Can you beat 5.47482014?

December 28, 2009 9:59 AM

What thread has the highest T number, where T is 'number of favorites given to comments' divided by 'number of comments'?

That Sexy Geeks thread got me thinking about it. There's a comment with 53 favorites right out of the gate, a bunch of big ones and a 100 something a bit later on. It seemed like a lot of favorites getting thrown around.

I clumsily crunched the numbers on threads on the front page with more than 100 comments (an arbitrary limit, but since I'm doing this sort of manually I wanted SOME threshold).

Sexy Geek: 139 comments, 761 favorites, 5.47482014
Dragon Sex: 149 comments, 489 favorites, 3.28187919
I am Tiger Woods: 157 comments, 412 favorites, or 2.62420382
Twitter: 113 comments, 221 favorites, or 1.95575221
Dr Who: 111 comments, 137 favorites, or 1.23423423

Nothing serious,and I'm not trying to start any sort of favorites war, this was just an "having a hard time working on work while everyone else is still on vacation" thing. The thematic similarities among those first two (three?) threads do crack me up though. I wonder if that sort of tendency would stand up to a larger study by someone who can actually use the infodump.

Oh, the T is for tehloki.

posted by dirtdirt to MetaFilter-Related at 9:59 AM (64 comments total)

It'd be cool to factor in unique favorites vs. overall favorites. The former tells you something about the level of activity/quality of discourse in the thread itself, while the latter tells you something about possible outliers (highly favorited items) that add to the overall assessment.
posted by iamkimiam at 10:04 AM on December 28, 2009

I'm kind of surprised by what comments get 50+ these days. In my rose-colored memories, 50+ means something like a unique personal story, not some iffy joke near the start of a long thread.

I wonder if it's due to people hiding favorite counts.
posted by smackfu at 10:05 AM on December 28, 2009 [1 favorite]

smackfu: "I wonder if it's due to people hiding favorite counts."

The favorites number crunching thread made it pretty clear that favoriting has been growing in a pretty consistent manner for at least the past year. In Jan/Feb 2009, there was about 1.3 faves per comment on average. By October, that was 1.8. I'd bet if you looked at the beginning of 2008, it was well below 1.0.
posted by Plutor at 10:15 AM on December 28, 2009

Yeah, but comments with over 100 favorites tend to be interesting personal stories. Could it be that our community weblog is getting bigger?
posted by iamkimiam at 10:19 AM on December 28, 2009 [3 favorites]

I'm kind of surprised by what comments get 50+ these days. In my rose-colored memories, 50+ means something like a unique personal story, not some iffy joke near the start of a long thread.

I wonder if it's due to people hiding favorite counts.

That's only been happening for less than 60 days. Not a statistically significant amount of time to determine rosy pigment infiltration.
posted by oneirodynia at 10:25 AM on December 28, 2009

Excellent point Iamkimiam. An interseting forinstance is Wall-E (with Astro Zombie's 863 favorite behemoth), which has a line of: 252 comments, 1283 favorites, or 5.09126984. Still lower than the Sexy Geek thread, but without that one gargantuan outlier is only a 1.66666667
posted by dirtdirt at 10:27 AM on December 28, 2009

Also, having a couple heavily favorited comments nested in a big thread actually changes favoriting behavior somewhat. We don't and can't know whether or not the Wall-E thread would have a number higher or lower than 1.67 if Astro Zombie's 863 favorited-comment weren't there, both drawing attention to the thread overall, but possibly drawing attention away from others' stories/comments in it. People only have so much time to read, and we're constantly evaluating what's worth our time to do so.
posted by iamkimiam at 10:35 AM on December 28, 2009 [1 favorite]

Also, compare audiences. The sexy geek thread is very likely attracting a lot of sexy (imho) geeks. Sexy geeks and favoriting activity are probably positively correlated behaviors (and why I love you so).

The audience for the Wall-E thread is probably much wider, includes a bigger sample of users with a possibly more varied set of behaviors. Also, a geek thread is semantically much more geared to the hearts of the participants in it (hey, these are my peeps we're talking about here!) versus Wall-E, which is a movie about robots - people might be less invested.
posted by iamkimiam at 10:40 AM on December 28, 2009

The "Neighborhood of Make-Believe" thread has 131 comments, and 981 comment favorites* for a score of 7.4885. (2.7862 without pastabagel's comment)

What do I win?

* Not counting the favorites it has accumulated since linking there just now
posted by sambosambo at 10:51 AM on December 28, 2009 [1 favorite]

I think this thread from AskMe may be a candidate.
posted by Gyan at 10:53 AM on December 28, 2009

And as of my previous comment, there were 616 favorites on pastabagel's comment.
posted by sambosambo at 10:54 AM on December 28, 2009

Yeah, but comments with over 100 favorites tend to be interesting personal stories. Could it be that our community weblog is getting bigger?

