Linest - Why did they do that?
 

I am having difficulty with the order of the coefficients outputed by linest
when doing a multivariable regression. My problem is that the order is
reversed from the input x variables. While for displaying the coefficients I
do not have a problem, the problem is that my sheet is setup so that the
column order is x1,x2,x3,x4... and I would like to use
mmult(inputs,transpose(output coefficient)) or Sum(inputs*ouput
coefficients) ) to fill in dataspace in other operations.

Any suggestions on a way reverse the order so that I can use it with matrix
operations without creating new columns in reverse order (involves reverse
thinking on my part).

Thanks in advance
cseeton

I don't know that my solution could even be called good, but I used a
modified identiy matrix
0 0 0 0 1
0 0 0 1 0
0 0 1 0 0
0 1 0 0 0
1 0 0 0 0
to flip a column of data to that it would work with the index() function. I
simply used mmult to multiply the column with the matrix and I had the data
in the form I needed it. It will also work if the data is in a row.
Unfortunately, this was the only work around I could find in Excel.

"cseeton" wrote in message
...
I am having difficulty with the order of the coefficients outputed by
linest
when doing a multivariable regression. My problem is that the order is
reversed from the input x variables. While for displaying the
coefficients I
do not have a problem, the problem is that my sheet is setup so that the
column order is x1,x2,x3,x4... and I would like to use
mmult(inputs,transpose(output coefficient)) or Sum(inputs*ouput
coefficients) ) to fill in dataspace in other operations.

Any suggestions on a way reverse the order so that I can use it with
matrix
operations without creating new columns in reverse order (involves reverse
thinking on my part).

Thanks in advance
cseeton

You could use the INDEX() function to reverse the coefficients, as in
=INDEX(LINEST(B1:B6,A1:A6),1,{2,1})

My guess as to why LINEST works this way, is that if you are doing model
selection (such as testing whether a quadratic term is a significant
improvement over a linear fit), you can do the test without displaying
all the columns of LINEST output. Yes, that's pretty weak, and I would
have preseverd the order of input columns, but I didn't write it ...

Jerry

cseeton wrote:

I am having difficulty with the order of the coefficients outputed by linest
when doing a multivariable regression. My problem is that the order is
reversed from the input x variables. While for displaying the coefficients I
do not have a problem, the problem is that my sheet is setup so that the
column order is x1,x2,x3,x4... and I would like to use
mmult(inputs,transpose(output coefficient)) or Sum(inputs*ouput
coefficients) ) to fill in dataspace in other operations.

Any suggestions on a way reverse the order so that I can use it with matrix
operations without creating new columns in reverse order (involves reverse
thinking on my part).

Thanks in advance
cseeton