Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

JOURNAL Of EXPERIMENTAL SOCIAL PSYCHOLOGY 16, 573-581 (1980)

Cross-Lagged Panel Analysis of the Relation between


Attraction and Perceived Simillarity
DONALD GRANBERG

Universiiy of Missouri-Cohmbia

AND

MICHAEL KING
California State University, Chico

Received December 17, 1979

Cross-lagged panel analyses of the relation between attraction and perceived


similarity to Richard Nixon and Gerald Ford were reported. The analyses were
based on a panel of U.S. adults for 1972-1974-1976. The cross-lagged analysis
involving Nixon was supportive of the causal flow being predominantly from
attraction to perceived similarity. This conclusion was sustained in multivariate
analysis. For Ford, the cross-lagged correlations between attraction and per-
ceived similarity were not significantly different. However, multivariate analyses
suggested that the effect of attraction to Ford on perceived similarity to him
remained significant even when numerous control variables were used. In con-
trast, the effect of perceived similarity to Ford on attraction to him was reduced to
nonsignificance when controls were entered in regression analyses.

One of the more reliable findings in experimental social psychology is


that similarity in attitudes leads to attraction. In experiments demonstrat-
ing this, subjects are exposed to a stimulus person with attitudes that are
at a given level of similarity to the subject’s atttitudes. The dependent,
variable is how attracted the subject feels toward the stimulus person.
Although some of the experiments on this topic are quite contrived, and

The data used in this article were obtained through the Inter-University Consortium for
Political and Social Research and were originally collected by the Center for Political Studies
of the University of Michigan. The authors are solely responsible for the analysis and
interpretation. The authors wish to thank David Leuthold for assistance in obtaining and ac-
cessing the data and Harris Cooper and Ed Btent for very helpful methodological advice and
direction. Requests for reprints should be sent to Donald Granberg, Center for Research in
Social Behavior, University of Missouri-Columbia, Columbia, MC 652 1I.

573

fx122-lo3 1/80/0~~573-a~~o2.oa~~
Copyright 0 1960 by Academic Press, Inc.
AI1 rights of reproduction in any form reserved.
574 GRANBERG AND KING

their results may possibly be explained in terms of compliance with


demand characteristics, others have been done in field settings with a
higher degree of mundane realism (Byrne, 1961,1969,1971; Byrne, Ervin,
& Lamberth, 1970; Griffith & Veitch, 1974). In the experiment most
closely related to this paper, subjects were informed of a hypothetical
political candidate whose positions on issues were manipulated to be at
various levels of similarity to the subjects’ attitudes. Attraction to
hypothetical candidates was shown to vary directly as a function of
similarity (Byrne, Bond, & Diamond, 1969). Such experiments, however,
can only show that this is one plausible way that it could work in the real
world.
An alternative, not often considered in experimental settings, is that
there is actually a nonrecursive (bidirectional) relationship between at-
traction and perceived similarity. Furthermore, it is possible that under
natural circumstances, the causal flow is predominantly in the direction
opposite from that asserted by Byrne. That is, it could be that attraction
causes a presumption of similarity to self or an actual convergence toward
increased similarity .(Sherif, 1935; Sampson & Insko, 1964; Aronson,
Turner, & Carlsmith, 1963; Feather, 1967; Wyer, 1974).
A sizeable body of literature using correlational analyses indicates that
people attribute similar attitudes to individuals and groups they like. In
the political realm, several articles have reported a tendency for people
to assume that a preferred politician’s positions on issues are similar
to their own. This is implied by a positive correlation or regression
coefficient between a person’s own attitude on an issue and the position
attributed to a preferred political candidate. This is often interpreted as
evidence that attraction to the candidate leads to the presumption of
similarity to self. Although this finding has been repeatedly documented
(Berelson, Lazarsfeld, & McPhee, 1954; Sherrod, 1972; Page & Brody,
1972; Granberg & Brent, 1974; Granberg & Seidel, 1976; Granberg &
Jenks, 1977; King, 1978; Kinder, 1978), it has been based on cross-
sectional correlational or regression analyses. Hence, the finding would
be equally compatible with the hypothesis that people become attracted to
political candidates who are seen as taking policy positions that are
similar to their own attitudes.
An analytical procedure that can be useful in the effort to clarify the
predominant flow of causality in a given context is cross-lagged panel
correlation (Kenny, 1975; e.g., Crano & Mellon, 1978; Kahle & Berman,
1979). If the cross-lagged correlation between attraction at Time 1 and
perceived similarity at Time 2 is significantly stronger than between
perceived similarity at Time 1 and attraction at Time 2, and if several
assumptions are met, this would be evidence that attraction causes per-
ceived similarity. This technique provides a potential for saying more
ATTRACTION AND PERCEIVED SIMILARITY 575