I think there's some confirmation bias at work, here. certainly some interesting personal stories get plenty of favorites, but I'd say we get a lot fewer of those types of comments than you imagine. I'd like to see how many 100+ comments are one-liner snark, to be honest, since that's the type of thing I notice getting favorited more than anything else.
posted by shmegegge at 11:01 AM on December 28, 2009

thread 71507 (DELETED) comments=1 favorites=22 t-number=22
thread 86695 comments=99 favorites=825 t-number=8.33333333333333
thread 82929 comments=46 favorites=368 t-number=8
thread 74818 comments=53 favorites=422 t-number=7.9622641509434
thread 73480 comments=72 favorites=559 t-number=7.76388888888889
thread 85667 comments=845 favorites=6293 t-number=7.44733727810651
thread 61699 comments=131 favorites=964 t-number=7.3587786259542
thread 76440 comments=133 favorites=968 t-number=7.2781954887218
thread 87016 comments=87 favorites=630 t-number=7.24137931034483
thread 86424 comments=169 favorites=1203 t-number=7.11834319526627
thread 79749 comments=58 favorites=408 t-number=7.03448275862069
thread 83821 comments=74 favorites=486 t-number=6.56756756756757
thread 70365 comments=129 favorites=828 t-number=6.41860465116279
thread 77114 comments=67 favorites=427 t-number=6.37313432835821
thread 82525 comments=94 favorites=593 t-number=6.30851063829787
thread 77735 comments=21 favorites=131 t-number=6.23809523809524
thread 85384 comments=85 favorites=527 t-number=6.2
thread 84093 comments=347 favorites=2148 t-number=6.19020172910663
thread 83597 comments=201 favorites=1231 t-number=6.12437810945274
thread 85961 comments=186 favorites=1115 t-number=5.99462365591398
thread 87172 comments=160 favorites=948 t-number=5.925
thread 84993 comments=696 favorites=4056 t-number=5.82758620689655
posted by Rhomboid at 11:03 AM on December 28, 2009 [11 favorites]

FYI, that was from this sunday's dump and only covers MeFi (not AskMe or MeTa.)
posted by Rhomboid at 11:05 AM on December 28, 2009

Gyan, he single best intro book thread has a line of 237 comments, 356 favorites for a T number of 1.5021097.

Of course, there are those 1281 favorites on the post itself. But they don't count in this metric.
posted by dirtdirt at 11:08 AM on December 28, 2009

this comment wins the internet.
posted by shmegegge at 11:10 AM on December 28, 2009

This is an awesome comment and you should favorite it.
posted by ob at 11:11 AM on December 28, 2009

FYI, that was from this sunday's dump

*giggle*
posted by sambosambo at 11:13 AM on December 28, 2009 [4 favorites]

If this were the future, then the 53 line perl program I wrote to give the above output could be condensed down into, "Computer, list me some threads that have comments of particularly heartwarming or endearing personal revelation."
posted by Rhomboid at 11:15 AM on December 28, 2009 [1 favorite]

Rhomboid, how brutal would it be to ask you to gin up another line of perl that tells what the MOST heartwarmingest comment in each of those threads is? In other words, how many of those have a favorites economy being driven by a single comment, a-la Wall-E and AZ's ladyfriend?
posted by dirtdirt at 11:25 AM on December 28, 2009

Wow. Most of those threads in rhomboid's list have one giant comment, but the "Hi. Whatcha Reading?" thread has 7 comments with over 100 favorites.
posted by sambosambo at 11:28 AM on December 28, 2009

The subjects of the top five threads from Rhomboid's comment in ascending order (excluding the deleted thread):

Rape
Cenocide
Schizophrenia
Alcoholism
Fonts

Find the one which doesn't belong.
posted by Kattullus at 11:43 AM on December 28, 2009 [4 favorites]

I ran the numbers, and the number of comments getting x favorites, for any value of x, has definitely grown over the years. Like there are more > 100 favorite comments in the Blue in 2009 (165) than in all the years before 2009 combined (138). Same with > 50 favorite comments (539 vs. 839). Here's a graph

So it seems like it is true that 50 favorites used to be more "special" than it is now.
posted by smackfu at 11:53 AM on December 28, 2009

My numbers are a bit different from Rhomboids. First of all, I used the 100-comments in a thread threshold suggested by the OP. Also, these are run on a slightly older comment dump (two or three weeks old)

