Download presentation

Presentation is loading. Please wait.

1
The STARTS Model David A. Kenny December 15, 2013

2
2 Overview u STARTS Model u Stationarity Assumption u Multivariate Generalization

3
3 The STARTS Components u Stable Trait or ST (trait) u Unchanging component u Autocorrelations of one u Autoregressive Trait or ART (state) u Slow-changing component u State or S (error) u Fast-changing, random component

4
4 Over-Time Correlations Assuming Equal Variances

5
5 Over-Time Correlations Large Stable Trait Variance

6
6 Over-Time Correlations Large ART Variance

7
7 Over-Time Correlations Large State Variance

8
8 The STARTS Model 1 1 1 1 b 11 11 1 1 11 1 1 1 b b

9
9 Complexity Mixed with Simplicity Complexity More latent variables (11) than variances and covariances (10) Simplicity Only 5 parameters (regardless of the number of waves) 4 variances: ST, ART, S, and U 1 path: ART path all loadings fixed to 1

10
10 Ensuring Stationarity Variance of ART at time 2 equals Var(ART 2 ) = b 2 [Var(ART 1 )] + Var(U 2 ) Note for the ART variances to be stationarity, it follows that: Var(U 2 ) = Var(ART 1 )[1 - b 2 ] This nonlinear constraint must be made and an SEM program is needed to do so. Thus the total number of parameters for STARTS is four, regardless what the number of waves are.

11
11 Unequally Spaced Measurement Assume age at each wave is denoted as A t. ART Model for time t-1 to t: ART t = b (At-At-1) ART t-1 + U t For the self-esteem study, we can use in the actual ages and set the time unit for b as one year (autocorrelation for one year).

12
12 Identification See Cole, Martin, and Steiger (Psychological Methods, 2005). Four waves is the very minimum, but many more (perhaps at least six) are necessary. Estimation is much better with many waves and large N.

13
13 Problems in Estimation if the AR coefficient is too small (looks like State) or too large (looks like Stable Trait) if a variance component small (explains less than 10% of the variance)

14
14 Stability of the ART Component There can be a high one-year stability of the ART component but the stability over a long period of time. For example, if the.766 is the year to year stability, the correlation across 11 years is only.053 (.766 11 ).

15
15 Relaxing the Stationarity Assumption All of the equality assumptions require that the variances of the measures not change over time. Seems rather implausible. Model can be modified to allow for latent stationarity with T – 1 parameters and so T + 3 parameters in total.

16
16 Differential Variances

17
17 Multivariate Generalization TSO Model (Trait, State, and Occasion) of Cole, Martin, and Steiger Create a latent variable for each time Two factors cause the latent variable Stable Trait (Trait) Autoregressive Trait (Occasion) State: Error Variance of each measure Really a START not a STARTS model Can be estimated with 3 waves.

18
18

19
19 Multivariate STARTS Implemented by Donnellan et al. in a study of self-esteem. Add the true State Factor (S). Correlate errors of the same indicator at different times. Requires at least four waves of data.

20
20

Similar presentations

OK

FACTORING Think Distributive property backwards Work down, Show all steps ax + ay = a(x + y)

FACTORING Think Distributive property backwards Work down, Show all steps ax + ay = a(x + y)

© 2017 SlidePlayer.com Inc.

All rights reserved.

Ads by Google

Free ppt on germination of seeds Ppt on wireless home security system Ppt on leadership qualities of bill gates Ppt on global business communication Ppt on brand marketing services Ppt on stock exchange-bse-nse Ppt on surface water definition Can you run ppt on ipad Show ppt on second monitor Ppt on id ego superego video