Backward Binary logistic regression: effect of the "indicator" value (first vs. last)

#1
I have performed a backward binary logistic regression in SPSS (20).

The dependent variable is nominal, and there are 5 explanatroy variables, of which one is numeral and the rest nominal. For the nominal ones I indicated that they are categorical variables. Then, I switched the Indicator to "first", given that the first values were the default values so to speak, so I wanted to compare the other values against that value. So far, I think this was the right way to go. The analysis came up with a model which gave me some maine effects and some interactions.

Then I did the same analysis but now I switched the Indicator to "last" again. Surprisingly, the results were quite different! I had a main effect that was not there in the first analysis and also some different interactions.

Can anyone tell me why changing the Indicator value from fort to last can make such a difference, and also whether I am right in setting it to "first" if this is a sort of default value?