How do we know which variables (X1,X2,....,X6) are affecting the outcomes (y)? The data I am using are sales person data. The data contain six independent variables (six vulnerabilities) scored 1-20 and the output is win/loss of the deal. My question is how do we know which variable or combination of variables are affecting the output i.e. win/loss, and what should be the methods or step to come up with my objective?
 A: The question is a common problem in statistics known as model selection. A concept that has gained a lot of popularity recently is penalized regression. It is similar to regular regression, except that some of your coefficients might be shrunk down to zero. These are the ones that 'don't matter'. 
Here is an example using R:
#My variables (X1...X20)
x=matrix(rnorm(100*20),100,20)

#My response (0 or 1)
g2=sample(0:1,100,replace=TRUE)


#Fit model using cross validation to choose tuning parameter
modcv <- cv.glmnet(x,g2,family="binomial")

#final model
 mod <- glmnet(x,g2, family="binomial", lambda = modcv$lambda.min)

#coefficients which my model identified as important
mod$beta > 0

With such a small number of variables you might want to try some of the other common model selection approaches (forward selection, backward selection, best-subset selection, etc). Look at this freely available online book: http://www.stanford.edu/~hastie/local.ftp/Springer/OLD/ESLII_print4.pdf
A: Since your outcome is binary (win/loss), you may want to fit a Logistic regression by including all or a subset of those 6 independent variables and decide which variable is significant based on reported $p$-values. Now there are different methods, to find a reasonable model including: r-squared ($R^2$) sometime called coefficient of determination, Adjustted r-squared, $C_p$ Mallows' statistic, Akaike information criterion (AIC), Bayesian information criterion (BIC). As said above, you may also try to use some automatic procedures such as (forward selection, backward selection, stepwise regression) to find a preferred model. However, these methods are not optimal in any sense and don't take into account your knowledge about the independent variables and should be used with care. Some examples of fitting a logistic regression in R have been given here.
