You are on page 1of 20

Cognition and Emotion

ISSN: 0269-9931 (Print) 1464-0600 (Online) Journal homepage: http://www.tandfonline.com/loi/pcem20

Comparing measures of approach–avoidance


behaviour: The manikin task vs. two versions of
the joystick task

Regina Krieglmeyer & Roland Deutsch

To cite this article: Regina Krieglmeyer & Roland Deutsch (2010) Comparing measures of
approach–avoidance behaviour: The manikin task vs. two versions of the joystick task, Cognition
and Emotion, 24:5, 810-828, DOI: 10.1080/02699930903047298

To link to this article: http://dx.doi.org/10.1080/02699930903047298

Published online: 16 Dec 2009.

Submit your article to this journal

Article views: 1233

View related articles

Citing articles: 58 View citing articles

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=pcem20

Download by: [King's College London] Date: 16 September 2017, At: 12:51
COGNITION AND EMOTION
2010, 24 (5), 810828

Comparing measures of approach avoidance behaviour:


The manikin task vs. two versions of the joystick task

Regina Krieglmeyer and Roland Deutsch
University of Würzburg, Würzburg, Germany

The present research compared three measures of approachavoidance behaviour with respect to
their sensitivity and criterion-validity: moving a manikin on the screen towards and away from
stimuli (manikin task), pulling and pushing a joystick (joystick task), and pulling and pushing a
Downloaded by [King's College London] at 12:51 16 September 2017

joystick causing the visual impression that the stimuli come closer or disappear (feedback-joystick
task). When participants responded to stimulus valence, the manikin task was more sensitive to
valence than the joystick task (Experiment 1). When participants responded to the grammatical
category of valent words, the manikin and the feedback-joystick but not the joystick task were
sensitive to valence (Experiment 2). Finally, the manikin task was more sensitive than the feedback-
joystick task in assessing approachavoidance reactions towards spiders, and it was more strongly
related to self-reported fear of spiders (Experiment 3). The likelihood of recategorisation of
approachavoidance responses and the means of distance change are discussed as possible
explanations for the differences.

Keywords: Approach-avoidance; Implicit measures; Unintentional evaluation; Fear of spiders.

Immediately escaping from dangers or quickly responses towards emotionally significant stimuli.
taking chances to gain food or other rewards are Supporting this theorising, numerous studies
among the most important regulatory needs of suggest that perceiving positive things immedi-
organisms. In line with this notion, several ately facilitates approach behaviour, whereas per-
psychological theories suggest a fundamental ceiving negative things immediately facilitates
link from evaluation to behaviour (e.g., Cacioppo avoidance behaviour. For instance, a seminal study
& Gardner, 1999; Fazio & Zanna, 1990; Lang, by Solarz (1960) demonstrated that participants
Bradley, & Cuthbert, 1990; Neumann, Förster, & are faster to pull cards with positive words towards
Strack, 2003; Strack & Deutsch, 2004). For them and to push cards with negative words away
instance, Lang and colleagues (Lang, 1995; from them than vice versa.
Lang & Bradley, 2008) argued that emotions are Recent research demonstrated this stimulus
action dispositions that serve the function of response compatibility effect with various measures
quickly preparing the organism for appropriate of approachavoidance behaviour (MAAB). For

Correspondence should be addressed to: Regina Krieglmeyer, Lehrstuhl für Psychologie II, Röntgenring 10, D-97070
Würzburg, Germany. E-mail: krieglmeyer@psychologie.uni-wuerzburg.de
This research was supported by a grant from the German Science Foundation (DFG) to Fritz Strack (STR 264/231).
We thank Jan De Houwer, Stefan Schmukle, and two anonymous reviewers for helpful suggestions.

# 2009 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business 810
http://www.psypress.com/cogemotion DOI:10.1080/02699930903047298
MEASURES OF APPROACHAVOIDANCE BEHAVIOUR

instance, Chen and Bargh (1999) asked partici- vioural component of emotional responding.
pants to pull a lever toward (i.e., approach) and to Particularly, MAAB have potential advantages
push it away (i.e., avoidance) as a response to compared to self-report measures or measures of
valent stimuli that appeared on a computer screen. real approachavoidance behaviour. First, they
In a similar vein, Rinck and Becker (2007) let might provide valid reflections of someone’s
their participants pull and push a joystick. How- behavioural reactions even if he or she is moti-
ever, to reduce the likelihood that participants vated to conceal it. Second, they might assess
would recategorise pull-responses as withdrawing behavioural reactions people are not aware of and,
the hand, i.e., avoidance, and push-responses as therefore, are unable to report. Third, they can be
grasping, i.e., approach (cf. Seibt, Neumann, applied more economically than measures of real
Nussinson, & Strack, 2008), they introduced a behaviour. Finally, MAAB often are ethically less
visual feedback to the task that created the illusion problematic than real behaviour measures. That
that the stimulus actually moved towards or away is, to assess natural approachavoidance behaviour
Downloaded by [King's College London] at 12:51 16 September 2017

from the participants. In particular, pulling the real stimuli have to be used that are more intense
joystick gradually increased the size of the stimu- than the symbolic stimuli used in MAAB. In sum,
lus on the screen, thus strengthening the sugges- MAAB might complement the catalogue of
tion that the pull response led to the approach of various implicit measures (e.g., Wittenbrink &
the stimulus. Pushing the joystick decreased the Schwarz, 2007) by providing access to sponta-
size of the stimulus, thus suggesting that the push neous approachavoidance reactions towards
response actually resulted in the avoidance of the emotional stimuli.
stimulus. In a study by De Houwer, Crombez, So far, the different MAAB such as the joystick
Baeyens, and Hermans (2001), participants task1 (cf. Chen & Bargh, 1999; Fishbach & Shah,
moved a little figure on the screen towards or 2006), the feedback-joystick task (Rinck & Becker,
away from a stimulus by pressing buttons on the 2007) or the manikin task (De Houwer et al., 2001;
keyboard. In particular, the figure either appeared Mogg et al., 2003) have been employed as
above or below a stimulus that was shown in the equivalent measures. Surprisingly, research dedi-
centre of the screen. Depending on the position of cated to comparing these different variants has
the figure, moving upwards or downwards meant been lacking up to now. Yet, current explanations
approach or avoidance, respectively. In later of the mechanisms underlying MAAB effects
applications, MAAB have been used to assess suggest that these measures might differ with
approachavoidance tendencies towards social respect to their validity and sensitivity, depending
groups (e.g., Neumann, Hülsenbeck, & Seibt, on the operationalisation of the approachavoid-
2004), goals and temptations (e.g., Fishbach & ance behaviour. Therefore, the present research
Shah, 2006), phobic stimuli (e.g., Rinck & aimed to compare often-used MAAB with respect
Becker, 2007), or drugs (e.g., Mogg, Bradley, to their sensitivity and criterion-validity. We
Field, & De Houwer, 2003; Schoenmakers, define the sensitivity of a measure as the magnitude
Wiers, & Field, 2008). of a measurement outcome given that the to-be-
The potential value of MAAB is that they may measured construct was present. The measurement
provide a distinguished tool to assess the beha- outcome in MAAB is the size of the compatibility

1
Whereas Chen and Bargh (1999) let their participants push and pull a vertical lever 91 cm in length, in later studies a joystick
was used to assess push and pull movements (e.g., Fishbach & Shah, 2006; Markman & Brendl, 2005; Seibt, Neumann, Nussinson,
& Strack, 2008). The main difference between lever and joystick responses is the distance participants have to move their hand,
with lever movements requiring a greater distance change (approximately 16 cm in Chen and Bargh’s studies) than joystick
movements (approximately 3 cm in the present experiments). However, as recent research revealed that compatibility effects in
MAAB do not depend on the movement itself but on the cognitive categorisation of the response (Markman & Brendl, 2005; Seibt
et al., 2008), this difference can probably be disregarded. Therefore, in the present article, lever and joystick responses are treated
equivalently.

