developmental models/ longitudinal data analysis
DESCRIPTION
Developmental Models/ Longitudinal Data Analysis. Danielle Dick & Nathan Gillespie Boulder, March 2006. Why conduct longitudinal analyses?. Can improve power by using multiple observations from the same individual Can examine time-dependent genetic and environmental effects - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/1.jpg)
Developmental Models/Longitudinal Data Analysis
Danielle Dick & Nathan Gillespie
Boulder, March 2006
![Page 2: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/2.jpg)
Why conduct longitudinal analyses?
Can improve power by using multiple observations from the same individual
Can examine time-dependent genetic and environmental effects
• Changing magnitude of genetic/environmental influence across development
• Same versus different genes across development
![Page 3: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/3.jpg)
Methods for Longitudinal Analysis
Cholesky Models
Simplex Models
Growth Curve Models
![Page 4: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/4.jpg)
Cholesky Model
![Page 5: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/5.jpg)
Cholesky Model Advantages:
• Logical: organized such that all factors are constrained to impact later, but not earlier timepoints
• Requires few assumptions, can predict any pattern of change
Disadvantages:• Not falsifiable
• Does not make predictions about what will happen in the future (as yet unmeasured timepoints)
• Only feasible for limited number of measurements
![Page 6: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/6.jpg)
Cholesky Model
Questions you can address:
• Magnitude of genetic/environmental influence at each time
• Extent to which genetic/environmental influences overlap across time
![Page 7: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/7.jpg)
A1 A2 A3
C1 C2 C3 E1 E2 E3
Age 16 Substance Use
Trivariate Cholesky Decomposition
Age 17 Substance Use
Age 18 Substance Use
![Page 8: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/8.jpg)
0
0.1
0.2
0.3
0.4
0.5
A C E
Age 16
Age 17
Age 18.5
X2 = 57.462, df = 45, p = .101, AIC = -32.538, RMSEA = .011
Partitioning of Variance from Longitudinal Drinking Data
Rose, Dick 2001, ACER
![Page 9: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/9.jpg)
Age 16Drinking
Age 17Drinking
Age 18.5Drinking
A1
C1 C2 C3
.58(.51 - .66)
.70(.62 - .78)
.71(.63 - .79)
.61(.53 - .67)
.22(.10 - .32)
.39(.29 - .45)
.37(.16 - .47)
Best-fitting model for drinking
![Page 10: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/10.jpg)
Cholesky Model
Questions you can address:
• Magnitude of genetic/environmental influence at each time
• Extent to which genetic/environmental influences overlap across time
• Other standard multivariate extensions (e.g., multiple group models)
![Page 11: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/11.jpg)
Methods for Longitudinal Analysis
Cholesky Models
Simplex Models
Growth Curve Models
![Page 12: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/12.jpg)
What aresimplex/autoregressive models?
Models with a specific structure of association whereby correlations are largest between adjoining measures, and drop off systematically as distance between variables increases
This structure is frequently observed with longitudinal data:
• Classic example: weight measurements• 2 year intervals from age 5 to 15
• you would expect that weight at age 5 would be most highly correlated with weight at 7, somewhat less correlated with weight at 9, less correlated with weight at 11, etc.
![Page 13: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/13.jpg)
Simplex Model
(Boomsma & Molenaar, 1987)
![Page 14: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/14.jpg)
Genetic Simplex Model
A A AA
BAC2 BAC6BAC4BAC3
E E EE
n21 n32 n43
p21 p32 p43
x11 x22 x33 x44
z11 z22 z33 z44
1 111
1 111
u11u22 u33 u44
x and z = genetic and nonshared environmental innovations respectivelyn and p = genetic and nonshared environmental transmission respectivelyu = error variances
![Page 15: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/15.jpg)
Simplex Model
Advantages:
• Makes restrictive predictions about covariance pattern
• Falsifiable
![Page 16: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/16.jpg)
![Page 17: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/17.jpg)
![Page 18: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/18.jpg)
![Page 19: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/19.jpg)
Today’s example Grant et al., 1999, Behavior Genetics, 29, 463-472. Australian alcohol challenge data, collected between 1979 and 1981
• Mean age = 23.5 years Subjects drank 0.75 g/kg alcohol at a steady rate over a 20-minute period.
