Are we becoming more tolerant of nepotism?
Republicans have a 54 percent chance of taking the Senate
The denominator fallacy rears its ugly head
How better educated whites are driving political polarization
Controversial claims about marriage promotion break the statistical rules of evidence
I was puzzled by that 54% post. One often simplifies a model by leaving out some effect (e.g. friction) that has a known sign but an essentially unknown magnitude. The predictions of models like this are stringent on one side but loose on the other; perhaps they are better thought of as bounds than estimates: if the data are off to one side that just gives you an estimate of the friction coefficient, if they are off to the other side then there’s something wrong with your model. (I thought John Sides’s case was somewhat similar.) Are you claiming that the right thing to do in these cases is to make up a range of friction coefficients so that you can symmetrize your estimate? This doesn’t make sense to me.
Sarang:
John was giving two estimates, 44% (based on all the data) and 64% (based only on recent data). He presented reasons for each. I think it reasonable, based on this information, to consider John’s forecast to be something in between. His forecasts are not bounds in any sense; they are just forecasts from two different regression models.
A rather sad case of denominator fallacy:
In the recent discussion in of the criticism of the UN towards the Vatican, the director of a research institute, Chrisitian Pfeiffer, is quoted with an estimation that only 0,1% – 0,3% of the cases of sexual abuse of children over the last 15 were committed by catholic priests (numbers refer to Germany). According to Pfeiffer, the catholic church does not have a “quantitative problem” (cf., http://www.sueddeutsche.de/politik/missbrauch-in-der-katholischen-kirche-drei-promille-aller-taeter-1.24359).
However, if you take into account that catholic priests represent only 0.05% of the male adult population in Germany, then you end up with a relative risk around 2 (to 6 for the higher estimate).
Does look like a quantitative problem to me.
But you’re committing the very same fallacy! The appropriate denominator is not the male adult population as a whole, but rather the population that has contact with children (teachers, scout leaders, etc).
This is probably not the right forum to discuss it.
But anyway:
I’m not on expert on the subject so I didn’t assume any theory. And, of course, there can be some debate what would be the most informed denomitator. However, your suggestion makes some theoretical assumptions that I seriously doubt. And also empirically there seem to be no hints in this direction. As far as I know, the biggest part is friends and relatives of the family.