never rat on your friends, and always keep your mouth shut

Tuesday, January 12, 2010

re: mo

"It's facile and leads to self-satisfied conclusions that the data doesn't, actually, support."

this seems to me to be the weakest part of your argument (and it's sort of unfortunately located up in the thesis area). it's actually the "self-satisfied" conclusions that are most supported in the data.

rachel getting married is not a popular movie in compton. it is a very popular movie in lower manhattan. evidence: there is a compton-sized hole in the LA renting patterns map. whereas the lower manhattan map is colored bright red.

brains might be "notoriously poor" at teasing out fine correlations in spatial data, but this one's pretty easy.

can you elaborate on this? there's other stuff in your post, but this seems pretty central and also strikes me as just wrong, regardless of whether or not i like the implications of some of what you're trying to say below (and i think i do).

3 comments:

Moacir said...

Argh, I started a whole comment only to realize that I misunderstood your point.

Self-satisfied conclusions are always inherently "supported" by the data, since you're cherry picking to make your point. You're trading on fantasies of what Compton is, what Lower Manhattan is, and what Rachel Getting Married is in order to make a point.

So yes, in that sense the data in the maps precisely does flatter our prejudices (or confuse us when we can't understand challenges to the prejudices).

But then look at the conclusions that the data supposedly support:

"No one's renting RGM in Compton is proof of ______." Whatever you fill in the blank with, other than a tautology, is unsupported by this data. So if you have any interest in filling in the blank--and my point is that the map, by being all interactive and "informational" and shit, invites you to fill in the blank--you have to turn elsewhere. And since the map doesn't bother providing a single other variable (per capita income, race, whatever) that one could regress against, you're left to your prejudices.

And that sucks. And is bad.

There are basically three responses to the fact that RGM wasn't rented in Compton:

1. "Why?" Dead end. Map has no idea and nothing resembling a guide to start answering.

2. "I knew it!" Because of some sort of prejudice. Bad science.

3. "I'm ambivalent about this information, but it may come in handy at a cocktail party or in solving a crossword puzzle [straight Postman]." Then why hype it?

The potential fourth response, "this changes what I thought about a ZIP code" is, in my opinion, unrealistic, since we are equipped enough to rationalize away challenges. I'm open to being proven wrong on this, but I'm skeptical that the maps will be able to do it.

I guess there's a fifth answer, "how can I monetize this?" but I'm not going to address it.

Moacir said...

I guess I can criticize the Very Large Array map I lauded yesterday for relying on viewer prejudice to fill in "why" on the Williamsburg map. I dunno.

Moacir said...

Finally, the reason I brought up the other stuff in my post was related to efforts at analysis that stretch past "Compton-sized hole." That is, A regular physical map tells you if there is water at a certain point or not. This can be helpful as a tool. But it can't tell you what percentage of the surveyed area is water, if the area is mostly water (or not), or anything more potentially interesting than "is there water or no?"

If one tried to use the Netflix maps to answer similarly more complex questions, then they'd run into the same problems, and that's where issues like ordinal data, ZIP codes, autocorrelation and the rest start being real issues. For binaries, I'll grant, it's not that big a deal.