You are on page 1of 18

207

British Journal of Social Psychology (2004), 43, 207–224


© 2004 The British Psychological Society
www.bps.org.uk

Affective and cognitive control of persons and


behaviours

David Trafimow1*, Paschal Sheeran2, Bridget Lombardo1,


Krystina A. Finlay1, Jennie Brown1 and Christopher J. Armitage2
1
New Mexico State University, USA
2
University of Sheffield, UK

Three studies assessed the relative contribution of affect and cognition to


determining behavioural intentions for a variety of behaviours using both between-
participants and within-participants analyses. The between-participants analyses
showed that affect tends to make more of a contribution than does cognition for
more behaviours. However, the within-participants analyses indicated that there are
strong individual differences among people. Some people are more under affective
control, across behaviours, whereas other people are more under cognitive control.
The most interesting finding was that, despite the potential independence of
between-participants and within-participants analyses (Mischela, 1990), between-
participants analyses on subsamples created from the within-participants analyses
showed significant dependence. The predictive validity of affect vs. cognition
depended upon whether participants were affectively or cognitively controlled.

During the last couple of decades, the classical dichotomy between feeling and
knowing as two facets of human experience has been invoked in order to understand
better why people perform behaviours. More specifically, it has become fashionable
for researchers to distinguish between ‘affective’ (i.e. feeling) and ‘cognitive’ (i.e.
knowing) determinants of behaviours. Affect seems to be more important for
determining some behaviours whereas cognition seems to be more important for
others, though there is a lack of systematic research pertaining to this issue. There may
also be individual differences in whether, across a wide range of behaviours, people
are more under affective or cognitive control. This article is concerned with both
issues: do behaviours tend to be more under affective or cognitive control (across
people) and are particular people more under affective or cognitive control (across
behaviours)?

*Correspondence should be addressed to David Trafimow, Department of Psychology, MSC 3452, New Mexico State
University, PO Box 30001, Las Cruces, NM 88003-8001, USA (e-mail: trafimow@crl.nmsu.edu).
208 David Trafimow et al.

Affect, cognition and behaviours


A good deal of evidence exists that affect and cognition are distinguishable from each
other, and that behaviours can be driven more by affect or more by cognition
(Abelson, Kinder, Peters, & Fiske, 1982; Breckler, 1984; Breckler & Wiggins, 1989;
Crites, Fabrigar, & Petty, 1994; Eagly, Mladinic, & Otto, 1994; Mann, 1959; Millar &
Tesser, 1986; Ostrom, 1969; Pfister & Böhm, 1992; Trafimow & Sheeran, 1998;
Verplanken, Hofstee, & Janssen, 1998). For example, researchers have induced partici-
pants to focus more on affect or more on cognition; participants who focus on affect
show greater affect-behaviour (or affect-attitude) consistency whereas those who focus
on cognition show greater cognition-behaviour (or cognition-attitude) consistency
(Crites et al., 1994; Millar & Tesser, 1986). The primacy of affect or cognition for
determining different behaviours can also be demonstrated without presenting special
instructions to participants. For example, Trafimow and Sheeran (1998) measured
affect, cognition, and intentions to smoke cigarettes and to study during the winter
vacation. They found that although affective beta-weights were greater than cognition
beta-weights for predicting intentions to smoke cigarettes; the reverse was true for
predicting study during the winter vacation.
Given that affect and cognition both seem to be important determinants of at least
some intentions, it seems desirable to know which of these two variables is more
important for predicting intentions to perform most behaviours. The theory of rea-
soned action tradition (e.g. Fishbein, 1980) suggests that cognition is more important.
As the name of the theory implies (i.e. ‘reasoned action’), intentions are presumed to
be based on an implicit expectancy-value formulation that takes both the probability
and the value of various potential consequences into account. Although people may
make wrong assumptions about either probabilities or values of potential conse-
quences, the theory of reasoned action assumes that given the assumptions people
make, their intentions are the result of a reasoned process. Thus, cognition would
seem to be implicated as the primary determinant of most behavioural intentions. Eagly
et al. (1994) also suggest that cognition is more important than affect.
On the other hand, research from an evolutionary perspective suggests that affect
might be more important than cognition as a determinant of most behaviours
(Johnston, 1999). To understand Johnston’s reasoning, let us consider the issue of why
affect would have evolved in humans. If the environment had been constant in our
evolutionary history, a set of fixed action patterns could have evolved that would have
allowed us to respond appropriately to stimuli in our environment, and the ability to
adapt to changing environments would not have provided a survival and reproductive
advantage. However, previous environments were not constant and thus the ability to
adapt to change, within an organism’s lifetime, did provide such an advantage.
Johnston argued that affect provides meaning to stimuli, which aids in learning.
Specifically, affect functions as an amplification device that increases the speed with
which learning can take place and also decreases the amount of time necessary for
adaptation. In support of his theorizing, Johnston performed simulations that, though
too complex to describe in detail here, provide a compelling case for the importance
of affect in learning and decision making. Reinforcing Johnston’s argument, Damasio
(1994) has shown that patients with brain lesions in ‘affective’ areas (particularly in the
frontal cortices), even when they have no detectable cognitive deficits, are unable to
respond appropriately to social events and are unable to make good decisions. Thus,
the simulations performed by Johnston and the data obtained by Damasio converge to
suggest that affect is primarily responsible for behavioural intentions (and ultimately,
Affective and cognitive control 209

