From pg 258 of his Twenty Lectures in Social Theory:
For a theorist to maintain individualism in a clear and honest way, he must introduce fantastic randomness into his picture of how the world comes to be orderly. Basically, he must deny that patterning exists outside of any specific situation. Most theorists, however, precisely because they are sociologists, will not be satisfied with such a position, and they will move, more or less hesitantly, towards embracing some element of the collectivist alternative. Yet, as long as formal commitments to individualism are maintained – as long as the theorist continues to call himself an ‘exchange theorist’ or a ‘symbolic interactionist’ or a ‘phenomenologist’ – this collective reference can only be introduced in a residual way. because it must be ad hoc, it will inevitably be indeterminate, hence theoretically and empirically frustrating. The dilemma, then, is produced by being confronted with the mutually unsatisfactory choices or randomness and residual category. The tension which s produced by being stranded on these horns often leads the theorist to resort to ‘last instance’ arguments. he suggests that, while collective dimensions may exist, in the last instance individual and contingent negotiation still creates order.