about the predominant causal flow than a simple cross-sectional COIT&-


tion. It also has a higher degree of external validity than a laboratory
experiment, although the latter permits a stronger ~on~~usi~~ about
causal relation.
This article reports analyses of panel data that clarify somewhat the
causal flow between attraction and perceived similarity that occurs under
natural circumstances. The analyses are based on the University of
Michigan’s Center for Political Studies Panel of the U.S. electorate for t
years 1972-1974-1976. The data contain measures of attraction toward
politicians and indicators from which we can infer perceived simi~ar~ty~
The hypothesis to be examined is that variation in the level of attra~~~~~
leads to change in perceived similarity.
METHOD
A total of 4455 U.S. adults were interviewed at one or more of the following five times,
before and after the 1972 presidential election, after the 1974 congressional election, and
before and after the 1976 presidential election. A relatively small number of people were
interviewed at a11five times. Given our interests and the limitations of the data, we shall deal
with two separate analyses, those pertaining to Richard Nixon in 1972 and 1974 and to
Gerald Ford in 1974 and 1976.
The measure of attraction is a lOO-degree feeling thermometer that respondents use to
indicate their affective feelings about a candidate, 100 being the most favorable possible, 50
neutral, and 0 the most unfavorable possible (Weisberg & Rusk, 1970). Attraction to Nixon
w-as measured in 1972 and 1974, and attraction to Ford was measured in 1974 and 1976.
Our measure of perceived similarity is based on a composite of several items. In 1972,
1974, and 1976, respondents were presented with five 7-point scales pertaining to (a) the
government guaranteeing a job and a standard of living, (b) urban unrest, (c) aid to minority
groups, (d) busing to achieve racial balance, and (e) a self-designated ideology scale. On
each issue, the end alternatives were explicitly defined with au underlying continuum from
the most liberal alternative (1) to the most conservative (7)’ In 1972 and 1974, in addition to
indicating their own attitudes, respondents were asked to estimate Nixon’s position on each
scale, and in 1974 and 1976, they were asked to place Ford on each scale. We first computed
the absolute difference between a person’s own attitude and the position attributed to a
politician on each scale. We next recoded each variable so that higher scores meant greater
perceived similarity. We then summed across issues to obtain a perceived similarity scale
with a possible range of O-30.

’ Each end of the scale was given a verbal description pertaining to that issue. The words
liberal and conservative were only used on the ideology scale. The intermediate positions
were not labeled, but respondents were instructed to think of them as gradations between
the two extremes. For examples of the wording in scales of this type, see Granberg and
Seidel(l976) or Granberg and Jenks (1977). The complete wording of the five scales used in
this article is available in any of the Center for Political Studies codebooks for 1972, 1974, or
1976. PeopIe placed themselves and Ford on an additional scale dealing with the rights of the
accused in legal situations. However, since comparable data did not exist in regard to Nixon,
it was decided not to include it in order to keep the analyses for Ford and Nixon more
comparable. Similar placement judgments were made of McGovern in 1972 and Carter in
1976, but no panel type data for these variables are available for any Democratic candidate in
this file.
576 GRANBERG AND KING

