Column headings
---------------
order A unique identifier for each participant
sub_no The number handed out by the server against which cond was set. Not guaranteed unique because a timed out sub_no might still get submitted after that sub_no has been remarked as free.
cond 0 = positive skew condition; 1 = other condition
trial The ordering of trials
id A unique identifier for each choice
p, x, q, and y For a choice of (p, x) vs. (q, y). p < q and x > y in the non-catch choices without dominance
or
t, x, y, and u For an intertemporal choice between x at time t or y at time u. x > y and t > u for non-catch trials
choice 1 = qy/uy chosen, 0 = px/tx chosen
Sample sizes
------------
We chose a number of subjects for an experiment, scheduled sessions, tested those who arrived, and only then looked at the data. We anticipated a large effect size for our first experiment, Experiment 1A, as range-frequency like effects tend to be quite robust. The effect size estimated from Experiment 1A (by using the random effects to give estimates of alpha for each participant) was Cohen's d = 1.5, which is indeed "large" by Cohen's labelling. A power calculation gives a sample size of 16 for a power of .8 or 26 for a power of .95. So we aimed for about 30 participants in each study. For Experiment 1C, which was the only on-line and non-incentivised study, we used a larger sample size.
> Cohen.d
1.501593
> pwr.t.test(d=Cohen.d, power=.8)
Two-sample t test power calculation
n = 8.045739
d = 1.501593
sig.level = 0.05
power = 0.8
alternative = two.sided
NOTE: n is number in *each* group
> pwr.t.test(d=Cohen.d, power=.95)
Two-sample t test power calculation
n = 12.5744
d = 1.501593
sig.level = 0.05
power = 0.95
alternative = two.sided
NOTE: n is number in *each* group