In one of his biographies, Richard Feynman recounts the tale of being contacted by a civilian seeking advice on how to join the rarefied members-only club of the physicist. Where does one begin? the man asked. What should he study first? Is it best to start with differential equations? Ordinary or partial or stochastic? What about Lie algebra or algebraic topology? Or perhaps group theory or measure theory or number theory?
Feynman replied the true path to physics was none of those. Instead, he recommended the man simply think about the things he sees every day. Why are the colors of a rainbow in the order they are? Why does touching the bare ends of a wire together make sparks? Why does a magnet attract iron but not aluminum? Why is the sky blue? That, more than any mathematics, is what physics is really about (Feynman conveniently overlooking the infinite dimensional Banach space path integrals he used to snag a Nobel. But I digress).
I was reminded of this anecdote after recently stumbling across a copy of Stats: Data and Models. For I have spent many dark nights grappling with the statistics equivalent of path integrals in infinite dimensional Banach space -- sometimes successfully, sometimes not. Richard De Veaux and his coauthors provided me my own personal Ratatouille moment. A return to the simple roots of statistics, before it became a freak show of covariate matrices and Markovian bootstraps and what all else.
It's a revolutionary idea: Statistics should make sense, and if it doesn't you're doing it wrong. Such blasphemy was never dared spoken aloud in any of my courses. I took hardcore engineering stats as an undergrad and my graduate follow-on was rumored to be the toughest class at the university (the claim turned out to be hyperbole from people who never tried Orgo or E&M). Always the mantra was: Statistics will never make sense, and if you think it's starting to here comes more bad news. Richard De Veaux (Williams College), Paul Velleman (Cornell), and David Bock (also Cornell) have taken up the mantle to rage against this machine, the rallying maypole being Stats: Data and Models, currently in its fourth edition.
To be sure, their text isn't for everyone. The presentation suffers from dinosaur-pill syndrome (one of those capsules you buy at the museum and put in water and it expands into a giant sponge dinosaur. How amusing). The chapters can seem repetitive. There's side boxes and glossy pictures of happy people doing things related to the topic of the accompanying example (for a counterpoint thumb through Zar sometime, which contains zero pictures of anything, and if it did it would be a Puritan woodcarving of the reader's soul being dragged to hell while Jerrold stands at the edge of the pit cackling maniacally). S:D&M only gets as far as regression and t-tests and ANOVA, and these only the cases that don't go off any rails, and only after a protracted set-up that can only be described as Jacksonian (I mean Peter, but I guess John David also works in this context). If the material were mapped onto a calendar year, the good stuff would only start happening in December. The text also comes pre-highlighted (I thought our library's copy had been defaced until I noticed the Sharpie work was too precise to be mere rat bastardy). I'm uncomfortable when an author tells me what to think -- if they took the time to identify the important bits, why include the rest?
But there is method to this madness, and that method is making converts out of the would-be stats-o-phobe. If you only take one mathy course in your academic life it will probably be statistics, a freshman detour through Nerdland required by many a university before you get to be a dance major or neurosurgeon. Here, De Veaux and colleagues are ambassadors of goodwill, taking an opportunity to put the captivate in captive audience. Hunting the beast methodically, systematically, logically, until you know its secrets as readily as the back of your hand. This is something often missed in introductory treatments, the whatevers for Dummies and the Idiot's guides. The key to making technical material accessible isn't bright colors or cartoon characters in the margins, it's rock-solid, crystal-clear explanation. Repeat until it becomes second nature. An idea so simple it's revolutionary. Why did none of my stats professors understand this?
That's when it hit me -- Stats: Data and Models is aimed at high schoolers. So of course I never saw statistics presented in such a light. My high school class offerings didn't go beyond Why Athletes are Better Than You and Why Athletes are Better Than You II: The Date Rapening. And by the time uni rolled around, we were neck deep in covariate matrices and Markovian bootstraps and what all else.
Is the end result successful? I can't answer from the perspective of the intended audience -- my days of being a innocent LabKitten facing a world of possibility and wonder got beaten out of me by grad school like Anjelica Huston getting worked over with a pillowcase full of oranges. All I know is I went to the library one fine morning looking for a clarification of some statistics blah-blah, S:D&M caught my eye, and by the time I looked up it was dark outside. However, lots of pre-curmudgeony waifs staring down the barrel of AP stats have left nice comments about the book on Amazon, and I'll take them at their transparent word (lying convincingly on the Internet only happens after you start writing a blog). Also, the Textbook Industrial Complex being what it is, new editions of S:D&M appear pretty regularly, which means you can pick up an old edition used for a couple of bucks. Win/win. If you don't like it, use the book for riprap or self-defense (the hardcover has enough heft to kill something thick-skulled and charging. Which I guess is also win/win).
