Implicit Learning of Evaluative vs Non-Evaluative Covariations : The Role of Dimension Accessibility

Journal of Experimental Social Psychology, in press





Implicit Learning of Evaluative vs. Non-Evaluative Covariations:
The Role of Dimension Accessibility

Michael A. Olson
University of Tennessee

Richard V. Kendrick
University of Tennessee

Russell H. Fazio
Ohio State University


Address Correspondence to:
Michael A. Olson
Department of Psychology
Austin Peay Building
University of Tennessee
Knoxville, TN 37996
Email: olson@utk.edu
Phone: 865-974-8264
Fax: 865-974-3330

Word count: 4989

---

Implicit Learning 2
Abstract
Implicit covariation learning, the development of simple associations without awareness,
has been demonstrated repeatedly along the evaluative dimension (De Houwer, Thomas, &
Baeyens, 2001), but associations involving other dimensions appear more difficult to learn
implicitly. The present research highlights the unique properties of the evaluative dimension that
may predispose it to implicit learning. We provide evidence in the first experiment that implicit
covariation learning occurs along the evaluative dimension, but does not spontaneously occur
along non-evaluative dimensions. In Experiment 2, implicit learning along non-evaluative
dimensions occurred when participants were subliminally primed with the to-be-learned
dimension. In the discussion, we integrate findings from implicit evaluative conditioning
research with the broader implicit learning literature.


Key Words: Attitudes, Implicit learning, Evaluative conditioning, Priming

---

Implicit Learning 3
Implicit Learning of Evaluative vs. Non-Evaluative Covariations: The Role of Dimension
Accessibility
What are the limits to our ability to learn implicitly, without conscious awareness of what
was learned? The question is a controversial one, with some proponents claiming that much of
our general knowledge comes about through implicit learning processes (e.g., Frensch & Rünger,
2003; Reber, 1989, 1993), and others arguing that the evidence for implicit learning is limited, at
best (e.g., Shanks & St. John, 1994). The present research aims for a place in this debate by
investigating the boundary conditions of implicit covariation learning—the development of
associations on the basis of co-occurrences or contiguity between objects. We begin by
highlighting the evidence of implicit covariation learning on one dimension in particular:
evaluation. We then attempt to generalize this phenomenon to non-evaluative dimensions. In so
doing, we delineate some of the conditions that appear necessary for implicit covariation learning
to occur in non-evaluative domains.
Evidence for Implicit Attitude Learning

There has been a recent surge of interest in the origins of “implicit attitudes” (e.g.,
Rudman, 2004). Much of this research has centered on the phenomenon of evaluative
conditioning (EC), in which an object comes to take on the valence of items with which it is
repeatedly paired (De Houwer et al., 2001). Procedurally, EC largely resembles traditional
classical conditioning, in that a CS (the attitude object) is paired repeatedly with USs that already
evoke positive or negative reactions. Over time, the CS comes to elicit the response originally
elicited by the US; its presentation causes an evaluative response.
Research across varied EC paradigms indicate that attitudes can form and change through
conditioning-like procedures (De Houwer et al., 2001). For example, Olson & Fazio (2001,
2002) presented participants with a stream of non-rhythmic words and images under the guise of

---

Implicit Learning 4
an experiment about “attention and rapid responding.” Most of the items were filler words and
images that were unrelated to the conditioning itself; some appeared alone and others appeared in
pairs. Participants were instructed to be vigilant for a pre-specified target item (which was not
itself a CS) that appeared randomly throughout the procedure. Embedded in this task were
critical pairings of novel objects (CSs) and other valenced words and images (USs). One CS
consistently appeared with positive items, and another with negative items. Later in the
experiment, participants reported their evaluations of some of the items, including the two CSs.
On average, participants preferred the CS paired with positive items over the CS paired with
negative items (see also Baeyens, Eelen, Crombez, & Van den Bergh,1992; De Houwer,
Hendrickx, & Baeyens, 1997; Hammerl & Grabitz, 2000; Kim, Allen, & Kardes, 1996; Walther,
2002). Participants also completed a measure of their memory of the pairings after the
conditioning procedure and responded at chance levels. Moreover, open-ended questionnaires
also indicated that participants were unaware of the pairings.
There must certainly be limits to humans’ tendency to unconsciously absorb
environmental covariations into their psyches. One question that arises is whether implicit
covariation detection—like what is seen in EC research—occurs for non-evaluative dimensions.
In raising this question, it is important to note that implicit learning research in the cognitive
tradition tends to focus on novel dimensions with little hedonic meaning; it does not typically
address the importance or salience of the learning dimensions. Instead, most implicit learning
paradigms involve mundane perceptual dimensions, where emphasis is placed on the learning of
novel associations in artificial language, spatial locations of objects, and so on (Frensch &
Rünger, 2003; Seger, 1994). The question of what kinds of covariations might be best learned,
and along what kinds of dimensions, is seldom addressed (although one can find allusions into

