Toward an implicit measure of emotions: ratings of abstract images reveal distinct emotional states

ABSTRACT

Although implicit tests of positive and negative affect exist, implicit measures of distinct emotional states are scarce. Three experiments examined whether a novel implicit emotion-assessment task, the rating of emotion expressed in abstract images, would reveal distinct emotional states. In Experiment 1, participants exposed to a sadness-inducing story inferred more sadness, and less happiness, in abstract images. In Experiment 2, an anger-provoking interaction increased anger ratings. In Experiment 3, compared to neutral images, spider images increased fear ratings in spider-fearful participants but not in controls. In each experiment, the implicit task indicated elevated levels of the target emotion and did not indicate elevated levels of non-target negative emotions; the task thus differentiated among emotional states of the same valence. Correlations also supported the convergent and discriminant validity of the implicit task. Supporting the possibility that heuristic processes underlie the ratings, group differences were stronger among those who responded relatively quickly.

KEYWORDS:

• Emotion
• emotion induction
• misattribution
• implicit assessment

Acknowledgements

We thank Dr. E. Samuel Winer for his insightful comments on and editing of an earlier draft of this manuscript.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCiD

Gregory Bartoszek http://orcid.org/0000-0003-0814-5100

Notes

1. Research on the AMP findings is mixed in that some investigators (i.e. Blaison et al., 2012) conclude that only cold semantic priming underlies the AMP effects whereas others (Gawronski & Ye, 2014) argue that both semantic concepts and affective feelings drive the misattributions.

2. In order to detect at least a medium effect size (f = .25), power analyses determined a sample size for each experiment based on an alpha error set to 0.05 and a recommended statistical power of .8.

3. The scripts and audio-recordings of the stories are available from the first author upon request.

4. A median split procedure is sometimes associated with a loss of power, as it adopts the assumption that variances within each group (i.e. high and low) of the dichotomised variable is meaningless. While this assumption is often unwarranted, we believe that, in this case, the assumption is justified. That is, we do not expect that among slow respondents, there is a difference between a person who takes 8 seconds to rate an image and the person who takes 12 seconds. Both of these individuals would rely on more deliberate, controlled process than someone who takes 3 or 4 seconds to make such ratings. In other words, the person making the ratings uses either slow, controlled processes or fast, heuristic processes. Indeed, studies show that there are behavioural and neural differences between those responding fast or slowly to auditory or visual stimuli (e.g. Protzner & McIntosh, 2007), and thus median splits of reaction times are often preferred (e.g. Protzner & McIntosh, 2007; Smith, Johnstone, & Barry, 2006). Finally, we used the full range of our data rather than removing data of participants near the median to create more distinct groups.

5. The duration time of the math test (which was part of the anger-induction procedure) varied across participants and was used as a covariate in all analyses of between-condition differences.

6. By “implicit emotion” or “explicit emotion,” we are referring to (emotion) assessment procedures that are implicit or explicit. We are not proposing ontologically distinct realms of emotion, implicit versus explicit.

7. For example, in response to the question, “What do you think was the actual purpose of this study?” one participant in the anger condition wrote, “The real purpose was to give me a faulty file so I could waste an hour of my life and not receive credit for taking time out of my schedule to help some dude get a degree.”

8. Neutral pictures included: 7000, 7002, 7004, 7006, 7009, 7010, 7025, 7035, 7090, 7100, 7140, 7150, 7175, 7186, 7205, 7217, 7233, 7235, 7705 and 7950.

9. Participants’ response times were not measured in this online experiment, and thus we could not perform separate analyses for slow and fast respondents.

10. We collapsed data across the three control groups, as they neither interacted with nor differed in any of the implicitly measured emotions (ps > .10).