COGNITION AND EMOTION, 2010, 24 (5) 811


KRIEGLMEYER AND DEUTSCH

effect, that is, the facilitation of compatible For example, compared to an average person, a
responses (i.e., approach positive and avoid nega- person who has a strong fear of spiders may
tive) as compared to incompatible responses (i.e., evaluate spiders more negatively and respond with
avoid positive and approach negative). If a measure stronger avoidance tendencies. Applying our
is sensitive to variations in the to-be-measured definition of sensitivity, it follows that the
construct it also can be considered as a valid sensitivity of a MAAB to individual differences
measure of this construct (Borsboom, Mellen- is a function of the extent to which it reflects the
bergh, & van Heerden, 2004; De Houwer, impact of variations in the disposition variable on
Teige-Mocigemba, Spruyt, & Moors, 2009). the measurement outcome.
Thus, validity is a qualitive concept: a measure is Why might current MAAB differ with respect
either valid or not (Borsboom et al., 2004). In to their validity and sensitivity? Derived from
contrast, sensitivity is a quantitative concept: valid current explanations of MAAB effects, we pro-
measures can differ in their sensitivity. pose that two procedural features might be
Downloaded by [King's College London] at 12:51 16 September 2017

Figure 1 illustrates the processes that are responsible for differences in validity and sensi-
typically assumed to underlie MAAB effects. tivity: (1) the ease of recategorising the responses
The presentation of a stimulus leads to its in other terms than the instructed approach
evaluation, which results in the activation of avoidance responses; and (2) the means of dis-
compatible behavioural schemata. The measure- tance change. In the following, we will first
ment outcome is the facilitation of compatible describe the current explanations of MAAB
responses as compared to incompatible responses. effects and, then, present our analysis of how
This response facilitation is assumed to reflect the procedural differences of MAAB might be re-
degree to which approach or avoidance schemata sponsible for differences in their validity and
are activated. As the evaluation of the stimuli is sensitivity. While there is agreement that evalua-
assumed to be the main inner determinant of the tions result in the activation of compatible
activation of approachavoidance schemata, the behavioural schemata (cf. Figure 1), there are
measurement outcome also reflects the evaluation different views on which processes mediate this
of the stimuli. Taken together, the sensitivity of a link. According to one account, evaluations
MAAB is a function of the extent to which it induce motivational orientations (Neumann
reflects the degree of the activation of behavioural et al., 2003; Strack & Deutsch, 2004), which are
approachavoidance schemata. As is illustrated in conceptualised as states that prepare the organism
Figure 1, individual dispositions may moderate to decrease or increase the distance between the
how the stimulus is evaluated and to what degree self and an object. Motivational orientations are
approach or avoidance schemata are activated. assumed to have several consequences, the most

Figure 1. Schematic representation of the measurement model underlying our analysis of MAAB. Evaluation, disposition, and behavioural
schemata are hypothetical constructs.

812 COGNITION AND EMOTION, 2010, 24 (5)


MEASURES OF APPROACHAVOIDANCE BEHAVIOUR

important being a behavioural preparedness to away from yourself), they might reverse the
execute approach or avoidance reactions.2 In meaning of the responses in the incompatible
contrast, the evaluative coding view suggests that block only (e.g., withdraw your arm from negative
MAAB effects occur due to valence compatibility words and extend your arm to touch positive
rather than due to motivational orientations (Eder words). It is obvious that MAAB can only be a
& Rothermund, 2008; Lavender & Hommel, valid measure of the activation of approach
2007). Drawing on the theory of event coding avoidance schemata when the responses are not
(TEC; Hommel, Musseler, Aschersleben, & recategorised in other terms than the instructed
Prinz, 2001), the proponents of this approach approachavoidance responses. This is true from
argue that approach and avoidance behaviours are the perspective of the motivational orientation
mentally represented as being positive and nega- view as well as from the perspective of the
tive, respectively. Consequently, the perception of evaluative coding view. Thus, the possibility of
positive and negative stimuli may facilitate overt recategorisation undermines the validity of
Downloaded by [King's College London] at 12:51 16 September 2017

approach and avoidance behaviours because the MAAB because only responses that are coded in
evaluation of the stimuli overlaps with the valence terms of approachavoidance can capture the
of the response representations. activation of approachavoidance schemata.
What follows from these theories regarding Furthermore, the possibility of recategorisation
possible differences in the validity and sensitivity introduces error variance (as some participants
of MAAB as a function of their procedural might recategorise and others not) and, therefore,
differences? Both theories imply that the validity reduces the sensitivity of the measure. Note that,
of MAAB can be undermined when the responses according to the evaluative coding view, MAAB
are recategorised in other terms than the in- can still be a valid measure of valence (but not of
structed response labels. When recategorisation is the activation of approachavoidance schemata)
easy to accomplish participants are likely to when participants represent their responses with
recategorise their responses if it helps them to affective labels that are unrelated to approach
perform the task. Thus, they might recategorise avoidance (Eder & Rothermund, 2008). Neutral
the responses if other than the instructed labels response labels, however, would completely un-
map more naturally onto the requested motor dermine the validity of MAAB as a measure of
behaviours. For instance, the labels forward/back- valence.
ward might map more naturally onto joystick To what degree do current MAAB differ in the
movements than the labels towards/away. In ease with which the responses can be recate-
addition to a general recategorisation, participants gorised? Recent research suggests that simple
might recategorise the responses especially when joystick movements can easily be categorised in
the stimulusresponse assignment is incompatible different ways by giving respective instructions
with participants’ automatic tendency. Thus, (Eder & Rothermund, 2008; Markman & Brendl,
when participants complete two blocks, one with 2005; Seibt et al., 2008). Thus, it is very likely that
a compatible (e.g., pull positive words toward participants mentally categorise the responses in
yourself and push negative words away from other than the instructed terms as this can make
yourself) and another with an incompatible the task easier. In the feedback-joystick task,
stimulusresponse assignment (e.g., pull negative however, joystick responses lead to the visual
words toward yourself and push positive words impression that the stimulus comes closer or

2
It has also been suggested that evaluations directly activate concrete motor patterns such as arm flexion or arm extension
(Cacioppo & Priester, 1993; Coombes, Cauraugh, & Janelle, 2007). However, a considerable amount of research revealed that in
MAAB positive or negative stimuli can trigger the very same motor pattern depending on whether the behaviour increases or
decreases the distance between the participant and the stimulus in a given context (De Houwer et al. 2001; Markman & Brendl,
2005; Seibt et al., 2008; van Dantzig, Pecher, & Zwaan, 2008).

COGNITION AND EMOTION, 2010, 24 (5) 813


KRIEGLMEYER AND DEUTSCH

disappears. Under these conditions, reversing the defensive reflexes consist of moving parts of the
approachavoidance meaning seems to be very body away from the source of danger rather than
difficult, if not impossible (Rinck & Becker, 2007). causing the threatening stimulus to withdraw (cf.
Yet, participants still might represent their beha- Lang et al., 1990). Furthermore, the flexibility of
viour as merely increasing or decreasing the size of using one’s hands to manipulate the position of
the stimulus, which is unrelated to approach objects prevents the development of highly auto-
avoidance. Contrary to the joystick tasks, in the matised approachavoidance schemata (cf. Mark-
manikin task (De Houwer et al., 2001) recategor- man & Brendl, 2005; Seibt et al., 2008). From the
isation is rather unlikely. Although in principle the perspective of the motivational orientation view,
movements can be recategorised as moving down- highly automatised means of distance change
wards and upwards, this would make the task more might be activated more easily and strongly than
difficult instead of easier. The reason for this is less automatised means. In particular, positive
that the manikin either appears above or below the evaluations might activate the behavioural schema
Downloaded by [King's College London] at 12:51 16 September 2017

stimulus, and, therefore, up and down responses of moving towards to a stronger degree than the
are unrelated to the instructed approachavoidance behavioural schema of taking the object. Con-
responses. Thus, depending on the position of the versely, negative evaluations might activate the
figure, participants have to determine in each trial behavioural schema of moving away to a stronger
which response means approach or avoidance. degree than the behavioural schema of putting the
Consequently, the representation of approach object away. From this reasoning it follows that
and avoidance is activated in each trial. Taken the manikin task might have a greater sensitivity
together, the ease of a mental categorisation of the than the joystick tasks.
behaviours in other terms than approachavoid- One may question this conclusion because the
ance differs between the tasks, with the manikin relation between the concrete response and
task presumably being least susceptible for recate- the approachavoidance meaning is flexible in
gorisation and the joystick task without visual the manikin task. For example, pressing the up
feedback being most susceptible. button means approach when the figure appears
A second important procedural difference below the word, whereas it means avoidance when
between current MAAB is the means of distance the figure appears above the word. In the joystick
change. Whereas in the joystick tasks participants task, on the other hand, the relation between the
cause the stimulus to move, the manikin task concrete response (pulling and pushing) and the
requires them to move the figure. Thus, in the approachavoidance meaning remains constant
joystick task, the means of distance regulation is over the whole task. Given the more consistent
taking the object vs. putting it away, while in the mapping of concrete behaviours and their meaning
manikin task the means is running towards vs. in the joystick task as compared to the manikin
away from the object.3 Running towards/away task, one might suspect the behaviours being more
presumably is a more automatised means of automatised in the former than in the latter. For
distance change than manipulating the position our argument, however, it is important to distin-
of objects. For instance, highly automatised guish between the link from evaluation to the

3
In the manikin task, the means of distance regulation is represented at an abstract level rather than at a concrete motor level,
because the concrete response merely consists of pressing buttons. As several studies have shown, the meaning of the behaviour
rather than the concrete motor pattern is relevant for compatibility effects in MAAB (Markman & Brendl, 2005; Seibt et al., 2008).
An important question is whether participants represent the behaviour as moving themselves or as causing a figure to move. In other
words, do participants identify with the manikin? Two factors probably increased the likelihood of identification. First, we
instructed our participants that ‘‘they should move with the manikin towards or away from the stimulus’’. To the degree that
participants followed this instruction, identification with the manikin should have been increased. Second, the situation of the
participants is very similar to a computer game with a controllable figure. In such games, the figure typically represents the player,
thereby providing a reason for identification.

