Now, I’m no fan of baseball, but that does not prevent me from rooting for the Pittsburgh Pirates.  ‘Rooting’ might be too strong a word to characterize my support, which consists of checking to see if they won each night and reading a Post-Gazette article or two each week. But I’d like very much for the Pirates—who have been historically awful for a very very long time—to not be historically awful anymore.  So, I am, like the rest of my hometown, pretty jazzed that their win last night over the Red Sox put the Pirates two games over .500, 76 games into the season.

The above chart, although it’s not the easiest to read, documents just why Bucco fans might seem a little too exuberant for having more wins than losses before the all-star break.  Since being eliminated from the 1992 Playoffs, the Pirates have been over .500 for 257 games, which means over 90% of the last 18 years, they haven’t been.  As you can see in the chart, the Pirates haven’t been above .500 this late since before 2000 (graphing back to 1993 was a mess, but the latest they have had a winning record since making the playoffs was 133 games into the 1997 season).

So really, this is lightly trodden territory for the Pirates, and any fan is damn well justified in them being slightly above average.  It might be time for me to learn the names of a few players.

[Data comes from baseball-reference.com and does not include last night’s victory.  If someone knows a better way to graph this, or wants a .csv of the Pirates record each of the last 18 years, just let me know.]

View high resolution

So, yes, in that last “chart”, obviously popsicle remaining corresponds inversely to blowjob enthusiasm, as several of you felt the need to point out.  I got that, and it doesn’t matter.  The problem is that the image misleadingly purports to be a chart, but the categories are being expressed visually, rather than the values associated with the categories.  All it is is a fancied up table.  There’s nothing wrong with fancy tables, but if you’re gonna be a table, don’t try to trick people into think you’re a chart.

In case you were wondering, this is how you make a photographed chart.  This chart from Very Small Array, which Ben pointed me towards, is gorgeous and clever, but more than that: it is a clean, accurate representation of the data.  Well done, chart!

View high resolution

Dispatches from my education:

1. FrameNet (where the above part of a chart comes from) is pretty fun to play around with and probably not very useful.
2. I read a paper today in which the authors said the “learning curve” for a task was “steep”.  What they meant was that people learned the task quickly.  I’m not sure I’ve made reference to steep learning curves before, but if I have, I’ve done so incorrectly.  Now I know words and phrases can have more than one meaning (I probably wouldn’t be in school if that weren’t true), but I think it’s fair to say that using “steep learning curve” to mean “difficult to learn” is—if not outright wrong—definitely really stupid once you think about it.  Of course the x-axis on a learning curve is time.  What the hell else could it be?
3. I share an office with 11 other grad students.  There is a bell-like apparatus attached to the door.  It’s a hollow metal cylinder suspended from a string that fits into a metal loop and rings when someone comes in or out.  Or I should say that there was a bell-like apparatus attached to the door.  A lot of people come in and out of a twelve person office and I don’t really give a fuck about their presence (plus I can fucking hear them coming in anyways, even without a bell; it’s not like anyone is trying to sneak up on us).  So I took the bell down on Monday when nobody else was in the room.  I hope I’m not wading into some years old debate on the bell-like apparatus.  I mean, the bell-like apparatus couldn’t have always been there.  Someone once decided that that door needed to make a truckload of noise when people use it, and I just can’t envision a scenario in which the bell-like apparatus is helpful.  I decided against hiding it, so If someone puts it back up, I might need to ask to change offices to avoid confrontation.

View high resolution

If anyone knows anyone at NBC, I’ve got an awesome idea for a TV show:  It takes place at a fictionalized NBC that has a fictionalized Saturday Night Live show.  Then, they decide to develop two shows that take place behind the scenes of fictionalized fictionalized SNL at the same time.  After that things get pretty crazy.

