Scientific paradigms or "research programmes" (as Imre Lakatos calls them) survive because of vested interests in maintaining them rather because of their explanatory power. The facts do not fit to the theory - so much for the facts. There are elaborate progressive and degenerative "problemshifts," Lakatos claims, from saving theories from empirical falsification. Progressive problmeshift is a conceptual edifice that reconcile the core assumptions of the programme with the inconvenient empirical evidence - either by introducing auxiliary assumptions or "special cases.' Degenerative problemshifts boil down to reducing the scope conditions of the paradigm i.e defend the theory from refutation by limitn its empirical relevance.
When we are to economic modeling - most of them, especially those of the neoclassical variety - have little or no explanatory power, but that does not slate them for empirical falsification, because there are political interests in maintaining them as the "founding myths" of the established institutional order (the role served by the flat-earth, geo-centric astronomy in the middle ages). So despite their embarrassingly weak record in predicting novel empirical - the only proof of a theory that really matters - these models are resuscitated by degenerative problemshifts (focus on mathematical models rather than empirical data) and heroic assumptions about human behavior (see for example my piece "The death knell of the utilitarianism" _Voluntas_ 2000, 11(4): 375-388).
To be honest, the so-called Marxism was a research programme as well, and died not because it did or did not fit the facts, but because political-instituional interests that used it as their founding myth waned.
It is probably safe to say that in social and behavioral sciences, about 50% is science and 50% is politics and ideology. In economics, however, that ratio is closer to 25% science and 75% politics and ideology.
Wojtek