Winsteps
Sunday, April 02, 2006
Simple Rules for a Complex World
Richard Epstein's book, "Simple Rules for a Complex World" suggests that the legal system is overly complex. "Each additional lawyer reduces U.S. gross domestic product by $2.5 million."
We statisticians have also been guilty of adding excessive complexity to the world. The road to fame as a statistician is to have a new statistical index named after you, so there is great incentive to invent a new one - and then to have it reported by a major statistical software package.
Rasch measurement is not immune to this. We are encouraged to perform power analyes and test null hypotheses that we know can't be true. Real data never fit any theoretical model perfectly. If yours seem to, then you haven't collected enough data yet! So, the only simple rule that really matters is "Does the data fit the Rasch model well enough for practical purposes?" It's the same rule we apply when measuring our own heights and weights. We know that our height lessens during our waking day as we are compressed by gravity, and then increases during our nightly repose. We know that our weight decreases prior to meals and increases during them. But, for all our practical purposes, those instabilities don't matter. So it is with the Rasch model. Instabilities in the data don't usually matter unless they become huge (mean-squares > 2.0).
We statisticians have also been guilty of adding excessive complexity to the world. The road to fame as a statistician is to have a new statistical index named after you, so there is great incentive to invent a new one - and then to have it reported by a major statistical software package.
Rasch measurement is not immune to this. We are encouraged to perform power analyes and test null hypotheses that we know can't be true. Real data never fit any theoretical model perfectly. If yours seem to, then you haven't collected enough data yet! So, the only simple rule that really matters is "Does the data fit the Rasch model well enough for practical purposes?" It's the same rule we apply when measuring our own heights and weights. We know that our height lessens during our waking day as we are compressed by gravity, and then increases during our nightly repose. We know that our weight decreases prior to meals and increases during them. But, for all our practical purposes, those instabilities don't matter. So it is with the Rasch model. Instabilities in the data don't usually matter unless they become huge (mean-squares > 2.0).