RESULTS
Before presenting the cross-lagged comparisons, some overall results
are given. Not surprisingly, attraction to Nixon, as reflected in ratings on
the feeling thermometer, plummeted dramatically from 1972 to 1974. Re-
call that in 1972 he was elected in a landslide victory, but by the 1974
survey, he had resigned in disgrace to avoid being removed from office
by Congress over the Watergate affair. The feeling thermometer ratings of
Nixon were 65.3 in 1972 and 36.7 in 1974 (t(l551) = 39.97,~ -=c.OOl). By
contrast, the ratings of Ford were quite stable between 1974 (62.9) and
1976 (61.8), a decrease of only marginal significance even with a huge
N (t(1526) = 1.96, p = -05). Perceived similarity to Nixon also was re-
duced significantly (t(249) = 6.78, p < .OOl) from 1972 (21.4) to 1974
(19.3), while perceived similarity to Ford did not change significantly
(t(359) = 0.55, ns) from 1974 (21.4) to 1976 (21.3).
Figure 1 presents the cross-lagged panel correlations for the variables of
attraction and perceived similarity to Nixon and Ford. The synchronous
correlations (between attraction and perceived similarity at a given time)
are all moderately large and consistent with other evidence that these two
variables are not independent of each other. The relative stability of the
synchronous correlations (.65, .60, .57, .66) is consistent with the assump-
tion that the causal relationship between the variables is stable or station-
ary. The autocorrelations (between the same variable at different points in
time) are also quite high and tend to be slightly higher for perceived
similarity (.70 and .67) than for attraction (.59 and .57).’
For our purposes, however, the most important comparison is between
the cross-lagged correlations. In the case of Nixon, it is clear that attrac-
tion in 1972 was more strongly linked to perceived similarity in 1974 than
perceived similarity in 1972 was to attraction in 1974. The difference
between these two correlations (.58 and .41) was significant (z = 3.09,~ <
.Ol). This is a strong indication that predominant causal flow was from
attraction to perceived similarity. At the same time, it does not preclude
the possibility that there still could be some causal flow from perceived
similarity to attraction.
The results for Ford are much more equivocal. Figure 1 shows the
cross-lagged correlations between perceived similarity and attraction for
Ford in 1974 and 1976 were nearly the same 1.49 and .48, z = 0.21). Thus,
neither of the panel analyses produced evidence supportive of a predomi-
nant causal flow from perceived similarity to attraction.

* The high autocorrelation for attraction to Nixon is noteworthy because it indicates that
in spite of the large decrease in attraction to Nixon between 1972 and 1974, responses in 1974
were still strongly and positively correlated to those in 1972. A negative correlation could
have resulted if those who were most positive in 1972 felt especially betrayed and made
extremely low ratings in 1974, but it is clear that this did not occur on the whole.
ATTRACTION AND PERCEIVED SIMILARITY 577

FIG. 1. Cross-lagged panel correlations between attraction to a politician and perceived


similarity on issues, based on the University of Michigan Center for Political Studies
1972-1974-1976 Panel Study. For the correlations involving Richard Nixon, N = 250; for
the correlations invoking Gerald Ford, N = 359.

The validity of an inference based on a comparison of cross-lagged


correlations depends upon certain assumptions. Several of these will be
mentioned in the following section, but the assumption of stationarity is
brought up here because it is treated statistically through multivariate
analysis. A comparison of cross-lagged panel correlations is based on the
assumption that the causal structure of the variables does not change
across time. This assumption of stationarity is supported if the synchro-
nous bivariate correlations remain the same across time.
However, stationarity can be established more effectively by including
control variables in multivariate analyses that ideally are both stable
themsetves and are related to the variables in the cross-lagged compari-
sons. With these criteria in mind, we chose party identification (measured
on a 7-point scale from Strong Democrat to Strong Republican), educa-
tion, and sex as control variables. The least stable of these, party iden-
tification, has a correlation of .80 between 1972 and 1974 and .83 between
1974 and 1976. Of the three control variables, party identification was
more strongly correlated with the variables in the cross-lagged correlations
than the other two control variables, and thus it was important to include
party identification as well as sex and education as control variables in the
multivariate analyses.
In the case of Nixon, the p (standardized regression coefficient) for the
effect of perceived similarity in 1972 on attraction in 1974, controlling for
perceived similarity in 1974, was - .02, F( 1, 248) = . 11, ns. When controls
578 GRANBERG AND KING

