The FM Method: Covariance
Must be normalized by the variance in x.
...but the variance in x will increase with εx (e.g. if our weight scale is bad), so subtract off for that.
If the dispersion error is zero and Mbh error constant, FM and Nuker give same solutions!
Allows for intrinsic dispersion (covariance term need not be 1)
There could be an underweight/overtall and an overweight/undertall population
β goes to zero for zero covariance