2
$\begingroup$

I have 2 predictor variables which both have a low correlation value(0.007 and -0.017) with the response variable. What does this low correlation tell me?

Can I somehow alter the predictor variables by taking their log or some other function to increase the correlation value?

Improve this question
$\endgroup$
1
  • 3
    $\begingroup$ This information tells you nothing at all about the ability to predict the response based on both variables. It would be far more informative to regress the response on both variables in a multiple regression and then study the residuals. Searching our site for related keywords, including "residual" and "multiple regression" will turn up a great deal of advice, examples, and explanation. At the very least you should be looking at the data: use a scatterplot matrix of all three variables. $\endgroup$
    – whuber ♦
    Commented Nov 17, 2017 at 14:44

2 Answers 2

Reset to default
0
$\begingroup$

You should think about the process it is that you are modelling and its underlying mechanisms. The two predictors you have chosen appear to not have much of an influence on the modelled response. Transforming your predictors is a strategy for mutating data into a distribution required by a particular model. Are there other predictors which you can obtain and explore that may describe the process more accurately?

Improve this answer
$\endgroup$
2
  • $\begingroup$ There are no other predictors I can use else but ones I can make up from variations of the two given predictor variables. This is for a project were all I have been given is 1000 data points with the response variable and 2 predictor variables which were generated from a non-linear function plus an error term which is normally distributed. $\endgroup$
    – Jack
    Commented Nov 17, 2017 at 13:42
  • $\begingroup$ Is it a regression or classification problem? $\endgroup$
    – Seanosapien
    Commented Nov 17, 2017 at 21:24
0
$\begingroup$

In many circumstances that I have encountered unusual values lead to a down-weighting of the regression structure providing a false conclusion about significance. Often times empirical detection of unusual values can lead to a clearer picture of the relationship between variables as the standard error of coefficients can be markedly reduced. This is true for different kinds of data including longitudinal data.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.