100 Experiment E1: Length Assessment in Peripheral Vision
interaction between eccentricity and target line segment length prevails for DL (F(6; 66) = 2.86;p = 0.015), again confirmed by post-hoc comparisons of means using a Newman-Keuls test. In analogy to Figure 6.4, Figure 6.5 illustrates the relative length deviation DL as a function of eccentricity and target line segment length.
6.3 Discussion and Conclusions 101
influence on reaction times. The assessment of the lengths of the target line segments be-comes more and more difficult with increasing distance from the fixation point. Again, subjects obviously try to compensate for the increased assessment difficulty by prolonged reaction times in this Experiment E1. However, the following analyses of the assessment
“quality”, i.e. the length deviation DL(p) of the comparison line segment from the target line segment, rather questions the success of the “RT-compensation-strategy”: DL(p) in-creases for more peripheral line segment presentations, i.e. the length assessment quality deteriorates. As in Experiment E0, this might again be attributed to the required “men-tal translation”: Longer men“men-tal translation distances require more time, explaining the increase of RT when the lengths of line segments have to be assessed in more peripheral regions.
Before we turn to the discussion of the length deviation results, two more points have to be made with respect to the analysis of RT. First, the values of RT now show a pronounced difference between the eccentricity regions II and III. In contrast to the almost identical values measured in Experiment E0 for these two eccentricity regions, the observations here no longer suggest the existence of three designated eccentricity levels, but rather a continuous eccentricity band without clearly identifiable boundaries.
Second, the interaction effects of eccentricity and target line segment length on RT, al-though only tendential, encourage some speculation. We observed that RT yields relatively high values for long line segments – compared to RT measured for short and intermediate lengths – with only little divergence between the eccentricity levels I–IV. This can be un-derstood when recalling that the long line segments were chosen such that their lengths represent a “peripheral length” (due to the extent of the foveal eccentricity region I, line segments with a peripheral length could only be 7o long here (see Section 4.1)). Subjects obviously find the assessment of such length particularly difficult, even when the line seg-ments are presented in proximity to the fixation point, and take more time to accomplish this task. In summary, it could thus be that not only a far eccentric location of the stimu-lus, but also a line segment length that cannot be perceived foveally results in prolonged processing times. Furthermore, the relative stability of RT over the different eccentricities for assessing long line segments, i.e. the independence of RT from the eccentricity region, could indicate that, in this case, the (long) reaction times can mainly be attributed to the processing difficulties induced by peripheral length rather than by eccentricity.
One of the most striking findings of all the studies presented in this thesis is that the lengths of the target line segments are generally overestimated. Although a num-ber of recent studies arrive at similar results (see Section 2.2, e.g. Nakatani, 1995; Tsal &
Shalev, 1996; Prinzmetal & Wilson, 1998; Tsal, 1999), the majority of earlier investigations observed underestimation effects during size or length assessment in peripheral regions (see Section 2.2, e.g. Pearce & Taylor, 1962; Richards & Miller, 1971; Newsome, 1972;
Schneider, 1978; Thompson & Fowler, 1980). In Experiment E1, short and intermedi-ate line segments are overestimintermedi-ated in all but the foveal eccentricity region. Here, the assessment for intermediate line segments is highly accurate and shows a deviation of only -2%, i.e. a slight underestimation of the target line segment length. Such highly accurate assessment levels can also be found for long line segments, independent of the
102 Experiment E1: Length Assessment in Peripheral Vision
eccentricity region. The analysis of the “directional” length deviation DL in particular reveals that long line segments lengths are no less than 9% under- and no more than 5%
overestimated. This is significantly more accurate than is the case with the other two line segment lengths. Furthermore, these highly accurate assessments of long line segments can be maintained even in the more eccentric regions. A possible explanation for these observations could be the more salient image the longer segments yield on the retina, due to the excitation of more receptors as in the case of shorter line segments. Whereas for those, in particular for the short length in more peripheral eccentricity regions, only a
“blotch” with possibly no directional or length information might be perceived, the ori-entation and extent of longer line segments should still be visible, but blurred as well. For intermediate line segment lengths, a reasonable amount of length information seems to be available if not presented too far from the fixation point. For far peripheral locations then, the line segment shape becomes increasingly blurred and so does the length information, leading to greater uncertainty in the perceived size and thus less accurate assessments.
The fact that all assessments are independent of the orientation of the target line segment could be expected. As the comparison line segments were presented in the same orientations as the respective targets, possible interference from horizontal-vertical effects could be eliminated. These could have been expected if the comparison marker was pre-sented in a fixed, for example always horizontal orientation: Due to the horizontal-vertical illusion (see Section 2.2), the lengths of vertical target line segments should then have been overestimated. However, even with this factor being eliminated, it could have been that orientation affected the length assessment: When, for example, a vertical line segment has to be assessed that is located along the horizontal meridian, i.e. left or right of the fixation point, both its end points have more or less the same distance from the fixation point. When a horizontal line segment is now shown at the same meridial position, its end points have different distances from the fixation point, which might then complicate the length assessment. However, as the findings show, this is not the case. At least when, as in Experiment E1, the line segments lie entirely within one eccentricity region, the factor orientation does not affect the assessment of line segment lengths.
