The dependence of perceived speed upon signal intensity

In a recent paper, Edwards and Grainger (2006) manipulated the angle between transparent dot planes leads to increases in the coherence of random dot patterns and found that a reduction in coherence led to an increase in perceived speed; they took this to indicate that vector averaging is not employed in global speed calculations (Edwards & Grainger, 2006). We would like to take this opportunity to comment on the generality of their findings. As Edwards and Grainger note, the relationship between coherence and perceived speed has previously been studied (Zanker & Braddick, 1999). The latter found no effect of coherence on perceived speed. However, Zanker and Braddick used random re-plotting of noise dots, whereas, Edwards and Grainger used a random walk stimulus (Scase, Braddick, & Raymond, 1996). In the case of the Zanker and Braddick study the result is unsurprising, subjects were simply effective at separating signal from noise. However, in the light of the following studies, the results of Edwards and Grainger study are surprising. As part of a conference presentation we presented data which robustly indicated a reduction of perceived speed with a reduction of coherence (Benton & Curran, 2004). These data, and the method of their collection, are presented here as Supplementary Material. Our results showed a strong linear relationship between coherence and speed such as would be predicted if vector averaging formed a component of our speed judgements in patterns of this type. Additionally, a similar result has also previously been reported (Freeman & Sumnall, 2002)—as can be seen from Fig. 2 in their paper, these authors again found a linear decrease in perceived speed with decreases in coherence. The Benton and Curran (B&C), Freeman and Sumnall (F&S) and Edwards and Grainger (E&G) studies all used similar types of RDK— random walk with allocation to noise or signal for each dot on each frame. The RDK parameters that one might think critical for motion integration are dot speed, dot density and update rate. The E&G findings (and their subsequent conclusions) are based on a single point within this parameter space (6.7 deg/s, 1.3 dots/deg, 20 Hz). The other studies looked at a wider range of dot speeds (B&C: 1, 2, 4, 8 and 16 deg/s; F&S: 2.83, 5.66 and 11.31 deg/s), used markedly different dot densities from one another (B&C 63 dots/ deg; F&S 1.5 dots/deg) and employed slightly different update rates (B&C: 80 Hz; F&S 100 Hz). Given their findings, it is reasonable to conclude that the Edwards and Grainger result is not the general result. In contrast, it appears that reductions of coherence generally lead to linear reductions in perceived speed such as would be found if vector averaging formed an intrinsic part of our speed calculations (Watamaniuk & Duchon, 1992). E&G explain their result by proposing that perceived speed is influenced by the relative motion in the stimulus. They note that relative speed has previously been shown to influence speed judgements in dot displays containing opposite motions organised either to produce transparency or kinetic boundaries (De Bruyn & Orban, 1999). E&G extend this finding by showing that increasing