814 COGNITION AND EMOTION, 2010, 24 (5)


MEASURES OF APPROACHAVOIDANCE BEHAVIOUR

activation of approachavoidance schemata and debate: Do valent stimuli trigger approachavoid-


the translation of the activated approachavoid- ance behaviours even when participants do not have
ance schema into concrete behaviours. Whereas the intention to evaluate the stimuli? A large
the first presumably is a long-term link, the latter is amount of research measuring the startle reflex as
established during the measurement. MAAB aim an indicator of a preparedness to engage in
at assessing the long-term link between valence avoidance behaviour supports the notion that
and approachavoidance schemata, not the short- valent stimuli exert a significant influence on such
term association that translate such schemata into momentary dispositions even in the absence of
concrete movements. Importantly, our argument evaluation intentions (e.g., Bradley, Codispoti,
regarding the differential automaticity of the Cuthbert, & Lang, 2001; Lang et al., 1990; Lang,
means of distance change relates to the long- Greenwald, Bradley, & Hamm, 1993). In particu-
term links, not to the short-term associations. lar, when participants passively look at positive and
According to our reasoning, the long-term link negative pictures, the startle response is modulated
Downloaded by [King's College London] at 12:51 16 September 2017

between evaluation and moving the self might be by the valence of the pictures. In contrast to these
stronger than the long-term link between evalua- findings, studies using MAAB as indicators of a
tion and moving the object. Thus, given that the preparedness to approach or to avoid revealed
compatibility effect of the manikin task mainly mixed results. Some studies imply that stimulus
reflects the activation of behavioural schemata of valence affects approachavoidance reactions even
moving the self, and given that the compatibility without an intention to evaluate (Chen & Bargh,
effect of the joystick task mainly reflects the 1999; De Houwer et al., 2001). Other studies
activation of behavioural schemata of moving an suggest that stimulus valence has no effect under
object, the manikin task should reveal larger such conditions (Lavender & Hommel, 2007;
compatibility effects than the joystick task. In Rotteveel & Phaf, 2004). The present perspective
contrast, the fact that in the manikin task the may help to reconcile these findings. In particular,
relation between the concrete responses and their the procedural differences outlined above and their
approachavoidance meaning is flexible should hypothesised effect on the sensitivity of MAAB
result in an overall slowdown of the responses as may at least in part be responsible for diverging
compared to the joystick task. results on whether MAAB effects depend on
It is important to note that the means of intentions to evaluate. According to the above
distance change are less relevant from the per- analysis, MAAB might be most sensitive when
spective to the evaluative coding view. In parti- recategorisation is unlikely and when the means of
cular, the evaluative coding view suggests that the distance change is highly automatised. Given that
evaluative codes of the response representations the outcome of intentional evaluations is stronger
are determined by the labels used in the task than the outcome of unintentional evaluations, one
instructions (Eder & Rothermund, 2008). As the would most likely find an effect of valence even in
labels used in the manikin and the joystick tasks the absence of evaluation intentions in the manikin
are identical (i.e., towards vs. away), the concrete task as compared to the joystick tasks.
means of distance change should be irrelevant. To test our predictions, we first compared the
The above analyses suggest that the procedural sensitivity of the manikin and the joystick task
differences between the tasks might result in when participants responded according to the
differences in their sensitivity and criterion-valid- valence of normatively positive and negative words
ity. On one hand, the suspected differences are (Experiment 1). Then, we compared the manikin
important for practical reasons when it comes to and the joystick task when the valence of the words
selecting an optimal research tool. On the other was task irrelevant, and thus intentional evaluation
hand, the suspected differences are relevant for a of the stimuli was probably undermined (Experi-
theoretical question that has stimulated some ment 2a). Because Experiment 2a revealed a

COGNITION AND EMOTION, 2010, 24 (5) 815


KRIEGLMEYER AND DEUTSCH

compatibility effect in the manikin but not in the assigned to each of the counterbalancing groups
joystick task, we conducted Experiment 2b, in was approximately equal.
which we employed the feedback-joystick task
(Rinck & Becker, 2007). We hypothesised that Manikin task
the approachavoidance feedback provided in this Following De Houwer et al. (2001), participants
task would reduce the likelihood of recategorisa- moved the manikin upwards or downwards by
tion, and thus, we expected to observe an effect of repeatedly pressing the ‘‘8’’ or ‘‘2’’ keys, respec-
task-irrelevant valence in this measure. Finally, tively, of the numerical keyboard with their
Experiment 3 was geared towards comparing the middle finger. At the beginning of each trial, a
criterion-validity of the manikin and the feedback- cross was presented to centre participants’ atten-
joystick task in the domain fear of spiders. To this tion. Participants had to press the ‘‘5’’ key with
aim, we compared the correlations between self- their middle finger and keep it pressed until they
reported fear of spiders and the MAAB effects. began to move with the manikin. Thereby, at the
Downloaded by [King's College London] at 12:51 16 September 2017

Based on the above analysis, we hypothesised that beginning of a trial the middle finger was always
the manikin task has a higher criterion-validity at the same starting position. Triggered by the key
than the feedback-joystick task. In addition to press on the ‘‘5’’ key, the manikin appeared either
sensitivity and criterion-validity, we compared the in the upper or in the lower half of the screen.
split-half reliability of the MAAB in all experi- Then, 750 ms after the appearance of the mani-
ments. Although we did not have specific hypoth- kin, a word was presented in the centre of the
eses regarding differences in reliability, this screen. Depending on conditions, participants
comparison might still be interesting for practical were instructed to move with the manikin towards
reasons, that is, for gaining knowledge of the positive words and to move with the manikin
psychometric properties of the different measures. away from negative words, or vice versa. They
For the sake of brevity, we discuss the results of all were instructed to respond as fast and as accu-
experiments together in the general discussion. rately as possible. Participants had to press the
respective key three times to move the manikin
across the screen. Each key press moved the
EXPERIMENT 1 manikin 38 pixels. By alternating the length of
the manikin’s legs each time it appeared in a new
Method position, the impression of walking was evoked.
Participants and design 500 ms after the third key press all stimuli were
Thirty-eight students (32 female) from the Uni- deleted from the screen. The inter-trial interval
versity of Würzburg took part in the study. The was 1000 ms. The time between the onset of the
mean age was 22.16 years (SD5.01). Partici- word and the first key press was measured as
pants received course credit or t6 (approximately dependent variable.
US$9 at that time) as compensation, depending
on whether they were psychology or non-psychol- Joystick task
ogy students. The design was a 2 (Task: manikin Similar to Chen and Bargh (1999), participants
vs. joystick)2 (Compatibility: compatible vs. pulled and pushed a joystick. At the beginning of
incompatible)2 (Order of Compatibility: com- each trial, three asterisks were presented for 200 ms
patible first vs. incompatible first)2 (Order of to centre participants’ attention, followed by a
Tasks: manikin first vs. joystick first) factorial blank screen for 100 ms. Then a word appeared
mixed design with the last two factors manipu- in the centre of the screen. Depending on condi-
lated between participants. In all experiments tions, participants were instructed to move positive
reported in this article, the number of participants words towards themselves by means of pulling the

816 COGNITION AND EMOTION, 2010, 24 (5)


MEASURES OF APPROACHAVOIDANCE BEHAVIOUR

joystick, and to move negative words away from RESULTS


themselves by means of pushing the joystick, or vice
Data analysis strategy
versa4 (cf. Seibt et al., 2008). They were told to
Reaction-time analyses usually require corrections
respond as fast and as accurately as possible. After
of outliers, which can be achieved by various means
participants responded all stimuli were deleted
such as cut-offs or transformations (Ratcliff,
from the screen. The inter-trial interval was
1993). Unlike with other measures (e.g., Green-
1000 ms. The time between the onset of the word wald, Nosek, & Banaji, 2003) there exists no
and when participants had moved the joystick 50% convention how to correct for outliers in MAAB.
of the range in the respective direction was Therefore, we compared different correction
measured as dependent variable. methods. Following the suggestions of Ratcliff
(1993), we employed various general cut-offs
Materials (2000, 1500, and 1000 ms), an individual cut-off
Downloaded by [King's College London] at 12:51 16 September 2017

