Table 3.

Multilevel models predicting depression from cognitive functioning

Predictor variable	Concurrent model				Lagged model
	Fixed effects
	B	SE	df	t	B	SE	df	t
Within-person (level 1)
Linear time	−13.62	2.14	551.59	−6.35***	8.89	5.56	143.76	1.60
Quadratic time	1.30	0.21	95.71	6.24***	−0.63	0.53	63.31	−1.17
Age × linear time	0.16	0.03	579.78	5.20***	−0.12	0.08	155.10	−1.56
Age × quadratic	−0.02	0.003	99.86	−5.61***	0.01	0.01	67.33	1.13
Cognitive functioning	−0.14	0.04	123.27	−3.10**	0.01	0.07	236.49	0.17
Between-person (level 2)
Age	−0.36	0.08	581.16	−4.44***	0.23	0.19	310.62	1.22
Sex	0.98	0.77	210.99	1.27	1.74	1.08	110.84	1.61
Education	−0.05	0.15	236.04	−0.32	−0.01	0.25	125.42	−0.04
Cognitive functioning	−0.14	0.04	228.17	−3.97***	−0.09	0.05	142.42	−1.72
	Random effects
Covariance parameter estimate	B	SE		Z	B	SE		Z
Within-person
Linear time	0.08	0.10		0.76	0.03	0.10		0.29
Quadratic time	0.0004	0.001		0.47	0.000a	0.000a		—
Cognitive functioning	0.08	0.03		2.52*	0.000a	0.000a		—
		Within pseudo R2		33%		Within pseudo R2		59%
		Between pseudo R2		2%		Between pseudo R2		9%

Cognitive functioning is operationalized as cognitive test performance. Between-person cognitive function refers to person-mean level of the CERAD neuropsychological battery (i.e., average level of cognitive function over the entire study period). In the concurrent model, within-person cognitive function refers to person-centered level of CERAD neuropsychological battery (i.e., annual deviations in cognitive function centered around person-mean level of cognitive function). In the lagged model, within-person cognitive function refers to lagged person-centered level of the CERAD neuropsychological battery (i.e., previous year annual deviations in cognitive function centered around person-mean level of cognitive function).

^aVariance too small to be estimated. The final Hessian matrix is not positive definite although all convergence criteria are satisfied. The test statistic and confidence interval cannot be computed.

^***Predictor is significant at 0.001 level;

^**Predictor is significant at the 0.01 level;

^*Predictor is significant at the 0.05 level.