• Keine Ergebnisse gefunden

6. Behavioral Analysis

6.4. Excursus: abandoning rigid stimulus categories

46 5 (both in the fMRI experiment). Both of these items cover trusting the forecast or the personal intuition when deciding about the weather, but are coded conversely. More correlations were found, but the overall picture is such that the construction of the TI as an indicator for trust in the different sources of information seems to have worked.

Thus the TI presents a good option to assess trust in information for experimental designs with two sources of information. The TI could have been instrumental in separating the participants into different groups based on their strategy, to compare neural processing between different applied strategies. Alas, the strategies were too varied to create conclusive groups with a sample of only 30 persons.

Clearly, these three aspects of the participants' strategy encompass only some parts of it. Other facets, like the application of rules or heuristics or the motivation of the participant would have to be assessed with different approaches.

47 - Conversely, the High-High category was assumed to lead to a certain decision to take the umbrella along.

- The High-Low and Low-High conditions were assumed to cause uncertain decision making because of the contradicting information from the two sources of information.

- The Middle-Middle category was assumed to prompt uncertain decision making due to the ambiguity in the information available.

- The remaining four categories were also assumed to cause uncertain decision making due to varying degrees of contradiction and ambiguity.

Fig. 12: Visualization of the stimulus space, without division into categories on the left, and with them on the right. The abbreviations found in the right diagram stand for low (L), middle (M), or high (H) predicted probability of rain; The first letter describes the information obtained from the

forecast, and the second letter describes the information obtained from the sky picture.

After the experiment had been conducted and the analyses commenced, the performance of the participants in each category was explored with different means.

Box plots were created for the mean response times, the PI, and the UI (See figures 13 and 14 [A and B] for the corresponding plots).

48 Fig. 13: Box plots visualizing the mean response time in each experimental category (L = low-,

M = middle-, H = high probability of rain; the first letter describes the forecast and the second the sky picture). Black dots denote outliers.

The box plot of the response times showed huge variance in the mean response times between participants, in most of the categories. This becomes evident in the difference between the 1st and 3rd quartiles, and in the high number of outliers.

For example, the High-High and Low-Low categories seemed to have the shortest response times, thus inducing certain decision making. However, an ANOVA and successive post-hoc t-tests (alpha = 0.05, bonferroni-holm corrected) revealed that, while there were some significant differences between the conditions, the response times in LL did not differ significantly from ML, and HH did not differ significantly from ML, HM, or LM.

49 Fig. 14: Box plots visualizing the PI (A) and the UI (B) in each experimental category (L = low-,

M = middle-, H = high probability of rain; the first letter describes the forecast and the second the sky picture). Black dots denote outliers.

A

B

50 The picture painted by the box plot for the PI is a diverse one. On the one hand, the categories assumed to be certain are indeed leaning strongly to one end of the spectrum (either 1 in the case of High-High, or 0 in the case of Low-Low). On the other hand, each category has a huge variance between participants. That is especially true for the 'uncertain' categories like High-Low and Low-High, but even the 'certain' categories vary more than one would have expected, if they truly reflected a clear and certain decision situation. Apparently, the participants' response to the categories differed greatly.

The same conclusion can be drawn from the box plots depicting the UI. The amount of incoherent decision varies strongly between the participants, even in supposedly certain conditions, where twice a participant had more than 50% incoherent decisions.

To investigate these interindividual differences, the BSPs are an excellent tool (see figures 9 and 10). After consulting them it became apparent that the participants followed widely different strategies in their approach to the given problem. Therefore, only some adhered to the behavior laid out in the predefined categories, while other differed somewhat from them. Still others had a strategy that completely contradicted all assumptions of certainty and uncertainty within the stimulus space. See figure 15 for examples of all three types of decision makers.

In the end, it was decided to abandon the previously established categories. While they may work on average, there were too many participants that deviated from the expected pattern of decisions. Additionally, even those participants that - unknowingly - conformed to the general structure of the categories did not do so for the entirety of each category. Even if, for example, the Middle-Middle category contained many trials with incoherent decisions, this was never true for every trial in the category. So if the fMRI or EEG analyses aim to compare uncertain and certain decision making, the trials being used for that analysis have to be selected individually for each participant. This was considered to be the only way to ensure that, despite the

51 participants' individual decision strategies, the basis for following analyses remained as sound as possible.

Fig. 15: BSPs for three participants, illustrating a good, a mediocre and a bad fit between the decision pattern and the experimental categories.