Defeasible reasoning with variable degrees of justification