historical dates
indicated.Quite in accord with our earlier crude estimate, Figure 2
suggests thatmaturity is essentiallyapproached afterabout two and
one-half gener-ations (considering a generation, as we have, to be
30 years). Of coursethe logic of the model issuch that maturity
becomes &dquo;complete&dquo; in anabsolutesense after the
electorate iscleared by population turnover ofany members suffering
a deficit because of fathers not maximally experi-enced, etc.Such
an absolute maturity would arrive only more slowly ifgrandfathers
were considered to make any significant contribution to theprocess
in egos generation. However, such a variation of the model,
ifreasonably done, would make littledifference in our assessment of
thetimetable to quasi-maturity.We assume that the stateof a
democratic system in this regard-thedegree towhich
itisfrozenorremainsfluidwith respect tomassloyalties-is a
significant datum. We see these loyalties as having some-thing of
thesame conservative or preservative influence on democraticsystem
stability as doessocializationintotherulesof the democraticgame
atan elite level. This is not to imply that the possibility of
radicalchange in party structure, oreven inthenature of the regime,
disap-pears as a system approaches this type of
&dquo;maturity.&dquo; But itis to arguethatthe
probabilities declineina significantdegree or, if one wishes,that
the severity of shock necessary to induce such system change mustbe
progressivelygreater themore completely the system has jelled
inthese terms.NOTES1.Such a conclusion has been discussed in recent
years by (among others) V.O. Key (1961) and H. McCloskey
(1964).2.It might be noted that we are using terms such as
"stability" and "insta-bility" asflat descriptors,despite their
frequent evaluativeconnotations intheliterature.The text will have
no more and no less intended meaning if the reader[168]wishes
tosubstitute polar terms keyed to opposite evaluative connotations,
such as"flexibility"(forinstability), and "rigidity" or even
"ossification" (forstability).3.The germinal work here was Herbert
Hyman (1959). An excellent recentbibliography reflecting growth of
interest is contained in J.Dennis (1967).4.For such a graphic
contrast, see Newcomb et al. (1965:116). Subsequent(still
unpublished) evidence suggests moreover that the party
identification measureon which these arrays are based does not
represent the situation quite as clearly asone might wish. That is,
careful comparisons of defection rates (a behavioral votingmeasure)
withinclassesof party identifiers (an attitudinal
"loyalty"measure)across age cohorts shows that citizens in their
first voting years tend to pull out oflinewith the norms
characterizing the rest of the array, defecting somewhat
morefrequently than their statements of party identification
strength would predict. Inother words, they tend to overstate their
level of party loyalty, particularly if theycome from
stronglypartisanfamily backgrounds. Thus a measurement
difficultyblurs a relationship which remains nonetheless quite
strong.5.The most relevant discussions are presented in A. Campbell
et al., 1960 (seeespecially 161-165 and 497-498).6.TheMcPhee model
again hasan ingenious setof provisions toreflectpopulation
turnoverand learning of partisanship by children (McPhee and
Glaser[eds.] 1962: 151-152).7.The reader unfamiliar with this
theorem may enjoy experimenting with ithimself. Thus, for example,
application of this particular transitionmatrixtoaparental
generation with only 20% of party loyals produces a much more
heavilyidentifiedsecond generation56% with party identifications
above this threshold,rather than 20%. This figure thrusts rapidly
upward toward the fixed point of thetable. On the other hand, ifwe
imagine a hypothetical population where identifi-cations,
forwhatever reason, have surpassed the equilibriumlevel-say, 90%
ofparents areidentified-thenthesecond generation willhave slipped
"back" to a77% level, or again much nearer to the fixed point of
the process. This is the senseinwhich a very precise level
represents an "equilibrium" for the process. For anextremely
accessibletreatmentofsuch processconfigurations, the
readerisre-ferred to J. Kemeny et al., 1957 (Chapter 5).8.The
assumption of constancy seems a littlesafer than it might otherwise
beinviewof thebrief number of cycles required and the modest
tolerancethatmight be allowed in variationsof theentries (temporal
or cross-national) whileapproximating thesame result.It might be
noted in this regard that a seeminglycomparable parent-child matrix
for Great Britain pointed out to me by MichaelKahan has quite
differentinternalentries-and fixed point-but roughly a
two-generational solution nonetheless, at the level of convergence
required above. Ontheother hand, a parellelparent-child matrix
prepared by Akira Kubota on thebasisof national surveys of the
Japanese electorateare strikingly similarto theFrench and American
cases.9.Notethe reproduction hereof the slightly varied procedure
used intheUnited States sample.10.The agecoding inThe CivicCulture
study issomewhat coarse but by nomeans inadequate for our purposes,
there being seven classes distinguished:18-25,26-30, 31-35, 36-40,
41-50, 51-60, 61 and over.11.The slight decline in the oldest age
