How to use SmartPLS software_Assessing Measurement Models_3

Download Report

Transcript How to use SmartPLS software_Assessing Measurement Models_3

Using the SmartPLS Software
Assessment of Measurement Models
Joe F. Hair, Jr.
Founder & Senior Scholar
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
2
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
3
Extended Reputation Model Constructs
Outcome Reputation Constructs (endogenous)
CUSL
COMP
CUSA
LIKE
=
=
=
=
loyalty (3 items)
competence (3 items)
satisfaction (1 item)
likability (3 items)
Driver Constructs (exogenous)
QUAL = quality of a company’s products/services and
customer orientation (8 items)
PERF = economic and managerial performance (5 items)
CSOR = corporate social responsibility (5 items)
ATTR = attractiveness (3 items)
All rights reserved ©. Cannot be reproduced or distributed without express written permission from
Prentice-Hall, McGraw-Hill, Sage, SmartPLS, and session presenters.
To evaluate reflectively measured models, we
examine the below:
• outer loadings
• composite reliability
• average variance extracted (AVE = convergent
validity)
• discriminant validity
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
5
To access
the
information
to evaluate
reflective
models
select one of
the reports
under this
tab.
All outer loading
When you select the Default Report
reflective construc
this is the screen you will get.
CUSL, and LIKE
above the minimum
To eliminate the unnecessary options
value of .70
on the navigation tree click on the
The loadings range
minus sign on the left side. You will get
of 0.7985 to a high
the simplified screen on the next slide.
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
6
All outer loadings of the
reflective constructs COMP,
CUSL, and LIKE are well
above the minimum threshold
value of .708.
The loadings range from a low
of 0.7985 to a high of 0.9173.
The “Toggle Zeros” button in the
task bar (top left of screen) was
used to improve the readability of
the results table above. This
button suppresses the zeros in
All rights reserved ©. Cannot
reproduced or distributed without express written permission from
the be
table.
Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
7
Composite Reliability
vs. Cronbach Alpha?
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
8
Composite Reliability
All three reflective constructs have
high levels of internal consistency
reliability, as demonstrated by the
above composite reliability values.
To obtain the above table that
shows the AVE, Composite
reliability, Communality,
Redundancy, etc., left click on the
Overview tab under the Quality
All rights reserved ©. Cannot be reproduced or distributed without express written permission from
Criteria.
Prentice-Hall, McGraw-Hill,
SmartPLS, and session presenters.
9
Discriminant validity is not present in the above constructs. Correlation
squared (variance shared between constructs = 64%) is larger than the
AVE of Y1 (only 0.55 – variance shared within construct = 55%).
Average Variance Extracted = AVE
The AVE values (convergent validity)
are well above the minimum required
level of .50, thus demonstrating
convergent validity for all three constructs.
To obtain the above table that
shows the AVE, left click on
the Overview tab under the
Quality Criteria.
All rights reserved ©. Cannot be reproduced or distributed without express written permission from
Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
12
Discriminant Validity
The off-diagonal values in the above
matrix are the correlations between
the latent constructs.
To obtain the above table that
includes information to
determine the Fornell-Larcker
criterion for discriminant
validity, left click on the Latent
Variable Correlations tab under
the Quality Criteria.
To obtain the shared values between
the constructs you must square these
correlations. See next slide where
this calculation is shown.
The results on the next slide
indicate there is discriminant
validity between all the constructs.
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
13
Discriminant Validity – Fornell-Larcker Criterion
Interconstruct Correlations
COMP
CUSA
CUSL
LIKE
COMP
1
0
0
0
CUSA
0.4356
1
0
0
CUSL
0.4496
0.6892
1
0
LIKE
0.6452
0.5284
0.6146
1
Squared Interconstruct Correlations
COMP
CUSA
CUSL
LIKE
COMP
0.6806
0
0
0
CUSA
0.1897
Single-Item Construct
0.0000
0.0000
CUSL
0.2021
0.4750
0.7484
0.0000
LIKE
0.4163
0.2792
0.3777
0.7471
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
Note: diagonal = AVEs
14
Discriminant Validity – Cross Loadings Criterion
Comparing the loadings across the
columns in the above matrix indicates
that an indicator’s loadings on its
own construct are in all cases higher
than all of its cross loadings with
other constructs.
To obtain the above table that
shows the cross loadings to
The results indicate there is
assess discriminant validity, left
click on the Latent Cross
discriminant validity between all
Loadings tab under the Quality
the constructs based on the cross
Criteria.
All rights reserved ©. Cannot be reproduced or distributed without express written
15
loadings criterion.
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
17
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
18
Empirical assessment of formative measurement models is not the
same as with reflective measurement models. This is because the
indicators theoretically represent the construct’s independent causes and
thus do not necessarily correlate highly. As a result, internal consistency
reliability measures such as Cronbach Alpha are not appropriate.
Instead, researchers should focus on establishing content validity
before empirically evaluating formatively measured constructs. This
process requires ensuring that the formative indicators capture all (or at
least major) facets of the construct.
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
19
All rights reserved ©. Cannot be reproduced or distributed without express written
permission from Sage, Prentice-Hall, McGraw-Hill, SmartPLS, and session presenters.
20
Corporate Reputation Extended Model
The extended corporate reputation model has three
main conceptual/theoretical components:
(1) the target constructs of interest (i.e., CUSA and
CUSL);
(2) the two corporate reputation dimensions, COMP and
LIKE, that represent key determinants of the target
constructs; and
(3) the four exogenous driver constructs (i.e., ATTR,
CSOR, PERF, and QUAL) of the two corporate reputation
dimensions.
Indicators for SEM Model Exogenous Constructs
– Assessing Content Validity –