1 85667 7.37988165680473
2 61699 7.34351145038168
3 76440 7.2781954887218
4 86424 7.1301775147929
5 70365 6.41860465116279
6 84093 6.19020172910663
7 83597 6.12437810945274
8 85961 5.99462365591398
9 87172 5.8875
10 84993 5.82758620689655
11 79486 5.61313868613139
12 78669 5.55865921787709
13 70997 5.52352941176471
14 87219 5.48507462686567
15 81263 5.46491228070175
16 83991 5.40944881889764
17 82958 5.37795275590551
18 86900 5.32352941176471
19 84268 5.26571428571429
20 72632 5.20858895705521
21 72958 5.1031746031746
22 81047 5.08695652173913
23 87071 5.03658536585366
24 85228 5.02678571428571
25 85692 5.00992907801418
posted by chrisamiller at 12:07 PM on December 28, 2009 [1 favorite]

smackfu - can you scale that graph by the number of registered users on mefi? (or by the active users, as described by FishBike's stuff in the previous infodump thread?)
posted by chrisamiller at 12:12 PM on December 28, 2009

Well yeah, if you only include posts with >100 comments that's really going to change things, as you can see that more than half in my list have less than 100 comments. I don't think OP was actually suggesting that criterium for infodump analysis but rather as a quick way to keep the amount of manual labor manageable. If the computer's doing it though I don't see how that's relevant.

Here's my list again but this time with a listing of the 50th percentile of highly favorited comments for each post. (I.e. the list of comments after each post is displayed in descending order of number of favorites until more than half of the the total number of favorites for that thread has been accounted for.) It contrasts the posts with a single megacomment vs the longer threads with multiple not-quite-so-highly favorited comments.

thread 71507 (DELETED) comments=1 favorites=22 t-number=22.000 top 1 comments: [22 favs]
thread 86695 comments=99 favorites=825 t-number=8.333 top 1 comments: [493 favs]
thread 82929 comments=46 favorites=368 t-number=8.000 top 1 comments: [277 favs]
thread 74818 comments=53 favorites=422 t-number=7.962 top 1 comments: [270 favs]
thread 73480 comments=72 favorites=559 t-number=7.764 top 1 comments: [351 favs]
thread 85667 comments=845 favorites=6293 t-number=7.447 top 12 comments: [573 favs], [193 favs], [163 favs], [135 favs], [117 favs], [103 favs], [101 favs], [89 favs], [74 favs], [67 favs], [60 favs], [59 favs]
thread 61699 comments=131 favorites=964 t-number=7.359 top 1 comments: [616 favs]
thread 76440 comments=133 favorites=968 t-number=7.278 top 1 comments: [648 favs]
thread 87016 comments=87 favorites=630 t-number=7.241 top 3 comments: [234 favs], [68 favs], [35 favs]
thread 86424 comments=169 favorites=1203 t-number=7.118 top 12 comments: [66 favs], [62 favs], [46 favs], [37 favs], [37 favs], [37 favs], [36 favs], [33 favs], [31 favs], [31 favs], [30 favs], [29 favs]
thread 79749 comments=58 favorites=408 t-number=7.034 top 3 comments: [112 favs], [76 favs], [32 favs]
thread 83821 comments=74 favorites=486 t-number=6.568 top 2 comments: [128 favs], [122 favs]
thread 70365 comments=129 favorites=828 t-number=6.419 top 1 comments: [596 favs]
thread 77114 comments=67 favorites=427 t-number=6.373 top 1 comments: [266 favs]
thread 82525 comments=94 favorites=593 t-number=6.309 top 3 comments: [163 favs], [91 favs], [59 favs]
thread 77735 comments=21 favorites=131 t-number=6.238 top 1 comments: [125 favs]
thread 85384 comments=85 favorites=527 t-number=6.200 top 3 comments: [126 favs], [85 favs], [83 favs]
thread 84093 comments=347 favorites=2148 t-number=6.190 top 12 comments: [197 favs], [191 favs], [118 favs], [110 favs], [91 favs], [85 favs], [62 favs], [53 favs], [52 favs], [50 favs], [46 favs], [44 favs]
thread 83597 comments=201 favorites=1231 t-number=6.124 top 3 comments: [517 favs], [71 favs], [49 favs]
thread 85961 comments=186 favorites=1115 t-number=5.995 top 12 comments: [128 favs], [65 favs], [60 favs], [49 favs], [46 favs], [35 favs], [34 favs], [34 favs], [27 favs], [27 favs], [27 favs], [26 favs]
thread 87172 comments=160 favorites=948 t-number=5.925 top 12 comments: [58 favs], [56 favs], [46 favs], [43 favs], [39 favs], [38 favs], [34 favs], [33 favs], [32 favs], [31 favs], [30 favs], [26 favs]
thread 84993 comments=696 favorites=4056 t-number=5.828 top 12 comments: [280 favs], [155 favs], [124 favs], [120 favs], [103 favs], [95 favs], [70 favs], [66 favs], [65 favs], [64 favs], [63 favs], [58 favs]
posted by Rhomboid at 12:19 PM on December 28, 2009 [7 favorites]

Kattullus: Rape
Cenocide
Schizophrenia
Alcoholism
Fonts

Find the one which doesn't belong.

