How to know which variables influence a Bayesian model Is there any way to know what variables in a predictive model are influencing the prediction result of a particular row in a dataset?
For example, I have a student dataset and my target variable is whether or not the student will evade school. My model predicted that student X has an 85% chance of leaving school. How do I know which of the variables are influencing the outcome for that particular student?
 A: It depends on what sort of model you're using.  I imagine that if your predictions are coming out as probabilities then you're using a logistic regression such as:
$Pr[\text{Student Evades School}|\vec{x}] = \frac{1}{1-\text{Exp}[\beta_0 + \sum\limits_{i=1}^n \beta_i x_i]}$
Where the $x_i$ are your predictor variables.
The simplest way to assess the relative predictive-ness of different $x_i$ is to consider the magnitude of the $\beta_i$.  When the coefficients are large, this means that small changes in the corresponding variable produce big changes in the predicted probability.  This would be a good approach if you wanted to, for example, find the variable that best reduced the probability of truancy.  Note that the sign of the $\beta_i$ reflects what an increase in $x_i$ will do to the probability; negative coefficients mean increasing the variable will decrease the probability.  Keep in mind that you will need to combine this information with the specific predictor variable values for a given predictor.  If $\beta_{10}$ is way bigger than every other $\beta_i$, but for specific prediction $x_{10} = 0$ then you will need to look for the next biggest $\beta_i$ and so on.
