Statistical Confusion
A statistical generalization is a statement which is usually true, but not always true. Very often these are expressed using the word "most", as in "Most conservatives favour welfare cuts." Sometimes the word "generally" s used, as in "Conservatives generally favour welfare cuts." Or, sometimes, no specific word is used at all, as in: "Conservatives favour welfare cuts."
Fallacies involving statistical generalizations occur because the generalization is not always true. Thus, when an author treats a statistical generalization as though it were always true, the author commits a fallacy.
A general rule is applied when circumstances suggest that an exception to the rule should apply. "The fallacy of accident occurs when we suppose that what is true of a term used in a relatively indeterminate or unconditioned sense, is true of the term when used in a relaticely determinate or conditioned sense. The Latin name of the fallacy is, Argumentum a dicto simpliciter ad dictum secundum quid." (Davies, A Textbook of Logic, p. 520.)
An exception to a generalization is applied to cases where the generalization should apply.