Data Analysis help?

S

sherifffruitfly

Hi all,

I've got some time series data which is approximately constant for
awhile, and then drops off as -(a exp(b x)). Is there a way I can have
Excel help me to determine what the "best" place to consider the
constant section as stopping and the exponential section as beginning?

It should be straighforward to determine the exponenetial's parameters
after that.


Thanks for any ideas,

cdj
 
G

Gary''s Student

Rather than considering the range as the spline of two separate processes,
can you construct a single mathematical model that covers both parts of your
data?

If it is possible, then your data analysis will be independent of an
arbitrary separation point.
 
M

Mike Middleton

cdj or sherifffruitfly -

If you can specify a numerical way of determining what you regard as "best,"
then there may be a way to use Excel to solve your problem.

But you might find it easier to "look at the data" and apply your judgment
directly.

And you might include your judgment about external events that may have
affected the data.

- Mike
http://www.mikemiddleton.com
 
S

sherifffruitfly

cdjor sherifffruitfly -

If you can specify a numerical way of determining what you regard as "best,"
then there may be a way to use Excel to solve your problem.

But you might find it easier to "look at the data" and apply your judgment
directly.

And you might include your judgment about external events that may have
affected the data.
Click to expand...

The data is a time series - I wasn't around back then to have any
feelings about when the change might have formally taken place.

Doing something like a "running chi-square test", occurred to me,
something along the following lines:

(a) from left to right, loop through the independent variable, x.
(b) each x splits the data into "the left half" and "the right half".
(c) with the left half of the data, calculate its linear regression
model.
(d) with the right half of the data, calculate its exponential
regression model
(e) use chi-square on the Current Point to decide which model it
belongs to.
(f) when I get a certain number in a row that chi-square says belong
to the exponential model, stop

If this is a reasonable thing to do, then I'm left with the tasks:
(1) calculating a linear model based upon a col of data
(2) calculating an exponential model based upon a col of data
(3) constructing a chi-square test deciding between two models.

Time to hit the excel book. lol!

Thanks for your reply,

cdj
 
Top