View high resolution

A histogram of average review score minus Metacritic score.  This pertains to the footnote in the last post, where I mentioned that well over half the publications averaged scores below Metacritic.  Metacritic says they take a weighted average (the weights are the significance of the publication; the weights are also a secret) and then the weighted averages are normalized.  I don’t quite understand the normalization process, but Metacritic explains that it causes scores to spread out over a wider range.  But between the weighting and the normalizing, the Metascore is shifting up the unweighted mean just about every time (89%).  As you can see from the histogram, this difference is almost always between 0 and 5 points.  I can’t imagine the weighting is causing this: although I don’t have a great grasp on which publications are best regarded by Metacritic, I have a tough time believing the systematically positive reviewers (Entertainment Weekly, Billboard.com, Rap Reviews) are treated as having more clout than Rolling Stone and Village Voice, who are, on average, below the Metacritic score.

It’s not universal, but the unweighted average of all the review scores is usually a little more than 2 points below the Metacritic score (mean: -2.21, median: -2.33).  It’s between 1 and 3 points lower a little less than half the time, although it is negatively correlated with Metacritic score.

Luckily, this does not change anything I’ve said in the last posts.  Aside from people in the middle swapping a position, the rankings of most negative publications come out the same whether you are looking at difference from Metacritic score or difference from the unweighted average, but the actual numbers get shifted (Pitchfork averages 5.5 points below the Metacritic score, but 3 points below the unadjusted average, which is a typical change).

Yesterday, something called Flavorwire had a post comparing the Metacritic and Pitchfork scores of albums recently released by women over forty.  The post suggested that since Pitchfork gave lower scores than Metacritic’s aggregations to all but one of these, they must be be biased against old women.  I came across it via a post by Barthel, who (along with A Grammar) rightfully pointed out how poor the statistical analysis was, since it lacked any control comparisons.  Barthel then used unspecified and “extremely narrow cherry-picked sample”, to conclude Pitchfork is just biased against everyone, and old women aren’t special.

With my Jeopardy! stuff starting to bore me, I thought this would be a good new project.  So today, I created a nice little 5442x89 dataset (available on request) using Metacritic.  Each observation is an album, and 85 of the columns are the scores Metacritic attributed to the reviews of different publications.  So basically, we have several (nowhere near 85) scores for each album.

Although we are going to hold off on Pitchfork’s age and gender effects (for which I would need to get some age and gender data), this is exactly the right data to examine whether Pitchfork’s scores are systematically lower than Metacritic’s.

As you can see from the above Hexagonal Binning Plot, they totally are.  I hand-drew in the boulevard of equality, where all the points would be located if Pitchfork and Metacritic agreed every time.  As things skew well below the line, that means, Pitchfork gives lower ratings than Metacritic.  This is based on 3,467 reviews of albums covered by both Pitchfork and Metacritic.

Here are a few means to drive this home:  In the overlapping 3,467 entries: Pitchfork reviews average a score of 67.67 (scaled to 100) versus an average Metacritic score of 73.16.  That average difference is 5.49 points.  If, as Barthel suggested, you factor in the standard deviations of all the reviews that Metacritic uses for each album, Pitchfork scores albums .293 standard deviations below the critical consensus represented in the Metacritic score.  In fact, Pitchfork is lower than Metacritic 67% of the time.

Footnotes:

1. There is some obvious endogeneity here, since a Pitchfork score factors into a Metacritic score.  If I wasn’t interested in preserving Metacritic’s weights, and if I were more proficient in R, it would probably make sense to take the Pitchfork-less average of the other reviews.  Of course, that would only show a larger negative bias in the Pitchfork reviews.
2. Fun fact: the largest discrepancy between Pitchfork and the critical consensus was Northern State’s 2003 release Dying In Stereo, which Pitchfork gave a 0.8 (8 on our scale) versus the 77 Metacritic score (that was 4.23 SDs).
3. After collecting and organizing the data in Python, I was again working in R (including for the chart).  I’m a complete novice in R, so it is likely I have made a mistake.
4. The hexagonal binning plot is just a fancypants scatter plot for when the data overlaps too much to see regular dots.  The darker the hexagon the more observations are concentrated there.  The scale on the left lays this out for you.
5. Barthel and A Grammar are both excellent tumblrs.
6. Clearly, this is a dataset I’m going use more than once.  So, look out for that.