We selected 30 positive and 30 negative nouns as based on 1.5 standard deviations above the in-
test stimuli and 10 positive and 10 negative nouns dividual mean, log-transformation, and inverse-
as practice stimuli (Hager & Hasselhorn, 1994; transformation. Additionally, we analysed the
Klauer & Musch, 1999). Words were presented in median. Previous to applying the outlier correction
uppercase letters with a font size of 32 pt. For the methods, we inspected the distribution of the
manikin task, we draw a simple figure consisting of reaction latencies to check if there were obviously
invalid responses. This was the case in Experiment
a circle for the head, a square for the body, and lines
1 for one response (20823 ms), and it was the case
representing the arms and legs. The manikin was
in Experiment 3 for one response (47 ms). These
about 2.8 cm high and 1.9 cm wide. All stimuli
two reaction latencies were excluded. We report
were presented in white font on a black back-
the impact of the outlier correction methods on the
ground. The screen had a resolution of 1024768
compatibility effects as well as on split-half
pixels. The joystick was a Logitech Attack 3.
reliability. For the sake of conciseness, we report
the inference-statistical comparisons of the
Procedure MAAB based on the correction method that
Participants learned that the experiment was revealed the strongest effects over all experiments
about sensorimotor co-ordination and reaction reported in this article, namely the 1500 ms cut-
rate. Depending on conditions, participants first off. In all experiments reported in this article, we
completed the manikin task and then the joystick conducted preliminary analyses, in which the
task or vice versa. In each task, participants method factors (i.e., order of tasks, order of
completed one compatible block consisting of 60 compatibility, or response mapping) were in-
test trials, and one incompatible block, consisting cluded. As these analyses did not reveal significant
of 60 test trials. Each block was preceded by 20 main effects or interactions of the method factors
practice trials. All trials were presented in random we excluded them from the main analyses.
order. The order of the compatible and the
incompatible block was counterbalanced between Sensitivity
participants. Yet, the same order was administered We subjected latencies of correct responses of
in the manikin and the joystick task. compatible and incompatible trials (incorrect

4
While Chen and Bargh (1999, p. 219) instructed their participants ‘‘to push the lever forward’’ and ‘‘to pull the lever back
toward them’’, we decided to follow Seibt et al.’s (2008) instructions in order to increase the likelihood that participants represented
the behaviours as approach and avoidance behaviours.

COGNITION AND EMOTION, 2010, 24 (5) 817


KRIEGLMEYER AND DEUTSCH

responses: manikin 5.8%; joystick 4.6%) to a 2 trials in order of appearance, separately for
(Compatibility)2 (Task) analysis of variance stimulus valence, compatibility with response,
(ANOVA) for repeated measures. This analysis and participant. Then, we calculated a compat-
yielded a significant main effect of Task, F(1, ibility score (i.e., latencies of correct incompatible
37)20.67, pB.001, d0.81, a significant main responses minus latencies of correct compatible
effect of Compatibility, F(1, 37)32.69, pB.001, responses) based on the odd trials, and a compat-
d0.60, and a significant interaction of Task and ibility score based on the even trials. Thereby,
Compatibility, F(1, 37)21.61, pB.001, d each score was based on an equal number of
0.95. Post hoc contrasts indicated that the differ- positive and negative stimuli. We correlated these
ence between latencies of compatible and incom- two scores by applying a SpearmanBrown cor-
patible trials was significant in the manikin task, rection. The reliability was r.91 for the manikin
t(37)6.27, pB.001, d0.74, and marginally task and r.85 for the joystick task. To test
significant in the joystick task, t(37)1.90, p whether the reliability coefficients of the tasks
Downloaded by [King's College London] at 12:51 16 September 2017

.065, d0.19 (Table 1). As an effect-size measure, differed significantly, we followed the suggestions
we calculated Cohen’s d of the compatibility effects of Steiger (1980) for testing differences between
(i.e., latencies of correct responses in incompatible dependent correlations. In particular, we trans-
trials minus latencies of correct responses in formed the correlation coefficients into z-values
compatible trials). The impact of the outlier and then calculated the difference of the z-values
correction methods on the compatibility effects is by using the formula suggested by Steiger (1980),
presented in Table 2. Whereas the compatibility which takes into account the sample size and the
effect in the manikin task (d0.500.79) was covariance of the correlation matrix. This analysis
between medium and large according to Cohen’s suggests that the two tasks are equally reliable,
(1988) standards, the compatibility effect in the zdiff 1.12, p.26. The reliability coefficients are
joystick task (d0.150.22) was small. The presented in Table 2 as a function of the outlier
manikin compatibility effect was statistically sig- correction method. The reliability of the manikin
nificant with all correction methods, whereas the
task ranged between .84 and .95. The reliability of
joystick compatibility effect was only significant
the joystick task ranged between .77 and .91.
with a 1000 ms cut-off and with an inverse- as well
Thus, irrespective of the outlier correction
as log-transformation. The compatibility effects of
method, the reliability of both measures was good.
the tasks were positively, but not significantly,
correlated, r.22, p.19.
EXPERIMENTS 2A AND 2B
Reliability
We calculated the split-half reliability of the Experiments 2a and 2b were focused on comparing
compatibility effect scores of the manikin task MAAB under conditions in which participants
and the joystick task. To this end, we listed all were not required to process the valence of the

Table 1. Mean response latencies of compatible and incompatible trials as a function of task and experiment.

Manikin Joystick Feedback-Joystick

Experiments Compatible Incompatible Compatible Incompatible Compatible Incompatible

Exp. 1 740.22 (139.24) 843.76 (139.20) 685.19 (112.86) 706.20 (109.06)


Exp. 2a & 2b 828.03 (114.51) 842.63 (107.06) 755.64 (114.52) 755.85 (108.97) 795.86 (99.08) 804.97 (91.93)
Exp. 3 643.01 (94.33) 676.82 (106.66) 613.55 (71.70) 625.77 (69.60)

Note: Means are based on the 1500 ms cut-off criterion. Values enclosed in parentheses represent standard deviations.

818 COGNITION AND EMOTION, 2010, 24 (5)


MEASURES OF APPROACHAVOIDANCE BEHAVIOUR

Table 2. Effect sizes (Cohen’s d) and reliability coefficients (split-half SpearmanBrown) of the compatibility effects as a function of task,
outlier correction method, and experiment.

Manikin Joystick Feedback-Joystick

Experiments d r d r d r

Exp. 1 No correction 0.63*** .90 0.21  .91


Cut-off 2000 ms 0.65*** .94 0.18  .85
Cut-off 1500 ms 0.74*** .91 0.19  .85
Cut-off 1000 ms 0.77*** .84 0.17* .77
Log(RT) 0.72*** .92 0.21* .90
1/RT 0.79*** .90 0.22* .89
1.5 SD 0.50*** .95 0.15 .86
Median 0.63*** .93 0.17 .83
Exp. 2a & 2b No correction 0.10* .03 0.03 .21 0.03 .44
Downloaded by [King's College London] at 12:51 16 September 2017

Cut-off 2000 ms 0.10** .65 0.02 .31 0.06 .20


Cut-off 1500 ms 0.13** .11 0.002 .27 0.10* .23
Cut-off 1000 ms 0.05 .04 0.04 .53 0.04 .59
Log(RT) 0.11** .22 0.04 .29 0.05 .30
1/RT 0.12** .22 0.04 .32 0.06 .07
1.5 SD 0.09** .10 0.01 .44 0.01 .28
Median 0.08  .09 0.01 .28 0.03 .33
Exp. 3 No correction 0.29** .81 0.14 .78
Cut-off 2000 ms 0.29** .80 0.16  .80
Cut-off 1500 ms 0.34*** .78 0.17  .69
Cut-off 1000 ms 0.29** .75 0.19* .66
Log(RT) 0.28** .80 0.15  .79
1/RT 0.25** .79 0.15  .79
1.5 SD 0.22** .72 0.11 .58
Median 0.32** .79 0.16  .66

Note: p B.10; *p B.05; **p B.01; ***pB.001.

stimuli. To this end, we instructed participants to took part in the study. The mean age was 23.49
respond according to the grammatical category of years (SD3.73). We excluded one participant
the words (De Houwer et al., 2001; De Houwer & from the analyses because he did not follow
Eelen, 1998). In Experiment 2a, we compared the instructions, but moved the manikin only in one
manikin and the joystick task. Because we did not direction. Participants received t6 (approximately
find a compatibility effect in the joystick task, we US$9 at that time) as compensation. The
conducted Experiment 2b, in which we employed design was a 2 (Task: manikin vs. joystick)2
the feedback-joystick task (Rinck & Becker, 2007). (Compatibility: compatible vs. incompatible)2
If providing approachavoidance-relevant feed- (Response Mapping: noun-approach, adjective-
back reduces the likelihood of recategorisation of avoidance vs. vice versa)2 (Order of Tasks:
the responses, then the feedback-variant should manikin first vs. joystick first) factorial mixed
reveal a compatibility effect when valence is task design with the last two factors manipulated
irrelevant. between participants.

