STRUCTURAL
EQUATION MODELING
(SEM)
Wei-Jiun, Shen Ph. D.
Purpose
 To analyze whether a combination of the DVs
varies as a function of IVs
 Multiple DVs
Advantage of MANOVA
 Improve the chance of discovering what it is that
changes as a result of different IVs and their
interaction
 ↓type 1 error
 Reveal difference not show
in separate ANOVA
Kinds of research questions
 The goal of MANOVA is to discover whether DVs
is changed by manipulation
 Main effect of IVs
 Interaction among IVs
 Importance of DVs
 Parameter estimates (marginal, cell and adjusted mean)
 Specific comparisons and trend analysis
 Effect size
 Effects of covariates
 Repeated-measures analysis of variance
Limitations to factor analysis
 Theoretical issues
 Causality
 Choice of variables
 generalizability
 Practical issues
 Unequal sample sizes, missing data and power (relations among DVs)
 Multivariate normality
 Absence of outliers
 Homogeneity of variance-covariance matrices (Box’s M)
 Linearity
 Homogeneity of regression (covariate; MANCOVA)
 Reliability of covariance
 Absence of multicollinearity and singularity
Fundamental equation for MANOVA
 ANOVA to MANOVA
𝑆𝑆𝑡𝑎𝑡𝑎𝑙 = 𝑆𝑆 𝑏𝑔 + 𝑆𝑆 𝑤𝑔
Y11mean
level1
Y1∙1∙
1
Y1∙1∙
2
Y1∙1∙
3
Y1∙1∙
4
Y1∙1∙
5
Y1∙1∙
6
Y1∙1∙
7
Y12mean
level2
Y1∙2∙
1
Y1∙2∙
2
Y1∙2∙
3
Y1∙2∙
4
Y1∙2∙
5
Y1∙2∙
6
Y1∙2∙
7
Y13mean
level3
Y1∙3∙
1
Y1∙3∙
2
Y1∙3∙
3
Y1∙3∙
4
Y1∙3∙
5
Y1∙3∙
6
Y1∙3∙
7
Y21mean
level1
Y2∙1∙
1
Y2∙1∙
2
Y2∙1∙
3
Y2∙1∙
4
Y2∙1∙
5
Y2∙1∙
6
Y2∙1∙
7
Y22mean
level2
Y2∙2∙
1
Y2∙2∙
2
Y2∙2∙
3
Y2∙2∙
4
Y2∙2∙
5
Y2∙2∙
6
Y2∙2∙
7
Y23mean
level3
Y2∙3∙
1
Y2∙3∙
2
Y2∙3∙
3
Y2∙3∙
4
Y2∙3∙
5
Y2∙3∙
6
Y2∙3∙
7
Y31mean
level1
Y3∙1∙
1
Y3∙1∙
2
Y3∙1∙
3
Y3∙1∙
4
Y3∙1∙
5
Y3∙1∙
6
Y3∙1∙
7
Y32mean
level2
Y3∙2∙
1
Y3∙2∙
2
Y3∙2∙
3
Y3∙2∙
4
Y3∙2∙
5
Y3∙2∙
6
Y3∙2∙
7
Y33mean
level3
Y3∙3∙
1
Y3∙3∙
2
Y3∙3∙
3
Y3∙3∙
4
Y3∙3∙
5
Y3∙3∙
6
Y3∙3∙
7
Fundamental equation for MANOVA
 Variance
𝑖 𝑗
𝑌𝑖𝑗 − 𝐺𝑀
2
= 𝑛
𝑗
𝑌𝑗 − 𝐺𝑀
2
+
𝑖 𝑗
𝑌𝑖𝑗 − 𝑌𝑗
2
𝑆𝑆𝑡𝑎𝑡𝑎𝑙 = 𝑆𝑆 𝑏𝑔 + 𝑆𝑆 𝑤𝑔
Fundamental equation for MANOVA
 > one IV
𝑛 𝑘𝑚
𝑘 𝑚
𝐷𝑇𝑘𝑚 − 𝐺𝑀 𝑘𝑚
2 = 𝑛 𝑘
𝑘
𝐷 𝑘 − 𝐺𝑀 2 + 𝑛 𝑚
𝑚
𝑇 𝑚 − 𝐺𝑀 2
+ 𝑛 𝑘𝑚
𝑘 𝑚
𝐷𝑇𝑘𝑚 − 𝐺𝑀 2 − 𝑛 𝑘
𝑘
𝐷 𝑘 − 𝐺𝑀 2 − 𝑛 𝑚
𝑚
𝑇 𝑀 − 𝐺𝑀 2
