Edited by:
Reviewed by:
*Correspondence:
This article was submitted to Cognition, a section of the journal Frontiers in Psychology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Ambivalence refers to a psychological conflict between opposing evaluations, often experienced as being torn between alternatives. This dynamic aspect of ambivalence is hard to capture with outcome-focused measures, such as response times or self-report. To gain more insight into ambivalence as it unfolds, the current work uses an embodied measure of pull, drawing on research in dynamic systems. In three studies, using different materials, we tracked people’s mouse movements as they chose between negative and positive evaluations of attitude objects. When participants evaluated ambivalent attitude objects, their mouse trajectories showed more pull of the non-chosen evaluative option than when they evaluated univalent attitude objects, revealing that participants were literally torn between the two opposing evaluations. We address the relationship of this dynamic measure to response time and self-reports of ambivalence and discuss implications and avenues for future research.
Imagine walking across campus after a long day of work. While you are heading to the train station, you notice a fast-food truck. While your preference for the ease of a quick and tasty snack might pull you toward the truck, your aversion to unhealthy foods may pull you away from it. And even though you keep heading for the train, your final path may show a curve, reflecting the pull of the fast-food truck.
The psychological state of being torn between opposing evaluations, or ambivalence, is an inescapable part of human life. People can be ambivalent about a wide range of topics including fast food, abortion, out-group members, organ donation, euthanasia, and alcohol (for overviews, see
Conceptually, ambivalence involves a psychological conflict between opposing implicit or explicit evaluations about an attitude object (e.g., Meehl, 1964 [in
How might behavioral scientists be able to assess this psychological tug-of-war between opposing evaluations? Most indices of ambivalence are either assessed after an evaluation (i.e., response time based measures) or rely on people’s subjective assessment of their psychological state (i.e., self-report based measures), both of which may have their limitations. Whereas the former presents us with the problem of a black box (i.e., we cannot know what goes on during the formation of the response), the latter relies on people’s (often biased) insight into their own thoughts and feeling. Below, we discuss both approaches in turn.
Self-report based measures of ambivalence ask how conflicted individuals feel (e.g.,
Researchers have also inferred ambivalence from response times (e.g.,
Thus, although ambivalence can be understood as a tug-of-war between evaluations, to our knowledge, no research has systematically addressed the dynamic evaluative processes that underlie ambivalence. The present research begins to fill this gap. Drawing on an approach developed in dynamic systems research, we propose that motor outputs during the evaluation process may provide a window into the unfolding of ambivalent evaluations.
Research has shown that motor output can readily reveal people’s evaluative stance toward attitude objects. For instance, subtle changes in specific muscles in the face (e.g.,
As a consequence, ongoing mental processes can be observed in even simple behaviors, such as mouse movements on a computer screen. In the mouse tracking paradigm people engage in a computer task in which they perform certain actions, for instance categorization of human faces as male or female, by moving their mouse toward specific response locations on the screen. While they do so, the path of their mouse is traced and recorded. The characteristics of the trajectories, such as its curvature for instance, can be used to give insight into evaluative conflict
For instance, in stereotyping research, when people were asked to evaluate whether masculine or female faces fitted a specific trait by dragging the mouse over the screen to a certain response, arm movements showed pull to both responses when the face was
Capitalizing on the embodied expression of ongoing evaluations, we conducted three studies in which we employed a mouse tracker paradigm, allowing us to assess motor output during an evaluative task. In accordance with previous operationalizations of cognitive competition (
It is worth noting that one earlier study has related mouse locations on a computer screen to assess mixed feelings toward a target person (
Aside from our primary measure of pull, we also assessed response times. As noted above, previous work found that people take longer to evaluate ambivalent attitude objects, compared to non-ambivalent attitude objects (e.g.,
In all studies, we report all manipulations, all participants and exclusions, and all dependent measures. To reduce the influence of outlier responses, latencies under 300 ms as well as above 3,000 ms were excluded from analyses in all studies, however, including these observations did not alter our results. Cohen’s
Forty-nine students (34 females, 3 unreported,
We used 12 attitude objects that have successfully been used in prior research to induce ambivalence (
Upon entering the lab, participants were seated in individual cubicles. All instructions were computer-administered. In the first part of the experiment, participants were asked to evaluate different words as positive or negative. Participants were instructed to respond as quickly and accurately as possible. At the start of each trial, a start button appeared at the bottom center of the screen and the response options (positive and negative) appeared in the top left and top right corner of the screen. When participants clicked the start button at the center bottom of the screen, the cursor was relocated to the center starting position (to ensure all trajectories start from the same location) and the attitude object appeared. Participants then moved the mouse to one of the two response buttons located in the top right and top left of the screen. After this, the attitude object disappeared and a new trial started.
At the start of the experiment, participants completed two neutral practice trials to familiarize themselves with the task. Each participant evaluated each attitude object twice in two separate blocks of trials, including 12 attitude objects each, resulting in 24 trials. In each block, all attitude objects were presented once, in a random order. The location of the positive and the negative response option was reversed between blocks. Unfortunately, due to a technical error, only one counterbalancing condition functioned, and block order was not counterbalanced. For all participants, in the first block, the positive response option appeared on the right and the negative on the left. In the second block, this location was reversed and the negative response option appeared on the left and the positive response option on the right.
During each trial, streaming
Ambivalence was measured with three established measures of ambivalence. First,
Second, participants’
Third, we measured participants’
Due to a technical error, the data of one participant were lost. One participant did not provide any self-report ratings. To reduce the influence of outlier responses, latencies under 300 ms as well as above 3,000 ms (1.9% of trials; cf.
Paired sample
As predicted, participants’ mouse trajectories showed a greater pull toward the opposite evaluation when they evaluated ambivalent (
Moreover, the moment at which the pull of the opposing evaluation was largest, occurred later for ambivalent stimuli (
Responses were faster in the second than in the first block; however, the pattern of results was the same and we collapsed the data over blocks. Replicating numerous previous findings (e.g.,
The measure of pull did not correlate with self-report measures of ambivalence. Because we orthogonally manipulated ambivalence, perhaps reducing variance, we separately examined correlations between our measures for ambivalent and univalent stimuli. These analyses did not reveal significant relations between the response times and pull on the one side, and self-report measures of ambivalence on the other side. This is surprising because earlier work (e.g.,
Finally, exploratory analyses showed that for ambivalent stimuli, trajectories showed more directional reversals over both the horizontal axis (i.e., x-flips;
As hypothesized, during evaluation of ambivalent attitude objects, pull toward the non-chosen evaluation was greater than during the evaluation of univalent attitude objects, indicating that ambivalence literally pulls people in two directions. Shedding more light on the unfolding of ambivalence during evaluation, we also found that maximum conflict occurred later in time for ambivalent evaluations, compared to univalent evaluations. Additionally, we replicated the effect of ambivalence on response times (
Study 2 replicates and extends the first study by examining the process of evaluating health-related ambivalent attitude objects. Unhealthy eating and drinking behavior are known to be strong elicitors of ambivalence (
Thirty-eight students (28 females,
In this study, we used images depicting different kinds of foods and beverages (for clarity we will refer to ‘foods’ for this category throughout the manuscript). The ambivalent attitude objects depicted: beer, a hamburger, chocolate, and fries. The univalent attitude objects depicted vegetables, an apple, bread, and orange juice, resulting in a total of eight trials. The attitude objects were presented in random order. The images were acquired from the Internet and, keeping original constraints, scaled to 300 pixels wide and presented against a white background.
The experiment was administered in the same manner as Study 1, using the same measures, with the exception that the location of the response buttons on the screen was counterbalanced between participants. All attitude objects were presented only once and no habituation could ensue. Finally, the neutral practice trials consisted of an image of baskets and an image of a gray desk cabinet.
Responses under 300 ms and above 3,000 ms were removed from the data (2.6%) as well as errors (i.e., negative responses to the generally positive univalent stimuli, 1.6%; cf.
Ambivalence scores were higher for ambivalent attitude objects than for univalent attitude objects on objective ambivalence (
Confirming our hypotheses, we found that pull was greater for ambivalent attitude objects (
The data revealed slower responses for ambivalent stimuli (
Examining the correlations between the mouse tracker variables and the self-report variables, we found that pull was positively correlated with subjective ambivalence
The results of Study 2 replicated the core findings of Study 1. Ambivalent objects were associated with stronger pull indicating more ambivalence. Additionally, ambivalent and univalent stimuli did not only differ in amount of pull and magnitude of curves, but also in the temporal unfolding. Finally, pull and response time were not correlated.
However, the pattern of correlations between self-reported and pull and response times differed from Study 1, such that subjective and objective ambivalence did have a positive relationship with both the magnitude of pull as well as the time people took to respond. These differences across studies may be due to differences between the stimuli used (abortion and euthanasia in Study 1; foods in Study 2). To test this possibility we directly compared the two types of stimuli in one design in Study 3.
In Study 3, we used the attitude objects from Study 2 (foods) and Study 1 (abstract topics) in a single design. To make sure that possible differences were not due to mode of presentation (i.e., Study 1 stimuli presented as words vs. Study 2 stimuli presented as images) we now presented the attitude objects from Study 2 also as words. Thus, instead of showing a picture of a hamburger, we now presented the word “hamburger.” Second, because one of the attitude objects in the category foods was “beer,” we decided to drop “alcohol” from the Study 1 attitude objects, to prevent stimulus overlap.