behaviours). Johnston does not eschew the importance of cognition—he suggests that
cognition increases our behavioural flexibility; however, it is affect that provides the
motivational power for behaviours. Thus, Johnston’s theorizing suggests that affect
should be more important than cognition for predicting most intentions.
The obvious way of testing these competing perspectives is to measure affect,
cognition and intentions to perform a wide variety of behaviours. If cognitive beta-
weights, on average, turn out to be larger than affect beta-weights, then the general
importance of cognition over affect would be supported. If the reverse were true, then
the general importance of affect over cognition would be supported.
Unfortunately, even if this analysis were carried out, it would not provide an
unambiguous answer. This is because there would be at least two possible expla-
nations for the findings. Specifically, one of the variables might, on average, receive
larger beta-weights because it actually is more important for most behaviours (i.e. a
behaviour effect). However, an alternative possibility is that one of the variables
might, on average, receive larger beta-weights because it is more important for more
people (i.e. a person effect). It is also possible that there is some truth to both
possibilities.
The possibility of person effects, as well as behaviour effects, can be tested by
performing within-participants analyses to determine for which people cognition or
affect is more important (see Trafimow & Finlay, 1996, for similar tests pertaining to
attitudes and subjective norms). These within-participants analyses, which are per-
formed across behaviours, result in cognition and affect beta-weights for each partici-
pant. If the cognition beta-weight for a particular participant is greater than his or her
affect beta-weight, then the person can be considered to be more under cognitive than
affective control across the behaviours. On the other hand, if the affect beta-weight is
greater than the cognitive one, then the person can be considered to be more under
affective than cognitive control across behaviours.
What is gained by categorizing individual people as being affectively or cognitively
controlled? The answer lies in the analyses of subsamples that can be performed.
Between-participants analyses can be performed on a subsample of cognitively
controlled participants, or on a subsample of affectively controlled participants. If the
results of the between-participants analyses are similar in both subsamples, the implica-
tion is that the type of person is not an important factor for determining between-
participants beta-weights. However, if the results of the between-participants analyses
differ for the two subsamples, the implication is that the type of person does matter.

STUDY 1
Study 1 was performed to test whether cognition or affect was a better predictor of
intentions for a variety of behaviours. Analyses of subsamples of cognitively or
affectively controlled participants were also conducted to determine whether the type
of person, as well as the type of behaviour, contributed to between-participants beta-
weights.

Method
Participants
The participants were 87 undergraduate students at a North American university who
were volunteers.
210 David Trafimow et al.

Procedure
A group of undergraduates who were blind to the purpose of the study generated 14
behaviours in as random a manner as possible for use in Study 1 (i.e. they were asked
to generate behaviours at random and no reason for doing so was given). Then, in the
main part of the study, participants responded to items measuring their intentions to
perform these behaviours, affect towards the behaviours, and cognition towards the
behaviours (Table 1 contains a list of all of the behaviours). For example, on an
intention item, participants responded to a 7-point scale ranging from −3 (extremely
not intend) to 3 (extremely intend) to behaviours such as ‘keep my room clean’. Affect
was measured on 7-point scales ranging from −3 (extremely unpleasant) to 3
(extremely pleasant), and cognition was measured on 7-point scales ranging from –3
(extremely foolish) to 3 (extremely wise). Trafimow and Sheeran (1998) demonstrated
that items such as these provide reliable and valid measures of affect and cognition.
Further, a variety of studies (Finlay, Trafimow, & Jones, 1997; Finlay, Trafimow, &
Moroi, 1999; Trafimow & Finlay, 1996; Trafimow & Finlay, 2001) have demonstrated
that similar single item scales can be used reliably and validly.

