Summary

Using the General Loglinear Analysis procedure, you have paired responses from a survey. By comparing results from symmetry and quasi-symmetry models, you were able to test the marginal homogeneity model and found that it did not hold, thus a candidate "won" the debate in the sense that a statistically significant change in preferences occurred. Moreover, since the quasi-independence model holds, no candidate gained (or lost) an especially large number of supporters from any particular competitor, as compared to their gains from (losses to) other competitors.