Books to Read While the Algae Grow in Your Fur, June 2017

Attention conservation notice: I have no taste.

Peter Frase, Four Futures: Life After Capitalism

An expansion of his essay of the same name. This short book is very much worth reading if you like my blog at all. (Unless you're only here because you wish I'd write more about theoretical statistics, in which case you may be disappointed on many levels.)

Zeynep Tufekci, Twitter and Tear Gas: The Power and Fragility of Networked Protest

Tufekci is one of all-too-few social scientists and humanists studying computer technologies who actually understands, at a technical level, how they work, meaning that she is capable of actual critiques, rather than mere complaints. (Thus the only time I have ever recommended a TED talk, and probably the only time I ever will, is is this one by her about on-line advertising *.)

This book is the outcome a major area of Tufekci's research, which is studying contemporary more-or-less leftist protest movements and how they use on-line communications. My account will not do this rich book justice, but I will attempt it anyway. Unfortunately, even my summary effort have already grown past 800 words, so it will need to be a separate review.

ObLinkage: Tufekci has a website for the book, with a free, creative-commons copy there. But if you can afford it, I encourage buying a copy, as the proceeds will be donated to supporting refugees.

*: I can't resist adding some caveats, though. In her talk, Tufekci is essentially taking companies with Facebook at their word about their ability to influence behavior, and I am more skeptical about their current capabilities. For example, the infamous study about the spread of negative emotions used software for sentiment analysis, LIWC, which is very common, but also so bad ** that, without exaggeration, I have no idea what we can conclude about the relationship between network neighbors' emotions from a small relationship between their LIWC scores. For another, and more consequential, example, the equally famous Facebook voter-encouragement experiment doesn't actually show that Facebook can mobilize social influence to get Americans to vote, because of poor experimental design ***. But "the evidence for these claims is weaker than it looks" isn't the same as positive reason to think "this doesn't work", much less "this can never work". And even if companies like Facebook **** are engaging in pure investor story-time now, it would be imprudent to think that they, or their successors, will never be able to manipulate behavior, so Tufekci's point stands. ^

**: For example, as of 7 December 2017, putting "I can't complain" into their free demo scores the sentence as entirely negative in sentiment. Even if we could treat the gap between LIWC scores and actual sentiment (whatever that is) as random measurement noise (which would itself have to be carefully established), the magnitude of the noise is clearly huge. When looking at the influence of Irene's emotions on the emotions of their friend Joey, the noise would appear not only in the measurement of Joey's emotions (the regressand), but also in the measurement of Irene's (the regressor), making any estimate of the relationship (the regression curve) extremely imprecise. At the very least, one would need to do an error-in-variables analysis, rather than a straightforward regression --- and that's assuming the measurement noises in the regressor and the regressand were independent of each other and of the true values. ^

***: More specifically, the design they used confounded direct exposure to a pro-voting message (which they randomized), indirect exposure through social influence, and whatever characteristics of users lead to American accounts having more or fewer American Facebook friends. (As I once heard Cyrus Samii put it, "Randomization for treatment does not randomize influence.") And a confounded design does not get more in