## Introduction

### The reasoning process

### Visual attention

### Research objectives

### This study

### Using eye movements as a method to assess memory retrieval

### Hypothesis 1: differences experienced in task difficulty

### Hypothesis 2: elements of the situation model

### Hypotheses 3: integrative solutions

## Method

### Participants

### Apparatus

### The black box task

^{1}

### Procedure

### Pairwise comparisons

## Results

### Performance

_{A&O}= 92% (SD = 13); M

_{A}= 90% (SD = 11); M

_{O}= 86% (SD = 15); M

_{N}= 79% (SD = 22); F

_{ACC}(2.02, 60.46) = 4.52, p = 0.02, η

_{p}

^{2}= 0.13, BF

_{10}= 10.99], there are no significant differences in the Bonferroni pairwise comparisons between conditions regarding the percentage of trials solved correctly. As this shows that participants were generally able to solve the task across conditions, in the following we will focus on how the task was solved. We will take a closer look at trials that were solved integratively in Hypothesis 3a.

_{VT}(3, 90) = 1.66, p = 0.18, η

_{p}

^{2}= 0.05, BF

_{10}= 0.29] or in the time participants took for an entire trial [F

_{T}(3, 90) = 1.44, p = 0.24, η

_{p}

^{2}= 0.05, BF

_{10}= 0.22].

_{VT}(3,90) = 11.44, p < . 001, η

_{p}

^{2}= 0.28, BF

_{10}> 1000; F

_{T}(3,90) = 10.50, p < 0.001, η

_{p}

^{2}= 0.26, BF

_{10}> 1000]. Both viewing time (VT) and time (T) show significant results for the Bonferroni pairwise comparisons of Block 4 with each of the other three blocks. Accuracy even seems to drop slightly over time. However, the decrease in ACC is not a significant statistical result [F

_{ACC}(1.70, 51.03) = 2.01, p = 0.15, η

_{p}

^{2}= 0.06, BF

_{10}= 0.53]. None of the Bonferroni comparisons yields a p value below 0.05. When looking at the conditions independently, VT and ACC also show no significant change over time (all ps > 0.05, all BF

_{10}< 3). That is, none of the conditions show significantly differing results depending on the point in time they were presented throughout the experiment. As for the time participants needed to solve all four observations, the ANOVA yields a significant result for the condition in which atoms as well as observation locations remained visible [F

_{T.A&O}(3, 34) = 3.86, p = 0.02, η

_{p}

^{2}= 0.25, BF

_{10}= 4.44], indicating that participants solved trials faster when this condition was presented later in the experiment. Regarding all other conditions, ANOVAs show no meaningful results (all ps > 0.05, all BF

_{10}< 3).

Viewing time in ms | Time in ms | Accuracy in % | Trials solved integratively in % | |||||
---|---|---|---|---|---|---|---|---|

M | SD | M | SD | M | SD | M | SD | |

Block 1 | 12,060 | 3077 | 48,042 | 12,138 | 92.6 | 9.8 | 8.5 | 20.5 |

Block 2 | 13,014 | 3292 | 50,741 | 13,218 | 85.4 | 13.7 | 13.3 | 26.2 |

Block 3 | 12,804 | 3582 | 51,188 | 14,342 | 84.7 | 14.1 | 15.3 | 32.5 |

Block 4 | 10,143 | 2458 | 40,258 | 9850 | 83.7 | 24.0 | 16.7 | 34.5 |

### Gaze analysis

### Hypothesis 1: differences experienced in task difficulty

A&O | A | O | N | |
---|---|---|---|---|

A&O | 27 | 28 | 29 | |

A | 4 | 13 | 28 | |

O | 3 | 18 | 27 | |

N | 2 | 3 | 4 |

### Hypothesis 2: elements of the situation model

_{p}

^{2}= 0.74] and information type [F (1.19, 35.63) = 78.80, p < 0.001, η

_{p}

^{2}= 0.72]. The first main effect indicates that explanation location receive significantly more attention (higher proportion of fixation times) than observation location. The second main effect shows that participants looked more to current information location (because they are visible on the screen) than to previous information location and irrelevant spatial areas. That is, data support a step of comprehension and integration into a situation model of new information as assumed by TAR (Johnson & Krems, 2001; see also Klichowicz, Strehlau, Baumann, Krems, & Rosner, 2020). Even more important is that participants looked also more to previous information location (that no longer contained visible information) than to irrelevant spatial areas. This indicates that the information is still part of the mental representation. The differences between looks to current, previous, and irrelevant areas are also supported by post–hoc Bonferroni pairwise comparisons (all ps < 0.001).

_{p}

^{2}= 0.64, BF

_{10}> 1000] indicates that participants spent more time looking at the atom locations than at the observation locations regardless of what was visible on the screen. This favors the hypothesis that explanation locations are more important than previous observation locations in the process of abductive reasoning. This is further supported by the fact that there is no significant difference between the time participants spent looking at atom locations depending on whether atom locations are still visible or have to be retrieved. As the absence of significance does not provide statistical support, note that the BF

_{01}speaks in favor of the null hypothesis [Raftery 1995; F(1, 30) = 0.16, p = 0.70, η

_{p}

^{2}= 0.005, BF

_{01}= 6.82]. It follows that data show no statistical difference between memory indexing and spatial indexing referring to the atom location. Especially the value of the Bayes Factor BF

_{01}supports this suggestion.

_{p}

^{2}= 0.30, BF

_{10}= 44.15] supports the hypothesis that participants only look at observation locations when they are visible.

### Hypothesis 3: integrative solutions

_{p}

^{2}= 0.02, BF

_{01}= 12.07]. When all explanation and observation information stayed visible, 14% (SD = 30) of the test trials were solved integrativley. With 16% (SD = 30) when only atoms remained and 13% (SD = 30) when only observation locations stayed visible, all conditions produced more integrative solutions than the strictly memory based one (M

_{N}= 11%; SD = 25). Even though this result is in the expected direction, none of the Bonferroni pairwise comparisons between conditions are statistically meaningful.

_{p}

^{2}= 0.04, BF

_{01}= 4.13]. Participants needed more time for the last observation M

_{N}= 6.3 s (SD = 3.5) when nothing remained visible compared to when atoms and observations (M

_{A&O}= 5.0 s, SD = 1.9), atoms (M

_{A}= 5.5, SD = 2.3) or observation locations (M

_{O}= 5.2, SD = 2.2) remained visible. Even though this shows a trend in the right direction, none of the pairwise comparisons yields significance (all ps > 0.05).