112 Experiment E2: Orientation Assessment in Peripheral Vision
eccentricity regions, RT shows very little variation. Significant changes occur only in the
“border regions”. It must be remembered, however, that the analyses of Experiment E1 did not support this hypothesis.
Apart from the eccentric region, RT is significantly influenced by the length of the target line segment in this experiment. In Experiment E1, the length of the target line seg-ments had no such effect on RT, but it subsequently emerged that the length of short(er) line segments could be less accurately assessed. If we conclude that these lengths must thus have been more difficult to judge, the lack of a corresponding effect on RT, i.e. longer RT for short(er) line segments, suggests that subjects did apparently not attempt to compen-sate for the increased complexity of assessing short line segment lengths in Experiment E1 through an increase in RT. This, however, is the case in Experiment E2. The orientation assessment of short line segments appears to be a far more complex task than that of intermediate or long line segments, in particular in the far peripheral regions. Subjects try to compensate for the higher orientation assessment complexity of short line segments through the increase in RT. In the gaze-contingent stimulus presentation condition this is done again within the limited (time) range of still acceptable concentration effort.
The subjects’ tendency – only marginally short of significance – to produce shorter reaction times when the target line segments are displayed in (near-) horizontal rather than in (near-) oblique or (near-) vertical orientations hints at a preference of horizontal over vertical information processing strategies. Indeed, the tendency towards the facilitatory effects of processing horizontal line segments as visible in RT is supported by the significant effect that target line orientation exerts on the orientation deviation DO. The fact that the orientation of horizontal line segments can be more accurately assessed than those of oblique and vertical line segments clearly favours horizontal processing. This is possibly due to the anatomy/physiology of the eye and the retina and, probably even more so, to the fact that most visual information processing requires a horizontally oriented strategy, such as reading – at least in Western cultures. These requirements are supposed to have influenced preferred visual strategies in many other tasks as well and thus, as a habituation effect, resulted in theimproved “visual performance” along this orientation. In eye-tracking studies at the University of Bielefeld this was shown, for example, in comparative visual search (e.g. Pomplun, 1998) and numerosity estimation tasks (Koesling, 1997; Koesling et al (submitted)).
Although it takes subjects longer to assess the orientation of the target line segments in more peripheral regions, the extra time does not help to improve the assessment ac-curacy in general. In some cases, however, there appears to be such a facilitatory effect;
for example for short line segments: While RT is about equal for short and intermediate line segments in the eccentricity regions I and II, the orientation deviation DO is signif-icantly larger for short line segments. However, with the sharp increase of RT for short line segments that are presented in the eccentricity regions III and IV, no such difference in DO between short and intermediate lengths is visible in the eccentricity regions III and IV any more. These observations might indeed suggest that for specific lengths an increased reaction time helps to more accurately assess line segment orientation. Never-theless, subjects do not succeed in achieving the same assessment accuracy level for short
7.3 Discussion and Conclusions 113
(and intermediate) length as for long line segments. The accuracy differences regarding the orientation assessment between these two types of line segment length, i.e. short(er) and long, remain about the same, independent of the eccentricity region and unaffected by RT differences.
As before, the decreasing accuracy of the orientation assessments must probably be attributed to a more and more “blotch”-like retinal image the further away the target stimulus is displayed from the fixation point. Such a blurred representation is then laid off as an equally blurred mental image or model. This obviously renders the accurate assessment of line segment lengths difficult as the end points of line segments and thus the line segment’s extent cannot exactly be determined (see Experiment E1, Chapter 6).
Furthermore, the orientation of a line segment can only vaguely be assessed when only a rather diffuse image of the stimulus is perceived in the periphery. It could even be expected that, at some stage, either in very far eccentric locations or for very short line segments, all directional information is lost. The only information available to the visual system then would be the existence of some object in the periphery, but none of its features such as its orientation and length relevant here.