category for both the United States andGreat Britain isthe only
mildly unusual feature, and even this turns up (throughsampling
error?) in occasional U.S. samples with our standard party
identificationquestion.Summing across many such studies, however,
the terminal age categorymatching the Almond-Verba divisionsdoes
show a very faint continuing positiveslope.[169]12.In view of the
extremely minor error involved, the reaggregation for thefirstfour
relatively narrow age cohortsassumeda rectangular distribution
ofpopulation across the cohort. For the fifth and sixth cohorts an
appropriate grossweighting wasintroducedtoreflectthe decliningslope
of population, and theweighting was done on a careful year-by-year
basis in the critical seventh category(over 61 years of age),
within which the range of annual population proportions isgreatest
and the independent variables were often showing quite disparate
values aswell.The basis,however, wasnota country-specificagetable,
butratherarepresentative table thought to be a reasonable average
for the types of countriesinvolved. The riskof any serious error
here is very slight, as the reader may cometorealizeon
reflection.That is,although the age tables for these
fivecountriesmay well have been quite disparate in1959-the year
most of the Ahnond-Verbadatawere collected-countrydiscrepancies in
age distributionsare largely con-trolled by virtue of our
predictions being made within fairly narrow age cohorts.13.The 70
estimates of partisanship intensity available from the
Almond-Verbamaterials rest on a range of case numbers from 37 to
156, with the median valuebeing 71 cases.14.Another
operationalization that might recommend itself in the preparationof
an indicator of accumulated voting experience for a cohort, would
be to sumacross"elections experienced," ratherthan simple years. In
point of fact, twoparallel variables developed on thisbasis (E eand
Ev: elections eligible and elec-tions voted) were the subject of
some experimentation. Although in one sense suchvariableswouldseem
to bring us another step closertothe concept "electionexperience,"
the shift of base opens a Pandoras box of practical woes. When is
anelection an election? Do we count minor partisan elections or
only national ones?Do off-yearCongressional
electionsintheUnitedStates"count" thesame aspresidential elections?
Indeed, itwas something of a relieftodiscoverthatthepredictivepower
of the E variables experimented with was slightly less than
thatoftheY variables, fortheY variables represented such simple and
readilyaccessibly information on polities.15.Itsrelative efficacy
shouldnot be neglected even atthis point, incom-paring Yvand Ip.
That is, in view of the very high correlation between the
twomeasures (actually.994), themaximal amount of further variance
that Ip couldaccount for beyond whatever Yv already accounts for in
a dependent variable islimitedand readily calculablein any given
instance.InlaterrelationstotheAlmond-Verba partisanship variable as
dependent variable it could be seen that themodificationaccounts
for nearly half (.41) ofthismaximum possiblefurthervariance. In
view of the fact that the need for a modification of this type was
feltbefore the Almond-Verba materials were first examined, and the
function proposedfrom other information without recourse to any
close dissection of the Almond-Verba data, the utility of the
modification can scarcely be questioned.16.The McPhee simulation
iscontrived to represent some temporal dispersionof impact on the
child (see McPhee and Glaser [eds.] 1962: 152).17.Theactual
forgetting function employed was asfollows.Considertheamount of
partisan experience, both direct and inherited (Ip and
If,respectively),that a citizen has accumulated up to the time when
partisan process is suspended.Let b represent the proportion of
that experience which would be actively retainedwhen partisan
process is restoredt years later. Thenb = r /log t + r[170]wherer
=c + 0.05A sand (As being the chronological age of thec = a
constant, 2.9 subject at the time of suspension)18.One exception
involves differentialrates of voting turnout within cohort.At this
point a fully elaborated model would indeed make differential
predictionsatan individuallevelwithinthe cohort. For simplicitys
sake, however, we havehere treated turnout asa cohort-level rate,
and hence bypassed the opportunity totreat within-cohort
variability.19.In this regard, it may occur to the reader that the
model does imply somedifferential national predictions that we have
not chosen to exploit here. Thus, forexample, ourinitial
analysessuggestedstrongly thatwhat is important forin-creasing
levelsof partisan identificationisnot somuch generalparticipation
indemocratic processperse, butrather persistentvoting for
particularpoliticalparties. As a corollary, a polity withinwhich
politicalparties have frequentlydisappeared with new ones emerging
(for whatever reason) should be predicted toshow lowerlevelsof
partisanship,everything else equal, than the polity withinwhich the
same constellation of parties has endured for a long time (for
whateverreason). Therefore more incisive predictions could be made
by taking into accountthe continuity of voter-partyrelationships,
ratherthan the simple voter-processrelationship we have used here.
