wjhopper / bayes_bars Goto Github PK

A user has reported when the "top" bar representing the conditional probability has focus, clicking the "bottom" bar representing the marginal probability doesn't consistently change the focus onto the "bottom" bar. I cannot reproduce this issue on my computers or Starn's lab computers (all running Windows 7) .

At the time of writing (latest commit 59ffdc6) the code governing changing focus from rects is here:

If this issue can be reproduced, this code should be revisited.

A user has proposed adding an extra visual guide for bar width, in the form of a dashed line connecting the right edge of the bar to the x axis that would appear when adjusting the bar and disappear when the adjustment is finished.

There are several reasons why this feature may be helpful. First, it would help subjects precisely match their bar widths to the proportions given in the problem. Second, if participants knew the precise numeric widths of the bars representing the conditional probabilities p(B|A) and p(B|Ā), this could aid them in estimating the ratio of their lengths (i.e., they could use numeric estimation, rather than pure visual perception).

There is one reason not to like this idea though, which is it's potential to encourage matching the width of the bar to the conditional probability given in the problem. For example, if a problem says "20% of people eat cookies, and 10% of people who eat cookies eat brownies", then the correct width of the p(brownie | cookie) bar is .2 ✕ .1 = .02. However, having the visual guide connecting the bar to the axis might encourage matching the width of the bar to the value given in the problem (10%), rather than making the width a proportion of the bar representing the marginal probability. This concern is speculative, but I think is important to consider in light of the anticipated difficulty participants will have in understanding the conditional probability bars should be adjusted to be proportions of the marginal probability bars.