This package is designed to make dealing with eye-tracking data easier. It addresses tasks along the pipeline from raw data to analysis and visualization.
But the formula for the weights given there is
Weights = 1/(y + .5) + 1/(N - y + .5)
Which produces a distribution which is symmetrical around proportions of .5, which as I understand it fits with the rationale of weights, namely that the variance of the empirical log estimate is greater close to proportions of 0 or 1.
However, on line 55 of helpers.R, it looks like the formula is rather:
Weights = 1/(y + .5) / 1/(N - y + .5)
Which means that it's just the exp() of the Elog value. Again, if I'm understanding, that would mean that proportions closer to 1 get weighted more, which does not seem to be the point of the weights here.
So... unfortunate typo (substituting a / for a +), or have I misunderstood?
In familiarizing myself with the way that the Weights are supposed to work, I noticed that the reference points to this page: http://talklab.psy.gla.ac.uk/tvw/elogit-wt.html
But the formula for the weights given there is Weights = 1/(y + .5) + 1/(N - y + .5)
Which produces a distribution which is symmetrical around proportions of .5, which as I understand it fits with the rationale of weights, namely that the variance of the empirical log estimate is greater close to proportions of 0 or 1.
However, on line 55 of
helpers.R
, it looks like the formula is rather: Weights = 1/(y + .5) / 1/(N - y + .5)Which means that it's just the
exp()
of the Elog value. Again, if I'm understanding, that would mean that proportions closer to 1 get weighted more, which does not seem to be the point of the weights here.So... unfortunate typo (substituting a
/
for a+
), or have I misunderstood?Thanks!