Friday, Part 2

10 October, 2004. 715 words.

I had planned to write long saga about Friday, but now it’s Sunday night and I don’t really want to think about it anymore. So this is an abbreviated rendition.

When last we left our story, the BEAVA test specificity was coming up at 12% using the new computations Julio requested earlier in the week.

Needless to say, Julio freaked out. He had expected the specificity result to be about the same (actually so had I), having convinced himself that his computational changes would have a negligable impact on the test results. (In fact, I had observed that while most of the time the impact was negligable, sometimes the impact was quite dramatic.)

The next seven hours of my life consisted of a concentrated assault of non-stop Julio-style problem solving.*

At first, I thought there was a major bug in my code, so I didn’t complain about staying late to try to fix it. (I hate unfixed bugs. That being said, however, had Julio not been standing right there, I would have simply made a note of it to work on first thing Monday morning.) As time wore on, however, I began to suspect that the problem was probably not in my code at all, but rather that Julio’s computational changes had had a much more pronounced effect than he anticipated.

Were I to have suggested such a thing to him without quantifiable proof, he would probably have scoffed. Julio hardly ever takes anybody else’s word on anything related to statistics. (He has a PhD, after all.) So I was stuck until Julio ruled out every other possible explanation for the change in specificity (that is, until he ruled out the possibility that I had fucked up somewhere).

First, we checked the code that generated the specificity statistics. No problems found.

Then, just to make sure, we ran the statistics using the computations from the last version of BEAVA to see if we still got 10.47%. We did. This categorically ruled out any problems in the calculation of the specificity, which meant the problem had to be in the new score computations.

Julio, of course, assumed the problem must have been with my calculations, and not with his changes. A long ordeal of double-checking my calculations ensued. No problems were found.

At one point, Julio found that I had expressed normative error data as a percentage of correct responses instead of a number of errors. I chose to use this method based on his kevetching several months earlier about my expressing normative data as a number of errors instead of a percentage. This time, Julio declared that I should NOT have used a percentage and should have used the number of errors. Julio was convinced this was the cause of the problem. After a long time of recalculating the normative tables, there was no appreciable difference in the specificity outcome.

We spent an enormous amount of time comparing my computer-calculated numbers to Julio’s hand-calculated numbers. Any differences were examined in excruciately close detail. He found that certain numbers did not match up no matter how he checked them.

Some hours later, close to midnight, in fact, Julio declared that the reason the numbers didn’t match was because of my calculation of the normative data. (I have used these same calculations for the past six versions of BEAVA without any problems.) He insisted I fix this problem Monday. I shall do so diligently, and ensure that whenever he is looking over my shoulder that I use the “fixed” code, whilst continuing to use the CORRECT code behind his back. In any case, that “problem” was completely unrelated to the specificity problem.

In the end, Julio resigned himself to the fact that the specificity had indeed gone up because of his new computations. This realization took seven hours to reach.

I expect to receive a phone call from Julio Monday morning around 9:30-10:00, where I will have to rehash everything we did Friday night. Most likely, this conversation will take place when the Merry Maid people are running the vacuum cleaner in my ear.

Julio-style problem solving involves a great deal of panic, thinking out loud, talking to himself, walking back and forth, ridiculously unfounded assumptions, blaming everyone but himself, jumping to conclusions, and talking in gibberish.

Related