I run a multiple regression using the Least Squares Method that showed good fit although one of the variables showed a weak power of explanation. When running a simple regression with the same variable, the influence of this shows high significance. Why does it happen?
That means some other variable[s] in the multiple regression model is accounting for the same variance that your variable of interest was accounting for; the variable of interest does not uniquely account for much variance. Identifying these kinds of relationships (i.e. cases where your variable of interest does or does not have a substantial effect over and above the other variables) is one of the main purposes of multiple regression
There are many, many factors which could influence this. The previous 2 answers are both quality responses...to give additional feedback, I have some questions for you first.
1) How many data points are you working with?
2) Is the dependent variable a continuous variable, ordinal variable, or a binary variable? Following from that, are you analyzing your data with linear regression, logistic regression, or some other model?
3) Is the independent variable in question a continuous variable or categorical variable? If it's a categorical variable, are you actually analyzing it as a categorical variable? Don't laugh - I have seen many people include a categorical variable in a model, but they unwittingly treat it as a continuous variable (because the variable is coded as 1, 2, 3, 4, etc...)
4) How many other independent variables are you adding to the multiple regression model? Again, often I see people include FAR more independent variables than their sample size can support. If you have 12 data points and your multiple-regression model includes 7 variables, I have your answer for you.
5) Finally, and perhaps most importantly, what is the real PURPOSE of your multiple-regression model? As Stephen alludes above, broadly speaking we often use multiple regression to tease apart whether a certain variable has an effect on the dependent variable "over and above" the other variables - for example, showing that pulmonary hypertension is truly an "independent" predictor of cardiovascular disease mortality when accounting for confounders such as age, obesity, etc. I've seen many cases where a PI does not even understand what they're trying to do in their own model. So plainly speaking, give us a flavor of your research question, and explain why you're using the multiple regression.
Fleshing out these questions will give us a better ability to determine what may be causing this in your specific case.
- Badges
- Science topic