Schützler, OleOleSchützler0000-0001-8868-0191Cummins, ChrisElder, Chi-HéGodard, ThomasMacleod, MorganSchmidt, ElaineWalkden, George2019-09-192015-08-102011https://fis.uni-bamberg.de/handle/uniba/39298Hierarchical data structures in variationist linguistics are given when, for example, each of a number of speakers makes several utterances of interest and individual observations are therefore not independent. Structures of this kind are often not fully reflected in the analytic tools used to model language variation. Against this background, the paper investigates the presence or absence of coda-/r/ in a hierarchically structured dataset produced by 27 middle-class speakers of Scottish Standard English (SSE). Two regressionbased statistical techniques are compared: multiple logistic regression and a hierarchical generalized linear model (HGLM). The latter emerges as a model that is not only more correct in theoretical terms because it takes the nested structure of the data into account, but also detects cross-level effects that go unnoticed in multiple logistic regression. The most striking finding is, for example, that in prepausal position coda-/r/ is less likely to be deleted, an effect, however, that is not general but depends critically upon the age of the speaker and the dialect contact to which he or she has been exposed. The paper thus provides one example of the potential of HGLM to explain complex patterns of variation in hierarchically structured data.engHierarchical ModelsMultilevel modelsScottish EnglishSociophoneticsRhoticityconferenceobjecthttps://www.srcf.ucam.org/camling/proceedings/schuetzler.pdf