5
$\begingroup$

I am using a scale which consists of discrete values 0 (normal), 1 (mild), 2 (moderate), 3 (severe). I have used this scale for 200 patients.

I am going to find the correlation of this scale with some continuous variables, these variables also measured for 200 patients.

I don't think I can use multivariate or multiple regression because the dependent variable is discrete and the independent variables continuous. Can you please suggest a method to find this correlation?

$\endgroup$
1
  • 3
    $\begingroup$ Have you thought instead about using techniques that are more powerful and insightful than a correlation coefficient? For instance, have you drawn side-by-side boxplots of the continuous variables, partitioned by their position on your scale? Such graphics will not only display the nature of the association, they can also suggest ways to re-express both the continuous values and the numeric codes in your scale to make subsequent linear modeling more applicable. $\endgroup$
    – whuber
    Jun 27 '13 at 14:49
7
$\begingroup$

The discreteness is not an issue, so much as the ordinal (ordered, graded) scale used for your assessment from normal to severe. That indeed implies something different from standard linear regression, namely some ordinal regression method such as ordered logit or ordered probit.

Note incidentally that multivariate regression is not the same as multiple regression.

$\endgroup$
4
  • $\begingroup$ I am going to generate these discrete values by machine learning algorithm and they are the target . The continues variables are the features for machine learning. I was thinking maybe Spearsman-correlation or regression can be a good method to find the more effective features in predictor of discrete values. I have to know which features can be the best option ? $\endgroup$
    – user27379
    Jul 10 '13 at 13:55
  • $\begingroup$ This comment does not illuminate your question further so far as I am concerned. If your data are synthetic, what is there is to discover? Standard correlation and regression are poor methods for a graded or ordinal response with four distinct levels. $\endgroup$
    – Nick Cox
    Jul 10 '13 at 14:03
  • $\begingroup$ 0(normal), 1 (mild), 2 (moderate), 3 (severe) and these values are for 4 components. Each patients has four discrete numbers. I am going to find the correlation of this scale with some continuous variables, these continues variables also measured for 200 patients. Target: To generate these discrete values by machine-learning algorithm based on continues variables. 1.what is best method for feature selection ( continues variable). Spearsamn-Correlation? If the continues variables were highly correlated with discrete values. it means I should consider them as features? $\endgroup$
    – user27379
    Jul 10 '13 at 14:25
  • $\begingroup$ The same question implies the same answer. I recommend to you some ordinal regression method such as ordered logit or ordered probit. $\endgroup$
    – Nick Cox
    Jul 10 '13 at 14:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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