Is it 'cenocide'?
posted by shakespeherian at 12:22 PM on December 28, 2009 [1 favorite]

dirtdirt: how many of those have a favorites economy being driven by a single comment, a-la Wall-E and AZ's ladyfriend?

1 85667 (7.37988165680473) - top comment: 571 (9.16%)
2 61699 (7.34351145038168) - top comment: 615 (63.93%)
3 76440 (7.2781954887218) - top comment: 648 (66.94%)
4 86424 (7.1301775147929) - top comment: 67 (5.56%)
5 70365 (6.41860465116279) - top comment: 596 (71.98%)
6 84093 (6.19020172910663) - top comment: 197 (9.17%)
7 83597 (6.12437810945274) - top comment: 517 (42.0%)
8 85961 (5.99462365591398) - top comment: 128 (11.48%)
9 87172 (5.8875) - top comment: 57 (6.05%)
10 84993 (5.82758620689655) - top comment: 280 (6.9%)
11 79486 (5.61313868613139) - top comment: 184 (11.96%)
12 78669 (5.55865921787709) - top comment: 120 (12.06%)
13 70997 (5.52352941176471) - top comment: 76 (8.09%)
14 87219 (5.48507462686567) - top comment: 79 (10.75%)
15 81263 (5.46491228070175) - top comment: 92 (14.77%)
16 83991 (5.40944881889764) - top comment: 167 (24.31%)
17 82958 (5.37795275590551) - top comment: 130 (19.03%)
18 86900 (5.32352941176471) - top comment: 99 (13.67%)
19 84268 (5.26571428571429) - top comment: 295 (16.01%)
20 72632 (5.20858895705521) - top comment: 525 (61.84%)
21 72958 (5.1031746031746) - top comment: 863 (67.11%)
22 81047 (5.08695652173913) - top comment: 189 (14.69%)
23 87071 (5.03658536585366) - top comment: 92 (11.14%)
24 85228 (5.02678571428571) - top comment: 189 (16.79%)
25 85692 (5.00992907801418) - top comment: 235 (6.65%)


So 5 of the top 25% have a single comment responsible for over 50% of the favorites.
posted by chrisamiller at 12:23 PM on December 28, 2009

So 5 of the top 25 % have a single comment responsible for over 50% of the favorites.
posted by chrisamiller at 12:28 PM on December 28, 2009

Also I think it would be kind of neat to re-do the list but exclude the influence of comments that were sidebarred, because they tend to just blow up with favorites. Unfortunately that data isn't in the dump so it would be a little harder to manually scrape and extract those from the sidebar archives.
posted by Rhomboid at 12:40 PM on December 28, 2009

smackfu - can you scale that graph by the number of registered users on mefi?

I tried but it doesn't really matter since the number of active users has been pretty stable over the range of that graph. Under 10% annual increases. It's really just that people are giving way more favorites now.
posted by smackfu at 12:41 PM on December 28, 2009

Two things:
1. A plug: beanplate, an Infodump filter. I think you can specify an expression like "favorites/comments" as output, though I haven't documented it & I can't test it right now because
2. I'm on the road during an Infodump exploration thread? This sucks. SO MUCH. Grar grar grar.
posted by Pronoiac at 12:41 PM on December 28, 2009 [1 favorite]

Neat! Just by eyeballing I see that in Rhomboid's list the first item that would STAY on that list without its own biggest comment, is 86424. Tenth place. Chrisamiller's list bears this out, but in a different way.

Megafavorites warp the t-number. What is the solution to find a more illustrative metric? A curve? Chop off the top?

And although the 100 comment threshold was implemented just for practical reasons, I think there has to be something to account for threads like 71507. Although in that case, anything that trapped for megafavorites would catch it as well, so maybe not.
posted by dirtdirt at 12:44 PM on December 28, 2009

So here's a similar calculation, but instead it's average favorites/comment excluding the most favorited comment in the thread.

Ask MetaFilter post 137148 (11.20)
MetaTalk post 18437 (9.80)
MetaTalk post 17309 (6.80)
MetaFilter post 85667 (6.77)
MetaTalk post 17431 (6.75)
MetaFilter post 86424 (6.73)
Ask MetaFilter post 65662 (6.49)
MetaTalk post 17107 (6.03)
MetaFilter post 84093 (5.62)
MetaFilter post 87172 (5.56)
Ask MetaFilter post 140973 (5.53)
MetaTalk post 17363 (5.50)
Ask MetaFilter post 133062 (5.43)
MetaFilter post 84993 (5.43)
Ask MetaFilter post 85255 (5.42)
MetaFilter post 85961 (5.31)
Ask MetaFilter post 131875 (5.30)
MetaFilter post 79749 (5.10)
MetaFilter post 70997 (5.08)
MetaFilter post 79486 (4.94)
MetaFilter post 87219 (4.90)
MetaFilter post 78669 (4.89)
MetaFilter post 83821 (4.84)
Ask MetaFilter post 113960 (4.84)
MetaTalk post 16805 (4.83)

posted by FishBike at 12:48 PM on December 28, 2009 [2 favorites]

Megafavorites warp the t-number. What is the solution to find a more illustrative metric?