Results
We performed between-participants analyses on each of the behaviours across all of
the participants and within-participants analyses on each of the participants across all
Affective and cognitive control 211

of the behaviours. Finally, we also performed between-participants analyses on


subsamples obtained from the within-participants analyses.

Between-participants analyses
Table 1 summarizes the between-participants analyses including all of the participants.
Affect and cognition were moderate predictors of intention when considered
separately (median affect–intention correlation = .56, median cognition–intention cor-
relation = .48), and the prediction of intention improved when affect and cognition
were considered together in multiple regression analyses (median multiple correlation
was .65). In addition, the median affect beta-weight was .46 and the median cognition
beta-weight was .25. Finally, the median unique variance in intentions accounted
for by affect (R2 − r2cog–int) was .15 and the median unique variance in intentions
accounted for by cognition (R2 − r2aff–int) was .05.
There is some evidence that the behaviours were more influenced by affect than by
cognition. First, the median affect–intention correlation was greater than the median
cognition–intention correlation (medians are .56 and .48 across the 14 behaviours)
(Wilcoxon test: mean ranks are 7.86 and 6.17, Z = 2.13, p < .04). Secondly, the median
affect beta-weight was greater than the median cognition beta-weight (medians are .46
and .25) (Wilcoxon test: mean ranks are 8.05 and 5.50, Z = 2.26, p < .03). Finally, the
median unique variance in intentions accounted for by affect was greater than that
accounted for by cognition (median proportions are .15 and .05) (Wilcoxon test: mean
ranks are 7.77 and 6.50, Z = 2.07, p < .04).

Within-participants analyses
The within-participants prediction of intentions from affect and cognition was even
better than was true when between-participants analyses were conducted. Affect and
cognition were good predictors of intention both singly (median affect–intention
correlation was .75 and median cognition–intention correlation was .75), and in combi-
nation in a multiple regression analysis (median multiple correlation was .84). The
median affect beta-weight was .41 and the median cognition beta-weight was .44.
Finally, the median unique variance in intentions accounted for by affect and cognition
were .07 and .07, respectively.

Separate analyses for people under affective or cognitive control


Is there any evidence that the influence of affect or cognition on behaviours in the
between-participants analyses is due, in part, to the inclusion of people under affective
or cognitive control in the sample? To address this question, we categorized people
as being generally under affective or cognitive control on the bases of the within-
participants analyses, and then performed separate between-participant analyses on
each subsample. Thus, if a particular participant’s affect beta-weight was greater than
his or her cognition beta-weight, then this participant was categorized as being more
under affective than cognitive control and was included in the subsample containing
participants under affective control. If the reverse was true, then the participant was
included in the subsample containing participants under cognitive control.
To test whether the affective or cognitive control of persons influenced the
between-participants contributions of affect or cognition to behavioural intentions,
we simply compared the unique variance in intentions accounted for by affect or
212 David Trafimow et al.

cognition across the two subsamples. A glance at Tables 2 and 3 shows that for 13 of
the 14 behaviours, the unique variance in intentions accounted for by affect was
greater for people under affective than cognitive control; and the unique variance
accounted for by cognition was greater for the people under cognitive than affective
control (binomial test: p < .01 in both cases). A Mann–Whitney test across the two
subsamples results in a similar conclusion. The median unique variance in intentions
accounted for by affect was .42 vs. .05 for the subsamples under affective or cognitive
control, respectively (mean ranks are 7.92 and 2.00, Z = 3.17, p < .01); and the median
unique variance in intentions accounted for by cognition was .16 and .01 for the
subsamples under cognitive or affective control, respectively (mean ranks are 7.77 and
4.00, Z = 3.04, p < .01). Clearly, whether people are under affective or cognitive
control does affect the contributions of affect and cognition in predicting specific
behavioural intentions.

STUDY 2
Study 2 was performed as a conceptual replication of Study 1 in order to determine
whether the findings would hold up with a new set of randomly selected behaviours.

Method
Participants
The participants were 116 undergraduates at a North American university who
volunteered to take part in the study.
Affective and cognitive control 213

Procedure
The procedure was similar to that in Study 1 except for two changes. First, participants
responded to 19 randomly selected behaviours rather than 14 as in Study 1 (see
Table 4). The reason for using different numbers of behaviours in different studies was
to show that the actual number of behaviours used does not affect the findings in an
important way. Secondly, to ensure that the greater importance of affect compared to
cognition in the between-participants analyses of Study 1 was not due to measuring
cognition in a less valid way than affect, we used a different measure of cognition in
the present study (see Trafimow & Sheeran, 1998). Specifically, participants responded
on 7-point scales ranging from −3 (extremely harmful) to 3 (extremely beneficial)
rather than on scales ranging from ‘extremely foolish’ to ‘extremely wise’.

