In a thought-provoking article, Guy (2013:63) claims that “lectal coherence ... [implies] that variables are correlated; if they are not, the cognitive and social reality of the ‘sociolect’ is problematic.” Considerable linguistic research has established that, for structural reasons, variables are correlated; however, structural correlation does not imply sociolectal coherence. Thus the question arises: do multiple variables cluster or correlate, loosely, tightly, or not at all, based on social factors, such as age, gender, community belonging, salience, prestige/stigma, or other external factors (Guy and Hinskens 2016)? This paper explores the unresolved question of cognitive sociolectal coherence by investigating a panel of 20 speakers of Swabian, a dialect spoken in southwestern Germany, across a 35-year lifespan. The corpus consists of Labovian-style sociolinguistic interviews, and the data comprise 20 phonological and morphosyntactic linguistic features, coded for a binary distinction between the dialect and standard variant. In concurrence with Guy (2013), the findings support the notion that sociolectal coherence is more multi-dimensional than previously believed. However, clear patterns have emerged, indicating that sociolectal coherence may lie in more cognitive constructs such as the type, status, and stigmatisation of the variable itself.