Method Experiment 2b. Forty-eight non-psychology stu-


dents (31 female) from the University of Würzburg
Participants and design
took part in the study. The mean age was 23.35
Experiment 2a. Forty-nine non-psychology stu- years (SD3.57). Participants received t6 (ap-
dents (34 female) from the University of Würzburg proximately US$9 at that time) as compensation.

COGNITION AND EMOTION, 2010, 24 (5) 819


KRIEGLMEYER AND DEUTSCH

The design was a 2 (Compatibility: compatible vs. for 500 ms. This created the illusion that the word
incompatible)2 (Response Mapping: noun-ap- moved away from the participant and finally
proach, adjective-avoidance vs. vice versa) factorial disappeared. The inter-trial interval was 1000 ms.
mixed design with the second factor manipulated
between participants. Materials
We selected 30 positive and 30 negative nouns as
well as 30 positive and 30 negative adjectives as test
Measures
stimuli, and 4 positive and 4 negative nouns as well
Experiment 2a. The manikin and the joystick as 4 positive and 4 negative adjectives as practice
task were the same as in Experiment 1 with the stimuli (Hager & Hasselhorn, 1994; Klauer &
exception that participants responded to the Musch, 1999). All other materials were the same as
grammatical category of the words. In the man- in Experiment 1.
ikin task, participants were instructed to move
Downloaded by [King's College London] at 12:51 16 September 2017

with the manikin to the word, when it was a Procedure


noun, and to move with the manikin away from In Experiment 2a, participants first completed the
the word, when it was an adjective, or vice versa, manikin task and then the joystick task, or vice
depending on conditions. In the joystick task, versa, depending on conditions. In each task,
participants were instructed to pull the word with participants first completed 16 practice trials, and
the joystick towards them, when it was a noun, then 120 test trials with a break after 60 trials. All
and to push it away from them, when it was an trials were presented in random order. The
adjective, or vice versa, depending on conditions. procedure of Experiment 2b was similar to
Experiment 2a with the exception that only the
Experiment 2b. In Experiment 2b, we used a feedback-joystick task was administered.
feedback-joystick task similar to that introduced
by Rinck and Becker (2007). At the beginning of Results
each trial, three asterisks were presented for
Sensitivity
200 ms to centre participants’ attention, followed
by a blank screen for 100 ms. Then a word Experiment 2a. We subjected latencies of correct
appeared in the centre of the screen. Depending responses of compatible and incompatible trials
on response-mapping conditions, participants (incorrect responses: manikin, 5.6%; joystick,
were instructed to move nouns towards them- 4.5%) to a 2 (Compatibility)2 (Task) ANOVA
selves by means of pulling the joystick, and to for repeated measures. This analysis yielded a
move adjectives away from themselves by means significant main effect of Task, F(1, 47)22.68,
of pushing the joystick, or vice versa. They were pB.001, d0.72, a marginally significant main
told to respond as fast and as accurately as effect of Compatibility, F(1, 47)3.29, p.076,
possible. After participants responded the word d0.08, and a significant interaction of Task and
increased or decreased in size, depending on Compatibility, F(1, 47)4.26, p.045, d
whether participants pulled or pushed the joy- 0.39. Post hoc contrasts indicated that the
stick. Immediately after pulling, the font size of difference between latencies of compatible and
the word changed to 40 pt for 50 ms, then to incompatible trials was significant in the manikin
48 pt for 50 ms, then to 56 pt for 50 ms, and task, t(47)3.03, p.004, d0.13, but not
finally to 64 pt for 500 ms. This created the significant in the joystick task, tB1 (for means
illusion that the word moved towards the parti- and standard deviations see Table 1). As Table 2
cipant. Immediately after pushing, the font size of shows, the manikin compatibility effect (d
the word changed to 24 pt for 50 ms, then to 0.050.11) was small, but significant with all
16 pt for 50 ms, then to 8 pt for 50 ms, and finally outlier correction methods, except from the
disappeared. Then a blank screen was presented 1000 ms cut-off and the median. The joystick

820 COGNITION AND EMOTION, 2010, 24 (5)


MEASURES OF APPROACHAVOIDANCE BEHAVIOUR

compatibility effect, however, did not reach sig- negative words, it remains an important question
nificance. The compatibility effects of the tasks whether the measure would be capable of mapping
were positively, but not significantly correlated, stable differences between people. Experiment 3
r.16, p.28. was designed to compare the sensitivity and the
criterion-validity of the manikin and the feedback-
Experiment 2b. We subjected latencies of correct joystick task in the domain fear of spiders. Towards
responses of compatible and incompatible trials this end, we estimated the disposition variable by
(incorrect responses: 2.8%) to an ANOVA for administering a questionnaire measure. The corre-
repeated measures. This analysis yielded a sig- lation between the compatibility effect of the
nificant main effect of Compatibility, F(1, 47) MAAB and the questionnaire served as an index
4.54, p.038, d0.10 (for means and standard of criterion-validity. Based on the analysis of the
deviations see Table 1). As Table 2 shows, the procedural differences between the tasks, we
compatibility effect depended strongly on the expected the manikin task to reveal a higher
Downloaded by [King's College London] at 12:51 16 September 2017

outlier correction method, reaching significance sensitivity and criterion-validity than the feed-
only with a 1500 ms cut-off. back-joystick task.

Method
Reliability
Participants and design
Experiment 2a. We calculated oddeven split-
Sixty-one non-psychology students (40 female)
half SpearmanBrown reliability similar to the
from the University of Würzburg took part in the
previous experiment. The reliability was r .11
study. The mean age was 22.98 years (SD2.45).
for the manikin task and r.27 for the joystick
Participants received t6 (approximately US$9 at
task. The reliability coefficients did not differ
that time) as compensation. The design was a 2
significantly, zdiff 0.98, p.33. As can be seen
(Task: manikin vs. feedback-joystick)2 (Com-
in Table 2, the reliability of the manikin compat-
patibility: compatible vs. incompatible)2 (Or-
ibility effect was negative with all outlier correction
der of Compatibility: compatible first vs.
methods, ranging from .65 to .03. The relia-
incompatible first)2 (Order of Tasks: manikin
bility of the joystick compatibility effect was positive
first vs. feedback-joystick first) factorial mixed
but not satisfactory, ranging between .21 and .53.
design with the last two factors manipulated
between participants.
Experiment 2b. Split-half reliability of the feed-
back-joystick task was r .23. Table 2 shows
that the reliability coefficient was affected strongly Materials
by the outlier correction methods, ranging from We used 12 spider and 12 butterfly pictures as test
.59 to .44. Yet, over all correction methods, the stimuli and 4 spider and 4 butterfly pictures as
reliability was not satisfactory. practice stimuli (Gerdes, Alpers, & Pauli, 2007).
The pictures had a size of approximately 200
250 pixels. For the feedback-joystick task, seven
EXPERIMENT 3 additional picture sizes were created by increasing
the size by 1.3 four times, and decreasing the size
Researchers are often interested in assessing stable by 0.6 three times. The resolution of the screen
individual differences in approachavoidance was 1024768 pixels.
reactions towards various stimuli (e.g., Mogg et
al., 2003; Rinck & Becker, 2007; Wiers, van Measures
Woerden, Smulders, & de Jong, 2002). Even if a The manikin and the feedback-joystick task were
measure has been proven to be sensitive towards similar to the previous experiments. Yet, we pre-
normative stimuli such as clearly positive or sented pictures instead of words, and participants

COGNITION AND EMOTION, 2010, 24 (5) 821


KRIEGLMEYER AND DEUTSCH

were instructed to respond according to the d0.19, being significant at the p.05 level only
category of the picture (i.e., spider vs. butterfly). with a 1000 ms cut-off.