Using the median number of favorites, rather than the average is one way to minimize the impact of outliers. That said, there are only five threads on the blue where the median number of favorites is greater than one:

1 86424 (3)
2 79764 (2)
3 84904 (2)
4 85961 (2)
5 85667 (2)
posted by chrisamiller at 12:56 PM on December 28, 2009

Also if anyone else is feeling bored I think it would be super to do a sort of "favorites cost of living" where you compute an inflation curve and then be able to say e.g. "getting 10 favorites on a comment in 2006 is equivalent to getting 87 today". To do this correctly though would be tricky because people can go back and favorite old comments and you don't want that to skew the numbers so you'd have to do the computation based on timestamps of the favorites not timestamps of the comments/posts the favorites are attached to.
posted by Rhomboid at 1:04 PM on December 28, 2009 [1 favorite]

Aw. My megafavorite just got lopped off. Typical.
posted by greekphilosophy at 1:12 PM on December 28, 2009

So 5 of the top 25 % have a single comment responsible for over 50% of the favorites.

Rooting these out might make a nice list of hidden gems as well. Not that gemlike, but for example, my comment in this thread accounts for 100% of the n=130 favourites. The post itself had 24 comments.
posted by Rumple at 1:17 PM on December 28, 2009

a sort of "favorites cost of living" where you compute an inflation curve and then be able to say e.g. "getting 10 favorites on a comment in 2006 is equivalent to getting 87 today".

The tricky bit is that favorites seem to be increasing linearly. Graph! So a traditional inflation model wouldn't apply.

Incidentally, the first person other than mathowie to give a favorite was togdon, to this question. He beat the announcement of the feature by 5 minutes.
posted by smackfu at 2:13 PM on December 28, 2009

(P.S. If anyone can tell me how to get graphs out of Mac Excel 2008 without them ending up huge, I would love to know.)
posted by smackfu at 2:17 PM on December 28, 2009

So here's a similar calculation, but instead it's average favorites/comment excluding the most favorited comment in the thread.

Why are excluding you my comment in each thread?
posted by orthogonality at 3:19 PM on December 28, 2009

(P.S. If anyone can tell me how to get graphs out of Mac Excel 2008 without them ending up huge, I would love to know.)

When it comes to exporting images at screen resolution from the recalcitrant, spit-happy llama that is Mac Office 2008, plain vanilla screenshots are your friend.
posted by killdevil at 3:39 PM on December 28, 2009

You shouldn't be excluding, but redistributing the wealth, comrade.
posted by UbuRoivas at 3:41 PM on December 28, 2009

Ah, more favorite fetishizing. Go us.
posted by Joseph Gurl at 5:45 PM on December 28, 2009

Awesome thread.
posted by fourcheesemac at 6:53 PM on December 28, 2009

Mac Excel 2008

Put it down. Go to another computer. Open any other version of excel, ever made, for anything.

There, isn't that better?
posted by pompomtom at 7:25 PM on December 28, 2009

It's true that there are a good number of things I just can't figure out at all, so I end up having to just delete what I did and start over and do it right. And it's pretty awesome how they're just like "oh, this version doesn't have Pivot Graphs, suck it!" OTOH, it makes much prettier graphs than Excel for Windows, without much tweaking at all. And if you copy and paste the graphs into Preview, you can save them straight to scalable PDF. So, I've been using it for graphs.
posted by smackfu at 7:58 PM on December 28, 2009

Man, you guys are nerds.

