One thing to watch out for. Your residual by predicted plot is pretty telling about how your data is distributed, it looks more like a binomial distribution than a continuous range. There is some scatter at the high end of the response, but virtually none at the low end. You would like to see a random scatter along the continuum to have a better feel for whether or not you have the "best" possible model based on the predictors you have used to build the model. You may want to convert your response to a categorical (low and high) and redo the model as logistic regression if nothing else than for comparison sakes.
Looking at your Prob > ItI values you could say that Mode(major) is the most important variable and may have the largest influence. You also have a very important interaction term that will influence your overall decision. Go the red hotspot by Response NH... and go to Factor Profiling. Turn on the Profiler. The Prediction Profiler will show up at the bottom of the fit report. Click on the red hotspot by Prediction Profiler and go to Assess Variable Importance and select Independent Uniform Inputs. Your Prediction Profiler will be reordered in the order of importance for your predictors. You will also get a report that shows the order. Check these features out and see if this is the direction you were looking to go in.
HTH
Bill
