21 - Repeated Assessments Flashcards
Benefits of Repeated Assessments
- Examines within-person change to evaluate causality
- Examine time effects (development, sensitive periods)
Possible Reasons for X-Y Correlation
- X causes Y
- Y causes X
- Confound: a third variable influences both X, Y
- Spurious: false
Benefit of Longitudinal Design
- Whether predictor predicts outcome
- Uses person as their own control (within-person)
Internal Validity
Extent to which a causal inference is justified from the research
Longitudinal Approaches, Ranked Worst to Best
- Cross-Sectional: one timepoint, cannot establish causality
- Lagged Association: X at T1 predicts Y at T2
- LA Controlling for Prior Levels: X at T1 predicts Y at T2 controlling for Y at T1 (predicts change)
- LACfPL, Testing Both Direction of Effects Simultaneously: demonstrates chicken/egg and directionality (if bidirectional, magnitude of each)
Test of Mediation
- Evaluate whether X predicts M at a later time point when controlling for earlier M; and whether X and Y are better explained by M
- Use MEM or SEM
- Challenge: time points on scales may differ
Use if you want to observe growth over time
RAW SCORES (not tranformed/normed scores)
Score Tranformations/Norms
T: mean 50, SD 10
Z: mean 0, SD 1
Standard: mean 100, SD 15
*DON”T ALLOW YOU TO OBSERVE GROWTH: for that, you need raw scores
Strengthening Inference of Change
- Large magnitude between scores
- Measurement error (unreliability) at both timepoints is small (reduce unreliability by combining multiple measures)
- Measurement is invariant across time (same meaning, comparable scale)
- Assess using SEM (intercept, factor loading) or IRT (differential item functioning, or difficulty/discrimination)
-No evidence for confounds of change (practice effects, cohort effects, time-of-measurement effects)
Difference Scores
-Less reliable than each individual measure; more so if they’re correlated
- Depends on:
1. reliability of individual measures
2. if variability of true individual differences is large
Nomothetic vs Idiographic
Nomothetic: population approach; assuming homogeneity (more accurate, more generalizable)
Idiographic: individual approach
Structural Equation Modeling
Allows for multiple dependent variables to help assess causality
Cross-Sectional Design
- Multiple participants at one timepoint
- Interest: age differences
- Confounds: cohort differences
- Limitation: does not show change over time
Cross-Sectional Sequence Design
- Successive studies of different participants at different times
- Interest: age differences
- Confounds: cohort differences, time-of-measurement
Time-Lag Design
- Participants from different cohorts assessed at same AGE
- Interest: cohort differences
- Confounds: time-of-measurement
Longitudinal Design
- Same participants measured at multiple timepoints
- Interest: age differences
- Confounds: cohort effects and time-of-measurement
Longitudinal Sequences Design
- Following multiple cohorts across time
1. Time sequential: multiple ages assessed at multiple times
2. Cross sequential: multiple cohorts assessed at multiple times
3. Cohort sequential: multiple cohorts assessed at multiple ages
*Remember the triangle!
Heterotypic, Homotypic, Phenotypic Continuity & Discontonuity
- Homotypic: same process, same behavior across time
- Heterotypic: same process, different behavior
- Phenotypic: different process, same behavior
- Discontinuity: different process, different behavior
Identifying Heterotypic Continuity
- Rank order stability: changes in degree of stability
- Content level on the construct (IRT-difficulty, SEM-intercept)
- How strongly the content reflects the construct (IRT-discrimination, SEM-factor loading)
Assessing a construct across development
- All possible content
- Pros: comprehensive, allows you to examine change in each facet
- Cons: Inefficient, intrusions
- Only the common content
- Pros: efficient, may exclude inappropriate content, same measure for easy interpretation
- Cons: fewer items, loss of info, gaps
- Only the construct-valid content
-Pros: efficient, retains construct & content validity
-Cons: time-intensive, uses different measures (so harder to compare)
RECOMMENDED!
Ensuring Statistical Equivalence
- AKA: same mathematical metric
- Use Developmental scaling: measures that differ in difficulty and discrimination are on same scale with all content (age-common and age-different) to estimate each person’s score on that scale
Developmental Scaling Approaches:
- SEM: allows estimation of latent variable with different content across time
- IRT: links measures’ scales based on difficulty and discrimination of common content
Ensuring Theoretical Equivalence
- Construct validity invariance = content reflects the construct at each age
- Should show content validity (sample all aspects of the construct)
- Test-retest reliability in short term
- Convergent and discriminant validity of measures
- Similar factor structure across time based on factor analysis
- High internal consistency reliability
Confounds of Change
- Practice effects
- Cohort effects
- Time of measurement
Inferring Development
Would need to do all 3 longitudinal sequential designs and see that age-related differences were stronger than cohort- and time-of-measurement differences
Establish longitudinal measurement invariance:
- SEM: same intercepts and factor loadings across ages
- IRT: same difficulty and discrimination across ages
