Two-wave Two-variable Models

1 / 11

# Two-wave Two-variable Models - PowerPoint PPT Presentation

Two-wave Two-variable Models. David A. Kenny. The Basic Design. Two variables Measured at two times Gives rise to 4 variables Say Depression and Marital Satisfaction are measured for wives with a separation of one year. We have D 1 , D 2 , S 1 , and S 2.

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.

## PowerPoint Slideshow about 'Two-wave Two-variable Models' - lazar

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
The Basic Design
• Two variables
• Measured at two times
• Gives rise to 4 variables
• Say Depression and Marital Satisfaction are measured for wives with a separation of one year.
• We have D1, D2, S1, and S2.
Standard Cross-lagged Regression Model
• Time 1 variables cause Time 2 variables.
• Two stabilities
• S1  S2
• D1  D2
• Two cross-lagged effects
• S1  D2
• D1  S2
• Time 2 disturbances correlated.
• Inadvisable to have paths between Time 2 variables (S2  D2 or D2  S2)
Causal Preponderance
• Is S a stronger cause of D than is D of S?
• No easy way because the units of measurement of S and D are likely very different.
• Can standardize all the variables, but as will be seen this is more difficult when S and D are latent.
Assumptions
• No measurement error in S1 and D1. (Ironically, OK if there is measurement error in S2 and D2.)
• Nothing that causes both the time 1 variable and the time 2 variables. Such a variable is sometimes called a confounder. So if there is a gender (and gender is not controlled) difference at time 1, once we control for S1 and D1, there are no remaining gender differences at time 2 in S or D.
• The lagged effect of variables is exactly the length of measurement.
What to Do about the Assumptions?
• Measurement error in S1 and D1:
• Latent variable analysis (discussed in a latter slide).
• Confounders
• Measure them.
• Sensitivity analysis: See how the results change assuming confounders.
• Wrong lag
• Multi-wave study can be used to establish the optimal lag.
Latent Variables
• Can have as few as two indicators per latent variable.
• Correlate errors of the same indicator measured at different times.