In fact, some analyses were conducted within theAlmond-Verba data
on this score, and showed positive results. That is,among truemass
parties (asopposed to ideological splinter groups) there issome
mild corre-lation between party age and intensity of popular
support within the same system.However, theseresultsare
sufficientlyweak, when othermain effectsarecon-trolled, and the
problems of ascertaining party "age" in view of the continuum
ofpossible transformations of a political party as a perceived
object (mere change inname as one extreme, for example) are
sufficiently difficult that itseemed wise toproceed on the simpler
base.20. Thus, for example, the most glaring known defect of the
data is the severeunderrepresentation ofmales inthe Mexican
sample.However, sinceallof ourestimates have been carried out
within sex categories, this defect is irrelevant.21.One
particularly interesting reflection of these processes isof a
behavioraland historical nature. Itcan be shown that temporal
variation in party voting andticket splitting-bothsupposed to be
symptoms of weakly developed partisanshipin
aggregates-areremarkablyhigh inthe"new votingsystems" represented
byAmerican states freshly admitted tothe union inthe latenineteenth
and earlytwentieth centuries.22.Problemsofsubstantive
interpretation intertwinewiththese ambiguities.For example, the
decrement in partisanship among the older German and Italiancohorts
afterthe initial test, which we then handled with a "forgetting
function,"might be given amore vigorousinterpretation, asan
activealienationfrom thepoliticalprocess onthe part of cohorts that
had participated inthesuccessivedemisesof democraticandfascist
regimes. Such an interpretationmight havesuggested adifferent type
of functionto predict thedecrement. The residualdeviations atthis
point were not sufficiently robust or patterned to permit such
adelicate judgment inthis regard. We simply recognize here
competing interpre-tations.23. Although themodel implies that
progress of identification among femaleswill be more rapid than
isassumed by Figure 2, due to their being embedded in apartisan
atmosphere before suffrage.[171]REFERENCESALMOND, G.and S.VERBA
(1963) The CivicCulture.Princeton:PrincetonUniv. Press.CAMPBELL,
A., P.E. CONVERSE, W. MILLER, and D.STOKES (1960) TheAmerican
Voter. New York: John Wiley.CONVERSE, P.andG.DUPEUX (1962)
"PoliticizationoftheelectorateinFrance and the United States."
Public Opinion Q. 26 (Spring).DENNIS, J. (1967) Recent Research on
Political Socialization:A Bibliography ofPublished, Forthcoming,
and Unpublished Works, Theses and Dissertations, anda Survey of
Projects in Progress.Medford, Mass.:Lincoln-Filene Center
forCitizenship and Public Affairs (July, mimeo.).HYMAN, HERBERT
(1959) Political Socialization. New York: Free Press.KEMENY, J., J.
SNELL, andG.THOMPSON (1957) IntroductiontoFiniteMathematics.
Englewood Cliffs, N.J.: Prentice-Hall.KEY, V.O. (1961) "Public
opinion andthe decay of democracy." VirginiaQuarterly Review 37
(Autumn): 481-494.LIPSET, S.and S.ROKKAN [eds.] (1967) Party
Systems and Voter Alignments.New York: Free Press.McCLOSKEY, H.
(1964) "Consensus and ideology inAmerican politics." Amer.Pol. Sci.
Rev. 58 (June): 361-382.McPHEE, W. N. (1963) Formal Theories of
Mass Behavior. New York: Free Press. and W. GLASER [eds.] (1962)
Public Opinion and Congressional Elections.Glencoe, Ill.:Free
Press.NEWCOMB, T. M., R.H. TURNER, and P.E.CONVERSE (1965) Social
Psy-chology. New York: Holt, Rinehart &Winston.PROTHRO, JAMES
W.and C.W.GRIGG (1960) "Fundamental principles ofdemocracy: basesof
agreement and disagreement." J.of Politics22 (May):276-294.