Loading presentation...

Present Remotely

Send the link below via email or IM


Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.


Examining Changes Over Time: A Selection of Studies and Methods

2011-02-18 Created

Jamie DeCoster

on 23 March 2011

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Examining Changes Over Time: A Selection of Studies and Methods

Examining Changes Over Time:
A Selection of Studies and Methods Jamie DeCoster Repeated Measures ANOVA Hierarchical Linear Modeling Linear Mixed Models Latent Growth Curve
Modeling Elliott, Burgio, & DeCoster (2010)
Journal of the American Geriatrics Society Sun, Kosberg, DeCoster, & Hong (2009)
China Journal of Social Work Hilgeman, Allen, DeCoster, & Burgio (2007)
Psychology and Aging Jeon, Peterson, & DeCoster (under review)
Developmental Psychology The Study The Decision The Method The Method The Study The Decision The Study The Decision The Method The Decision The Study The Method Treats time point as a fixed within-subjects factor
Examines whether the outcome means vary between groups
Can be tested using either univariate or multivariate statistics Pros Cons Simple to conduct and explain
Easy to incorporate moderating variables
Can work directly with change scores for each participant
Readers will be familiar and comfortable with the method Most implementations require complete data
All participants must be measured at the same timepoints
Limited ability to model the correlations among timepoints
Difficult to examine predictors that vary over time
Does not treat time numerically Difficult to do without HLM software
Equations are hard to follow if you don't know HLM
Improper centering can alter the interpretation of the findings Allows you to examine time-varying predictors
Number and timing of assessments can vary between subjects
Allows you to have additional levels above the subject (like classroom)
Alows you to test both overall trends and the extent to which the trends vary by subject Pros Cons Cons Pros Many of the same advantages as HLM Number and timing of assessments must be the same across subjects
Requires SEM software
May need to modify model to get the solution to converge Allows you to work with latent variables
Allows you to independently examine the effects of baseline and change scores
Allows you to examine longitudinal changes in the context of a larger model Pros Cons Based on the idea of having the regression coefficients for one equation be the outcome measures for another equation
In longitudinal analysis, Level 1 is the timepoint and Level 2 is the subject
Trajectories (Level 1 coefficients) are separately estimated for each subject The HLM Example Equations Mathematically the same as HLM, but the model is defined differently Does not naturally group the variables into hierarchical levels
Often involves complicated models that can be difficult to set up appropriately
Estimated iteratively, so the analysis will sometimes fail to reach a solution
Readers are often not familiar with the method Can be performed in SPSS and SAS
Framing of results is similar to ANOVA
Allows you to specify the covariance structure Time-varying predictors
Assessments can vary between subjects
Accommodates multiple levels
Overall trends + variability Designed as a generalization of the General Linear Model
Data set is defined at the observation level
Subjects and other grouping factors are added as random effects
Allows fixed and random effects of subject-level predictors The LMM Fixed effect tells you the overall relation between predictor and outcome
Random effect tells you the extent the relation varies between subjects LGCM
Coefficients The LGC Model Uses SEM
Treats observations as indicators, extracting latent estimates of intercept and slope coefficients Intercept = Baseline value
Slope = Change score Lets you relate the intercepts and slopes of your variables to each other Objective: To evaluate an intervention designed to improve the health and psychological wellbeing of people caring for family members with Alzheimer's disease (REACH) Longitudinal Assessments: Prior to intervention
At the end of the intervention (6 months later) Test whether the intervention changed caregiver health
Test a mediation model Significant main effect of time (F[1, 492] = 12.70, p < .001)
Significant main effect of treatment (F[1, 492] = 7.08, p < .01)
Significant time by treatment interaction (F[1, 492] = 4.71, p < .05)
Significant change for the intervention condition but not for the control condition IV: Changes in caregiver health
Mediator: Changes in depression
DV: Changes in perceived burden Sobel test of the mediation model was significant (Z = 6.11, p < .001) Improvements in caregiver health were associated with decreases in depression
Decreases in depression were associated with decreases in the perceived burden of caregiving We only had two time points to consider Small amount of missing data
No concerns about different correlations between time points Target was a medical journal Wanted to use the change scores in a mediation analysis afterwards Did not want to use complicated analyses Wanted the longitudinal analysis to focus on change scores for consistency with the mediation
Needed complete data for the mediation analysis Objective: Results - Trajectories: Results - Associations with
Better Wellbeing at Baseline: Wave 1: 1998
Wave 2: 2000
Wave 3: 2002 Analysis Aims: Examine how emotional wellbeing changes over time
Examine how the trajectories differ based on demographic variables Analysis Aims: ANOVA Results: Mediation Results: Longitudinal Assessments: To understand how the emotional wellbeing of the oldest old (80+ years) in mainland China changes over time HLM Equations: Being male (β=.42, SE=.13, p ≤ .01)
Higher self-rated health status (β=.66, SE=.09, p≤.001)
Greater involvement in social activities (β=.09, SE=.03, p≤.001)
Higher cognitive functioning ((β=.14, SE=.07, p≤.05)
Living in an urban area (β = .82, SE = .14, p ≤ .001) Overall longitudinal trend not significant (π = -.05, SE = .24, p > .05)

However, there were several moderators A number of participants had incomplete data HLM allows you to include participants with incomplete data in the sample
Makes use of whatever data you have without requiring imputation The target was a social work journal, where people are familiar with HLM
The first author had experience working with HLM Baseline
Post-intervention (6 months)
Follow-up (12 months) Objective: To determine how the effect of a caregiver training program (REACH again) on psychological outcomes is influenced by the degree a participant perceives there to be positive aspects of caregiving Results - Predicting Burden Longitudinal Assessments: Results - Predicting Depression Choosing the Analysis Fit between analysis and research question
Fit between analysis and the data
Comfort of the PI with the analysis
Acceptability of the analysis among the target audience Persons with higher values of PAC had lower levels of depression across time (F[1,447] = 17.12, p < .0001)
Effect of the treatment was not significantly influenced by PAC There was a significant 3-way interaction between PAC, time, and treatment (F[2, 89.2] = 5.94, p = .004) Used PAC as a time-varying predictor of depression and caregiver burden
Included a number of demographic covariates
Used a first-order autoregressive covariance structure Analysis Aims: A number of participants had incomplete data LMM provides the same analytic benefits as HLM regardinng missing data Expected the correlations between nearer time points to be greater Can be accurately modeled using an autoregressive covariance structure Was aimed at a psychology journal Tolerant of new methods The PI was interested in learning LMM Analysis Aims: To understand the longitudinal relations of parent interaction style and child emotion regulation with cognitive development for children with developmental risks Overall SEM fit measures indicate the model has adequate or good fit
Baseline cognition was positively related to baseline emotion regulation
Changes in cognition were positively related to baseline supportiveness and changes in emotion regulation Results: Latent Growth Curve Model: Longitudinal Assessments: Objective: LGCM Estimated Coefficients: Time 1: child at 14 months of age
Time 2: child at 24 months of age
Time 3: child at 36 months of age Used interlocking growth curve models
Predicted the baseline and changes in cognition from the baseline and changes in parental supportiveness and child emotion regulation Had multiple indicators for supportiveness and emotion regulation Can be handled better using SEM Assessments were consistent between participants Meets the assumption of LGCM Targeted at a psychology journal Tolerant of new methods PI had experience with LGCM Analysis as a Multiply Collaborative Process The PI and analyst collaborate when determining what questions to address and what analyses to perform
The researchers collaborate with the reviewers when determining what methods are acceptable to the field
There are better and worse analyses, but there is no single "right" way to analyze a given project
Full transcript