Great thread, btw. Is there a way to also include the favorites of the thread itself?
posted by molecicco at 8:14 PM on December 28, 2009

FOR ME?
posted by tehloki at 1:39 AM on December 29, 2009 [1 favorite]

Ceno = To Dine
Cide = To Die

Cenocide = HAMBURGER
posted by Potomac Avenue at 1:53 AM on December 29, 2009 [3 favorites]

cue megafavorites
posted by Potomac Avenue at 1:54 AM on December 29, 2009

OK, what we really need is for someone to calculate VORT (Value Over Replacement Thread) by determining the average t-number for non-deleted post-favorites threads, and using that as the stand-in number, the Replacement Thread, that we compare the t-number of any given thread to (minus itself?). This should probably be calculated by subsite.
posted by dirtdirt at 5:19 AM on December 29, 2009 [2 favorites]

Be careful with those sabermetrics or else somebody will produce data showing that we should trade mathowie for Cory Doctorow.
posted by Rhomboid at 6:06 AM on December 29, 2009 [1 favorite]

No way, Rhomboid. Doctorow looks good at first glance, but when you dig a little deeper you see his numbers are padded by that field he plays in, so away from home he's awful. He's a statue in the field, too, with no range. His overall line belies his poor situational hitting, and his failures close and late. He's got a nifty inside-out swing, but has a hard time committing on good breaking stuff. He's got soft hands though in the field, and good instincts. He gets on base a lot.

But Mathowie? Although it's impossible to quantify, Mathowie is clutch. No trade.
posted by dirtdirt at 7:13 AM on December 29, 2009 [1 favorite]

dirtdirt: "OK, what we really need is for someone to calculate VORT (Value Over Replacement Thread) by determining the average t-number for non-deleted post-favorites threads"

Hugs.
posted by Plutor at 7:48 AM on December 29, 2009

howie's on the juice you didn't hear it from me
posted by found missing at 7:48 AM on December 29, 2009

Not to mention that not all threads can be substituted for any other thread. You wouldn't trade Ortiz for K-Rod, would you? They're not interchangeable parts. We could probably use tags to figure out which posts would fall into another post's replaceable class.
posted by Plutor at 7:51 AM on December 29, 2009

ME
posted by Damn That Television at 8:30 AM on December 29, 2009

OK, what we really need is for someone to calculate VORT (Value Over Replacement Thread) by determining the average t-number for non-deleted post-favorites threads, and using that as the stand-in number, the Replacement Thread, that we compare the t-number of any given thread to (minus itself?). This should probably be calculated by subsite.

Surely we need to calculate the replacement thread value by league (sub-site) and season (year)? In which case the top 25 threads look like this:

Ask MetaFilter post 68352 (vort=20.77 f/c above 2007 Ask MetaFilter avg of 0.12)
Ask MetaFilter post 100880 (vort=20.47 f/c above 2008 Ask MetaFilter avg of 0.19)
Ask MetaFilter post 102361 (vort=14.81 f/c above 2008 Ask MetaFilter avg of 0.19)
Ask MetaFilter post 137148 (vort=12.68 f/c above 2009 Ask MetaFilter avg of 0.28)
Ask MetaFilter post 120624 (vort=12.38 f/c above 2009 Ask MetaFilter avg of 0.28)
Ask MetaFilter post 112933 (vort=11.40 f/c above 2009 Ask MetaFilter avg of 0.28)
Ask MetaFilter post 104881 (vort=11.29 f/c above 2008 Ask MetaFilter avg of 0.19)
Ask MetaFilter post 133062 (vort=11.04 f/c above 2009 Ask MetaFilter avg of 0.28)
Ask MetaFilter post 88726 (vort=11.03 f/c above 2008 Ask MetaFilter avg of 0.19)
Ask MetaFilter post 119868 (vort=9.93 f/c above 2009 Ask MetaFilter avg of 0.28)
Ask MetaFilter post 65244 (vort=9.88 f/c above 2007 Ask MetaFilter avg of 0.12)
Ask MetaFilter post 88686 (vort=9.81 f/c above 2008 Ask MetaFilter avg of 0.19)
Ask MetaFilter post 64076 (vort=9.63 f/c above 2007 Ask MetaFilter avg of 0.12)
Ask MetaFilter post 97076 (vort=9.44 f/c above 2008 Ask MetaFilter avg of 0.19)
Ask MetaFilter post 97078 (vort=9.12 f/c above 2008 Ask MetaFilter avg of 0.19)
Ask MetaFilter post 106005 (vort=8.81 f/c above 2008 Ask MetaFilter avg of 0.19)
Ask MetaFilter post 126923 (vort=8.72 f/c above 2009 Ask MetaFilter avg of 0.28)
MetaTalk post 15962 (vort=8.66 f/c above 2008 MetaTalk avg of 0.47)
Ask MetaFilter post 91148 (vort=8.56 f/c above 2008 Ask MetaFilter avg of 0.19)
Ask MetaFilter post 52074 (vort=8.53 f/c above 2006 Ask MetaFilter avg of 0.06)
Ask MetaFilter post 134344 (vort=8.23 f/c above 2009 Ask MetaFilter avg of 0.28)
MetaTalk post 16610 (vort=8.03 f/c above 2008 MetaTalk avg of 0.47)
Ask MetaFilter post 122931 (vort=7.85 f/c above 2009 Ask MetaFilter avg of 0.28)
Ask MetaFilter post 111325 (vort=7.72 f/c above 2009 Ask MetaFilter avg of 0.28)
Ask MetaFilter post 119433 (vort=7.72 f/c above 2009 Ask MetaFilter avg of 0.28)

Hmm, those look to be mostly short-ish threads with one very highly favorited comment. Note that the hypothetical "replacement thread" has a favorites/comment of much less than 1, so this list probably doesn't differ at all from a list of threads with highest favorites-per-comment.