Results
As in Study 1, we performed between-participants analyses on each of the behaviours
across all of the participants, within-participants analyses on each of the participants
across all of the behaviours, and between-participants analyses on the individual
subsamples of participants who were under affective or cognitive control.

Between-participants analyses
Table 4 summarizes the between-participants analyses including all of the participants.
Affect and cognition were moderate predictors of intention when considered
separately (median affect–intention correlation = .63, median cognition–intention
correlation = .40), and the prediction of intention improved when affect and cognition
214 David Trafimow et al.

were considered together in multiple regression analyses (median multiple correlation


was .64). In addition, the median affect beta-weight was .53 and the median cognition
beta-weight was .17. Finally, the median unique variance in intentions accounted
for by affect (R2 − r2cog–int) was .17 and the median unique variance in intentions
accounted for by cognition (R2 − r2aff–int) was .03.
As in Study 1, and despite the use of a different measure of cognition, behaviours
were again more influenced by affect than by cognition. First, the median affect–
intention correlations were greater than the median cognition–intention correlations
(medians are .63 and .40) (Wilcoxon test: mean ranks are 14.25 and 9.50, Z = 2.68,
p < .01). Secondly, the median affect beta-weights were greater than the median
cognition beta-weights (medians are .53 and .17) (Wilcoxon test: mean ranks are 11.50
and 9.82, Z = 2.90, p < .01). Finally, the median unique variance in intentions
accounted for by affect was greater than that accounted for by cognition (median
proportions are .17 and .03) (Wilcoxon test: mean ranks are 10.35 and 7.00, Z = 3.26,
p < .01).

Within-participants analyses
The median within-participants correlations between cognition and intention and
between affect and intention were .57 and .52, respectively. The median within-
participants multiple correlation was .67; affect and cognition were both reliable
predictors of intention (median betas = .42 and .34, respectively). Finally, the median
Affective and cognitive control 215

unique variance in intentions accounted for by cognition was .12 compared to .08 for
affect.

Separate analyses for people under affective or cognitive control


As in Study 1, we performed separate between-participants analyses on the affectively
or cognitively controlled subsamples. Inspection of Tables 5 and 6 shows that for 17 of
the 19 behaviours (there was one tie and one behaviour in the other direction), the
unique variance in intentions accounted for by affect was greater for people under
affective than cognitive control; and for 13 of the 19 behaviours (there was one tie
and five behaviours in the opposite direction) the unique variance accounted for by
cognition was greater for the people under cognitive than affective control. Thus, for
the 19 (behaviours) × 2 (variables: unique variance accounted for by affect and
cognition) = 38 cases, 30 comparisons were in the predicted direction, six were in
the other direction, and there were two ties (binomial test: p < .01). More precise
Mann-Whitney tests across the two subsamples imply a similar conclusion. The median
unique variance in intentions accounted for by affect was .27 vs. .13 for the
subsamples under affective or cognitive control, respectively (mean ranks are 7.92 and
2.00, Z = 3.17, p < .01); and the median unique variance in intentions accounted for by
cognition was .00 and .05 for the subsamples under affective or cognitive control,
respectively (mean ranks are 4.00 and 7.77, Z = 3.04, p < .01). In sum, the findings
from Studies 1 and 2 indicate that individual differences in whether people are
under affective or cognitive control have an important effect on how well affect and
cognition predict specific behavioural intentions.

STUDY 3
Three possible objections might be levelled at the analyses that we have presented so
far. First, although we used pilot studies to ensure that our selection of behaviours for
the main studies was random, it is possible that some systematic biases of which we
are unaware could have characterized the selected behaviours. To overcome this
potential difficulty, we decided to adopt a more systematic approach to selection. In
particular, we drew upon the classification of behaviour types employed in Kim and
Hunter’s (1993) comprehensive meta-analysis of the impact of topic on attitude-
behaviour relations (see also Canary & Seibold, 1984). Fifteen behaviours were chosen
to represent each of the following categories from Kim and Hunter’s taxonomy:
altruistic behaviour, consumer behaviour, ‘deviance’, environmental behaviours,
health behaviours, group participation, ‘race relations’, religion, drug and alcohol use,
class attendance, family planning, blood donation and classroom behaviour.1
A second potential objection to our analyses is that the measures of affect and
cognition involved single items which can be associated with poor reliability. The fact
that the bivariate and multiple correlations obtained in Studies 1 and 2 are large
(cf. Cohen, 1992) suggests that poor reliability did not characterize our measures
(since there is a direct relationship between reliability and the strength of correla-
tions). Nonetheless, just to be sure, we decide to use multi-item measures in our third
1
Kim and Hunter’s (1993) taxonomy involved 20 types of behaviour. Five of these behaviour types would not have been
meaningful or relevant to our sample and were, therefore, excluded. The excluded behaviour types were ‘voting’ (there
were no elections around the time of data collection), ‘game behaviour: game performance in laboratory’, ‘migration:
moving, applying for public housing’, ‘verdict: mock court decision’ and ‘miscellaneous; any other topics that do not belong
to any of the above categories’ (p. 115).
216 David Trafimow et al.

