Home > Statistics Every Writer Should Know > The Stats Board > Discusssion
Multiple Regression
Message posted by Gavin DevenMuieres on May 19, 2000 at 12:00 AM (ET)
1. When I performed multiple regression I got coefficient of constants which are not significant. How do I interprete this ?
2. I f a coefficient of a varible is negative but not significant what does it mean? how to interprete it?
READERS RESPOND:
(In chronological order. Most recent at the bottom.)
Re: Multiple Regression
Message posted by Phil on May 20, 2000 at 12:00 AM (ET)
If a variable is not significant you can omit it from the model. The sign of the coefficient is not important. However, if you are going to use your model to predict a value people will use, it helps to use a model that is intuitively correct. While "mathematically" sound, a negative coeficient can make the regression equation look suspicious to others. I had that happen once when I created a formula to estimate labor standards. After my first pass at it, I never released a formula with a negative coeficient again.
Re: Multiple Regression
Message posted by phil on May 20, 2000 at 12:00 AM (ET)
I just wanted to clarify my previous post with respect to negative coeficients. Many time you get a choice of models to use if used the "all possible models" feature of a software package. Many of the models have a very similar r-squared and other properties. I was only saying that some of these models have more intuitive appeal than others. Mathematically a negative coefficient is not an issue as long as you are predicting with the ranges of the x's used in the original data set. You should always examine the residuals to see if the model you choose has a "good fit" or not.
Re: Multiple Regression
Message posted by Gavin on May 20, 2000 at 12:00 AM (ET)
Thanks for your input.
Is it possible to use 'all possible models'command in SPSS? If so how does one go about it?
Re: Multiple Regression
Message posted by DONALD MCKENZIE on May 21, 2000 at 12:00 AM (ET)
y=aB1P B2AD - DO YOU THINK (Y) IS LINEARLY RELATED TO (P)AND (AD)?
Your $5 contribution helps cover part the $500 annual cost of keeping this site online.
