Abstract
Emotional picture databases are commonly used in emotion research. The databases were first based on ratings of emotional dimensions, and the interest in studying discrete emotions led to the categorization of subsets from these databases to emotional categories. However, to-date, studies that categorized affective pictures used confidence intervals in their analysis, a method that provides important data but also results in a high percentage of blended or undifferentiated categorization of images. The current study used 526 affective pictures from four databases and categorized the pictures to discrete emotions in two steps (Pre-testing phase & Experiment 1). First, clinical psychologists were asked to generate emotional labels for each picture, according to the emotion the picture evoked in them. This resulted in the creation of 10 emotional categories. These labels were presented to students who were asked to choose the emotional category that matched the emotion a presented picture evoked in them. Agreement levels on the emotional categories were calculated for each picture, and pictures were categorized according to the most dominant emotion they evoked. The analysis of agreement levels rather than confidence intervals enabled us to provide both dominance of emotional category and agreement in the population regarding the dominance. In Experiment 2, we asked participants to provide ratings of emotional intensity and arousal, in order to provide more detailed information regarding the database. This is the first study to provide agreement levels on the categorization of affective pictures, and may be useful in various studies which aim at generating specific emotions.