Reliability
Procedure
Split-half reliability was r.78 for the manikin
Participants first completed the MAAB. The
task and r.69 for the feedback-joystick task.
order of the tasks was counterbalanced across
The tasks were equally reliable, zdiff 0.88, p
participants. In each task, they completed one
.38. As can be seen in Table 2, the reliability of
block of 48 compatible trials and one block of 48
the manikin compatibility score ranged between
incompatible trials. The order of the blocks was
.72 and .81 as a function of the applied outlier
counterbalanced across participants. Each block
correction method. The reliability of the feed-
was preceded by eight practice trials. After
back-joystick compatibility effect ranged between
finishing the MAAB, participant evaluated the
.66 and .80.
pictures on a scale from 1 (very negative) to 9 (very
Downloaded by [King's College London] at 12:51 16 September 2017

positive). Respective Cronbach as for the spider


Criterion-validity
and the butterfly pictures were .94 and .93. Then,
To examine the criterion-validity of the measures,
they completed the German version of the Fear of
we first computed approachavoidance indices for
Spiders Questionnaire (FAS; Rinck et al., 2002;
spiders as well as for butterflies by subtracting
Szymanski & Odonohue, 1995) as well as the
latencies of approach responses from latencies of
Fear of Spiders Screening (SAS; Rinck et al.,
avoidance responses (cf. Rinck & Becker, 2007).
2002). Both scales had a satisfactory reliability.
These indices can be interpreted as approach
Respective Cronbach as were .90 for the FAS,
biases towards the respective stimulus category.
and .81 for the SAS.
Table 3 presents the correlations among the
explicit measures as well as the approachavoid-
Results
ance indices. Confirming the validity of the
Sensitivity explicit measures, the spider fear scales were
We subjected latencies of correct responses of positively correlated. Furthermore, the spider
compatible and incompatible trials (incorrect fear scales were negatively correlated with explicit
responses: manikin, 3.3%; feedback-joystick, positivity towards spider pictures but not with
2.3%) to a 2 (Task)2 (Compatibility) ANOVA explicit positivity towards butterfly pictures. Cor-
for repeated measures. This analysis yielded a relations of explicit measures with the approach
significant main effect of Task, F(1, 60)11.48, avoidance indices of the manikin task were as
p.001, d0.49, a significant main effect of expected. In particular, the spider fear scales were
Compatibility, F(1, 60)16.58, pB.001, d negatively correlated with approach-biases to-
0.32, and a significant interaction of Task and wards spiders and positively, but not significantly,
Compatibility, F(1, 60)4.58, p.036, d correlated with approach biases towards butter-
0.36. Post hoc contrasts indicated that the flies. Explicit positivity towards spiders was
difference between latencies of compatible and positively correlated with approach-biases towards
incompatible trials was significant in the manikin spiders and negatively correlated with approach
task, t(60)3.91, pB.001, d0.34, and margin- biases towards butterflies. The reverse pattern was
ally significant in the feedback-joystick task, true for explicit positivity towards butterflies.
t(60)1.93, p.059, d0.17 (for means and Correlations of explicit measures with approach
standard deviations see Table 1). As can be seen in avoidance indices of the feedback-joystick task
Table 2, the manikin compatibility effect was were only partly as expected. In particular, the
small, ranging from d0.22 to d0.34. The spider fear scales were not significantly correlated
compatibility effect in the feedback-joystick task with the approachavoidance indices. Yet, explicit
was smaller and ranged between d0.11 and positivity towards spiders was positivity correlated

822 COGNITION AND EMOTION, 2010, 24 (5)


MEASURES OF APPROACHAVOIDANCE BEHAVIOUR

Table 3. Correlations among the measures in Experiment 3

1 2 3 4 5 6 7 8

1. FAS * .85*** .45*** .07 .22  .18 .04 .01


2. SAS * .53*** .02 .29* .19 .08 .01
3. Explicit evaluation of spider pictures * .29* .43** .40** .29* .14
4. Explicit evaluation of butterfly pictures * .24  .22  .26* .09
5. Manikin approach bias towards spiders * .33* .12 .12
6. Manikin approach bias tw. butterflies * .09 .02
7. FB-Joystick approach bias towards spiders * .18
8. FB-Joystick approach bias towards *
butterflies

Note: pB.10; *p B.05; **pB.01; ***p B.001.
Downloaded by [King's College London] at 12:51 16 September 2017

with approach biases towards spiders and nega- 1) as well as when valence was task-irrelevant and
tively, but not significantly, correlated with ap- thus participants probably had no intention of
proach biases towards butterflies. The reverse processing the stimulus valence (Experiment 2a).
pattern was true for explicit positivity towards Also, the manikin task was more sensitive than the
butterflies. The manikin approach biases were not feedback-joystick task when participants re-
significantly correlated with the feedback-joystick sponded to spider and butterfly pictures (Experi-
approach biases. ment 3). Although we did not compare the
traditional and the feedback-joystick task within
one experiment, Experiments 2a and 2b indicate
GENERAL DISCUSSION that only the feedback-joystick but not the tradi-
tional joystick task led to compatibility effects
The present research compared the sensitivity and when participants are not required to process
criterion-validity of the joystick task (cf. Chen & stimulus valence. In all experiments, the compat-
Bargh, 1999), the feedback-joystick task (Rinck & ibility effects in the manikin task were significant
Becker, 2007), and the manikin task (De Houwer
with nearly all applied outlier correction proce-
et al., 2001). Based on an analysis of procedural
dures, whereas the compatibility effects in both
differences between the tasks, we expected the
joystick tasks were significant only with few
manikin task to outperform the joystick tasks.
correction procedures. In sum, the sensitivity
Within the joystick tasks, the feedback-joystick
results indicate that the manikin task is a valid
was expected to be superior to the traditional
measure of the valence-induced activation of
joystick task. In addition, we compared the
MAAB with respect to their reliability for approachavoidance schemata, even in the absence
exploratory reasons. of evaluation intentions. The same is true for the
feedback-joystick task, yet with the constraint that
the measure is highly susceptible for the outlier
Sensitivity correction method. The joystick task is only a valid
Across three experiments, the manikin task had measure of the valence-induced activation of
larger effect sizes than the joystick tasks. This approachavoidance schemata when participants
suggests that the manikin task is a more sensitive have evaluation intentions, yet with the constraint
measure in mapping the valence-induced activa- that the measure is highly susceptible for the outlier
tion of approachavoidance schemata. This was correction method.
true when participants responded according to the The greater sensitivity of the manikin task
valence of positive and negative words (Experiment might be due to two reasons. First, recategorisation

COGNITION AND EMOTION, 2010, 24 (5) 823


KRIEGLMEYER AND DEUTSCH

of the responses is less likely in the manikin than in conclusions of De Houwer et al. (2001) in
both joystick tasks. The motivational orientation demonstrating that the representation of the
view (Neumann et al., 2003; Strack & Deutsch, approachavoidance behaviour is also a crucial
2004) as well as the evaluative coding account factor for non-intentional influences of valence.
(Eder & Rothermund, 2008; Lavender & Hom- In particular, we observed significant effects of
mel, 2007) suggest that the sensitivity of MAAB stimulus valence on approachavoidance behaviour
should decrease with the likelihood of recategoris- even in the absence of evaluative intentions, but
ing the responses. Second, according to the mainly in the manikin task. Whereas in the feed-
motivational orientation view, MAAB can be back-joystick task the compatibility effect reached
expected to be more sensitive to the valence- significance with only one outlier correction
induced activation of approachavoidance sche- method, the traditional joystick task did not reveal
mata when the means of distance regulation is significant compatibility effects. Together with the
running towards/away as compared to moving the results of Experiment 1, these findings suggest that
Downloaded by [King's College London] at 12:51 16 September 2017