---

Implicit Learning 5
these issues in preparedness research (e.g., Ohman & Mineka, 2001). As we describe next,
unlike the sorts of learning dimensions typically considered in the cognitive tradition, the
dimension of evaluation has unique properties that might make it especially conducive to
implicit covariation learning.
Implicit Covariation Learning in the Evaluative Dimension
Attitudes are important; knowing whether something is good or bad can be critical to
survival. As summaries of positive and negative information about objects that can either hurt us
or help us, they provide perhaps the most important information one can know (Fazio & Olson,
2003). Indeed, evaluation captures more “meaning” of the objects in our worlds than other
dimensions do (Osgood, Suci, & Tannenbaum, 1957), and our evaluations of objects can be
automatically-activated upon perceiving them (Fazio, 2000).
In evaluative conditioning research, the USs are by definition attitude-evoking—they
automatically elicit evaluative responses. Two empirically established properties of such
attitude-evoking stimuli may facilitate implicit learning: their abilities to attract attention and to
facilitate categorization.

As evidence of attitudes’ attention-grabbing qualities, Roskos-Ewoldsen and Fazio
(1992) demonstrated that objects toward which participants held accessible attitudes were more
likely to attract attention when located in a complex visual array, even when their attention to
those objects was not relevant to their task. Research in the cognitive domain tends to show that
selective attention to the to-be-learned information enhances implicit learning (e.g., Jiang &
Chun, 2001; Jimenez & Mendez, 1999; Nissen & Bullemer, 1987; Pacton & Perruchet, 2008;
Stadler, 1995, but see Cohen, Ivry, & Keele, 1990). Thus, because evaluative information
automatically attracts our attention, it might be considered “pre-disposed” to implicit learning.

---

Implicit Learning 6
Second, valenced information also promotes categorization along the evaluative
dimension. Stimuli typically can be categorized or construed in multiple ways. However, of all
the potential categorizations of a given stimulus, the more attitude-evoking is at an advantage
and more likely to determine the categorization process (Smith, Fazio, & Cejka, 1996). For
example, a Black male professor is more likely to be categorized as a Black by someone who is
racially prejudiced (Fazio & Dunton, 1997). If some novel object is paired with clearly valenced
objects, we can be relatively certain that those valenced objects will be categorized according to
their valence, and hence that regularity in the environment will more likely be encoded. Thus,
two related processes facilitate the impact of the US in EC procedures: the US attract attention
and they are likely to construed or categorized in a hedonically meaningful manner.
Consider the example of a novel object paired with objects that share a dimension that
does not spontaneously attract attention or promote categorization, say, “large objects.” A real-
world covariation between some novel object and, say, an ocean liner, a truck, and other large
stimuli, might be presented, but those large objects may not be consistently categorized as large.
As a result, the novel object might merely appear to have been paired with a series of
unconnected stimuli, instead of “large” items. Hence, unlike valence, when the dimension along
which some novel object is paired repeatedly is not spontaneously categorized as such, we
cannot be certain that that regularity in the environment will be encoded.
Consistent with this reasoning, previous research using EC paradigms has shown that
people tend to learn covariations involving non-evaluative dimensions only when they become
aware of the contingencies. For example, Meersman, De Houwer, Baeyens, Randell, and Eelen
(2005) valiantly attempted, across numerous studies, to use an EC procedure to create
associations involving the gender of infants. Their procedure involved the repeated pairing of

---

Implicit Learning 7
images of gender-ambiguous infants with clearly male or female infants. However, both direct
and indirect measures of associations indicated that the gender-ambiguous infants failed to
acquire the gender of their paired associates in the minds of participants who were unaware of
the pairings. The same pattern of findings emerged when Japanese names (whose gender was
unknown to the participants) were associated with typical known male and female names.
Meersman and colleagues also cite numerous studies (many unpublished) that failed to
demonstrate implicit covariation learning in non-evaluative dimensions.
The Present Research

We have argued that the evaluative dimension is particularly conducive to implicit
covariation learning. However, we do not mean to imply that implicit covariation detection is
limited to the evaluative dimension. As we describe later, non-evaluative dimensions can lend
themselves to such learning under the appropriate circumstances. The point we wish to make is
that the evaluative dimension appears uniquely conducive to implicit covariation learning, and
the experiments we report aim to provide empirical evidence to this effect. Moreover, we also
hope to capitalize on the unique features of the evaluative dimension to demonstrate how implicit
covariation learning in non-evaluative domains can be enhanced.

