Hardship Pay

Some time ago I heard that diplomats who are posted to Ottawa get hardship pay. Apparently, of all the places where a diplomat can be sent, Ottawa has the second-lowest average winter temperature (the coldest city presumably being in Russia somewhere). It's true Ottawa can get quite cold, but on the other hand, where else can you skate to work and stop on the way for a beavertail and coffee? Instead of receiving hardship pay, diplomats should have to pay some sort of bribe (or in bureaucratese, a preferred posting premium) for the opportunity to live in one of the few places on earth where one can dine on a beavertail.

To my mind, having to work with statisticians is what truly merits hardship pay. They refuse to ever give a definite answer. (An interesting statistic would be the proportion of statisticians who overcome this inability to commit and get married.) Maybe it's because, like meteorologists who state the forecast in terms of probabilities, this means the probability they will ever be wrong is 0%. If you go to a statistician and ask: “What test should I use to analyze these data?” – a seemingly straightforward question – you will never get the straightforward answer you seek. Instead, you will be presented with a maze of options, usually accompanied by a stack of articles to read. In fact, I'm beginning to suspect that not even statisticians fully understand statistics. I don't blame them. Trying to decipher an article on statistics is like dealing with an algebraic system that has been off its medication for too long, or alternatively, has been taking steroids. Instead of just having to cope with the familiar faces of x and y and the occasional z, all sorts of letters will be randomly distributed throughout the text, sometimes with disturbing symbols beside them. By the time the article gets around to explaining what g and g' are, you've forgotten what a symbolizes. No wonder statistical errors were found in more than 50% of articles published in the British Medical Journal.

To “help” with an analysis I was working on recently, a statistician sent me an article titled Smearing estimate: A nonparametric retransformation method, which sounds like the Frankensteinish offspring of gynaecology and statistics. Here is an excerpt from the article:

Assume (i) the retransformation h is continuously differentiable, (ii) X contains the intercept, (iii) XtXIn+ I:positive definite, and (iv) the expectation (4.5)is finite for all M > 0. Then the smearing estimate (3.1)is weakly consistent. For most popular transformations, the supremum in(4.5) can be evaluated at the endpoints. For example, if I h' I is monotonic, the moment condition (iv) in Theorem 1 can then be replaced by (v') E [ht(c + €)I2 < +a for all c, which is usually easy to check under hypothesized true error distribution. For example, for the power transformations q = g(y) = Y", a # 0, the desired moment condition is that E(c + E)~[('/")- l1 < +w for all c, which is satisfied for the normal error distribution if 0 < ci < 1. For the logarithmic transformation q = log( Y), Y = exp(q), the desired moment condition reduces to E exp(2~)< +m, which is satisfied for the normal error distribution.

Can I rest my case? (This requires only a yes or no answer, like you’re supposed to give when you’re being examined in a court of law. Maybe not having become a lawyer was a mistake. Although I suppose a statistician could end up taking the stand, in which case they would no doubt alienate the jury with their evasive manner and tendency to hand out articles in response to every question.)

Maybe I should write the foreign service exam. What are the chances that I would end up being posted to the coldest place on earth? Maybe I’ll ask a statistician about this. Whatever answer I don’t get from them, at least I know that if I am sent there, I’ll be entitled to some hardship pay.