Reply #15: I don't weight the polls... [View All]

[Celebrandil]

they have equal weights, but once a low-dimensional curve is fitted to the data, the deviations between polls and the curve are measured.

However, after this it's possible to apply an M-estimator and curve fitting once and then you use different weights for different polls, but the weights are set automatically from the initial deviations. This is standard robust statistics.

Nor professor yet... just PhD in computer science