pdalgd at gmail.com
Wed Oct 19 19:18:40 CEST 2016
> On 19 Oct 2016, at 17:47 , Mike meyer <1101011 at gmx.net> wrote:
> Jf(x)'Jf(x) nonsingular, for all x, is a reasonable condition, m>=n is not.
If Jf(x) has more columns than rows, then Jf(x)'Jf(x) is certainly singular. The reverse is not true, but what's wrong with a simple pre-check?
What you possibly _could_ argue is that you want a (non-unique) solution even in the singular case. Presumably, that could give you the correct minimum sum of squares, the rank, and a degenerate variance-covariance matrix of the estimated coefficients. That could be useful, either if you just want the minimum or if you need to see the cov. matrix in order to see which parameters are unidentifiable.
Peter Dalgaard, Professor,
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Office: A 4.23
Email: pd.mes at cbs.dk Priv: PDalgd at gmail.com
More information about the R-help