Table 2. Summaries of the mixed-effects regression models presented in Figure 3b: Models for female speakers’ sibilant tokens. Models were made using a treatment coding scheme with affricate (vs. fricative) and /i/ (vs. /a/) as reference levels

	Word-initial tokens
	Lax		Tense
	Estimate (Std.E)	p-value	Estimate (Std.E)	p-value
(Intercept)	6,617 (361)	<.0001	7,397 (279)	<.0001
Manner(vs. Fricative)	–1,157 (421)	0.009	−28 (381)	0.942
Birthyear(year 2008=0)	9 (14)	0.531	24 (10)	0.032
Vowel (vs. /a/)	10 (454)	0.983	880 (800)	0.277
Manner×Birthyear	−18 (16)	0.26	9 (14)	0.55
Manner×Vowel	3,407 (504)	<.0001	512 (774)	0.509
Vowel×Birthyear	−9 (17)	0.598	16 (26)	0.551
Manner×Birthyear×Vowel	25 (19)	0.187	–47 (25)	0.064

	Word-medial tokens
	Lax		Tense
	Estimate (Std.E)	p-value	Estimate (Std.E)	p-value
(Intercept)	8,042 (444)	<.0001	7,048 (302)	<.0001
Manner(vs. Fricative)	–2,474 (408)	<.0001	−626 (355)	0.086
Birthyear(year 2008=0)	39 (16)	0.03	7 (11)	0.537
Vowel (vs. /a/)	−151 (471)	0.749	1,634 (400)	<.0001
Manner×Birthyear	–45 (15)	0.005	10 (13)	0.453
Manner×Vowel	3,768 (532)	<.0001	1,622 (475)	0.001
Vowel×Birthyear	–29 (17)	0.088	30 (14)	0.039
Manner×Birthyear×Vowel	31 (19)	0.115	–47 (17)	0.006