Table 8.

Model comparison for linguistic-cue based models, extended with (1) index and corpus frequency or (2) corpus and relative frequency

	df	AIC	BIC	logLik	Deviance	χ²	χ² df	Pr(>χ²)
prom	16	538126	538273	−269047	538094
prom.rel.freq	50	538042	538499	−268971	537942	152.68	34	<0.001
prom.freq.index	52	538034	538509	−268965	537930	11.77	2	0.00278

Note that the basic model is nested within both of the larger models, but the larger models are not nested and so the results of the likelihood-ratio test must be carefully interpreted. Model names reflect the predictor used; ‘rel’ refers to relative frequency and ‘freq’ refers to corpus frequency, while ‘prom’ indicates ‘prominence’, i.e. linguistic cues.