study. The items enjoyable–not enjoyable and pleasant–unpleasant were used to


index affect, and useful–useless and worthwhile-pointless were the cognition items
(a principle components analysis reported by Trafimow and Sheeran, 1998, demon-
strated that these items loaded highly on affect and cognition factors and demonstrated
good reliability and validity).
Because we are trying to show that the type of person affects the strength of
affect–intention and cognition–intention relations, we also sought to determine
whether there was an association between affective vs. cognitive control and a
relevant individual difference variable. The measure we chose was Maio and Esses’
(2001) Need for Affect scale. This scale measures the extent to which people are
motivated to approach emotion-inducing situations or to avoid such situations. We
predicted that affectively controlled participants would show greater need for affect
compared to cognitively controlled participants. Finally, to further enhance the gener-
ality of our analyses, Study 3 was conducted in a different country (the UK) than our
previous two studies (the USA).

Method
Participants
The participants were 89 undergraduates at a UK university who volunteered to take
part in the study.
Affective and cognitive control 217

Procedure
Participants responded to 15 behaviours selected to represent 15 categories relevant to
undergraduate participants from Kim and Hunter’s (1993) taxonomy (see Table 7).
Participants responded on 7-point scales ranging from −3 to +3 for all measures. The
intention measure was ‘Do you intend to do each of the following in the next month?’
(definitely yes–definitely no). Affect was measured by responses to the stems ‘How
enjoyable or not [pleasant or unpleasant] would it be to do each of the following in the
next month?’ (enjoyable–not enjoyable, pleasant–unpleasant) whereas cognition
was measured by responses to the stem ‘How worthwhile or pointless [useful or not]
would it be to do each of the following in the next month?’ (useful–useless,
worthwhile–pointless). The affect and cognition items proved highly reliable. The
median alpha for affect across the 15 behaviours was .85 (median inter-item r = .73)
and the median alpha for cognition was .83 (median inter-item r = .72). (There were no
differences in the alphas or inter-items rs for affect vs. cognition, both ps > .13). We
used the average of the two items to compute affect and cognition scales.
Participants also completed Maio and Esses’ (2001) need for affect scale. This
measure comprises 26 items to which participants respond on 7-point strongly
disagree–strongly agree scales (e.g. ‘It is important for me to be in touch with my
feelings’, ‘I would love to be like “Mr Spock” who is totally logical and experiences
little emotion’. The scale exhibited satisfactory reliability among the present sample
(alpha = .80).
218 David Trafimow et al.

Results
Again, we performed between-participants analyses on each of the behaviours across
all of the participants, within-participants analyses on each of the participants across all
of the behaviours, and between-participants analyses on the individual subsamples of
participants who were under affective or cognitive control.

Between-participants analyses
Table 7 summarizes the between-participants analyses for the whole sample. The
median affect–intention and cognition–intention correlations were .66 and .55,
respectively. The median multiple correlation was .76 when affect and cognition were
the predictors of intention in a multiple regression analysis. The median affect beta-
weight was .47 and the median cognition beta-weight was .25. Moreover, the median
unique variance in intentions accounted for by affect (R2 − r2cog–int) was .17 and
the median unique variance in intentions accounted for by cognition (R2 − r2aff–int)
was .05.
In the same manner as Studies 1 and 2, we used Wilcoxon tests to compare the
correlations, beta-weights and unique variance associated with affect vs. cognition.
First, the median affect–intention correlation was greater than the median cognition–
intention correlation (medians are .66 and .55) (Wilcoxon test: mean ranks are 8.60
and 4.75, Z = 2.10, p < .04). Secondly, the median affect beta-weight was greater than
the median cognition beta-weight (medians are .47 and .25) (Wilcoxon test: mean
ranks are 9.00 and 5.25, Z = 2.22, p < .03). Finally, the median unique variance in
intentions accounted for by affect was greater than that accounted for by cognition
(medians are .17 and .03) (Wilcoxon test: mean ranks are 8.86 and 5.63, Z = 2.13,
Affective and cognitive control 219

p < .04). These findings are virtually identical to those obtained in our previous two
studies.

