Finally I have found where an error was made in both the 8th and 9th editions, regarding lineal regression equations. In table 4.8 (page 50, 9th edition) the 3rd column (XY), was the sum of multipling the previous 2 columns. Column Y should have been divided by column X. That would have left a value of 17.156 in the "totals" row of 17.156, averaged out over 20 trees it is .8578, the q referred to in example 1a on page 51.
I have been confused by that for the longest time, today the light came on!!!!!

Yippeeee, James

Reply to post by James Causton, on June 25, 2002 at 20:37:37:

Might be a nice module to have included in the software Russ!!

James

Reply to post by James Causton, on June 25, 2002 at 20:37:37:

OK, at the risk of embarassing myself due to some very rusty math skills, let's look at this.

Regression analysis looks at correlations between related (or unrelated) factors to determine how well they correlate. Linear regression looks at the first level, for a straight comparison. It compares the two values against a straight line.

The formula for a linear regression determines where the line goes. There are two parts- the slope and the intercept. If we know these, we know exactly where the line is. This is then used to estimate, with some degrees of certainty, what the correlation is, so if you know one of the variables (independent variable) you can estimate what the other should be. To determine the slope (m) divide the sum of XY by the sum of XX (x squared). If you do this, you will note that it is close to the numbers given in the Example in the Guide, but not exactly the same. this is why you need XY (not X/Y).

Now using the slope m, you reduce X (independent variable) to zero and determine the intercept.

This only gives an estimate of the value of Y when X is known, but it will be the best statistical estimate without any other data. The certainty of the fit can also be calculated, so you can say that if X is this, we know Y is that, within a certain degree of error.

As I mentioned in my other post in this thread, there may be a better curve fit to stump/trunk measurements than a straight line. To get the best curve fit, a multiple regression analysis would be necessary, and that gets more complex (and beyond my rusty ol' brain right now).

Reply to post by Russ Carlson, on June 25, 2002 at 20:37:37:

Oops. I got my threads crossed (a bad thing to do for right-threaded fellow). My comments on regression were in the 'Linear Regression' thread in the Valuation topic section.