Regression Anlaysis without constant term
I know how to do simple linear regression by hand and multiple linear
regression using Excel.  How do I find and solve a linear regression
equation (with one independent variable) without a constant term?

---

Request for Question Clarification by supermacman-ga on 05 Dec 2002 19:36 PST

What exactly do you mean, "constant term"? The set of data do you have
- it's probably a set of ordered pairs, i.e. (5,9) and (2,3). I don't
see where the constant term fits in... can you provide an example and
elaborate?

(Do you mean you don't have an dependent variable?)

Thanks!

---

Clarification of Question by cyntlhiadiane-ga on 05 Dec 2002 22:56 PST

I should have written:  How do I find and solve a regression equation
(with one independent variable) without a constant term?  (The
equation is probably linear but may not be.)  I believe the constant
term would be the y-intercept (as someone noted).  Yes, the pairs are
ordered.

    x y
    4 5
    3 2
    9 4

I want to do this regardless of the consequences because I will be
making a comparison to an equation with the intercept.

---

Clarification of Question by cyntlhiadiane-ga on 06 Dec 2002 10:17 PST

The problem is:  Find the regression equation without the constant
term (or intercept).  Also called "regression through the origin".

"Tools", "Data Analysis", "Regression", check "Constant is Zero"

A linear question is of the form, y = mx + c
I'm assuming you're trying to do a regression with c = 0. Right?

An extra regression tip that I've found very useful...

Often when I need to manipulate the equation, I find myself having to
plot a trendline through a graph and copying and pasting the equation
into the worksheet. And when I change the points, I have to go through
the whole rigmarole again.

The SLOPE() and INTERCEPT() function saved me from doing this.

The "constant" term simply refers to the y-intercept of the regression
line, and therefore it is impossible to calculate a regression line
without an intercept value.  Every line must pass through the y-axis.

As zhiwenchong suggested, you can force the y-intercept to be zero,
but there are consequences to doing that, such as the sum of all the
residuals will no longer equal zero (no longer least squares
regression).  Here is a link to a short .pdf file on forcing the
y-intercept to be zero.

http://www.math.nus.edu.sg/~stasq/mq3205/mq3205ln10-12.pdf

If you want more information on this topic simply search google for:
regression through origin