determine the weight of a new attribute I have a dataset about the automobile credits. This data set contains information that I care about when giving credit. For instance age, year of experience, income etc. And my database includes values of them and i can use multiple regression analysis to give weight for each attribute. My formula can interpret them because i have their history. 
My question is here.
I want to add new attribute to dataset. Let we say it to X. But my dataset doesn't include any history of X and i want to add it. X represents any emergency situation. So it can not give any coefficient for it. But i want to new equation with X.  How can i solve it ? How can i give coefficient for X ?
 A: It sounds as if you want to quantify the effect of X without having any data on X.  Perhaps these ideas will help.
Option 1:  Are you looking to include in the equation some estimate of the effect of X?  If so, you could, for a small portion of the dataset, simulate each case's X value, inserting values that would be plausible given the corresponding values of all the other variables.  If X is unrelated to the other predictors, then compute X as some function of the outcome. Then run the regression. You'll obtain a simulated X coefficient as well as simulated coefficients for the other predictors.  (You'll probably want to do this many times to see how much results can vary, how sensitive they are to your choice of simulated values.)
Option 2:  Do you already have an idea of the magnitude of the effect of X?  If so, leave the regression as it is, but allow for X's role as a source of uncertainty by increasing the confidence intervals around each predicted value (prediction intervals), by an amount that seems appropriate to you.  Then, to evaluate results, study and plot these prediction intervals rather than focusing on coefficients.  
Option 3:  Do you believe you have no knowledge of X?  If so, you're wrong in the sense that with your regression equation X's effect is present in the residuals--the error terms for each case.  Studying these may give you some clues as to X's role.  You may even want to suppose that any case with a large residual, or a large positive one, is a case for which X takes a certain value.  It's possible this could move your analysis forward in some way.