View high resolution

Another possible Jeopardy! scenario you might as well be braced for:  Your show is coming up, and you’re fairly well prepared, having memorized all the most common responses, as well as which responses show up most often given various keywords in the clues.  You’ve even learned which parts of the board various presidents are most likely to be found.  But with all that trivia in your head, for the life of you, you just keep on confusing Norway and Richard Nixon.

It’s a common mistake, and while we could get into some of the subtle differences between Norway and Nixon, it’ll probably be faster if you just memorize the above chart of which answer shows up more frequently at each position on the Jeopardy board.  Maybe you should even get it tattooed on your forearm.  If a square is white, Nixon has appeared there more often.  Black is for Norway and the gray squares have had exactly the same number of Nixons and Norways (that could be 0, could be 5, it’ll vary by square.)

[This is hopefully the last Jeopardy update for a bit, until I decide I should waste lots of time on Daily Doubles.  It’s worth noting the responses data I’m using comes from The J! Archive.  I meant it to include every episode since the show doubled their dollar values, but it turns out that the archive isn’t organized as sequentially as one would hope, so there should be 2,330 shows worth of information in here, but I don’t know which shows exactly (well, http://www.j-archive.com/showgame.php?game_id=1062 through http://www.j-archive.com/showgame.php?game_id=3386, but I don’t know the dates).

I can’t think of any reason why whatever I grabbed wouldn’t be a representative sample, but this is, frankly, the type of shoddy science you’d expect from some inexperienced jerk messing around on his personal blog, and we should all be above that.  Which is all to say, if you do memorize the above chart to help distinguish between Nixon and Norway, there might be some sort of sampling bias affecting the results.  If I go forward with this (and I really hope I have the sense to call it quits), I’ll see what I can do about fixing this problem.]

View high resolution

So, you’re back on Jeopardy.  Trebek is reading the clue for the $1,000 square in the sixth column (bottom right corner of the board), and you think it might be “Jefferson” but you don’t know for sure.  Lucky for you, here is a helpful map of where “Jefferson” is most likely to show up on the board.  And while our third president is a common answer, he never shows up in the bottom right corner of the board.  Your hunch is statistically pretty wrong.  Jefferson is much hotter at the top of the board, especially in the first few columns.

We can compare that with “Polk”.  Since Polk is a president nobody bothers to know anything about, the Jeopardy organizers tend to bury his answers at the bottom of the board.  Although, it’s worth noting that since Polk is a president nobody bothers to know anything about, he doesn’t show up anywhere very often.  The red square corresponds to 4 results and the yellows are each 1.

Now I concede that this data is both less interesting and less helpful than some of my previous Jeopardy! work.  But the charts are a little pretty.  This is the first time I’ve done anything in R.  It involved lots of trial and error and some embarrassingly slow recall of elementary matrix operations.  Because of all the work involved in this, you can expect an extension of this project in the next few days.

View high resolution

Recursion In Fun Places:

Depending on how you count, every fifth or sixth step of the Hokey Pokey is to “do the Hokey Pokey”.

This means we’ve all been doing the Hokey Pokey wrong.  When you get to the first “do the Hokey Pokey” step, it seems as though the correct thing to is do is to follow instructions and put your right foot in all over again.  Because of this: 1) once you start the “Hokey Pokey” it never ends, and 2) you never even get to turn yourself around, let alone move any body parts other than your right foot.

View high resolution

Were you aware of the existence of Scotch Eggs?  I wasn’t until very recently, but my personal breakfast food dark ages are over.  As you can see from my helpful diagram, these things are fantastic.

I find it a little disappointing that more people aren’t sitting around thinking of things to cover in sausage—and then fry.  These scotch egg people are really onto something.