Forty-five students (29 females, 1 unreported,
In this experiment, we used attitude objects from both previous studies, resulting in 17 trials (7 ambivalent, 10 univalent, see below). From Study 1, we took the ambivalent abstract topics: abortion, euthanasia, and organ donation (we had only three stimuli here, because we dropped alcohol from this category, as it would already appear in the food stimuli category as “beer”). The univalent abstract topics were: depressed, abuse, unhappy, happy, sun, and holiday. We used the same food stimuli as in Study 2. The ambivalent food stimuli were: beer, hamburger, chocolate, and fries with sauce
The experiment was run in the same manner as Studies 1 and 2, using the same measurements.
One participant did not provide data on the self-report measures. Responses given under 300 ms as well as above 3,000 ms were removed from the data (3.5%) as well as errors (i.e., positive responses to negative stimuli and negative responses to positive stimuli, 0.4%; cf.
Ambivalent attitude objects were more ambivalent compared to univalent attitude objects for objective ambivalence,
Study 3 mean ambivalent ratings for objective, subjective and affective ambivalence for ambivalent vs. univalent stimuli, for abstract topics, foods, and overall.
Objective Ambivalence |
Subjective Ambivalence |
Affective Ambivalence |
||||
---|---|---|---|---|---|---|
Ambivalent | Univalent | Ambivalent | Univalent | Ambivalent | Univalent | |
Overall | 1.51 (0.87) | 0.18 (0.62) | 3.74 (1.34) | 2.02 (0.86) | 1.23 (0.98) | –0.16 (0.64) |
Abstract topics | 1.59 (0.92) | 0.13 (0.63) | 3.77 (1.95) | 2.20 (1.35) | 1.43 (1.12) | –0.46 (0.71) |
Foods | 1.43 (1.13) | 0.23 (0.77) | 3.72 (1.85) | 1.84 (0.86) | 1.03 (1.21) | 0.13 (0.74) |
Repeated measures analyses with valence (ambivalent vs. univalent) and topic (abstract vs. foods) as within-subjects factor and MD as dependent variable showed a main effect of topic and valence (see
Study 3 means (SD) for maximum deviation (MD), response time (RT), and maximum deviation time (MDT) for ambivalent vs. univalent stimuli, for abstract topics and foods separately.
Abstract Topics |
Foods |
|||
---|---|---|---|---|
Ambivalent | Univalent | Ambivalent | Univalent | |
MD | 0.70 (0.38) | 0.23 (0.21) | 0.26 (0.24) | 0.14 (0.23) |
RT | 1737.01 (354.71) | 1237.29 (206.13) | 1388.71 (306.11) | 1244.66 (263.83) |
MDT | 1063.11 (286.33) | 682.58 (128.29) | 840.08 (244.26) | 727.35 (185.61) |
The main effect of topic revealed that pull was greater for abstract topics (
Finally, there was an interaction between valence and topic such that the main effect of valence was also somewhat stronger for abstract topics, compared to foods,
A main effect of valence showed slower responses for ambivalent stimuli (
Overall, response times correlated marginally negatively with pull,
Although, overall, we obtained the same pattern of results for both foods and abstract topics, effect sizes between studies differed. To investigate whether these signify a structural difference between these topic domains we compared size of the effect of our manipulation for foods vs. abstract topics for all our measures. First, we calculated effect sizes for foods and abstract topics separately in Study 3. We then calculated a point estimate for each index and topic domain separately and tested whether the point estimates differed significantly (see
Point estimates [95% CI] for effect sizes for objective, subjective, and affective ambivalence, response times (RT), curvature (MD), and moment of occurrence of maximum conflict (MD time).
Abstract Topics | Foods | ||||
---|---|---|---|---|---|
Objective ambivalence | 1.66 [1.11, 2.22] | 1.08 [0.81, 1.34] | 3.47 | 1 | 0.06 |
Subjective ambivalence | 1.35 [0.50, 2.19] | 1.09 [0.82, 1.36] | 0.33 | 1 | 0.57 |
Affective ambivalence | 1.62 [1.02, 2.21] | 1.37 [0.73, 2.01] | 0.32 | 1 | 0.57 |
RT | 1.32 [0.47, 2.18] | 0.85 [0.60, 1.10] | 1.06 | 1 | 0.30 |
MD | 1.05 [0.66, 1.44] | 0.45 [0.23, 0.68] | 6.64 | 1 | 0.01 |
MD time | 1.15 [0.50, 1.80] | 0.81 [0.56, 1.05] | 0.94 | 1 | 0.33 |
In addition, the data of Study 2 and the food conditions of Study 3 can be brought to bear on whether pictures of foods or words describing foods are more likely to elicit ambivalence. Although both studies presented the same attitude objects, the visual presentation of foods caused stronger effects on the subjective measure of ambivalence, (point estimate = 1.75, 95% CI [1.28, 2.22]) than the presentation of food words (point estimate = 1.08, CI [0.71, 1.46],
As expected, we found that evaluating ambivalent attitude objects, compared to univalent attitude objects, caused more pull toward the non-chosen option, both for abstract topics and foods. Additionally, the moment of maximum pull occurred later for ambivalent attitude objects. These findings replicate the effect found in Studies 1 and 2. We also replicated the effect of valence on response times – ambivalent evaluations took longer. Pull was also positively correlated with subjective ambivalence. Comparing effect sizes across studies, we found that when it comes to different topic domains, abstract topics yield larger effect sizes on pull of opposing evaluations and objective ambivalence, but not on any of the other indices. Finally, pictures of foods elicit somewhat stronger effects on subjective ambivalence compared to food words, but not on any of the other indexes of ambivalence.