View Stats: Data and Models on Amazon. Take the LabKitty Challenge: See if you can read the whole thing in one sitting!
Feynman replied the true path to physics was none of those. Instead, he recommended the man simply think about the things he sees every day. Why are the colors of a rainbow in the order they are? Why does touching the bare ends of a wire together make sparks? Why does a magnet attract iron but not aluminum? Why is the sky blue? That, more than any mathematics, is what physics is really about (Feynman conveniently overlooking the infinite dimensional Banach space path integrals he used to snag a Nobel. But I digress).
I was reminded of this anecdote after recently stumbling across a copy of Stats: Data and Models. For I have spent many dark nights grappling with the statistics equivalent of path integrals in infinite dimensional Banach space -- sometimes successfully, sometimes not. Richard De Veaux and his coauthors provided me my own personal Ratatouille moment. A return to the simple roots of statistics, before it became a freak show of covariate matrices and Markovian bootstraps and what all else.
It's a revolutionary idea: Statistics should make sense, and if it doesn't you're doing it wrong. Such blasphemy was never dared spoken aloud in any of my courses. I took hardcore engineering stats as an undergrad and my graduate follow-on was rumored to be the toughest class at the university (the claim turned out to be hyperbole from people who never tried Orgo or E&M). Always the mantra was: Statistics will never make sense, and if you think it's starting to here comes more bad news. Richard De Veaux (Williams College), Paul Velleman (Cornell), and David Bock (also Cornell) have taken up the mantle to rage against this machine, the rallying maypole being Stats: Data and Models, currently in its fourth edition.
To be sure, their text isn't for everyone. The presentation suffers from dinosaur-pill syndrome (one of those capsules you buy at the museum and put in water and it expands into a giant sponge dinosaur. How amusing). The chapters can seem repetitive. There's side boxes and glossy pictures of happy people doing things related to the topic of the accompanying example (for a counterpoint thumb through Zar sometime, which contains zero pictures of anything, and if it did it would be a Puritan woodcarving of the reader's soul being dragged to hell while Jerrold stands at the edge of the pit cackling maniacally). S:D&M only gets as far as regression and t-tests and ANOVA, and these only the cases that don't go off any rails, and only after a protracted set-up that can only be described as Jacksonian (I mean Peter, but I guess John David also works in this context). If the material were mapped onto a calendar year, the good stuff would only start happening in December. The text also comes pre-highlighted (I thought our library's copy had been defaced until I noticed the Sharpie work was too precise to be mere rat bastardy). I'm uncomfortable when an author tells me what to think -- if they took the time to identify the important bits, why include the rest?
But there is method to this madness, and that method is making converts out of the would-be stats-o-phobe. If you only take one mathy course in your academic life it will probably be statistics, a freshman detour through Nerdland required by many a university before you get to be a dance major or neurosurgeon. Here, De Veaux and colleagues are ambassadors of goodwill, taking an opportunity to put the captivate in captive audience. Hunting the beast methodically, systematically, logically, until you know its secrets as readily as the back of your hand. This is something often missed in introductory treatments, the whatevers for Dummies and the Idiot's guides. The key to making technical material accessible isn't bright colors or cartoon characters in the margins, it's rock-solid, crystal-clear explanation. Repeat until it becomes second nature. An idea so simple it's revolutionary. Why did none of my stats professors understand this?
That's when it hit me -- Stats: Data and Models is aimed at high schoolers. So of course I never saw statistics presented in such a light. My high school class offerings didn't go beyond Why Athletes are Better Than You and Why Athletes are Better Than You II: The Date Rapening. And by the time uni rolled around, we were neck deep in covariate matrices and Markovian bootstraps and what all else.
Is the end result successful? I can't answer from the perspective of the intended audience -- my days of being a innocent LabKitten facing a world of possibility and wonder got beaten out of me by grad school like Anjelica Huston getting worked over with a pillowcase full of oranges. All I know is I went to the library one fine morning looking for a clarification of some statistics blah-blah, S:D&M caught my eye, and by the time I looked up it was dark outside. However, lots of pre-curmudgeony waifs staring down the barrel of AP stats have left nice comments about the book on Amazon, and I'll take them at their transparent word (lying convincingly on the Internet only happens after you start writing a blog). Also, the Textbook Industrial Complex being what it is, new editions of S:D&M appear pretty regularly, which means you can pick up an old edition used for a couple of bucks. Win/win. If you don't like it, use the book for riprap or self-defense (the hardcover has enough heft to kill something thick-skulled and charging. Which I guess is also win/win).
View Stats: Data and Models on Amazon. Take the LabKitty Challenge: See if you can read the whole thing in one sitting!
No comments:
Post a Comment