As already mentioned, the overestimation effect for most line segment lengths and the fact that this effect is more pronounced with increasing eccentricity could not always be observed in earlier experiments. With only few similar experimental findings, explanations for the causes of such overestimation effects are also sparse. Furthermore, the respective explanatory approaches put forward could only partially account for the experimental observations. They frequently failed to yield a detailed representation of how particular factors influence peripheral length assessment. Even the promising studies of Tsal (1999), Tsal and Bareket (1999) and Tsal and Bareket (2001) and their subsequently introduced concept of “attentional receptive fields” (ARFs), for example, can be considered an ap-propriate concept only for distinguishing between coarse unattended and fine attended perception (see Section 2.2). It is not clear in how far this concept is able to correctly ac-count for the quantitative aspects of the assessments, i.e. to correctly represent the ratios between the target and comparison dimensions, and how ARFs deal with complementary stimulus attributes, such as line segment orientation. However, it appears to be reasonable
6.3 Discussion and Conclusions 103
to share their view that there exists a significant effect of attention on peripheral stim-ulus assessment (see earlier studies by Butler (1980), Egyl and Homa (1984) and M¨uller and Rabbit (1989)), and that it is not a completely preattentive operation (e.g. Sagi &
Julesz, 1985).
The possible explanation proposed here for the overestimation effects reported in Ex-periment E1 is based mainly ontwo assumptions: First, we suggest that the line segment’s end points play an important role for the length assessment. Second, we assume that, for the peripheral length assessment, similar principles apply as in the peripheral location assessment task in Experiment E0. In detail, the explanation of the overestimation of line segment lengths in Experiment E1 is based on the finding of Experiment E0 that the actual position of the target marker is perceived as beingshifted towards the fixation point (underestimation of the radial target marker position, see Sections 5.2.1 and 5.3).
In contrast, very little divergence between the tangential position of the comparison and the target marker were noticed. Let us now transfer these observations to the current study, assuming that the principles of location assessment apply to the peripheral per-ception of line segment lengths as well. It then emerges that, when memorising a periph-erally perceived line segment, subjects lay off a mental representation (or mental model (Johnson-Laird, 1983)) of a line segment of approximately the original (target) length, but shifted towards the fixation point. This (mental) “shift” or dislocation of the line segment towards the observer leads to an elongation of the line segment when its mental model is recalled in the reconstruction phase of the experiment, i.e. when the length of the comparison line segment has to be adjusted. This might be due to the principle of size/length constancy which states that the size/length of objects seems to increase when they are moved towards the observer (cf. Section 2.2).
Although this approach intuitively appears plausible and well suitable for explaining the overestimation of line segment lengths, further support should be provided that the explanation holds and is based on sensible assumptions. The development of an according (computational) model appears to be promising in that respect. Such a model would enables us to test if and how the perceived length of peripherally presented line segments can be concluded as a result of the peripheral assessment of the location of its end points – as formulated in the (stimulus) decomposition hypothesis. This would then allow us to perform a quantitative analysis of the model data and compare the outcome with the empirical findings. In case of a positive correlation of these two data sets, not only support would have been presented for the correctness of the previous assumptions. The model would also suggest a possible explanatory approach that accounts for the major empirical observations. Finally, a model would have been implemented that adequately simulates the quantitative ratios of the assessment effects as well.
However, it would be desirable to develop a model that does not only account for the eccentricity effects on the assessment of line segmentlengths. Although no significant effect of the factororientation on the length assessment could be established in Experiment E1 – due to the specific experimental procedure – we consider this dimension another essential factor in the characterisation of line segments. Thus, Experiment E2 will be conducted in order to investigate how accurately subjects can assess the orientation of line segments
104 Experiment E1: Length Assessment in Peripheral Vision
that are again presented at different eccentric locations (Chapter 7). Subsequently, we should be able to discuss the development of a model under the premises that it ideally incorporates both line segment length and orientation assessment (Chapter 8).
Chapter 7
Experiment E2: Orientation
Assessment in Peripheral Vision
The last experiment in this series of three experiments investigates the perception of line segments orientation. The “reference” data obtained here allows for the testing of the ex-istence of correlations between the assessment error of peripherally perceived orientations of line segments and the mislocation of marker positions. In analogy to Experiment E1, if a correspondence can be found, this would indicate that observations in this experiment can possibly be attributed to mechanisms that were identified to influence location assess-ment. The peripheral assessment of line segment orientation might thus be accordingly decomposed and could be accomplished by peripherally assessing the locations of the end points of the line segment. The relative position of the end points to each other then yields the line segment orientation.
The following sections describe the experimental method for Experiment E2, based on the methodological preliminaries that were established in Chapters 3 and 4. Subsequently, the results will be presented and discussed which allows us to draw conclusions about the mechanisms that may govern peripheral orientation perception.