This question already has an answer here:

In real world example I have elemental composition data for groundwater, which is known to be coming from two sources. I have control data where all predictors are from source A and data where all predictors are from source B. I would like to be able to take an unknown water sample and predict what percentage is source A vs. B based on the elemental composition of that sample.

The problem is not obvious to me because it is not a categorical outcome, but rather the relative contribution of two potential categories - if that makes sense.

I've looked for similar questions but couldn't find one - be gentle! Thanks


marked as duplicate by Michael Chernick, gung - Reinstate Monica regression Feb 16 '17 at 0:39

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • $\begingroup$ You can make artificial training set, by taking an example from A and an example from B and making their linear combination. $\endgroup$ – user31264 Feb 15 '17 at 1:37
  • $\begingroup$ You can use a binary logit model. See this related post. $\endgroup$ – Tim Feb 15 '17 at 3:58
  • $\begingroup$ Thanks Tim, i'll take a look at that option and see if it will be appropriate. Definitely couldn't find that before i posted! $\endgroup$ – Marty Feb 15 '17 at 5:55
  • $\begingroup$ This does not appear to be about a fraction of two counts but a continuous proportion of 2 positive real numbers. $\endgroup$ – gung - Reinstate Monica Feb 16 '17 at 0:39
  • $\begingroup$ Do you then anticipate that the method that Tim referenced would work? $\endgroup$ – Marty Feb 16 '17 at 5:49