I tried changing two things to make a more interesting list. First, I excluded the most-favorited comment in each thread from the calculations. Second, I calculated the VORT as a ratio rather than a difference, since the replacement threads have such small averages.

That results in a list that is almost all Music posts where the comments have 1 favorite each (because average favorites/comment in Music is nearly 0). So I excluded the Music sub-site and ran it again. I got a list that was all Ask MetaFilter posts. At which point I figured maybe just running a separate list per sub-site was the way to go. Here are the top 10 for each:

Ask MetaFilter post 65662 (vort=139.13x 2007 Ask MetaFilter avg of 0.04665 f/c)
Ask MetaFilter post 46223 (vort=118.01x 2006 Ask MetaFilter avg of 0.01695 f/c)
Ask MetaFilter post 45898 (vort=91.46x 2006 Ask MetaFilter avg of 0.01695 f/c)
Ask MetaFilter post 78970 (vort=87.45x 2007 Ask MetaFilter avg of 0.04665 f/c)
Ask MetaFilter post 45997 (vort=80.92x 2006 Ask MetaFilter avg of 0.01695 f/c)
Ask MetaFilter post 53385 (vort=78.67x 2006 Ask MetaFilter avg of 0.01695 f/c)
Ask MetaFilter post 137148 (vort=77.98x 2009 Ask MetaFilter avg of 0.14359 f/c)
Ask MetaFilter post 75249 (vort=72.88x 2007 Ask MetaFilter avg of 0.04665 f/c)
Ask MetaFilter post 50418 (vort=67.64x 2006 Ask MetaFilter avg of 0.01695 f/c)
Ask MetaFilter post 76919 (vort=62.32x 2007 Ask MetaFilter avg of 0.04665 f/c)

MetaFilter post 56220 (vort=23.64x 2006 MetaFilter avg of 0.02776 f/c)
MetaFilter post 63163 (vort=21.45x 2007 MetaFilter avg of 0.18109 f/c)
MetaFilter post 53089 (vort=19.24x 2006 MetaFilter avg of 0.02776 f/c)
MetaFilter post 53349 (vort=17.74x 2006 MetaFilter avg of 0.02776 f/c)
MetaFilter post 57357 (vort=17.44x 2006 MetaFilter avg of 0.02776 f/c)
MetaFilter post 53593 (vort=17.15x 2006 MetaFilter avg of 0.02776 f/c)
MetaFilter post 57012 (vort=16.80x 2006 MetaFilter avg of 0.02776 f/c)
MetaFilter post 53502 (vort=16.37x 2006 MetaFilter avg of 0.02776 f/c)
MetaFilter post 56613 (vort=16.01x 2006 MetaFilter avg of 0.02776 f/c)
MetaFilter post 56851 (vort=15.94x 2006 MetaFilter avg of 0.02776 f/c)

MetaTalk post 12355 (vort=27.59x 2006 MetaTalk avg of 0.03556 f/c)
MetaTalk post 12802 (vort=15.84x 2006 MetaTalk avg of 0.03556 f/c)
MetaTalk post 14726 (vort=15.55x 2007 MetaTalk avg of 0.17970 f/c)
MetaTalk post 14658 (vort=15.36x 2007 MetaTalk avg of 0.17970 f/c)
MetaTalk post 12764 (vort=15.30x 2006 MetaTalk avg of 0.03556 f/c)
MetaTalk post 12625 (vort=12.96x 2006 MetaTalk avg of 0.03556 f/c)
MetaTalk post 13212 (vort=12.78x 2006 MetaTalk avg of 0.03556 f/c)
MetaTalk post 14796 (vort=12.72x 2007 MetaTalk avg of 0.17970 f/c)
MetaTalk post 12962 (vort=12.70x 2006 MetaTalk avg of 0.03556 f/c)
MetaTalk post 13623 (vort=12.49x 2007 MetaTalk avg of 0.17970 f/c)

Music post 1435 (vort=236.19x 2007 Music avg of 0.00169 f/c)
Music post 25 (vort=167.42x 2006 Music avg of 0.00041 f/c)
Music post 480 (vort=147.12x 2006 Music avg of 0.00041 f/c)
Music post 2380 (vort=143.39x 2008 Music avg of 0.00719 f/c)
Music post 23 (vort=134.86x 2006 Music avg of 0.00041 f/c)
Music post 287 (vort=113.79x 2006 Music avg of 0.00041 f/c)
Music post 1438 (vort=98.41x 2007 Music avg of 0.00169 f/c)
Music post 1457 (vort=98.41x 2007 Music avg of 0.00169 f/c)
Music post 2494 (vort=71.68x 2008 Music avg of 0.00719 f/c)
Music post 2999 (vort=71.06x 2009 Music avg of 0.00844 f/c)

