[quote=cheesehead;199649]
If 1/ssd(set) is not an appropriate weighting for the sample sets when combined, if what one wants is for each sample is that it have more weight if it has a smaller standard deviation, what would a more appropriate weighting be, [/quote]
IIRC the most appropriate weighting is 1/variance rather than 1/sd

And variance of sum = sum of variances

[QUOTE=cheesehead;199649]If one combines these three sets of samples, with equal weighting, into one sample set A+B+C with n(A+B+C) = 3*n(A), what is m(A+B+C) and ssd(A+B+C)?[/QUOTE]

From the definitions I compute:

m(A+B+C) = (m(A) + m(B) + m(C))/3
ssd(A+B+C) = sqrt( (ssd(A)^2 + ssd(B)^2 + ssd(C)^2)*(n-1)/(3n-1) )

[QUOTE=davieddy;199650]IIRC the most appropriate weighting is 1/variance rather than 1/sd[/QUOTE]

It depends on what he wants, doesn't it? Weighting with 1/ssd gives a dimensionless number, which is necessary to obtain e.g. the correlation ceofficient cov(x,y)/(sd(x)*sd(y)), etc.

[quote=__HRB__;199651]
It depends on what he wants, doesn't it? [/quote]

As Basil Fawlty said to Manuel when asking him to
remove the dead pigeon from the water tank before the hotel
inspectors arrived: "This isn't a proposition from Wittgenstein".

What he wants is a weighted mean with the smallest variance (and sd).

[QUOTE=davieddy;199659]As Basil Fawlty said to Manuel when asking him to
remove the dead pigeon from the water tank before the hotel
inspectors arrived: "This isn't a proposition from Wittgenstein".

What he wants is a weighted mean with the smallest variance (and sd).[/QUOTE]

Too much butter 1-2-3. :-)

Max (you, too, Ernst, if you wish),

A hypothetical:

Suppose there were a measured 1% increase, over a span of 50 years, in average energy, per unit of time, received by Earth's surface from the sun, with no other change (other than Earth's natural reactions to the increase, that is -- e.g., there could be an increase in the average water vapor content because of the warming, but no increase in anthropogenic GHGs, no change in the way climate subsystems operate, and so on).

Would you agree that that increase could be sufficient to cause measurable global warming (from the beginning of the 50-year period to the end) of 1 degree?

For this exercise, assume Earth's current average surface temperature at the beginning of the 50-year period was 15 degrees C = 288 Kelvin. (That is approximately what's shown at [URL]http://www.grida.no/publications/vg/climate/page/3070.aspx[/URL] for the 1961-1990 average.) So, I'm asking whether you think an increase from 15 to 16 degrees C would be a reasonable result of the 1% increase in solar energy flux at Earth's surface, given 50 years to adjust.

If you don't like my 1-degree hypothetical resultant increase in temperature, what amount of warming _do_ you think a 1% increase in solar energy flux could reasonably cause? I'm not asking for some exact calculation -- just your estimate of how much increase in average temperature the 1% energy flux increase might reasonably cause -- a figure you're comfortable with until/unless you have data that indicates otherwise.

[quote=davieddy;199659]As Basil Fawlty said to Manuel when asking him to remove the dead pigeon from the water tank before the hotel inspectors arrived: "This isn't a proposition from Wittgenstein".

What he wants is a weighted mean with the smallest variance (and sd).[/quote]Thank you all for your answers. In the meantime, I've abandoned the line of inquiry for which I asked about the statistics, so there won't be any followup.

[quote=__HRB__;199664]Too much butter 1-2-3. :-)[/quote]
?Que?

[quote=cheesehead;199712]Thank you all for your answers. In the meantime, I've abandoned the line of inquiry for which I asked about the statistics, so there won't be any followup.[/quote]
Your problem is lack of persistence:smile:

Since appropriate handling of data is very much on
topic for this thread, I shall take the opportunity to
refresh myself on the elementary principles involved.

Don't be deceived by the gay banter.

If you consider a large sample as comprising many small
samples of size n, it is only appropriate to give them equal weight
when "combining" them. The reducto ad absurdum is to consider
n=2 or 1.

More anon.
Feeling no pain: the local Paki shop is open all Christmas:smile:

David

[quote=davieddy;199791]Your problem is lack of persistence:smile:

Since appropriate handling of data is very much on
topic for this thread, I shall take the opportunity to
refresh myself on the elementary principles involved.

Don't be deceived by the gay banter.[/quote]I apologize for the brevity of my "abandoned the line of inquiry" post, which, I now see, could easily mislead a reader as to what I meant.

Thank you, David, for your kind advice.

However, what I had in mind (but failed to explain) was only one narrow line of inquiry I was following for research on a potential future discussion of one particular piece of evidence. It didn't concern anything I'd already written in this thread. I was thinking about drawing an analogy that would involve statistics. Then I discovered during further reading that the analogy idea I had would not be useful, so I won't pursue the statistics further now.

[QUOTE=davieddy;199784]?Que?[/QUOTE]

Stupidissimo! Continental cretin!

[quote=__HRB__;199862]Stupidissimo! Continental cretin![/quote]
I believe that in the Spanish version Manuel is Italian.