Blood Alcohol Concentration (BAC) was assessed at 6 points after consumption:
Minutes post-consump.
Mean BAC
# of individuals with data
MZM (43 prs)
DZM (37 prs)
BAC 1 56 89.0 83 72
BAC 2 68 88.9 83 74BAC 3 83 88.8 84 71BAC 4 123 80.9 86 74BAC 5 143 76.6 52 56
BAC 6 182 67.7 83 74
![Page 20: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/20.jpg)
A simplex correlation pattern…
BAC 2 BAC 3 BAC 4 BAC 6
BAC 2 1.00
BAC 3 0.90 1.00
BAC 4 0.69 0.84 1.00
BAC 6 0.58 0.77 0.93 1.00
Sample correlations (the DZM twin A quadrant of an intraclass correlation matrix)
![Page 21: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/21.jpg)
Practical - Simplex Model
A A AA
BAC2 BAC6BAC4BAC3
E E EE
n21 n32 n43
p21 p32 p43
x11 x22 x33 x44
z11 z22 z33 z44
1 111
1 111
u11u22 u33 u44
x and z = genetic and nonshared environmental innovations respectivelyn and p = genetic and nonshared environmental transmission respectivelyu = error variances
![Page 22: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/22.jpg)
Practical - Simplex Model
A A AA
BAC2 BAC6BAC4BAC3
E E EE
n21 n32 n43
p21 p32 p43
x11 x22 x33 x44
z11 z22 z33 z44
1 111
1 111
u11u22 u33 u44
x and z = genetic and nonshared environmental innovations respectivelyn and p = genetic and nonshared environmental transmission respectivelyu = error variances
1 0 0 00 2 0 00 0 3 00 0 0 4
![Page 23: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/23.jpg)
Practical - Simplex Model
A A AA
BAC2 BAC6BAC4BAC3
E E EE
?
1 111
1 111
x and z = genetic and nonshared environmental innovations respectivelyn and p = genetic and nonshared environmental transmission respectivelyu = error variances
? ? ?
? ? ? ?
? ? ? ?
![Page 24: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/24.jpg)
Full Genetic Simplex Model
A A AA
BAC2 BAC6BAC4BAC3
E E EE
0.9892 0.5332 1.3123
0.7001 0.5027 0.4130
12.0279 6.5234 .7036 4.2649
9.4624 5.0094 10.3650 5.1284
-0.18401 111
1 111
Basic_simplex.mxo -2*LL=4620.028, 23 est. parameters, 606 df
-0.1840 -0.1840 -0.1840
![Page 25: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/25.jpg)
Sub-Models
1) Is the error variance on individual variable assessments significant?
2) Is the genetic innovations on BAC6 significant? BAC4? BAC2?
![Page 26: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/26.jpg)
Sub-Models
1) Is the error variance on individual variable assessments significant?
- drop 200
2) Is the genetic innovations on BAC6 significant? BAC4? BAC2?
- drop 4, 3, 2
![Page 27: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/27.jpg)
Simplex Model
Advantages:
• Makes restrictive predictions about covariance pattern
• Falsifiable
Disadvantages:
• Makes restrictive predictions about covariance pattern (future depends on current state only)
• Number of parameters increases with number of measurements
![Page 28: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/28.jpg)
Methods for Longitudinal Analysis
Cholesky Models
Simplex Models
Growth Curve Models
![Page 29: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/29.jpg)
Latent Growth Curve Model(shown here as linear)
• Mean Level of the Trait (Intercept) • Rate of Change In Trait (Slope)
![Page 30: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/30.jpg)
Latent Growth Curve Model(shown here as linear)
![Page 31: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/31.jpg)
Genetically InformativeLatent Growth Curve Model
![Page 32: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/32.jpg)
Genetically Informative Latent Growth Curve Model
→Like a bivariate model!
