You would use multiple regression statistics when you are pretty sure that a set of data is a function of more than one variable, or can be better represented by a polynomial than by other functions. for example, the cost of owning an automobile can be represented as a function of initial price, time, distance traveled, hours of engine running time, and number of engine starts. Regressing on each if these variables independently will not give you a true picture of othe cost dependency, but a multiple linear regression will yield a clearer picture.
You can postulate that
F = a + bw + cx + dy + ez
where
F = running total of money spent
a = initial cost (including "hidden" initial costs of other variables)
w = mileage
x = time of ownership
y = engine hours (from an hour meter), and
z = engine starts (from a counter on the ignition)
You build your database, of course, by recording all the variables each time you have an expenditure, along with the expenditure.
solution of the regression is as follows:
| n.. ∑w... ∑x... ∑y... ∑z.. . | |a| |∑F|
|∑w ∑w^2 ∑wx. ∑wy. ∑wz.. | |b| |∑wF|
|∑x. ∑wx.. ∑x^2 ∑xy.. ∑xz.. | |c| |∑xF|
|∑y. ∑wy. ∑xy.. ∑y^2 ∑yz.. | |d| |∑yF|
|∑z. ∑wz.. ∑xz.. ∑yz.. ∑z^2 | |e| |∑eF|
This sets up easily on a spreadsheet for solution by elimination.
So far, I've not come up with an analog for the correlation factor, so I add columns to the spreadsheet to compare predicted values to actual values, and take the standard deviation of the error as a sort of measure of fit.
Graphing and visualizations are hard to come by.
edit:
Sorry, the SPSS didn't register until after I typed all the above drivel.