for party identification, sex and education were added, that P remained
nonsignificant at -.08, F(1, 246) = 1.43, ns.
On the other hand, the 0 for the effect of attraction to Nixon in 1972 on
perceived similarity to Nixon in 1974, controIiing for attraction in I974
was a highly significant .34, F(1, 248) = 32.03,~ < ,001. When the controls
for party, sex, and education were added, this /3 remained significant at
.30, F(1,246) = 22.19, p < .OOl.
In the case of Ford, the results of the regression analyses are especially
interesting, given that the bivariate cross-lagged correlations for attrac-
tion and perceived similarity were not significantly different. For Ford,
the ,6 for the effect of perceived similarity in 1974 on attraction in 1976,
controlling for perceived similarity in 1976, was .08, F(1, 357) = 2.33, ns.
When controls for party, sex, and education were added, that /3 remained
nonsignificant at .05 F(1, 355) = .81.
The /3 for the effect of attraction to Ford in 1974 on perceived similarity
to Ford in 1976, controlling for attraction in 1976, was .16, F(1, 357) =
10.70,~ < .OOl. That j3 remained significant at .17, F(1, 355) = 12.64,~ <
,001, when controls for party, sex, and education were added.
The results of the multivariate analyses support the earlier conclusion
that for Nixon, the predominant flow of causality was from attraction to
perceived similarity, These analyses also suggest that the same may have
even been true for Ford, although this was not apparent in the comparison
of the cross-lagged correlations.
DISCUSSION
In addition to the assumption of stationarity mentioned previously,
cross-lagged panel correlation also requires other assumptions. It assumes
synchronicity of measurement of variables and this assumption is well
met by the data. Also, if one variable increases in reliability over time, it
will appear to be the effect in cross-lagged analysis. We have no measure
of reliability, but there is no reason to think that the perceived similarity
scale is less reliable than the feeling-thermometer measure of attraction or
that perceived similarity increased in reliability from 1972 to 1974. Cross-
lagged panel analysis further requires that the variables do change some-
what over time, and no problem is encountered on that point. In spite of
the fact that the mean attraction to Nixon changed much more between
1972 and 1974 than the mean attraction to Ford changed between 1974 and
1976, the autocorrelations for attraction were nearly the same for Nixon
and Ford.
Another assumption is that of linear relationship. The possibility of
significant nonlinear relations has not been considered in this article, but
the assumption of linearity seems reasonable for the variables we used. It
is also necessary for our analysis to assume that the interval of 2 years was
appropriate for causal relations to become manifest.
ATTRACTION AND PERCEIVED SIi%IEARITY 579

Overall, the results in these panel analyses were consistent With the
assertions that attraction and perceived similarity are related and that one
is at least as justified in contending that attraction leads to perceived
similarity as the reverse. In the case of Nixon, a plausible j~terpretati~~
consistent with the evidence is that people, for reasons quite unrelated to
positions on political issues, became less attracted to Nixon and that this
change, in turn, led to and may have caused a decrease in perceived
similarity.
The results for Ford were different. Ford was ~nd~~bt~dly a less
familiar politician in 1974 to the U.S. electorate than Nixon was in 1972.
The possibility that the relationship between attraction and perceived
similarity may vary at different stages in the ac~~ai~ta~ce process is not
exactly novel (Newcomb, 1961). The evidence from the cross-lagged
comparisons involving Ford suggests the possibility that at the early
stages of getting to know a politician, the relationship between attraction
and perceived similarity may be nonrecursive. owever, when a politi-
cian falls from grace, as Nixon did between 1 2 and 1974, it appears
probable that changes in attraction lead to changes in perceived similar-
ity. (An exception to this may of course occur when a politician falls
disfavor with the public because of a departure from acceptable bo
on political issues.)
The question remains as to how the relatio~s~~~ between perceived
similarity and attraction works in the real world. No laboratory experi-
ment or series of experiments manipulating similarity has the external
validity that would offer assurance that under natural circumstances,
perceived similarity causes attraction (Byrne et al., 1969). The way it
probably works for the average person who is neither an ideologue nor
apathetic might be as follows. The person finds something appealing or
r;ttracGve about the politician; this could be a matter of appearance or
style, a shared membership or reference group, or it could be a position or
part of a position taken by the politician on some issue. Then> if one
asked to attribute positions to that politician on a series of issues, t
estimates are in the direction of one’s own attitudes. The ~~der~y~~~
process may work differently for people with differing levels of ~o~~tica~
involvement. A more complete theory than what we have now would be of
necessity mare complex. It would have to take into account the fact that
some people support politicians because of their position on the issues,
others do so in spite of the politician’s positions on the issues, and others
support a politician regardless of issue position. The latter is similar to the
sociological concept of a charismatic leader. There is evidence uric
indicates that supporters of Eugene McCarthy and obert Kennedy in
1968 were different in this regard. McCarthy’s supporters seem to have
been drawn to him because of his position, while Kennedy’s supporters
favored him regardless of his position and many of them would, indeed,
580 GRANBERG AND KING

have supported him had his position on the issues been the opposite of
what they were (Chester, Hodgson, & Page, 1969).