More specifically, we pursued two goals in the experiments we report here. First, we
hoped to demonstrate the superiority of the evaluative dimension in implicit covariation learning.
To this end, Experiment 1 compared implicit covariation learning along evaluative versus non-
evaluative dimensions. In choosing among the myriad non-evaluative dimensions, we reasoned
that a strong test of our predictions would entail the use of common and easily discernable non-
evaluative dimensions. We chose “size” in Experiment 1a, and “speed” in Experiment 1b.
Second, we hoped to use what we have argued is unique about the evaluative dimension to

---

Implicit Learning 8
uncover the requirements for implicit covariation learning in non-evaluative dimensions. To this
end, in Experiment 2 we attempt to create conditions that will foster non-evaluative covariation
learning.
Experiment 1
Participants
Eighty-nine (Experiment 1a) and 92 (Experiment 1b) undergraduate females at a large
Midwestern university completed the experiment in order to meet course requirements.
Materials and Procedure
Participants completed the experiment in groups of 1 to 4. After arriving, they were
seated in individual cubicles equipped with video monitors and response boxes. They were
informed that the experiment related to attention, vigilance, and rapid responding, and that their
role could be likened to a security guard whose job is to be alert for suspicious activity. They
were told that they would be viewing a series of random images on the computer screen, and that
their task would be to hit a response key as quickly as possible whenever a target item appeared.

After receiving these general instructions, participants were then shown a page that
depicted the name and image of their first target. They were further instructed that the target
would appear among a series of distractor images and words that were included to make the task
more challenging. In order to insure that participants attended to both the image and word
stimuli, they were instructed to respond whenever they saw either the name or a visual depiction
of the target. After answering any questions, the experimenter then initiated the vigilance task.

The task consisted of 5 blocks, each with 86 trials of 1.5 second duration (0 second inter-
trial interval). Each block utilized a different target, and participants were allowed to rest

---

Implicit Learning 9
between blocks, when they were shown the target for the next block. All targets were lesser-
known Pokemon cartoon characters which were not themselves the subjects of conditioning.
Each target appeared in both word and image form 10 times in its assigned block. Filler images
consisted of a variety of words (e.g., concrete, farming, books) and images (other Pokemon
characters, a woman writing a letter, a motorcycle, a fire hydrant), as well as 16 randomly
dispersed blank screens per block to reduce the rhythmic nature of the presentation. Images
appeared sometime alone, sometimes in pairs, and in varying locations on the screen.

Embedded in this random-seeming string of stimuli were critical pairings of 2 CS
Pokemon (Metapod and Shelder) and their associates. The evaluative condition followed the
procedure of Olson & Fazio (2001) described earlier. In this condition, one CS was paired with
US images and words that were clearly positive, and the other CS was paired with negative USs.
No single US was repeated. Each CS appeared 4 times in each block for a total of 20 CS-US
pairings per CS across blocks.

Size was chosen as the non-evaluative dimension for Experiment 1a, and speed was
chose for Experiment 1b. Prior to the experiment, pilot participants (n = 22) were asked to
generate a list of “things that are generally seen as small and large (and slow and fast).” The 10
most frequently listed items of each were selected, and appropriate visual representations were
used as USs (e.g., an ant, button, and chipmunk for small items, a truck, ocean-liner, and hippo
for large items, a snail, tugboat, and sloth for slow items, and a race car, marathon runner, and
motorcycle for fast items). Additionally, participants were asked to generate synonyms for
“small”, “large,” “slow” and “fast,” and the 10 most frequently mentioned words of each were
selected for word USs (e.g., tiny, miniscule, and little for small, huge, gigantic, and mammoth for
large, leisurely and unhurried for slow, and swift and rapid for fast). As in the evaluative

---

Implicit Learning 10
condition, the two CSs were paired with either small or large items in Experiment 1a, or slow or
fast items in Experiment 1b. Thus, the only difference between the two conditions was whether
the two CSs were paired with valence-related USs or USs of a non-evaluative dimension. Which
of the 2 CSs was assigned to negative/slow/small USs versus positive/fast/large USs was
counterbalanced.

After the conditioning procedure participants completed a series of unrelated
questionnaires for approximately 10 minutes. They were then told we were interested in their
impressions of some of the “filler” items from the vigilance task. Twenty-eight fillers and the 2
critical CS Pokemon were selected for them to rate. Participants were told that we were
interested in either how pleasant or unpleasant they found the images, or their impressions of
how large/fast or small/slow they were, depending on the condition. They were instructed to
“follow their gut impressions,” even if they weren’t sure how to respond. On a given trial, an
item was presented on the screen, and participants were required to make their response within a
5-second window on a 7-point scale, anchored by either extremely unpleasant and extremely
pleasant, extremely small and extremely large, or extremely slow and extremely fast, depending
on the condition.
After the evaluation task participants completed an open-ended measure of their
contingency awareness, just as in past research (e.g., Olson & Fazio, 2001). The first item
inquired as to whether they noticed anything unusual about the way the images were presented in
the vigilance task. The second item asked them to report whether they noticed anything in
particular about the words and images that were presented with the 2 CSs. The third item asked
participants whether they thought they were supposed to respond to the judgment items in a

---