I'm keeping it in two files, content_series_trials: more values of
direction content and fewer values of spacing were used in each
session
spacing_series_trials many values of spacing were used but only one
value of content
Subjects seem to adapt to the strength of direction content, so you will see a bigger effect of the magnitude of direction content with the first dataset than the second.
The columns are:
eccentricity the eccentricity of the spots
abs_direction_content, The strength of the carrier motion, ranges
from -1 (counterclockwise) to 1 (clockwise).
abs_displacement, the spatial step size between each motion
pulse. This variable was controlled by a staircase procedure.
abs_response_cw TRUE if the subject responded clockwise.
folded_direction_content, folded_displacement
folded_response_with_carrier same data folded over so that carrier
direction content is always positive. The response variable is TRUE if
it agrees with the sign of the carrier.
target_spacing The distance between targets along the circumference
of circle.
target_number The number of targets. target_spacing =
2pieccentricity/target_number
subject Subject initials.
Another note, this only includes trials where the subject gave a response within a certain window from stimulus onset. They had feedback on whether their response latency were in the window. Some subjects' responses varied with the response latency, which isn't included in this data.
These are fits for the same data in the *_trials.csv files.
Here's a couple of CSV files. There are two, spacing_series and
contrast_series. The difference is that in spacing_series the
directional content was held constant for each entire session and the
spacing varied, while in contrast_series I used only 2 values of
spacing and four values of directional content.
The two situations are not directly comparable because subjects appear to adapt to the average amount of directional content used in any session. So varying directional content within a session is much more effective than varying it across sessions, which is why I've separated out the data.
subject, folded_direction_content, and target_spacing should be
self explanatory.
The psychometric functions are fit by logistic CDFs with a 5% allowance for lapses in either direction.
bias is just the intercept coefficient you get from logistic
regression; that is, it measures (in log-odds) how often subjects
respond clockwise to a stimulus with clockwise carrier and no
envelope motion. bias_sem gives you a standard error of the bias
measurement.
sensitivity is the slope parameter for the logistic CDF and
sensitivity_sem its standard error.
pse is the envelope-displacement-per-step at which the subject
equivocates between directions. Since steps have intervals of 100 ms,
multiply by 10 to get envelope speed in degrees/sec.
pse_25 and pse_75 gives a confidence intervals on the PSE
measurement (these are calculated by parametric bootstrap; I draw
random psychometric functions using the fitted likelihood function,
and find the 25% and 75% quantiles of the PSE)
threshold measures how much envelope motion it takes to move between
50% and 75% on the psychometric function. threshold_25 and
threshold_75 give a confidence interval.
Note that when sensitivity is low enough that you aren't entirely sure if sensitivity is positive, the confidence intervals for PSE and threshold will blow up (It's basically the problem of estimating the x-intercept of a linear regression)