Let us now – at least partly – consider again the suggestions made at the end of the previous chapter. It was proposed that the end points of a line segment play an important role for its length assessment (Experiment E1) and that their mislocation (Ex-periment E0) towards the observer leads to the overestimation of line length. With regard to the assessment of the orientation of line segments, possibly similar processing principles apply.
Indeed, it appears to be a promising approach to think of orientation assessment as a process determined by the perceived locations of the line segments’ end points as well.
In detail, the computation of the relative spatial positions of the two end points to each other should yield the target orientation information. This could be laid off as a mental representation and subsequently be recalled during the adjustment of the orientation of the comparison line segment. If we take into account the findings for the location assessment from Experiment E0, in particular considering the eccentricity effects on the location accuracy along the radial and tangential axes, these could also be reflected in and probably account for some of the results of Experiment E2. For example, due to the greater variance along the radial rather than along the tangential axis in the location assessment task of Experiment E0, a considerable variation in the orientation assessment must be expected – and could indeed be observed in Experiment E2. The same is true for the increased uncertainty in the orientation assessment in more eccentric regions, which appears plausible when we take into account the increasing location uncertainty – mainly in the radial direction – with increasing eccentricity in Experiment E0. Thus, this approach again seems plausible and intuitively suggests to account for the orientation deviations observed in Experiment E2. However, if it really does, this must be validated.
Due to the close resemblance of the suggested explanatory approaches for the pe-ripheral assessment of the length and orientation of line segments, it now appears to be favourable to develop a common model. This should enable us to test if and how both the perceived length and orientation of peripherally viewed line segments can be concluded as
114 Experiment E2: Orientation Assessment in Peripheral Vision
a result of the peripheralassessment of the location of its end points. As suggested in the previous chapter, an appropriate model should not only present support for the adequacy of the assumptions made – thedecomposition hypothesis – and suggest an explanatory ap-proach that may account for the empirical observations. Furthermore, it should adequately simulate the quantitative ratios of the assessment effects as well. The attempts made to design, implement and validate such a model are presented in the following chapter.
Chapter 8
Modelling Eccentricity Effects
8.1 Why Modelling?
A close inspection of the psychological literature reveals that, in general, the majority of studies follows the same strategy to arrive at its conclusions. After the motivation of a study, the review of related works and the choice of the experimental conditions and setup, the actual experiment is conducted. Next, the collected empirical data is subjected to qualitative and quantitative, mainly statistical analyses. Characteristic values such as means or standard deviations are computed and statistical analyses are performed in order to relate the experimental observations – the dependent variables – to the systematically varied parameters – the independent variables (e.g. Sichelschmidt & Carbone, 2003).
All these steps are rather straightforward and present a more or less standardised procedure. The exciting, but often troublesome part of work only starts here. The crucial questions that scientific studies should attempt to find an appropriate answer for are:
Why did we get the results just as they are, what do the observations tell us and how can the findings be interpreted in the experimental and, preferably, in a more general context? One of the common problems of the discussions of experimental results is that conclusions drawn from the empirical data are not entirely conclusive. Many conclusions apparently rely on vague interpretations of the observations and sometimes incorporate a considerable amount of speculation.
In an attempt to provide more support their interpretations and conclusions, some authors propose models, formalised descriptions of their reasoning on the basis of as-sumptions the experimental data seem to suggest. Unfortunately, most of these models only unspecifically describe the (proposed) theory and rarely contain clear suggestions with respect to a concrete implementation of the model. Very few authors present models that can be parameterised so that an algorithmic implementation realises the reproduc-tion of empirical data. However, only such a simulation enables us to compare empirical and simulation data in order to test the correctness of the model. In return, this then al-lows for testing the validity of the initial premises and the suggested interpretation of the empirical data that led to the generation of the model. This closed loop of empirical data acquisition — interpretation — modelling — verification represents a promising strategy
116 Modelling Eccentricity Effects
for making reliable statements on the processes underlying specific (human) performance.
Exactly these are the premises for the development of the present model.