Street Address & Other Stuff

[timg]

Just an FYI, you guys can now enter the street address for a course in the course info if you know it. I spent the last 2 or 3 hours updating the ones that had that info in the directions so at least 600 courses have that info.

Also, I re-vamped the review tab a little yesterday adding some new filters and a slightly new layout where the Sort box used to be.

[jedwards]

Cool -- thanks :clap:

[howardb]

:clap::clap:what would we do without Tim.

LETS HERE IT FOR TIM!!! :clap::clap:

Another upgrade in the books...atta boy!!

[Dave242]

I noticed those filters yesterday. Love it. Great stuff - Thanks!!

[mashnut]

I love that you can see the breakdowns of the individual ratings numbers for all reviews and for tr reviews side by side, that's awesome. Thanks Tim!

[mashnut]

On that note, I noticed that after you apply a filter, the total number of reviews at the top of the page changes to the number of reviews in the filtered list. The number of reviews and exact rating in the mouseover text is still right though. Example: on Flip City if you filter for TR reviews, it says that there are 8 total reviews and 8 tr reviews, but when you mouse over the total rating it shows 71 reviews.

[timg]

I'll have to fix that.

[Lewis]

Tim that's outstanding! My favorite aspect of this is that you can sort by "recently updated." That way when someone updates their review, it will rise back to the top.

[Lewis]

Quick question though. When you sort by "most helpful," does it calculate the % of helpful marks, or the difference of helpful minus unhelpful? If it's by percent, then you'd see a review with 1:0 helpful:unhelpful (100%) higher than a review with 24:1 (96%). Is that the way you want to sort it?

[timg]

For the reviews I use something called a wilson score interval (http://en.wikipedia.org/wiki/Binomial_proportion_confidence_interval#Wilson_sco re_interval).

[jkdisc]

awesome!

[Lewis]

For the reviews I use something called a wilson score interval (http://en.wikipedia.org/wiki/Binomial_proportion_confidence_interval#Wilson_sco re_interval).

Holy statistics, batman! Can you translate what that accomplishes for you into layman's terms, or is there no word in English for all the Wilson score interval's extra savoir faire?

[timg]

Basically I weight each vote by how many helpfuls or unhelpfuls the review already has and then come up with a single number (the wilson score) that I can use to sort the reviews.

[Lewis]

So 1 helpful to 0 unhelpfuls still beats 24 helpfuls to 1 unhelpful?

[timg]

Here is a good explanation I found (which I modified to fit DGCR).


Let's say the course only had 3 reviews, +10 -2, +9 -1, and +8 -0 on it. and the +8 -0 was posted first, the +10 -2 second, and the +9 -1 third.

the helpful-unhelpful scoring would sort them as all the same (all +8). and the only real way to break a tie in scoring is by newest-posted-first. so that'd result in an order of:

+8 -0
+10 -2
+9 -1


But that doesn't really make sense. That -1 on the +9 -1 could be a slipped mouse click, or a revenge unhelpful vote that doesn't actually reflect the review *itself*.

With the +10 -2, it's less clear that the -2 is a mistake. we have more corroborating evidence this is about the review, not the reviewer or some mistake. But still not as much evidence as the +10.

So I think the proper sort order here is:

+10 -2 (score: 7.22 - 0.47 = 6.75)
+9 -1 (score: 6.31 - 0.21 = 6.10)
+8 -0 (score: 5.40 - 0 = 5.40)


Now, note this is just a simple example, with three reviews in it. Imagine doing this for all reviews simultaneously. That's what the math brings - a mathematical tool that allows us to compare otherwise incomparable examples relatively rationally and dispassionately.

eg: which is higher, +26 -9 or +35 -18? both are +17 on the helpful-unhelpful scale. It's totally not clear to me, but the formula says the latter, and given the logic presented behind the math, I'm willing to buy that.