At least that's the theory.
In a new study, the theory falls apart. The telling graph comes near the end:
The data examined in this paper, however, are not consistent with the theory's predictions. CM clearly increases the probability that a respondent expresses an attachment to a specific party, and this positive relationship does not change across cohorts. In fact, there is some evidence that the positive effect of CM on party attachments is growing even stronger over time. In sum, neither the micro- nor the macro-level components of the theory are corroborated by the data.
Data yet again get in the way of theory.