Within-participants analyses
The median within-participants correlations between cognition and intention and
between affect and intention were .67 and .71, respectively. The median within-
participants multiple correlation was .75; affect and cognition were both reliable
predictors of intention (median betas = .44 and .24, respectively). Finally, the median
unique variance in intentions accounted for by affect was .09 compared to .04 for
cognition.

Separate analyses for people under affective or cognitive control


We performed separate between-participants analyses on the affectively or cognitively
controlled subsamples. Tables 8 and 9 show that for 14 of the 15 behaviours (there was
one behaviour in the other direction), the unique variance in intentions accounted for
by affect was greater for people under affective than cognitive control; and for eight of
the 15 behaviours (there were seven behaviours in the opposite direction) the unique
variance accounted for by cognition was greater for the people under cognitive than
affective control. Thus, for the 15 (behaviours) × 2 (variables: unique variance ac-
counted for by affect and cognition) = 30 cases, 22 comparisons were in the
predicted direction and eight were in the other direction (binomial test: p < .01).
Mann–Whitney tests that compared the two subsamples implied a similar conclusion.
The median unique variance in intentions accounted for by affect was .23 vs .12 for the
subsamples under affective or cognitive control, respectively (mean ranks are 19.70
and 11.30, Z = 2.62, p < .01); and the median unique variance in intentions accounted
for by cognition was .02 and .12 for the subsamples under affective or cognitive
control, respectively (mean ranks are 12.20 and 18.80, Z = 2.07, p < .04). Thus, the
findings from Study 3 are consistent with our previous two studies and support the
idea that the type of person (affectively vs. cognitively controlled) has
an important effect on how well affect and cognition predict specific behavioural
intentions.

Need for affect and affective vs. cognitive control


A final set of analyses examined whether need for affect (Maio & Esses, 2001) could
discriminate between affectively vs. cognitively controlled participants. Because
only participants with more extreme levels of affective vs. cognitive control are
relevant to this analysis, we subtracted the affect–intention and cognition–intention
within-participants correlations from one another. We then used the 20th and 80th
percentiles to designate participants as affectively vs. cognitively controlled (see
Sheeran, Trafimow, Finlay, & Norman, 2002 for equivalent analyses). Consistent with
predictions, affectively controlled participants showed higher need for affect than
did cognitively controlled participants (Ms = 5.03 and 4.59, SDs = 0.60 and 0.73,
respectively), t(38) = 2.11, p < .05. Thus, there is some evidence that affectively
vs. cognitively controlled participants can be distinguished on a related individual
difference variable.

Discussion
We began our investigations with the traditional premise that people’s behavioural
intentions are the result of a reasoned process (e.g. Eagly et al., 1994; Fishbein, 1980).
220 David Trafimow et al.

It follows from this assumption that the cognitive beta-weights should have exceeded
the affective beta-weights for most behaviours. In contradiction to this traditional
premise, the between-participants analyses demonstrated that exactly the reverse was
actually true. This finding, which has been replicated across three studies, suggests
that researchers should give affect more credit as a determinant of behavioural inten-
tions than has been true in the past. Further, the within-participants analyses allowed
us to investigate the roles of affect and cognition as individual differences variables; i.e.
to distinguish between affectively vs. cognitively controlled participants. When
subsamples were created on the basis of this distinction, and between-participants
analyses were again conducted on each of the subsamples, there was a significant
amount of dependence: affect was a more important determinant of intentions, even in
between-participants analyses, when an affectively controlled subsample was analysed;
but the reverse was true for a participant subsample that was under cognitive control.
Thus, the data obtained from all three studies provided unambiguous answers to both
of the questions we posed in the introduction. First, affect is more important than
cognition for most behaviours. Secondly, there are reliable individual differences in
whether the people are under affective or cognitive control, and these individual
differences play an important role in people’s behavioural intentions.
It is possible to question whether the between-participants and within-participants
analyses were really independent. If not, it is possible that the dependence that we
actually obtained was an artifact of the paradigm rather than reflecting the true state of
the world. Although there are reasons for considering the two types of analyses to
be independent (e.g. Mischela, 1990), we admit that there is a slight amount of
dependence. To see why, consider Table 10. Each cell contains a predictor variable (X)
and a criterion variable (Y). Between-participants analyses are conducted across rows
Affective and cognitive control 221

and within-participants analyses are conducted across columns. Consequently, each