object. Particularly, movements of the self might unintentional evaluations result in weaker activa-
be a more universal and, hence, more automatised tions of approachavoidance schemata than inten-
means of distance change than moving the object. tional evaluations and that the more sensitive a
To what degree recategorisation or means of MAAB is the more likely it can detect weak
distance change caused the differences in the activations of approachavoidance schemata.
sensitivity has to be investigated in further experi- The present research suggests that the null-
ments. To this end, the likelihood of recategorisa- effects in Lavender and Hommel’s (2007) as well as
tion and the means of distance change should be Rotteveel and Phaf’s (2004) experiments might be
manipulated independently within the same task. due to the operationalisation of the approach
One central question of the present work was avoidance reactions. In particular, in Lavender
whether the different MAAB are differentially and Hommel’s task, participants moved a little
sensitive to valence in the absence of intentions to doll towards or away from the screen. Thus, the
evaluate the stimuli. Previous research revealed means of distance change is moving towards/away.
inconsistent findings regarding this question. For Although this implies a high sensitivity according
instance, Lavender and Hommel (2007) concluded to our analysis, the opportunity of recategorisation
that the activation of approachavoidance re- might have reduced the sensitivity of the task. In
sponses depends on the intention to process particular, in Lavender and Hommel’s experimen-
stimulus valence (cf. Roelofs, Minelli, Mars, van tal setup the relationship between the concrete
Peer, & Toni, 2009; Rotteveel & Phaf, 2004). response and the meaning of the behaviour in
Diverging from this view, De Houwer et al. (2001) terms of approachavoidance is fixed. Across all
suggested that stimulus valence is processed in the trials, moving forward means approach and mov-
absence of the intention to do so, given that the ing backward means avoidance. As a consequence,
semantic representation of the stimulus is acti- participants can easily recategorise the behaviour,
vated. In other words, if the experimental setup for instance, as moving forward and backward,
allows for processing the semantic meaning of the thereby undermining the internal validity of the
stimulus, then the valence is processed and compa- task. In Rotteveel and Phaf’s study, participants
tible behavioural tendencies are activated. Our move their hand upward or downward to press
findings support this notion, given that participants buttons that are located on a vertical stand. As
based their approachavoidance decision on noun Rotteveel and Phaf sought to test effects on the
vs. adjective categorisations of the valent words. activation of muscles that have been related to
Presumably, this requires participants to engage in approach (i.e., the flexor muscle) and avoidance
some semantic analysis of the words and should (i.e., the extensor muscle), they did not instruct
thus increase the chances of non-intentional influ- their participants in terms of approachavoidance
ences of valence. Moreover, our findings extend the movements. Thus, the responses are unlikely to be

824 COGNITION AND EMOTION, 2010, 24 (5)


MEASURES OF APPROACHAVOIDANCE BEHAVIOUR

mentally represented as approachavoidance showed the same background as the spider pictures
movements. Taken together, the operationalisa- but not the spider. One limitation of the present
tions of the approachavoidance behaviours might research is that we only used self-reported fear of
have decreased the sensitivity of the tasks em- spider as a validity criterion. As self-reports might
ployed in these two studies. Given our analysis, it be distorted by self-presentation motives, real
might seem surprising that Chen and Bargh (1999) behaviour measures such as the spider approach
found an effect of task-irrelevant valence on lever test (cf. Rinck & Becker, 2007) might be better
responses. However, our reasoning does not imply validity criteria. Indeed, Rinck and Becker (2007)
that it is impossible to find such effects with lever found that approachavoidance indices of the
movements, but only that the likelihood is reduced feedback-joystick task predicted real behaviour in
as compared to other measures such as the manikin the spider approach test over and above self-
task. reported spider fear. These issues notwithstanding,
In sum, the present findings suggest that our results indicate that the manikin task is more
Downloaded by [King's College London] at 12:51 16 September 2017

valence activates approachavoidance responses robust against the effects of variations in metho-
even in the absence of evaluation intentions, and dological details.
in order to find such effects in MAAB, the In addition to correlations with explicit mea-
responses have to be clearly represented as sures, we also examined the correlations among
approachavoidance behaviours. The present re- the various MAAB. In all experiments, the
sults are in line with research on defensive reflexes approachavoidance indices of the manikin and
showing that passively viewing valent pictures the joystick tasks were positively, but not sig-
modulates the startle response (e.g., Bradley et al., nificantly correlated. This might be due to several
2001; Lang et al., 1990). Together, these lines of reasons. First, if the likelihood of recategorisa-
evidence support the notion that the perception of tions of the responses influences the internal
emotionally significant stimuli immediately re- validity of the tasks then this feature should also
sults in the activation of action dispositions decrease the correlations among the measures.
towards approach and avoidance, respectively. Second, low internal consistencies may limit the
correlations. This explanation applies for Experi-
ment 2a, in which valence was task irrelevant.
Criterion validity
Third, given that the MAAB reflect motivational
With respect to criterion-validity, the manikin task orientations they might assess different subtypes
outperformed the feedback-joystick task in the of approachavoidance motivations. In particular,
domain of spider approachavoidance motivations there might be individual differences in the
(Experiment 3). In particular, whereas spider preferred means of distance regulation (i.e., run-
approachavoidance indices of the manikin task ning towards/away vs. changing the position of
correlated with self-reported fear of spiders as well the object) as well as personsituation interac-
as with explicit spider evaluations, spider ap- tions regarding the preferred means. This might
proachavoidance indices of the feedback-joystick be especially relevant in the domain of spider fear.
task correlated only with explicit spider evalua- Whereas individuals with high fear of spiders
tions. Although we expected the manikin task to would prefer to run away from the spider,
have a higher criterion-validity, we were surprised individuals with medium fear of spiders might
that the feedback-joystick was uncorrelated with prefer to take the spider and put it outside the
explicit fear of spiders because Rinck and Becker room.
(2007) found such a relationship. One reason for
this divergence might be slight differences in the
Reliability
experimental procedure. Rinck and Becker’s task
consisted of more trials than ours (320 trial as Across all MAAB, reliability was satisfactory when
compared to 96 trials), and their control pictures participants responded according to the to-be-

COGNITION AND EMOTION, 2010, 24 (5) 825


KRIEGLMEYER AND DEUTSCH

measured feature of the stimuli (Experiments 1 and Conclusions


3). When, however, participants were not required
In summary, the present findings suggest that the
to process the to-be-measured feature (Experi-
sensitivity and the criterion-validity of MAAB
ments 2a and 2b), reliability was low. The various
strongly depend on the operationalisation of the
MAAB did not differ in their reliability. Surpris-
approachavoidance behaviour. Across all experi-
ingly, the manikin and the feedback-joystick task
ments, the manikin task outperformed the joy-
even had negative reliabilities in Experiments 2a
stick tasks in this respect. The superiority of the
and 2b. Although reliability coefficients theoreti-
manikin task might be due to low recategorisation
cally vary between 0 and 1, it is still possible that,
of approachavoidance responses and/or due to
empirically, they are negative. According to Cron-
the means of distance change (i.e., running
bach and Hartmann (1954) negative coefficients
towards/away instead of moving the object).
can arise when the test has zero internal consistency
Importantly, given a sensitive MAAB, valence
because of ‘‘chance fluctuations in splitting the test’’
Downloaded by [King's College London] at 12:51 16 September 2017

can be shown to activate approachavoidance


(p. 343). According to these authors, negative
reactions, even in the absence of evaluation
reliability coefficients should be interpreted as
intentions.
zero reliability.
Our reliability findings are in line with what
Manuscript received 22 August 2008
has been reported in the domain of implicit Revised manuscript received 13 May 2009
attitude measures (De Houwer & De Bruycker, Manuscript accepted 14 May 2009
2007a, 2007b) as well as in the domain of implicit First published online 16 December 2009
self-concept measures (Teige, Schnabel, Banse, &
Asendorpf, 2004). It is plausible that at least some
processing of the to-be-measured feature of the
stimuli is necessary for its influence on task REFERENCES
performance. Consequently, reliability can be
expected to be reduced by compromising partici- Borsboom, D., Mellenbergh, G. J., & van Heerden, J.
pants’ processing of the feature of interest (cf. De (2004). The concept of validity. Psychological Re-
Houwer, 2008). Supporting this reasoning, some view, 111, 10611071.
studies found an increase of the reliability when Bradley, M. M., Codispoti, M., Cuthbert, B. N., &
the task was modified in such a way that Lang, P. J. (2001). Emotion and motivation I:
Defensive and appetitive reactions in picture proces-
participants were required to process the to-be-
sing. Emotion, 1, 276298.
measured feature (De Houwer & De Bruycker, Cacioppo, J. T., & Gardner, W. L. (1999). Emotion.
2007a; Olson & Fazio, 2003). A second reason for Annual Review of Psychology, 50, 191214.
the low reliability when valence is task irrelevant Cacioppo, J. T., & Priester, J. R. (1993). Rudimentary
might be that individual differences in approach determinants of attitudes: II. Arm flexion and
avoidance reactions towards normatively positive extension have differential effects on attitudes.
and negative words are small. If this reasoning is Journal of Personality and Social Psychology, 65, 517.
true, then the high reliability in Experiment 1 Chen, M., & Bargh, J. A. (1999). Consequences of
might be due to method-specific variance like automatic evaluation: Immediate behavioral
individual differences in executive control pro- predispositions to approach or avoid the stimulus.
Personality and Social Psychology Bulletin, 25, 215
cesses. In particular, as the design involved two
224.
blocks with different response rules (i.e., a Cohen, J. (1988). Statistical power analysis for the
compatible and an incompatible block), the task behavioral sciences (2nd ed). Mahwah, NJ: Lawrence
might also assess individual differences in the Erlbaum Associates, Inc.
ability to switch between different response rules Coombes, S. A., Cauraugh, J. H., & Janelle, C. M.
(cf. Mierke & Klauer, 2003). (2007). Emotional state and initiating cue alter