I spot-checked a few of these and they seem to be mostly threads with a lot of highly favorited comments. Ironically, the first and last ones on the MeTa list are discussions of favorites.
posted by FishBike at 9:36 AM on December 29, 2009 [1 favorite]

Well, they're almost all 2006 as well so probably something to do with introduction of favourites in the first place?
posted by Rumple at 9:59 AM on December 29, 2009

Some of those are in there just because someone went nuts with the + button. Like in 56220, sourwookie gave 40 or so comments a favorite.
posted by smackfu at 1:29 PM on December 29, 2009

Well, they're almost all 2006 as well so probably something to do with introduction of favourites in the first place?

Yeah, they launched in May 2006 and initial use was much lower volume than it is these days, so distribution effects would probably be a lot more pronounced for those first six months. I haven't really through through how that would work in specific, though.
posted by cortex (staff) at 1:35 PM on December 29, 2009

It might be interesting to look at how the distribution of favorite counts has changed over time, adjusted for the overall increase in favorites in absolute number terms. If it's the case all the numbers have just gotten bigger over time, you'd expect the lists of outliers above to feature posts from every year about equally.

But since they're almost exclusively showing 2006 and 2007 posts, that suggests some smoothing out of the distribution--the average favorites/comment has been going up, but the maximum hasn't gone up in proportion to it.

Hmm, why don't I just compute that for the metric used in my previous comment (average favorites per comment of each post, not including the most-favorited comment in the thread) and see how the biggest outlier for each year compares to the average of all posts that year:

Ask MetaFilter 2006: avg=0.0169 f/c, max=2.0000 f/c, ratio=118.01:1
Ask MetaFilter 2007: avg=0.0467 f/c, max=6.4906 f/c, ratio=139.13:1
Ask MetaFilter 2008: avg=0.0823 f/c, max=4.7857 f/c, ratio=58.18:1
Ask MetaFilter 2009: avg=0.1436 f/c, max=11.1964 f/c, ratio=77.98:1

MetaFilter 2006: avg=0.0278 f/c, max=0.6563 f/c, ratio=23.64:1
MetaFilter 2007: avg=0.1811 f/c, max=3.8837 f/c, ratio=21.45:1
MetaFilter 2008: avg=0.4089 f/c, max=5.0765 f/c, ratio=12.41:1
MetaFilter 2009: avg=0.6956 f/c, max=6.7692 f/c, ratio=9.73:1

MetaTalk 2006: avg=0.0356 f/c, max=0.9810 f/c, ratio=27.59:1
MetaTalk 2007: avg=0.1797 f/c, max=2.7937 f/c, ratio=15.55:1
MetaTalk 2008: avg=0.3286 f/c, max=2.3491 f/c, ratio=7.15:1
MetaTalk 2009: avg=0.5196 f/c, max=4.7115 f/c, ratio=9.07:1

Music 2006: avg=0.0004 f/c, max=0.0690 f/c, ratio=167.42:1
Music 2007: avg=0.0017 f/c, max=0.4000 f/c, ratio=236.19:1
Music 2008: avg=0.0072 f/c, max=1.0305 f/c, ratio=143.39:1
Music 2009: avg=0.0084 f/c, max=0.6000 f/c, ratio=71.06:1

So it isn't a totally consistent trend, but, for every sub-site the ratio of max:average is larger for 2006/2007 than it is for 2008/2009. So it probably is easier to be an outlier in relative terms in the first two years than in the last two.

I think that adjusting for year-over-year changes in distribution eventually just results in a 'top N for each year' type of list. Which I could do with this metric if anybody wants to see that.

Perhaps another way to find really good threads over time would be to list the record-holding posts over time for this metric. Which post had the highest favorites/commment (less most-favorited comment) the day the favorites feature was launched? And then what was the first post to beat that record, and so on. The initial few likely won't be that great, but after a while this seems like it would turn up a bunch of good stuff.

I think I will try to do that and see how it looks.
posted by FishBike at 2:24 PM on December 29, 2009

I'd just like to say that while, according to some earlier infodump calculations by FishBike, it looks to me and my insatiable ego like I have a fairly high personal T-value, I have never had a comment receive over 100 favorites.

To me, 50 favorites is still something very special. That's all I wanted to say. And happy New Year/Decade to you all.
posted by Navelgazer at 1:53 AM on December 30, 2009