DRAFT

Linares Holcombe & White (2009) and Goodbourn & Holcombe (2015) independently suggest that binding an exogenous transient with another visual stimulus is distributed ~ N(u=x,σ=75ms).

Variability is stable

X is delayed a bit on average, causing flash-lag and on average reporting of a delayed letter. Interestingly in the case of the flash-lag, σ varies less across participants than does μ. For RSVP binding with a single stream, σ varies about the same amount as μ, even though measurement of σ is less stable. Specifically, across 19 participants measured separately for left and right visual field, std dev of latency averaged 27 ms and precision 26.5 ms.

More common than getting a whole distribution is measuring a frequency threshold. Linares, Holcombe, & White (2009) checked that the distribution at low frequency predicted the threshold. In the paper we wrote that it checked out.

In our literature review, we * consider whether others’ RSVP data is consistent with σ ~= 70 * consider whether a common value of σ explains frequency thresholds found for other tasks.

If color is distributed with 70 ms

Color-motion binding

Holcombe & colleagues documented a rather general 3 Hz limit within vision whenever two features must be labeled (Clifford, Holcombe, & Pearson; ). The exceptions are few (Holcombe 2009). Fujisaki & Nishida found evidence that the 3 Hz limit extends cross-modally whenever one must identify two features.

If the 3 Hz limit with chance=0.5 were caused by a single feature being normally-distributed, would indicate a σ of 125 ms. This is shown in the plot below. (If chance=0, then σ of 70 ms.)

An interesting departure from human performance is the very gradual decline of the prediction. The slope of human performance is much steeper than this. DANI is this related to your unpublished modeling of the position distributions?

For that reason, we agree with Fujisaki & Nishida that the cause of 3 Hz limits is not as simple as Gaussian temporal uncertainty.

These limits are probably qualitatively different from the ~70ms limit on binding a feature with a transient. Likely limited by time for identifying and consolidating a label. In their mutual presence, color and motion changes do not jump out and do not seem to call attention, indeed can result in change blindness (Saiki & Holcombe; Suchow & Alvarez). Motion changes particularly hard to detect (Werkhoven et al. 1992). /—Predicts that other gradual changes like dynamic orientation will also be bound wrongly. Could have mean color changing gradually from red to green. Your task is to judge the mean orientation when the mean motion direction is leftward. Now, even color transient should not help and thus binding should be even slower. Both features should show middle-binding rather than going with beginning.
Actually, I don’t even know the temporal resolution of statistical perception like size. So if everything is always changing in size, at what rate can I detect that the mean is actually changing? –/

range = σ * 3.464

If σ = 74 ms, range = 256.3435195

1/12*38*38
## [1] 120.3333

Symmetry of distributions in various tasks

I could write a paper documenting the symmetry of the distributions in both the work with Goodbourn and the work with Linares. Also, look out for oscillations in Holcombe & Kristjansson.