Ambivalence is a psychological state in which people are being torn between “one side” and the “other side” when making evaluations. This dynamic aspect of ambivalence is hard to capture in response times or self-report based measures of ambivalence. To gain more insight into ambivalence as it unfolds, we used an embodied measure of pull, drawing on research on dynamic systems (
Several findings are worth highlighting. First, our data show that amount of pull as assessed with the embodied measure, is sensitive to differences in ambivalence. When participants evaluated an ambivalent attitude object, their mouse trajectories showed more pull of the non-chosen option than when they evaluated a univalent attitude object, revealing how they were literally torn between two opposing evaluations. Second, the temporal trajectory of the motor movements suggests that the peak of experienced conflict, as indexed by the moment of maximum pull, occurs later for ambivalent than for non-ambivalent attitude objects. Thus, our work shows not only a difference in degree of pull between opposing evaluations, but also sheds light on its temporal unfolding. Finally, across all three studies, we replicated earlier findings that evaluating ambivalent attitude objects takes more time overall than evaluating non-ambivalent attitude objects (
Within cognitive psychology, longer response times have been considered an indicator of cognitive competition and conflict (e.g.,
In the present studies ambivalence was not experimentally manipulated and, instead, we used pre-selected materials. One might argue that our stimuli also differed on other dimensions than just ambivalence, for instance familiarity. However, our materials were all based on previous research into ambivalence and have been tried and tested in an array of studies (
It is worth noting that we found some variation in the relationship between pull and self-report based measures of ambivalence. Ambivalence is theoretically conceptualized as a state of evaluative conflict, caused by simultaneous activation of opposing evaluations (e.g., Meehl, 1964 [in
Previous work has addressed a similar question with regard to relatively low correlations between different self-report based measures of ambivalence. Measures of subjective and objective ambivalence generally show only moderate correlations (e.g.,
A notable finding from our exploratory analyses showed that, at least for Studies 1 and 3, trajectories were more complex as a function of ambivalence. This suggests that online experiences of ambivalence are an important factor in the driving and altering of response planning and execution. Future work might extend these preliminary findings and further understanding of the embodied consequences of ambivalence. It is also worth exploring whether different types of ambivalent attitudes unfold differently over time. Conceptually, different types of ambivalence can be distinguished. Intracomponent ambivalence refers to ambivalence as the result of conflicting cognitions (cognitive/cognitive conflict) or conflicting affective responses (affective/affective conflict), (
Our work shows that ambivalence is readily captured in mouse movements. Together with previous work on whole body movement and ambivalence (
Finally, given that ambivalence has important consequences for information processing (e.g.,
In the present work, we took a first step in assessing ambivalence through an embodied measure of pull between opposing evaluations. We found that the measure of pull is sensitive to manipulations of ambivalence, showing that people are torn between evaluations when evaluating an ambivalent attitude object. Pull was mostly positively correlated with self-reports of subjective ambivalence. However, it was not consistently related to objective ambivalence and response time, suggesting that there is more to ambivalence than these measures capture. Future research should further address the degree to which the measure of pull differentially relates to self-report measures and predicts different outcomes. The embodied approach used here adds to the methodological repertoire that researchers can draw on to further unravel the dynamic and multifaceted nature of ambivalence.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The authors thank J. Sanchez-Burks for helpful comments on previous versions of this manuscript. This work was made possible by The Netherlands Organization for Scientific Research (NWO) Rubicon Grant awarded Iris K. Schneider (# 446-13-015), as well as by a Consolidator Grant from the European Research Council (ERC-2011-StG_20101124) awarded Sander L. Koole.
In the Netherlands, it is customary to eat fries with some sort of sauce. Mayonnaise is usually the sauce of choice (in fact, to the degree that “fries