The rhoticity isogloss (ignoring Martha's Vineyard) seems to set off Western from Eastern New England pretty well, but the isogloss is significantly west of the settlement boundary in Connecticut and part of Massachusetts. On the other hand, the isogloss swings noticeably east of the E/W boundary in sparsely populated northern Vermont. This suggests that we can regard nonrhoticity as an urban prestige feature having spread from Boston, penetrating slowly into rural areas out of touch with urban culture, but crossing regional boundaries to reach cosmopolitan Hartford and Springfield. Note also that New Haven, a university city, shows complete nonrhoticity despite being deep within the Western region.
An approach to syllable structure similar to that of Wells (1990) suggests that rhoticity can be understood as a motivating factor for TLN. Wells argues that, in English, stressed syllables tend to attract following consonants into them as codas. Then in a rhotic dialect, the /r/ in words like merry will be part of the preceding syllable. So: the fact that there are no words ending in /er/ or /ær/ can easily be reinterpreted as a constraint against syllables ending in /er/ or /ær/. Once such a constraint is established, in rhotic dialects, the marked syllable-final /er/ and /ær/ in words like marry and merry can be replaced with the acceptable tautosyllabic string /ehr/, merging them with Mary. In a non-rhotic dialect, on the other hand, there is a constraint against /r/ ever occurring in a coda at all. So in this case, the /r/ in marry and merry is never in the same syllable as the vowel that precedes it, Thus the constraint favoring tautosyllabic /ehr/ over tautosyllabic /er/ and /ær/, as described above, has no relevance, and marry and merry remain distinct from Mary in non-rhotic dialects.