between-participant analysis shares exactly one cell with each within-participant
analysis. Clearly, as the number of rows and columns (e.g. behaviours and persons)
increases, the percentage of shared cells, and therefore the amount of dependence,
decreases. In fact, Trafimow, Kiekel, and Clason (in press) proved that as the
number of rows and columns approach infinity, the amount of dependence
approaches zero. Further, these researchers conducted mathematical and computer
simulations showing that with a reasonable number of behaviours and participants
(e.g. 15 of each), the amount of dependence is so small that it can be considered to be
zero.
Actually, for our present purposes, it is not necessary to demonstrate independence.
Let us ignore the foregoing comments and pretend for the moment that the single
shared cell causes an important amount of dependence. In this case, our main concern
would be about whether the dependence increases or decreases the probability that
222 David Trafimow et al.

the two types of correlations would go together or go in opposite directions. In fact,


these probabilities are precisely equal! Thus, based strictly on probability, it should
have been just as likely for our analyses of subsamples to work out in the opposite
direction as to work out in the direction that they actually did. Consequently, given
both that the amount of dependence caused by a single shared cell is trivial to begin
with, and the equal probabilities regarding the direction of effects, it seems implausible
to argue that the impressive degree of dependence we obtained in all three studies is
an artifact of the methodology.

Implications for intervention


Although the basic thrust of our research is the theoretical issue of whether behaviours
and people are more controlled by affect or cognition, there are practical implications
too. Most obviously, for interventions, our research suggests that those based on
affect are more likely to be effective for affectively controlled behaviours and
interventions based on cognition are more likely to be effective for cognitively
controlled behaviours. In the absence of definite knowledge about whether the
behaviours under concern are more influenced by affect or cognition, the between-
participants analyses in all three studies suggests that affect is the better bet. However,
the within-participants analyses, and the strong evidence that the type of person is an
important determinant of the relative importance of affect and cognition even in
between-participants analyses, suggests that interventions should be based on the
types of persons involved as well as the types of behaviours of concern. Affect-based
interventions should work better than cognition-based interventions for affectively
controlled people, but not for cognitively controlled people. Thus, when both the
behaviours and participants are under affective control or when they are both under
cognitive control it seems clear that interventions should be designed accordingly.
However, when the behaviours of concern are under affective control, and the
participants involved are under cognitive control, then it is not clear what to do. In this
case, it might be worthwhile to have both a cognitive and affective component to the
intervention.
There are times when interventions are based on a set of problematic behaviours
that plague particular groups of people. Most research has tended to focus on single
behaviours and is consequently difficult to apply to sets of behaviours. However, the
present within-participants approach seems ideal for this purpose. Members of the
group of interest can be measured on affect and cognition (or any other variables, see
Finlay et al., 1997, 1999) and within-participants analyses can determine whether
these group members tend to be more under affective or cognitive control vis-à-vis the
behaviours of concern. The intervention can then be adapted accordingly.

Conclusion
The data bring up one puzzle that, although we do not have a good answer to it,
strongly underscores the importance of performing both between-participants and
within-participants analyses. Specifically, although affect was clearly much more
important than cognition (on average) when between-participants analyses were
performed, this difference was attenuated (on average) when within-participants
analyses were performed. The reason for these diverging effects is not clear. What is
clear, however, is that the two types of analyses test very different things, and can
suggest very different conclusions. Over 65 years ago, Allport (1937) warned that the
Affective and cognitive control 223

reliance of personality and social psychology on between-participants analyses had the


potential to be misleading. He pointed out that between-participants correlations tell
us nothing about any of the particular participants involved in the study at hand.
Rather, the most we can learn from between-participants correlations is evidence
about the nature of the ‘average’ mind—and this may not correspond to the mind of
any particular individual. Allport was not arguing that between-participants corre-
lations were completely uninformative, rather, he was arguing that such ‘nomothetic’
analyses should be complemented by within-participants or ‘idiographic’ analyses. The
combination of the two types of analyses provides a much more complete picture of
the phenomenon under investigation than does nomothetic analysis alone. The present
data provide a case in point. Had only between-participants analyses been conducted,
we would have concluded that cognition is not very important in the prediction of
behaviour—at least not when contrasted with the importance of affect. Instead, the
idiographic (within-participants) analyses tell us that cognition is just as important as
affect for influencing many intentions formed by particular individuals. We hope that
researchers will take this lesson to heart and that within-participants analyses will
become a standard tool for those who wish to gain a more sophisticated understanding
of the variables involved in social behaviours.

Acknowledgement
We thank Joshua Smith for his help in collecting and entering data.

