related permanent organization that
parented this project, which allowed for
using one source for the collection of
data at both levels.
The study’s minimum sample size
was calculated as suggested by Hair,
Hult, Ringle, and Sarstedt (2014) by being
ten times the largest number of indicators used to measure a single construct.
Corporate governance was measured on
12 indicators; thus, the minimum sample
size was 120, which was met with 179
observations in the sample.
All constructs were assessed using five-point Likert scales and semantic differential questions.
The construct for Frequency of Ethical Issues (taken from Müller et al., 2014)
was assessed by asking how often each
of the three ethical issue types (
transparency, relationship, and optimization)
occurred in the last project, mission,
or other endeavor of the TO. The scale
ranged from never ( 1) to always ( 5), with
the midpoint being regularly ( 3).
TO governance was assessed along
the mechanism for trust (divided into
system trust, internal people trust, and
external people trust) and control. System trust was assessed using Müller
et al.’s (2014) two questions to assess
the extent to which managers of TOs
are ( 1) authorized to decide on ethical
issues themselves and ( 2) to implement
decisions on ethical issues themselves.
Higher values indicated higher trust by
the system. Internal and external trust
was assessed using the scales developed
by Chen, Chen, and Xin (2004). For both
internal trust (the TO’s manager trusts/
is trusted by the team) and external trust
(external stakeholders trust manager/
team of the TO) we asked for level of
trust in the TO, confidence in integrity,
confidence in right decision making,
consistency of actions with words, and
guidance by correct principles. