Mutiple Regression output

[happycow]

I have to do a multiple regression in excel with regression tool(with
variables). I successfully got it to work, and a lot of charts an
graphs are out putted. My question has to do with one of the charts
One of the charts has columns labled "Coefficients", what does thi
mean? I thought it was the slop (b) of the least squares line for eac
variable. However some of the values in the coefficients are negative
but if you look at the output charts the correlation go in a positiv
direction for these variables. another question i have is on the graph
on noticed that there are two types output graphic in excel what are th
differences

 

[Mike Middleton]

happycow -

I have to do a multiple regression in excel with regression tool(with 4
variables). I successfully got it to work, and a lot of charts and graphs
are out putted. My question has to do with one of the charts. One of the
charts has columns labled "Coefficients", what does this mean? <

The summary output of the Regression tool has a worksheet column
"Coefficients." (There is no chart with such a column.)

I thought it was the slop (b) of the least squares line for each variable.
However some of the values in the coefficients are negative, but if you
look at the output charts the correlation go in a positive direction for
these variables. <

In a multiple regression analysis, the coefficients are "net regression
coefficients" or "partial slopes." Each coefficient measures the average
change in Y for a unit change in an X, assuming the other X values are held
constant (or, after taking into account the other X effects).

The overall pattern of Y vs X may be positive, but after taking the other X
variables into account, the net effect may be negative.

another question i have is on the graphs on noticed that there are two
types output graphic in excel what are the differences? <

One kind of chart shows Y vs X, along with fitted Y. Another shows residuals
(Y minus fitted Y) vs X.

- Mike
www.mikemiddleton.com

 

[Conrad Carlberg]

Mike's analysis and advice is dead on, per usual. If you want more
information on MR theory (as distinct from how Excel does MR) and how betas
differ from correlations, you'll find a wonderful discussion in Kerlinger &
Pedhazur, a classic published by Holt Rinehart, 1973 -- there's a more
recent edition dated 1986, I think.

BTW, I too have long felt that the coefficients were the slop of the least
squares line (g).

C^2
Conrad Carlberg