![Page 33: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/33.jpg)
Growth Model Questions
What is the contribution of genetic/environmental factors to the variation of α (intercept) and (slope)?
Same or different genes influencing α (intercept) and (slope)?
Same or different environments influencing α (intercept) and (slope)?
![Page 34: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/34.jpg)
Practical Mx latent growth curve example (script from http://www.psy.vu.nl/mxbib/)
Submodels to test:1. No covariance between slope and intercept2. No genetic effect on intercept3. No genetic effect on slope4. No common environmental effect on intercept5. No common environmental effect on slope6. Best fitting model? (i.e., ACE, AE, CE, E?)
![Page 35: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/35.jpg)
Slope
A
AC CE E
1.0 (0.5)1.0
Intercept
Time 1 Time 2 Time 3 Time 4
1.01.0 1.0
1.00
1 23
x11x21
x22
F11
F21F31
F41
LL11
LL11
LL11
LL11
![Page 36: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/36.jpg)
Practical Mx latent growth curve example (script from http://www.psy.vu.nl/mxbib/)
Submodels to test:1. No covariance between slope and intercept – signif decrease in fit2. No genetic effect on intercept – signif decrease in fit3. No genetic effect on slope – signif decrease in fit4. No common environmental effect on intercept -- ns5. No common environmental effect on slope -- ns 6. Best fitting model? (i.e., ACE, AE, CE, E?) -- AE
![Page 37: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/37.jpg)
Growth Curve Model Advantages:
• Very efficient: number of parameters does not increase with number of measurements
• Provides prediction about behavior beyond measured timepoints
Disadvantages:
• Note regarding slope parameters
• Can be computationally intense
• Assumptions to reduce computational burden
• Linearity, no genetic effects on residuals, equal variance among residuals at differing timepoints
![Page 38: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/38.jpg)
Latent Growth Curve ModelingAdditional Considerations
Standard approach assumes data are collected at identical set of fixed ages for all individuals (e.g., start at age 12, yearly assessments)
Age heterogeneity and unequal spacing of measurements can be handled using definition variables
• Mehta & West, 2000, Psychological Methods
![Page 39: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/39.jpg)
S1 I1 S2 I2
As1Cs1 Es1 Ai1
Ci1Ei1 As2
Cs2 Es2 Ai2Ci2
Ei2
lgdri171 lgdri181 lgdri162 lgdri172 lgdri182
T1 T2 T3 T1 T2 T3Ar
Cr ErAr
CrEr
ArCr
Er ArCr
ErAr
CrEr
Ar Cr
Er
M
Age
lgdri161
Latent Growth Curve Model with Measured Variable
![Page 40: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/40.jpg)
Extensions of Growth Curve Models
Incorporation of measured variables (genotype, environment)
Nonlinear growth• Neale, MC & McArdle, JJ (2000). A structured latent growth
curves for twin data. Twin Research, 3, 165-177.
SlopeIntercept
Time 1 Time 2 Time 3 Time 4
1.01.0 1.0
1.0 01 2 4
L L L L
![Page 41: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/41.jpg)
Latent Growth Curve Modeling
McArdle, JJ (1986). Latent variable growth within behavior genetic models. Behavior Genetics, 16, 163-200.
Baker, LA et al. (1992). Biometrical analysis of individual growth curves. Behavior Genetics, 22, 253-264.
McArdle, JJ et al. (1998). A contemporary method for developmental -genetic analyses of age changes in intellectual abilities. Developmental Neuropsychology, 14, 69-114.
![Page 42: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/42.jpg)
Summary of Longitudinal Models Cholesky Model
• Few assumptions, predict any pattern of correlations
• Not falsifiable
• Limited measurements Simplex Model
• Falsifiable
• Limited measurements Growth Curve Model
• G, E influences on initial level, rate of change
• Unlimited measurements
• Computationally intensive, assumptions
![Page 43: Developmental Models/ Longitudinal Data Analysis](https://reader035.vdocuments.site/reader035/viewer/2022062222/56816039550346895dcf5921/html5/thumbnails/43.jpg)