Table 5.

Stepwise multiple linear regression models predicting MPT-Naming item difficulty or SP item weights from lexical properties. The linear coefficient associated with each significant predictor in each of the regression models is reported in the corresponding cell. Item difficulty parameters are measured on a logit scale; the Word-T parameter is measured on a probability scale; and S-weight and P-weight are measured on a connection strength scale. All dependent and independent variables were z-score normalized for coefficient comparison; the intercept of all models is zero. LexFreq = the log transformed frequency of the target in American television and movie transcripts; PhonLeng = the number of target phonemes; PhonDens = the log transformed number of words differing from the target by a single phoneme; AoA = age of target word acquisition; df = degrees of freedom; rmse = root mean square error.

		Lexical Property				Model fit
		LexFreq	PhonLeng	PhonDens	AoA	df	rmse
MPT-Naming Item Difficulties	Attempt	−0.22		−0.39	+0.18	171	0.80
	LexSem	−0.19			+0.16	172	0.97
	LexPhon	−0.54				173	0.84
	LexSel	−0.22	+0.41			172	0.83
	Phon	−0.27	+0.24	−0.43	+0.15	170	0.48
	Word–T			+0.68		173	0.73
SP Item Weights	S-weight	+0.30	+0.33			172	0.96
	P-weight	+0.14	−0.35	+0.44		171	0.55