View Single Post
  #8   Report Post  
Biff
 
Posts: n/a
Default

Hi!

Let me see if I can come up with something.

Biff

"Tom" wrote in message
...
Biff,
Thanks AGAIN!
One easy way of weighting is to approach a 10-interval set of data as
follows [each interval is the same difference from the previous or next]:
Latest: (1.0) * value(Latest)
Latest-1: (1.0-.1) * value(Latest-1)
Latest-2: (1.0-.2) * value(Latest-2)
....
Latest-9: (1.0-.9) * value(Latest-9)

You can see that the weighting is based off the most recent measurement
times the value at the time of THAT measurement [value(Latest-i)]. Hence,
any weighting would have to take into account A) the number of
measurements
in a set of intervals [# of intervals] and B) the values during each
interval
[value(Latest-i)].

There are a lot of different ways to obtain a weighting factor, (1.0-i) in
the example above, so any method of calculating the weighting factor
should
keep this in mind. That is, you could some up with logarithmic weighting
factor which puts more weight on the first few intervals rather than the
assigning each weighting factor the same DIFFERENCE from the previous or
next
weighting factor like the above example. Does this help!?!?
TIA


"Biff" wrote:

Hi!

Do I have to create my own formula for my 'own' average methodology
(non-standard)?


Well, that would fall into the category of statistics which I really
don't
know much about!

Coming up with a "roll your own" weighting formula and applying it
against
the data should be easy enough. Can you give an example of what the
criteria
would be? I understand latest has greater weight, but how do you want to
weight that against the oldest?

Biff

"Tom" wrote in message
...
Great help, thanks!
Now, what if I'd like to use some exponential or weighted averaging?
That
is, in exponential/weighted averaging the most recent value is of great
weight than an older value. In the '10-day temperature' example,
yesterday's
temperature of 90-degrees is better and weighted more than the
temperature
of
58-degrees of ten days ago.
How do I do the exponential/weighted average? Do I have to create my
own
formula for my 'own' average methodology (non-standard)?
Thanks

"Biff" wrote:

Ooops!

Correction:

=AVERAGE(OFFSET(A$1,(ROW(1:1)-1)*10,,E$1))

Should be:

=AVERAGE(OFFSET(A$1,(ROW(1:1)-1)*E$1,,E$1))

Biff

"Biff" wrote in message
...
Hi!

Assume your values to average are in column A, starting in A1.

Enter the interval you want to use in a cell, say, E1.

E1 = 10

=AVERAGE(OFFSET(A$1,(ROW(1:1)-1)*10,,E$1))

Copy down until you get #DIV/0! errors meaning the data has been
exhausted.

Depending on the interval size, the last average may not be a full
interval.

Biff

"Tom" wrote in message
...
I have a need to calculate a 10-interval vs x-interval moving
average.
Without totaling the last '10' measurements and dividing by 10 to
get
an
average and then comparing it to, say, '13' by totaling the last 13
and
dividing by 13, how can I make the calculations vary by the '# of
intervals'?
For example, if I want to average the high temperatures from Jan
1-Dec
31, I
have 365 intervals. If I want to compare the 10 day average vs a
13
day
average I can establish the formula pretty easy for the SPECIFIC
number
of
days. I want to vary the # days, in this example, so I can see if
I
can
spot
trends in the data.
So how do I do the GENERAL case of x-intervals?
TIA