5. Study Design
• Random sample by house
• Oxford House residents
– 95% of houses agreed to assist
• nj = 82
– 48% individual response rate
• ni = 296
6. Why Multilevel
• Disaggregation of data (Byrne, 2012)
– Biased estimates and standard errors
• Aggregated data (Byrne, 2012)
– Lack of individual variance may exaggerate group
effects
7. MSEM
• Extension of Multi-Group SEM
– Covariance matrices at within and between level
instead of for different groups
• Assumptions
– General Linear Model assumptions
• Linearity
• Normality
• Homoscedasticity
• Independence
8. MSEM Alternatives
• Segregated Approach (Yuan & Bentler, 2007)
– More established
– Modification indices
• Partially Saturated Approach (Ryu & West, 2009)
– MLR
– Random coefficients
– 2-1-1, 2-1-2, 2-2-2, 1-1-2, 1-2-1, 1-1-1
– Better power for level 2
– Greater n at L1, so more influence on model fit
statistics (Hox, 2002)
10. Process (Stapleton, 2013)
1. Consider single-level model
2. Baseline models for w/i & b/t levels while
saturating the other level
3. Theoretical w/i model with b/t saturated for
model fit
4. Theoretical b/t model with w/i saturated for
model fit
5. Combined w/i & b/t theoretical model for
parameters
6. Evaluate random coefficients
11. 1. Consider Single Model
• Consider single-level model
– Null model for descriptives
• Define ICC < 0.02 as within unless very large
cluster sizes, then possibly lower threshold
WITHIN ARE OHFITVC;
12. Between-Group Variance
ANALYSIS: TYPE IS TWOLEVEL RANDOM;
MODEL:
%WITHIN%
OHFITVC;
%BETWEEN%
OHFITVC;
Observed Variables
Null
ICC
Value Congruence 0.03
Commitment 0.11
Citizenship Behavior 0.04
14. 1.1 Measurement Reliability
Measures ωw ωb
Satisfaction 0.77 0.92
Demands-Abilities Fit 0.78 0.97
Needs-Supplies Fit 0.79 0.92
Interpersonal Similarity 0.79 0.98
Social Desirability 0.88 0.96
Value Congruence 0.90 0.93
Tenure 0.91 0.93
Commitment 0.91 0.99
15. 1.2 Descriptive Statistics
DEFINE: SDESIR=SDESIR/3;
Observed Variables n Min Max Mean SD SE
Social Desirability 292 0 13 6.67 3.23 0.19
Tenure 291 1 5 1.78 0.81 0.05
Commitment 292 2 7 5.26 1.10 0.06
Satisfaction 293 2 7 6.14 1.01 0.06
Interpersonal Similarity 291 1 5 3.42 0.96 0.06
Value Congruence 291 1 5 3.88 0.76 0.04
Demands-Abilities 293 1 5 3.96 0.80 0.05
Needs-Supplies Fit 293 1 5 4.01 0.70 0.04
16. 2. Baseline Models
Baseline w/i Model
MODEL:
%WITHIN%
OHFITVC with OHCit@0 OHCOM@0;
OHCit with OHCOM@0;
%BETWEEN%
OHFITVC with OHCit OHCOM;
OHCit with OHCOM;
Baseline b/t Model
MODEL:
%WITHIN%
OHFITVC with OHCit OHCOM;
OHCit with OHCOM;
%BETWEEN%
OHFITVC with OHCit@0 OHCOM@0;
OHCit with OHCOM@0;
17. Commitment
3. Theoretical w/i Model
Value
Congruence
Citizenship
Behavior
Commitment
Value
Congruence
Citizenship
Behavior
18. 3. Theoretical w/i Model
ANALYSIS: TYPE IS TWOLEVEL;
MODEL:
%WITHIN%
OHCOM on OHFITVC;
OHCOM on OHCit;
%BETWEEN%
OHCOM with OHFITVC OHCit;
OHFITVC with OHCIT;
19. 3. Theoretical w/i Model
Chi-Square Test of Model Fit
Value 38.019*
Degrees of Freedom 1
P-Value 0.0000
Scaling Correction Factor
RMSEA
Estimate 0.359
CFI/TLI
CFI 0.895
TLI 0.373
SRMR
Value for Within 0.169
Value for Between 0.010
20. Commitment
4. Theoretical b/t Model
Value
Congruence
Citizenship
Behavior
Commitment
Value
Congruence
Citizenship
Behavior
21. 4. Theoretical b/t Model
ANALYSIS: TYPE IS TWOLEVEL;
MODEL:
%WITHIN%
OHCOM with OHFITVC OHCit;
OHFITVC with OHCIT;
%BETWEEN%
OHCOM OHFITVC OHCit;
22. 4. Theoretical b/t Model
Chi-Square Test of Model Fit
Value 5.192*
Degrees of Freedom 3
P-Value 0.1583
Scaling Correction Factor
RMSEA
Estimate 0.050
CFI/TLI
CFI 0.994
TLI 0.988
SRMR
Value for Within 0.022
Value for Between 0.693
23. Commitment
3. Theoretical Combined Model
Value
Congruence
Citizenship
Behavior
Commitment
Value
Congruence
Citizenship
Behavior
Commitment
Value
Congruence
Citizenship
Behavior
24. 5. Combined Model
ANALYSIS: TYPE IS TWOLEVEL;
MODEL:
%WITHIN%
OHCOM on OHFITVC;
OHCOM on OHCit;
%BETWEEN%
OHCOM OHFITVC OHCit;
25. 5. Combined Model
Two-Tailed
Estimate S.E. Est./S.E. P-Value
Within Level
OHCOM ON
OHFITVC 0.711 0.079 9.025 0.000
OHCIT 0.338 0.055 6.168 0.000
Between Level
Means
OHCOM 5.250 0.076 68.744 0.000
OHCIT 5.870 0.058 101.094 0.000
OHFITVC 3.887 0.046 83.593 0.000
27. 6. Random Coefficients
LISTWISE=ON; (or Monte Carlo integration)
ANALYSIS: TYPE IS TWOLEVEL;
MODEL:
%WITHIN%
OHCOM on OHFITVC;
s1 | OHCOM ON OHCit;
%BETWEEN%
OHCOM OHFITVC OHCit;
31. Unresolved Issues
• Sample & Power Estimation
• Model Fit Approaches
• Model Fit Statistics
• MLR Performance in Multilevel Model
• 3+ Levels
• Balancing of Cluster Sizes
32. Unresolved Issues - Saturation
• “In the MODEL command, the following variable
is a y-variable (endogenous) on the BETWEEN
level and an x-variable (exogenous) on the
WITHIN level.
• This variable will be treated as a y-variable on
both levels: OHGEFIS”
• Any discrepancy is treated as a y-variable on
both levels