I am struggling in finding a way to assign a 'weight' (if possible) to each of the independent variables in a logistic regression (I am using R). I explain my problem in a more detailed way:
I am working on a dataset of 900 samples. About half of them are affected with a disease, the other half are healthy (my dependent variable is indeed their 'status': disease or healthy). For each sample, I checked the presence of a certain type of variation in 32 different genes, that I intend to use as predictors in my regression.
The problem is that I don't want to treat these 32 predictors in an equal way: some genes, in fact, when harbouring variation, are more likely to cause the disease than others. I already have 32 pre-computed scores coming from a previous study that I would like to use as 'predictor weights' a priori in the model. Is there any way to do so?

  • 1
    $\begingroup$ Regression is about estimating those weights. If you already know the weights, there is nothing for regression to do. $\endgroup$ – Peter Flom Jan 21 '14 at 18:31
  • 1
    $\begingroup$ Maybe you want to do something Bayesian? The results from the previous study would be used to set your prior for the parameters in the regression, and then you update the priors with the data from the new study. $\endgroup$ – Bill Jan 21 '14 at 18:33
  • $\begingroup$ If the initial fit is $Ax = b, A$ 900 x 32, tack on a 32 x 32 matrix $\lambda I$ to $A$, and 32 x 1 weights to $b$, e.g. $\lambda [w_0 x_0, w_1 x_1 ... ]$. This minimizes a sum of two terms: fit to $b$, and $\lambda *$ fit to $w * x$. Look at the balance of these two terms, and play with $\lambda$. $\endgroup$ – denis Oct 27 '15 at 15:55

Your Answer

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

Browse other questions tagged or ask your own question.