Forum: developers
Monitor Forum | 
Start New Thread
| predicting with treatment models [ Reply ] By: Ott Toomet on 2015-02-22 23:42 | [forum:41947] |
|
What would be a good way to predict treatment model outcomes? predict(model, "outcome", "conditional") two sets of predictions, for yS=0 and yS=1. However, as now yS is also an explanatory variable, we have not just to correct for correlated disturbances but also write yS=0/1 according to the participation? predict(model, "outcome", "unconditional") should we use the expected yS value (expected probability) as the explanatory variable for outcome? In this case predict will rather return "expected value", not "predicted value". I am willing to "predict" means to answer just 1 or 0--participate or not... However, glm probit prediction uses that approach if type="response", i.e. it will give expected probabilities... In case of probit it is easy to convert the expected probabilities into 0/1. In case of treatment regression it is slightly more complex, as these values are multplied by the corresponding betas and added to the rest of linear predictor... Maybe to introduce a third prediction type? Should look at other similar solutions... |
|