𝑆𝑆 𝑏𝑔 = 𝑆𝑆 𝐷 + 𝑆𝑆 𝑇+𝑺𝑺 𝑫𝑻
Main effect
Interaction
Fundamental equation for MANOVA
Λ =
𝑆 𝑒𝑟𝑟𝑜𝑟
𝑆 𝑒𝑓𝑓𝑒𝑐𝑡 + 𝑆 𝑒𝑟𝑟𝑜𝑟
Approximate F = 𝑑𝑓1, 𝑑𝑓2 =
1−𝑦
𝑦
𝑑𝑓2
𝑑𝑓1
𝑦 = Λ
1
𝑆; 𝑠 = 𝑚𝑖𝑛 𝑝, 𝑑𝑓𝑒𝑓𝑓𝑒𝑐𝑡
𝑑𝑓1 = 𝑝 𝑑𝑓𝑒𝑓𝑓𝑒𝑐𝑡
𝑑𝑓2 = 𝑠 𝑑𝑓𝑒𝑓𝑓𝑐𝑡 −
𝑝 − 𝑑𝑓𝑒𝑓𝑓𝑒𝑐𝑡 + 1
2
−
𝑝 𝑑𝑓𝑒𝑓𝑓𝑒𝑐𝑡 − 2
2
Effect size
𝑝𝑎𝑟𝑡𝑖𝑎𝑙 η2
= 1 − Λ
1
𝑠
η2
= 1 − Λ
Discriminant function
 Special case of canonical analysis
χ1
χ2
X1
X2
X3
X4
X5
η1
η2
D1
D2
𝑟𝑐2
𝑟𝑐3
0
0
MANOVA
 Special case of canonical analysis
η1
η2
Y1
Y2
Y3
Y4
Y5
χ1
χ2
D1
D2
𝑟𝑐2
𝑟𝑐3
0
0
Some important issue
 MANOVA vs. ANOVA
 MANOVA
 rDVs = -.6
 ANOVA
 Highly postively correlated DVs
 Uncorrelated DVs
Some important issue
 Criteria for statistical inference
 Wilk’s lambda
 Error variance / (effect variance + error variance)
 Hotelling’s trace
 Effect variance / error variance
 Pillai’s criterion
 Pooled effect variance (general effect size)
 Roy’s gcr
 (Effect variance)1∙23…n / error variance
Some important issue
 Assessing DVs
 Univariate F with Bonferroni correction (rDVs = 0)
 Roy-Bargmann stepdown analysis (rDVs ≠ 0)
 Discriminant analysis (structure loading)
 Specific comparisons and trend analysis
 Mathematical resolution of combining variables
 Design complexity
 Within-subjects and between-within design
 Unequal sample size
rDVs ≠ 0
 Correlated DVs measure overlapping aspects of
same behavior
 Inflation of type 1 error
IV
DV
1
DV
2
Roy-Bargmann stepdown analysis
 A series of ANCOVA
 Assessing DV1
 DV1 as covariate, assessing DV2
 DV1 , DV2 as covariate, assessing DV3
 However,
 Order
 Homogeneity of regression
Homogeneity of regression
0
1
2
3
4
5
6
con
exp
Covariate variable
DV
 Heterogeneity → Johnson-Neyman correction
Procedure
1. Research question
2. Research design
3. Check the assumptions
4. Estimation of the MANOVA model and overall fit
5. Interpreting the effect of variables
6. Validating results
PRACTICE
心理技能訓練與團隊
 研究生雷齊家想知道不同心理技能訓練對團隊所
產生的效應。他的研究問題是,接受不同心理技
能訓練的運動員所知覺到的集體效能、團隊的正
確任務知識、能力評價與情緒覺察是否有差異,
其中,心理技能訓練包含團隊建立、意象訓練與
靜坐三種水準。
心理技能訓練與團隊
 研究生雷齊家想知道心理技能訓練對團隊所產生
的效應。他的研究問題是,運動員所知覺到的集
體效能、團隊的正確任務知識、能力評價與情緒
覺察在接受團隊建立訓練後是否有改變。

Manova