References
Abelson, R. P., Kinder, D. R., & Peters, M. D., & Fiske, S. T. (1982). Affective and semantic
components in political person perception. Journal of Personality and Social Psychology,
42, 619–630.
Allport, G. W. (1937). Personality: A psychological interpretation. New York: Holt, Rinehart &
Winston.
Breckler, S. J. (1984). Empirical validation of affect, behaviour, and cognition as distinct
components of attitude. Journal of Personality and Social Psychology, 47, 1191–1205.
Breckler, S. J., & Wiggins, E. C. (1989). Affect versus evaluation in the structure of attitudes.
Journal of Experimental Social Psychology, 25, 253–271.
Canary, D. J., & Seibold, D. R. (1984). Attitudes and behaviour: A comprehensive bibliography.
New York: Praeger.
Cohen, J. (1992) A power primer. Psychological Bulletin, 112, 155–159.
Crites, S. L., Fabrigar, L. R., & Petty, R. E. (1994). Measuring the affective and cognitive proper-
ties of attitudes: Conceptual and methodological issues. Personality and Social Psychology
Bulletin, 20, 619–634.
Damasio, A. R. (1994). Descartes’ error: Emotion, reason, and the human brain. USA,
New York: Putnam.
Eagly, A. H., Mladinic, A., & Otto, S. (1994). Cognitive and affective bases of attitudes toward
social groups and social policies. Journal of Experimental Social Psychology, 30, 113–137.
Finlay, K. A., Trafimow, D., & Jones, D. (1997). Predicting health behaviours: Between-subjects
and within-subjects analyses. Journal of Applied Social Psychology, 27, 2015–2031.
Finlay, K. A., Trafimow, D., & Moroi, E. (1999). The importance of subjective norms on
intentions to perform health behaviours. Journal of Applied Social Psychology, 29,
2381–2393.
Fishbein, M. (1980). Theory of reasoned action: Some applications and implications. In H. Howe
& M. Page (Eds.), Nebraska Symposium on Motivation, 1979, (pp. 65–116). Lincoln, NB:
University of Nebraska Press.
224 David Trafimow et al.

Johnston, V. S. (1999). Why we feel: The science of emotions. Reading, MA: Helix Books.
Kim, M. S., & Hunter, J. E. (1993). Attitude–behaviour relations: A meta-analysis of attitudinal
relevance and topic. Journal of Communication, 43, 101–142.
Maio, G. R., & Esses, V. M. (2001). The need for affect: Individual differences in the motivation to
approach or avoid emotions. Journal of Personality, 69, 583–615.
Mann, J. H. (1959). The relationship between cognitive, affective, and behavioural aspects of
racial prejudice. Journal of Social Psychology, 49, 223–228.
Millar, M. G., & Tesser, A. (1986). Effects of affective and cognitive focus on the attitude-
behaviour relation. Journal of Personality and Social Psychology, 51, 270–276.
Mischela, J. L. (1990). Within-person correlational design and analysis. In C. Hendrick & M. S.
Clark (Eds.), Research methods in personality and social psychology (pp. 279–311).
London: Sage.
Ostrom, T. M. (1969). The relationship between the affective, behavioural and cognitive
components of attitude. Journal of Experimental Social Psychology, 5, 12–30.
Pfister, H.-R., & Böhm, G. (1992). The function of concrete emotions in rational decision making.
Acta Psychological, 80, 105–116.
Sheeran, P., Trafimow, D., Finlay, K. A., & Norman, P. (2002). Evidence that type of person
affects the strength of the perceived behavioural control intention relationship. British
Journal of Social Psychology, 41, 253–270.
Trafimow, D., & Finlay, K. A. (1996). The importance of subjective norms for a minority of
people: Between-subjects and within-subjects analyses. Personality and Social Psychology
Bulletin, 22, 820–828.
Trafimow, D., & Finlay, K. A. (2001). Evidence for improved sensitivity of within-participants
analyses in tests of the theory of reasoned action. Social Science Journal, 38, 629–635.
Trafimow, D., Kiekel, P., & Clason, D. (in press). The simultaneous consideration of between-
participants and within-participants analyses in research on indicators of behaviors: The issue
of dependence. European Journal of Social Psychology.
Trafimow, D., & Sheeran, P. (1998). Some tests of the distinction between cognitive and
affective beliefs. Journal of Experimental Social Psychology, 34, 378–397.
Verplanken, B., Hofstee G., & Janssen, H. J. W. (1998). Accessibility of affective versus cognitive
components of attitudes. European Journal of Social Psychology, 28, 23–35.

Received 9 April 2002; revised version received 21 January 2003

You might also like