Assorted Links

  • fruit and diabetes. Blueberries good, cantaloupe bad.
  • R most popular language for “analytics/data mining/data science work” among survey respondents. I wish I could describe the respondents, but I can only say they are people who might call what they do “data mining” or “data science”. In addition, the use of R is growing. Most psychology departments teach SPSS or Matlab.
  • Thomas Frank criticizes universities, undergraduate education in particular. “An educational publisher wrote to me [asking] to reprint an essay of mine [that is freely available]. . . . The low, low price that students were to pay for this textbook: $75.95.”

13 Replies to “Assorted Links”

  1. Astonishing – an apple a day keeps the doctor away. I hope it applies also to British apples, which are more – how to phrase this delicately – flavoursome than American.

    Unfortunately the blueberries on sale in Britain are tasteless.

    If anyone thinks he can frighten me off our home-grown strawberries he is sadly delusional.

  2. Cantaloupe is cheap. Blueberries are expensive. This is likely significant.

    Also, dearieme: the apples in American grocery stores may not be tasty, but I assure you a bit of effort uncovers delicious apples, many of varieties unavailable in the UK.

    Seth: Interesting point about cost. Maybe blueberries are better than cantaloupe because they are more likely to supply a nutrient we are not already getting (because, due to price, we are eating fewer blueberry-like foods than cantaloupe-like foods).

  3. Or maybe people who can afford to eat more blueberries can also afford to take many other steps to protect their health. If you have access to the actual paper, just look at the comparison across fruit quartiles in Table 1. All *measured* health behaviours (smoking, drinking, exercise, vitamins) increase monotonically across the quintiles, and it’s remarkably consistent across the 3 cohorts. Simply put, the people who ate most fruit, on average, were inherently healthier, and that just cannot be controlled for using a statistical model.

    The same has been true of the previous BIG RESULTS to come out of this Harvard group using these very same data. Walter Willet is a hack.

    Seth: Good point about confounding. However, I don’t agree “that just cannot be controlled for” — why not? Why not control for smoking, for example?

  4. Because controlling for all “known” confounders does not necessarily reduce the net confounding bias. That is, bias could very well be increased by controlling for things like this… this is very easy to demonstrate through simulations. “Smoking” (as much as that measures the same thing across people) can be controlled for, but not all of the other health related decisions that go along with it.

  5. So the Blueberry-Kefirgurt I eat turns out to be healthy.
    1) Ferment 20%/80% Kefir/Heavy cream mixture for 3-4 days at room temp until very thick and tangy. Cool in a fridge.
    2) Stir in blueberries.
    3) Eat.

    Seth: Good idea.

  6. R is a horrible language. The more I use it the more I appreciate Python.
    But sometimes you need a manova and R’s got all the stat functions.

  7. Seth, while most psychology departments teach SPSS/Matlab/Excel (seriously), the use of R in psychology has been growing for a while. Also, lots of data science questions are merely repacked psychology datasets (certainly those relating to advertising and/or social networks).

    Anecdotally, I have noticed quite a large movement towards R even in my department, where no-one knew what it was when I started using it three year ago.

  8. I eat a pound of frozen blueberries a night. A two pound bag costs $10. I am not even going to pretend that that much blueberries is good for you. It does not seem to be all that bad for me though.

Comments are closed.