Transcript Kock_2012_HumanComConf_SEMwWarpPLS

Structural Equation Modeling in
Human-centric Computing Research:
A Study of Electronic
Communication in Virtual Teams
Using WarpPLS
Ned Kock
Texas A&M International University
Outline of presentation
•
•
•
•
•
•
Fundamental problem with tech research
Types of models, including SEM models
Challenges in SEM analyses
WarpPLS
Electronic communication in virtual teams
YouTube video and WarpPLS resources
Fundamental problem with tech research
• A technology research team
develops a new technology
• They say: We developed a new
technology, and it works!
• But often they cannot answer these
questions:
– Is the technology associated with an
increase in performance in a given
task?
– If yes, how much?
Two-variable model
Coefficient of
association (beta)
and related P value
(probability that the
association is due to
chance)
Variable reflecting
degree of use of a
technology by a team
Variable reflecting
performance in a
given team task
Two-variable model: Measurement
Variables frequently
measured as columns on
a table where each row
refers to an individual or
team
Multiple regression model
Variables Tech and
Effic “compete” with
one another for the
explained variance in
Perf
Variable reflecting
the efficiency of a
team, in terms of
speed and cost
Similar to multiple
regression model,
but indirect effects
are also included
Path model
SEM model
SEM = structural
equation modeling
Similar to path
model, but variables
are measured
through other
variables
Latent variable (LV)
with 8 indicators,
hence “8i”
The 8 indicators that
make up LV
SEM model with results
P values < 0.05
suggest significant
associations among
LVs
Generality of SEM
• Two-variable models and tests, such as
ANOVA, are special cases of multiple
regression models and tests
• Multiple regression models are special cases
of path models
• Path models are special cases of SEM models
• SEM models and tests are very general and
can be employed in most technology studies
Challenges in classic SEM analyses
• Classic SEM is also known as covariancebased SEM
• Multivariate normality is assumed, but
frequently not the case
• Linearity is assumed, but frequently not the
case
• Usually indirect and total effects, with
respective P values, are not calculated
Nonlinearity among variables
• Most relationships among variables describing
natural and behavioral phenomena are
nonlinear
• Particularly common are noncyclical nonlinear
relationships, such as logarithmic, hyperbolic
decay, exponential decay, exponential, and
quadratic relationships
• Trying to force-model nonlinear relationships
as linear may cause false positives (type I
error) and false negatives (type II error) in
multivariate analyses
Force-modeling as linear
J-curve pattern modeled as a linear relationship
(R = .582; Variance explained = 33.9%)
Nonlinear modeling
J-curve pattern modeled as a nonlinear relationship
(R = .983; Variance explained = 96.7%)
WarpPLS as a solution
• Uses resampling techniques (e.g.,
bootstrapping) for calculation of P values,
therefore not requiring multivariate normality
• Takes nonlinearity among latent variables into
consideration in the calculation of coefficients
of association
• Indirect and total effects, with respective P
values, are calculated automatically
E-communication in virtual teams
• Study of 290 new product development teams
• New product development examples
– Develop a new car part
– Develop a new chemical
• The teams were geographically distributed
• The teams used a variety of e-communication
technologies (e.g., email and
videoconferencing)
SEM model to be tested
Prjmgt = degree to
which project
management
techniques were
used by team
ECM = degree of
electronic
communication
media use by team
Effic = efficiency of
team, in terms of
speed and cost
Success= success of
team, in terms of new
product sales and
profits
Test results: Direct effects
More electronic
communication media used by
team -> more project
management techniques used
Test results: Direct effects (2)
More project management
techniques used -> more team
efficiency in terms of speed
and cost
Test results: Direct effects (3)
More team efficiency in terms
of speed and cost -> more
team success in terms of new
product sales and profits
Test results: Direct effects (4)
More project management
techniques used -> more team
success in terms of new
product sales and profits
Test results: Direct effects (5)
More electronic communication media
used by team -> no more or less success in
terms of new product sales and profits;
when Prjmgt and Effic are controlled for
Test results: Direct effects (6)
More electronic communication media
used by team -> no more or less success in
terms of new product sales and profits;
when Prjmgt and Effic are controlled for
This suggests that electronic
communication media use exerts its
effects in indirect ways
Test results: Total effect of ECM
β=0.15
(P=0.01)
More electronic communication
media used by team -> more
success in terms of new product
sales and profits
Test results: Total effect of ECM (2)
β=0.15
(P=0.01)
Each 3-point increment in
use of electronic
communication media on a
0-10 scale (one standard
deviation), is associated with
a 15 % increase in Success
Test results: Nonlinearity
WarpPLS: YouTube videos
• SEM Analysis with WarpPLS (all steps)
– http://www.youtube.com/watch?v=yUojJaV3jlA
• Other YouTube video resources:
– Perform SEM Analysis and View Results with WarpPLS
(Step 5)
• http://youtu.be/Srp9ewa1T8o
– Various video clips
• http://www.scriptwarp.com/warppls
WarpPLS.com
WarpPLS blog
Acknowledgements: Resources used
•
•
•
•
Mobilemadlibs.com
Scriptwarp.com
Warppls.blogspot.com
WarpPLS.com
Thank you!