826 COGNITION AND EMOTION, 2010, 24 (5)


MEASURES OF APPROACHAVOIDANCE BEHAVIOUR

central and peripheral motor processes. Emotion, 7, Journal of Personality and Social Psychology, 85, 197
275284. 216.
Cronbach, L. J., & Hartmann, W. (1954). A note on Hager, W., & Hasselhorn, M (Eds.). (1994). Handbuch
negative reliabilities. Educational and Psychological deutschsprachiger Wortnormen [Handbook of German
Measurement, 14, 342346. word norms]. Göttingen, Germany: Hogrefe.
De Houwer, J. (2008). Comparing measures of atti- Hommel, B., Musseler, J., Aschersleben, G., & Prinz,
tudes at the functional and procedural level: Analysis G. (2001). The theory of event coding (TEC):
and implications. In R. E. Petty, R. H. Fazio, & A framework for perception and action planning.
P. Brinol (Eds.), Attitudes: Insights from the new Behavioral and Brain Sciences, 24, 849937.
implicit measures. New York: Psychology press. Klauer, K. C., & Musch, J. (1999). Eine Normierung
De Houwer, J., Crombez, G., Baeyens, F., & Hermans, unterschiedlicher Aspekte der evaluativen Berwer-
D. (2001). On the generality of the affective Simon tung von 92 Substantiven [A standardization of
effect. Cognition and Emotion, 15, 189206. various aspects of the evaluation of 92 nouns].
De Houwer, J., & De Bruycker, E. (2007a). The Zeitschrift für Sozialpsychologie, 30, 111.
Downloaded by [King's College London] at 12:51 16 September 2017

identification-EAST as a valid measure of implicit Lang, P. J. (1995). The emotion probe: Studies of
attitudes toward alcohol-related stimuli. Journal of motivation and attention. American Psychologist, 50,
Behavior Therapy and Experimental Psychiatry, 38, 372385.
133143. Lang, P. J., & Bradley, M. M. (2008). Appetitive and
De Houwer, J., & De Bruycker, E. (2007b). The defensive motivation is the substrate of emotion. In
Implicit Association Test outperforms the extrinsic A. J. Elliot (Ed.), Handbook of approach and
affective Simon task as an implicit measure of inter- avoidance motivation (pp. 5165). New York: Psy-
chology Press.
individual differences in attitudes. British Journal of
Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1990).
Social Psychology, 46, 401421.
Emotion, attention, and the startle reflex. Psycholo-
De Houwer, J., & Eelen, P. (1998). An affective variant
gical Review, 97, 377395.
of the Simon paradigm. Cognition and Emotion,
Lang, P. J., Greenwald, M. K., Bradley, M. M., &
12(1), 4561.
Hamm, A. O. (1993). Looking at pictures: Affec-
De Houwer, J., Teige-Mocigemba, S., Spruyt, A., &
tive, facial, visceral, and behavioral reactions.
Moors, A. (2009). Implicit measures: A normative
Psychophysiology, 30, 261273.
analysis and review. Psychological Bulletin, 135, 347
Lavender, T., & Hommel, B. (2007). Affect and action:
368. Towards an event-coding account. Cognition and
Eder, A. B., & Rothermund, K. (2008). When do
Emotion, 21, 12701296.
motor behaviors (mis)match affective stimuli? An Markman, A. B., & Brendl, C. M. (2005). Constrain-
evaluative coding view of approach and avoidance ing theories of embodied cognition. Psychological
reactions. Journal of Experimental Psychology: Gen- Science, 16, 610.
eral, 137, 262281. Mierke, J., & Klauer, K. C. (2003). Method-specific
Fazio, R. H., & Zanna, M. P. (1990). Multiple variance in the Implicit Association Test. Journal of
processes by which attitudes guide behavior: The Personality and Social Psychology, 85, 11801192.
mode model as an integrative framework. Advances Mogg, K., Bradley, B. P., Field, M., & De Houwer, J.
in Experimental Social Psychology, 23, 75109. (2003). Eye movements to smoking-related pictures
Fishbach, A., & Shah, J. Y. (2006). Self-control in in smokers: Relationship between attentional biases
action: Implicit dispositions toward goals and away and implicit and explicit measures of stimulus
from temptations. Journal of Personality and Social valence. Addiction, 98, 825836.
Psychology, 90, 820832. Neumann, R., Förster, J., & Strack, F. (2003). Motor
Gerdes, A. B. M., Alpers, G. W., & Pauli, P. (2007). compatibility: The bidirectional link between beha-
Toward and away from spiders: Attentional engage- vior and evaluation. In J. Musch & K. C. Klauer
ment and disengagement in spider phobic patients. (Eds.), The psychology of evaluation (pp. 371391).
Psychophysiology, 44, S35. Mahwah, NJ: Lawrence Erlbaum Associates, Inc.
Greenwald, A. G., Nosek, B. A., & Banaji, M. R. Neumann, R., Hülsenbeck, K., & Seibt, B. (2004).
(2003). Understanding and using the Implicit Attitudes towards people with AIDS and avoidance
Association Test: I. An improved scoring algorithm. behavior: Automatic and reflective bases of behavior.

COGNITION AND EMOTION, 2010, 24 (5) 827


KRIEGLMEYER AND DEUTSCH

Journal of Experimental Social Psychology, 40, 543 ments. Journal of Experimental Social Psychology, 44,
550. 713720.
Olson, M. A., & Fazio, R. H. (2003). Relations Solarz, A. K. (1960). Latency of instrumental responses
between implicit measures of prejudice: What are as a function of compatibility with the meaning of
we measuring? Psychological Science, 14, 636639. eliciting verbal signs. Journal of Experimental Psy-
Ratcliff, R. (1993). Methods for dealing with reaction- chology, 59, 239245.
time outliers. Psychological Bulletin, 114, 510532. Steiger, J. H. (1980). Tests for comparing elements of a
Rinck, M., & Becker, E. S. (2007). Approach and correlation matrix. Psychological Bulletin, 87, 245
avoidance in fear of spiders. Journal of Behavior 251.
Therapy and Experimental Psychiatry, 38, 105120. Strack, F., & Deutsch, R. (2004). Reflective and
Rinck, M., Bundschuh, S., Engler, S., Muller, A., impulsive determinants of social behavior. Person-
Wissmann, J., Ellwart, T., et al. (2002). Reliability ality and Social Psychology Review, 8, 220247.
and validity of German versions of three instruments Szymanski, J., & Odonohue, W. (1995). Fear of
measuring fear of spiders. Diagnostica, 48, 141149. Spiders Questionnaire. Journal of Behavior Therapy
Downloaded by [King's College London] at 12:51 16 September 2017

Roelofs, K., Minelli, A., Mars, R. B., van Peer, J., & and Experimental Psychiatry, 26, 3134.
Toni, I. (2009). On the neural control of social Teige, S., Schnabel, K., Banse, R., & Asendorpf, J. B.
emotional behavior. Social Cognitive and Affective (2004). Assessment of multiple implicit self-concept
Neuroscience, 4, 5058. dimensions using the Extrinsic Affective Simon
Rotteveel, M., & Phaf, R. H. (2004). Automatic Task (EAST). European Journal of Personality, 18,
affective evaluation does not automatically predis- 495520.
pose for arm flexion and extension. Emotion, 4, 156 van Dantzig, S., Pecher, D., & Zwaan, R. A. (2008).
172. Approach and avoidance as action effects. Quarterly
Schoenmakers, T., Wiers, R. W., & Field, M. (2008). Journal of Experimental Psychology, 61, 12981306.
Effects of a low dose of alcohol on cognitive biases Wiers, R. W., van Woerden, N., Smulders, F. T. Y., &
and craving in heavy drinkers. Psychopharmacology, de Jong, P. J. (2002). Implicit and explicit alcohol-
197, 169178. related cognitions in heavy and light drinkers.
Seibt, B., Neumann, R., Nussinson, R., & Strack, F. Journal of Abnormal Psychology, 111, 648658.
(2008). Movement direction or change in distance? Wittenbrink, B., & Schwarz, N. (Eds.). (2007). Implicit
Self and object related approachavoidance move- measures of attitudes. New York: Guilford Press.

828 COGNITION AND EMOTION, 2010, 24 (5)

You might also like