Abstract
Treatment of collinearity in multiple regression analysis. Collinearity among predictors in a regression model is a very frequent problem, specially in Human Sciences. There are several procedures for diagnosing collinearity, but it cannot be easily solved. However, if it is caused because wrong data or observations were collected, then it is possible to omit them and, in this way, the problem is automatically solved. To introduce new data or to select another subgroup of predictors is perhaps the best solution, but this procedure is not always possible to apply because of experimental setting. However, there are some alternative methods which allow to use previous information and to explain a similar percentage of response variability (or even greater than the preceding one). In this work, we use -among other procedures- Principal Component Analysis and Ridge Regression. We remark their implications for collinearity treatment as a consequence of their mathematic properties and, simultaneously, we expose which are the advantages and disadvantages when these procedures are used.
