BBS: Inland Empire Archive
Date: 04-14-92 (14:11) Number: 196
From: DON KROUSE Refer#: NONE
To: JEAN CREPEAU Recvd: NO
Subj: Regression Conf: (2) Quik_Bas
JC>DK=> Does anyone know were I might find source code for a multiple
JC> regression routine?
JC>DK=> I am particularly interested in a routine that does Linear, Log,
JC> Exponential, Power, etc. and that optimizes the order in which the
JC> dependent variables are included.
JC>Use the following formula
JC> m=(n*SUM(x*y) - SUM(x)*(SUM(y))/(n*SUM(x*x) - SUM(x)*SUM(x))
JC> b=(SUM(y)-m*SUM(x))/n
JC> where n=# of samples
JC> Your linear equation is y=mx+b
Hi Jean,
Thank you for taking the time to answer my request for help. Your approach
and code are "very good" for simple regression.
What I am looking for though is a "Multiple Regression" routine. That is a
routine that will determine the regression formula for a dependent
variable with multiple independent variables.
Silly Example:
y = Store profit
x1 = Store sq. ft
x2 = no. of employees
x3 = Store hours
x4 = Sales
The data is used to calculate the summary statistics, i.e: sum(y),
sum(x1), sum(x2), sum(x1y), etc.
The summary statistics are used in a set of "normal equations" which
much be solved simultaneously to determine the regression coefficients.
The linear regression equation would be similar to the one that you show
above.
y = m1x1 + m2x2 + m3x3 + m4x4 + b
PROBLEMS:
- The number of summary statistics that must be computed grows
unmanagable as predictors are added.
- The order in which the predictor variables are included in the
regression affects how well the regression line will fit the data.
- The number of normal equations which must be solved grows as predictors
are added. i.e. no. of equations = no. of predictors + 1
- For non linear regression a method of simultaneously solving non-linear
equations is necessary.
I'm trying to write a business program not a stastics program.
This would be an important part but not the essence of the program.
If I have to do all of this from scratch I'll probably put it off for
a long time. (Forever's a long time isn't it?)
I believe that BASIC is/was used by a lot of scientific types who must
have crossed this bridge before. I was hoping to find someone who might
have or know of some off the shelf code.
It might not sound like it, but I really do appreciate your taking the
time to help.
Don
---
* Origin: Silver Lake Systems * Bothell, WA (1:343/58)
