Chocolate Versus Vanilla


Roscoe showed up at my house awhile back with a copy of USA Today under his arm. "Looks like even McPaper is starting to understand," he said, knowing I would be unable to resist taking the bait. When I asked him what he was talking about, he showed me one of the little squibs in the corner of the page: 52 percent of the population prefers vanilla ice cream, 48 percent chocolate. From my perspective, that registered pretty high on the "So what?" meter.

Then Roscoe offered the following. "Notice what it says below the little bar graph: Accuracy is plus or minus 3 percent.[2] Now that is significant. For years, they published these factoids without any indication of the error bars. We must be getting less stupid about polls, if they think they have to tell us the likely error."

[2] Sometimes the term precision is used instead, but precision and accuracy are not the same. Precision indicates the degree to which the measurement is reproducible. If you have a high precision measurement method and measure many times, most of the measurements will fall within narrow margins. On the other hand, accuracy is a measure of how close the measurement is to reality. Precision talks about measurements on the sample; accuracy reflects how close your measurement of the sample is to defining the properties of, in this case, the underlying population.

Roscoe may appear to be sloppy in his use of these terms, but he is not; we shouldn't be either. Shooting for high precision is silly in the face of low accuracy, because high precision costs money, and we should never pay lots of money for an inaccurate result. Just always remember that high precision does not necessarily imply high accuracy.

I had to admit I had never thought about it that way. I also was intrigued by why Roscoe thought it was interesting.




The Software Development Edge(c) Essays on Managing Successful Projects
The Software Development Edge(c) Essays on Managing Successful Projects
ISBN: N/A
EAN: N/A
Year: 2006
Pages: 269

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net