Table 5.

Fit statistics for specifications of gender, period, and cohort effects.

	Model specification	L²	df	BIC	p	D
Gender differences^a
A	No differences (vertical, horizontal, channels/barriers)	11,170	8748	−81,917	0.000	33
B	A with unidiff (vertical)	11,140	8747	−81,936	0.000	33
C	A with unidiff diagonal and off-diagonal (vertical)	11,136	8747	−81,940	0.000	32.9
D	A with unidiff diagonal (vertical)	11,127	8746	−81,939	0.000	32.9
E	A with heterogeneous differences (vertical)	11,104	8742	−81,919	0.000	32.7
F	D with unidiff (horizontal)	11,133	8746	−81,932	0.000	32.9
G	D with heterogeneous diff (horizontal)	11,065	8730	−81831	0.000	32.8
H	F with unidiff (channels/barriers)	11,135	8746	−81,931	0.000	32.9
I	F with heterogeneous diff (channels/barriers)	11,037	8740	−81,964	0.000	32.5
Period differences^b
P1	Heterogeneous (vertical, horizontal, channels/barriers)	14,031	12,978	−1,24,081	0.000	35.4
P2	P1 with transition unidiff (horizontal)	14,089	13,011	−1,24,373	0.000	35.7
P3	P2 with no difference (vertical)	14,168	13,023	−1,24,422	0.000	35.8
P4	P2 with transition model (vertical)	14,111	13,018	−1,24,426	0.000	35.7
Period and cohort differences^c
PC1	Period differences	21,128	30,436	−3,00,640	0.999	42.8
PC2	Cohort differences	21,174	30,436	−3,00,595	0.999	42.9
PC3	PC2 with period differences for middle cohort	21,085	30,410	−3,00,409	0.999	42.6
PC4	Full period/cohort differences	21,072	30,384	−3,00,147	0.999	42.6

Models A–I include gender-specific marginal destination effects and gender-common marginal origin effects. Gender-specific marginal origin effects were tested but did not differ by gender.

Models P1–P4 are estimated with period-specific origin and destination marginals. Transition model on vertical effects in P4 is specified as follows: Power: transition effect, Autonomy parameters: no effect before transition, Capital parameters: linear change over period; Education: no period change.

Models PC1–PC4 are estimated on with period-cohort specific marginals. The satisfactory model fit is due to the large degrees of freedom relative to the (reduced) sample size. In all models Education cohort and/or period differences are constrained to zero, and horizontal effects are constrained to unidiff change over period and/or cohort. The middle cohort (born 1950–1959) was observed in all three periods.