[Numpy-discussion] weighted mean; weighted standard error of the mean (sem)

[josef.pktd at gmail.com]

On Thu, Sep 9, 2010 at 10:22 PM, cpblpublic <cpblpublic+numpy at gmail.com> wrote:
>
>
> I am looking for some reaally basic statistical tools. I have some
> sample data, some sample weights for those measurements, and I want to
> calculate a mean and a standard error of the mean.
>
> Here are obvious places to look:
>
> numpy
> scipy.stats
> statsmodels
>
> It seems to me that numpy's "mean" and "average" functions have their
> names backwards. That is, often a mean is defined more generally than
> average, and includes the possibility of weighting, but in this case
> it is "average" that has a weights argument. Can these functions be
> merged/renamed/deprecated in the future?  It's clear to me that "mean"
> should allow for weights.

I think of weighted mean and weighted average, pretty much as
synonyms, changing names would break backwards compatibility without
any real benefit, in my opinion.

>
> None of these modules, above, offer standard error of the mean which
> incorporates weights. scipy.stats.sem() doesn't, and that's the closest
> thing. numpy's "var" doesn't allow weights.
> There aren't any weighted variances in the above modules.

for weighted statistics, I usually refer to ticket 604

http://projects.scipy.org/scipy/attachment/ticket/604/Statistics.py

but I didn't see weighted sem in it

>
> Again, are there favoured codes for these functions? Can they be
> incorporated appropriately in the future?
>
> Most immediately, I'd love to get code for weighted sem. I'll write it
> otherwise, but it might be crude and dumb...

just a thought, I still have to check the details:

Estimating statsmodels.WLS with just a constant should give all the
result statistics on the mean, e.g. bse for variance of constant, t()
for t-statistic

Josef


>
> Thanks!
> Chris Barrington-Leigh
> UBC
>
> _______________________________________________
> NumPy-Discussion mailing list
> NumPy-Discussion at scipy.org
> http://mail.scipy.org/mailman/listinfo/numpy-discussion
>
>