C´alculo del perfil de l´ınea

Two pilot studies were used to investigate which lower probability distributions are able to be learned or too close to chance. In total, three different lower probability distributions were tested. In the first pilot study participants had to learn a 52-22-22 and a 48-24-24 probability distribution. Results showed a high error rate with regard to the correctness of the predictions and highly significant prediction errors for the estimated probabilities via the Concept Awareness Questionnaire. Thus, participants did not seem to learn the probability distributions. In the second pilot study, a 59-18.5-18.5 probability distribution was tested which seems to be easier to learn than the probability distributions in the first pilot study. Finally, this probability distribution was chosen for the final experiment.

Besides the learnability, an object bias was found in the first pilot study. This bias was probably not existing in Experiment IV while using the same Gabor figures due to the larger sample size and the

Chapter 7: Experiment V – Learning Different Probabilities 105 lower uncertainty. The newly developed Gabor figures in the second pilot study with no directional information seem to be less confounding and thus were also used in the final experiment.

In conclusion, the setting used in the second pilot study worked well and can be applied in the final experiment. In more detail, the 59-18.5-18.5 probability distribution should be used for low probability concept and Gabor figures with few circles, many circles and a symmetric pattern as target objects.

7.3 Method

7.3.1 Participants

26 students participated in the study at the IfADo. Two participants had to be excluded due to technical issues and language barriers. Finally, 24 participants (14 female) with mean age 24 years (SD=3 years) entered data analysis. All of them had a dominant right hand and no vision impairment.

7.3.2 Procedure

Participants in Experiment V had to performer a 59-18.5-18.5 lower and a 74-11-11 higher probability condition due to the results of the pilot studies. The procedure was the same as in the second pilot study. In general, there were three main differences of Experiment V in comparison to the previous experiments. First, participants had to learn two different probability concepts separately each within a test session. Thus, participants had to perform two sessions on different dates. The high probability concept lasted four blocks as in the previous experiments whereas the low probability concept lasted six blocks as few information was available about how much time participants need to learn a concept with a lower probability concept. At the end of each session participants had to complete the Concept Awareness Questionnaire for the learned concept. The sessions were counterbalanced in the way that all participants with even number started with the lower probability concept and all participants with odd numbers started with the higher probability concept. Second, for the first time participants were explicitly instructed that every object is associated with the exits to a distinct probability to control prior knowledge. Participants had to complete the Concept Awareness Questionnaire after each session and would be biased concerning a possible probability concept in the second session. Therefore, we decided to inform them about the presentation of a probability structure. Thus, the initial uncertainty regarding the task was reduced. Third, all object-exit associations were counterbalanced across participants in order to avoid potential confounding effects caused by an object bias.

As the findings of the D2 attention test were consistent in all of the previous experiments, this test was excluded in the current experiment. Instead, the “Inventar zur Messung der Ambiguitätstoleranz” by Reis (1996) was used at the end of the experiment to control for ambiguity tolerance which might

Chapter 7: Experiment V – Learning Different Probabilities 106 influence the learning process. Tymula et al. (2012) already reported in their study that participants with a higher tolerance for ambiguity, i.e. options with consequences inhering unknown probabilities, showed increased risk-seeking behavior. Thus, participants’ performance might also be influenced by tolerance of ambiguity in the current experiment. Participants filled in the questionnaire with 40 items on a scale ranging from 1 (“trifft sehr zu”, ”I strongly agree”) to 6 (“trifft gar nicht zu”, ”I strongly disagree”). This inventory contains five subscales measuring ambiguity tolerance for problems that seems to be unsolvable (1), ambiguity tolerance for social conflicts (2), ambiguity tolerance of the parents’ image (3), ambiguity tolerance for stereotyped roles (4) and ambiguity tolerance for new experiences (5). The internal consistency for these subscales lies between Cronbach’s alpha α = .74 and α = .86 and for the full scale α =.87.

7.3.3 Data Analysis

Before analyzing data statistically, again systematical drifts in eye movement data were checked and drifts were found for the first time. Thus, a correction was necessary by calculating the mean of the scatter plot per block and then, centering the scatter plot. 2.2% of the trials were excluded from data analysis of the high-probability concept due to missing predictions and additionally, 0.5% of the trials were excluded as less than 65 % of the eye movement data of the trial were valid. In the low-probability concept condition, 1.9% of the trials were excluded as predictions were missing and additionally, 0.2% of the trials had to be excluded due to insufficient validity of the data (less than 65%). Distinct AOIs were defined as shown in table 7.3. These additional classifications of AOIs were necessary due to the counterbalancing.

The further analyses are comparable with those in the previous experiments. Two-way repeated measures ANOVAs were calculated with the within-subject factors block (1-4) and judgment (correct, incorrect) for each condition in order to analyze the effect of the learning development. Dependent variables were fixation frequency, fixation duration, number of gaze shifts, scanpath length, saccadic

Table 7.3: AOI classifications and definitions in Experiment V

AOI Definition

AOItarget AOI around the exit at which the target object appears AOIpredict AOI around the exit which is predicted by the participant AOIlow AOI around the exits with the lower probability

Chapter 7: Experiment V – Learning Different Probabilities 107 velocity as well as judgment time and reaction time. Now, saccadic velocity could be calculated as eye movements were recorded with a higher sampling rate of 500 Hz. Additionally, task performance, viz. the number of correct predictions was used as dependent variable. ANOVAs were also run for the newly defined AOIs to analyze the participants visual search behavior within the distinct AOIs. For this reason, fixation duration, fixation frequency and a new variable, the dwell time were analyzed. The dwell time is the time participants spend in the AOI independent of the fixations (see Chapter 1). For the comparison of both conditions, the high and low probability concept, planned t-tests were run as well as repeated measures ANOVA with the between factor probability condition.

Stimuli used in the experiments (Experiment I-V) were gradually developed due to object biases. The type of object (geometric figures, Gabor figures with lines, Gabor figures symmetric patterns) as well as the background (patterned vs. single-colored) differed in the experiments. Planned t-tests were used to investigate the influence of the stimulus presentation on the subjective probability concept, i.e. the prediction error, eye movements, i.e. fixation frequency, as well as performance. Finally, correlation analyses were used to test the impact of the confounding variables on task performance.

7.4 Results

In the following, only significant (p<.05) results or trends (p<.10) were reported, except if the results were relevant for the aforementioned research questions.