I am running a logistic regression to find how font manipulation (x1) and people perceptions over how easy it is to process the information (x2) influence the online buying behaviour (y). I was especially interest to see if x2 acts like a moderator. Initially I have got to the following quadratic regression equation (with all terms significant):
(Purchase)=0,3-0,07objective-0,1subjective-0,1objective*subjective +0,01subjective^2 +0,02objective*subjective^2

But the results differed quite a lot from what was initially expected. Looking more in depth, I got to this 6 degree polynomial, with all terms significant and a better AIC (by 100):
(Purchase)=-5,9-0,06objective+16,05subjective-0,3objective*subjective-14,9subjective^2+0,06objective*subjective^2+6,5subjective^3-1,4subjective^4-0,0003objective*subjective^4+0,1subjective^5-0,006subjective^6

Is this a realistic model to work on? If so, is it possible to interpret it any other way that graphing the predicted probability of purchase (y) against the subjective level of understanding the website (x2) and grouped by the objective conditions (x1)?