REFERENCES
Aronson, E., Turner, J., & Carlsmith, J. Communicator credibility and communication
discrepancy as determinants of opinion change. Journal of Abnormal and Social
Psychology, 1963, 67, 31-36.
Berelson, B., Lazersfeld, P., & McPhee, W. Voting: A study of opinion formation in a
presidential election campaign. Chicago: Univ. of Chicago Press, 1954.
Byrne, D. Interpersonal attraction and attitude similarity. Journal of Abnormal and Social
Psychology, 1961, 62, 713-15.
Byrne, D. Attitudes and attraction. In L. Berkowitz (Ed.), Advances in experimental social
psychology. New York: Academic Press, 1969. Vol. 4.
Byrne, D. The attraction paradigm. New York: Academic Press, 1971.
Byrne, D., Bond, M., & Diamond, M. Response to political candidates as a function of
attitude similarity-dissimilarity. Human Relations, 1969, 22, 251-62.
Byrne, D., Ervin, C., & Lamberth, J. Continuity between the experimental study of
attraction and real-life computer dating. Journal of Personality and Social Psychology,
1970, 16, 157-6.5.
Chester, L., Hodgson, G., & Page, B. An American melodrama: The presidential campaign
of 1968. New York: Viking, 1%9.
Crano, W., & Mellon, P. Causal infIuence of teachers’ expectations on children’s academic
performance: A cross-lagged panel analysis. Journal of Educational Psychology, 1978,
70, 39-49.
Feather, N. A structural balance approach to the analysis of communication effects. In L.
Berkowitz (Ed.), Advances in experimental so&f psychology. New York: Academic
Press, 1967. Vol. 3.
Granberg, D., & Brent, E. Dove-hawk placements in the 1968 election: Application of social
judgment and balance theories. Journal of Personality and Social Psychology, 1974,29,
687-95.
Granberg, D., & Jenks, R. Assimilation and contrast effects in the 1972 election. Human
Relations, 1977, 30, 623-40.
Granberg, D., & Seidel, J. Social judgments of the urban and Vietnam issues in 1968 and
1972. Social Forces, 1976, 55, l-15.
Griffith, W., & Veitch, R. Preacquaintance attitude similarity and attraction revisited: Ten
days in a fall-out shelter. Sociometry, 1974, 37, 163-73.
Kahle, L., & Berman, J. Attitudes cause behaviors: A cross-lagged panel analysis. Journal
of Personality and Social Psychology, 1979, 37, 315-21.
Kenny, D. Cross-lagged panel correlation: A test for spuriousness. Psychological Bulletin,
1975, 82, 887-903.
Kinder, D. Political person perception: The asymmetrical influence of sentiment and choice
on perceptions of presidential candidates. Journal of Personality and Social Psychol-
ogy, 1978, 36, 859-71.
King, M: Assimilation and contrast of presidential candidates’ issue positions, 1972. Public
Opinion Quarterly, 1978, 41, 515-22.
Newcomb, T. The acquaintance process. New York: Holt, Rinehart & Winston, 1961.
Page, B., & Brody, R. Policy voting and the electoral process: The Vietnam war issue.
American Political Science Review, 1972, 66, 979-95.
Sampson, E., & Insko, C. Cognitive consistency and performance in the autokinetic situa-
tion. Journal of Abnormal and Social Psychology, 1964, 68, 184-92.
Sherif, M. A study of some social factors in perception. Archives ofPsychology, 1935,
27(No. 187), l-60.
ATTRACTION AND PERCEIVED SIMILARITY 581

Sherrod, D. Selective perception of political candidates. Public Q&ion Quarterly, 1972.35,


554-62.
Weisberg, FL, & Rusk, J. Dimensions of candidate evaluation. American Political Science
Review, 1970, 64, 1167-85.
Wyer, R. Cognitive organization and change: An information processing approach. New
York: Wiley, 1974.

You might also like