Are you “good with numbers”? Occasionally someone will give me credit for having that skill. Sometimes they’re offering a compliment, and sometimes I think they’re excusing themselves from needing to understand what I’m trying to say. “Predictive modeling? Ha ha, sorry mate, I’m not good with numbers.”
The truth is, although I’m not afraid of numbers, I’m no whiz. I imagine number people are good with Sudoku puzzles — I don’t like the look of them and have never even tried one. Simple math is not a favourite activity either; when I play any kind of game that requires keeping score, my wife does it — I’m just too slow. I also have no memory for numbers. I know my Phonathon stats pretty well, but if it’s something I don’t have my hands on every day, such as our overall totals in Annual Giving or our capital campaign figures, it’s in one ear and out the other. I’m easily impressed by people from other institutions who “know their numbers.”
I’m having these thoughts as we sift through the Annual Giving results from the fiscal year just ended, trying to nail down what happened, not just overall (which we already know) but split along important distinctions such as donor status (renewed vs. acquired) and solicitation method (mail, phone, etc.).
Important work, to be sure. But getting the sum of the last column to agree with the sum of the last row is one of my definitions of Hell.
I’ve written about the difference between “everyday” data and “insight” data. Year-end annual fund reporting falls squarely in the “everyday” category. By everyday I don’t mean trivial. “Everyday” means immediate — the forms and usages of data that are most familiar, the current data used for today’s purposes. Everyday data must meet a high standard: if it’s not up to date, has missing values or is in any way “messy,” then it’s invalid. If the sum of the last column fails to agree with the sum of the last row, there’s a problem somewhere, and you can be expected to beat your head against the keyboard trying to figure it out.
“Insight” data — that’s different. It may be messy, full of holes and old, but that’s okay. When I’m cobbling together a dependent variable for a Phonathon giving model, I don’t fret over whether I’ve managed to capture every last dollar from every last donor who ever gave in response to phone solicitation. I do my best to put together a decent variable that looks like a reasonable representation of “giving by phone” and leave it at that. Striving to meet the standards of an accountant in Finance is not going to make one bit of difference in a model that assigns a score between one and ten.
Yes, there are ways to make serious errors in predictive modeling, and I am not careless. But what is most important, what excites me, is not the numbers and the math. What excites me is the search for patterns and tendencies in a swirl of messy data that to other eyes appears formless, and turn those patterns into something actionable. It’s my understanding that x-rays and other forms of medical imaging are difficult to interpret — it takes skill and a knowledge of what to look for, and it’s something no machine can do. Data mining is a bit like that.
Are you good with numbers? If you are, I admit you’ve got a leg up. The question is, are you also good with meaning?