21. PLS估計方法選擇
• Factor Weighting Scheme
– the estimation of LV will be estimated
using correlations (inner estimation)
• Path Weighting Scheme
– the estimation of LV will be estimated
with regression and correlations
(depends on the arrows)
• Centroid Weighting Scheme
– used when the other does not run
(it is an older procedure)三星統計服務有限公司
32. 三星統計服務有限公司
SEM兩大家族
• Covariance-based SEM (CBSEM)
– Goal:樣本矩陣與模型期望共變異數最接近
– Data source: raw data, covariance matrix
or correlation matrix with standard
deviation
– Software: Amos, LISREL, EQS, Mplus…
33. 三星統計服務有限公司
SEM兩大家族
• Variance-based SEM (PLS)
– Goal:內生變數的被解釋能力最大
– Data source: raw data (format: .csv
或.txt)
– Software: SmartPLS, PLS-Graph,
VisualPLS (中文), PLS-GUI, SPAD PLS…
50. 形成型與反映型指標的差異
“Whereas reflective indicators are essentially interchangeable (and
therefore the removal of an item does not change the essential nature of
the underlying construct), with formative indicators ‘omitting an indicator
is omitting a part of the construct’.”
(Diamantopoulos / Winklhofer, 2001, p. 271)
反映型指標重點在指標
之間重疊的最大化
形成型指標重點
在互補的指標之
間重疊的最小化
Formative
Construct
Reflective
Construct
69. 三星統計服務有限公司
EXAMPLES from MISQ
An Empirical Examination of Individual
Traits as Antecedents to Computer
Anxiety and Computer Self-Efficacy
Managing Client Dialogues during
Information Systems Design to
Facilitate Client Learning
77. Path coefficients
三星統計服務有限公司
• PLS Algorithm HTML Report
Path Coefficients
Y
EI LOY SAT SC SQ
X
EI 0.428 0.176
LOY
SAT 0.263
SC 0.151 0.090
SQ 0.534
79. 結構係數t值
Path Coefficients (Mean, STDEV, T-Values)
• Report Html Report
• t >1.96 at p < 0.05, t > 2.58, at p < 0.01,
t > 3.29 at p < 0.001 for two-tailed tests
Original
Sample (O)
Sample
Mean (M)
Standard
Deviation
(STDEV)
Standard
Error
(STERR)
T Statistics
(|O/STERR|)
CI -> ATT 0.326 0.326 0.062 0.062 5.273
CI -> SAT 0.197 0.191 0.072 0.072 2.745
SAT -> ATT 0.304 0.304 0.057 0.057 5.349
SC -> ATT 0.204 0.206 0.063 0.063 3.260
SC -> SAT 0.088 0.090 0.053 0.053 1.654
SQ -> SAT 0.473 0.480 0.062 0.062 7.616
The bootstrap procedure is just used to compute standard error and
t-values of outer loadings, outer weights, path coefficients.
For these reasons: t = original / std error
91. 中介效果的分析
Original
Sample (O)
Standard
Error
(STERR)
CI -> ATT 0.284 0.069 b1
SAT -> ATT 0.233 0.073 b2
SC -> ATT 0.167 0.061 b3
SQ -> ATT 0.500 0.061
SQ -> CI 0.439 0.059 a1
SQ -> SAT 0.601 0.039 a2
SQ -> SC 0.415 0.052 a3
三星統計服務有限公司
92. 中介效果的解釋
• Sobel test z>1.96表示在α=0.05時,結果
是顯著的,表示中介效果存在.
•直接效果如果也是顯著,則為部份中介,若
不顯著則為完全中介.
• Variance Account For (VAF)=IE/TE
• VAF<20%表示無中介效果
20%<VAF<80%表示部份中介效果
VAF>80%表示完全中介效果
三星統計服務有限公司
99. PLS兩組模型係數
low interactivity high interactivity
Difference
between
coefficient
Original
Sample (O)
Standard
Error
(STERR)
Original
Sample (O)
Standard
Error
(STERR)
t df p
EFF -> ATT -.051 .134 -.047 .128 .018 75 .985
EFF -> VALUE .234 .114 .233 .138 .011 75 .991
SAT -> ATT .221 .153 .181 .201 .159 75 .874
SAT -> VALUE .438 .123 .655 .097 1.415 75 .161
VALUE -> ATT .621 .173 .632 .161 .047 75 .962
Path Coefficients (Mean, STDEV, T-Values)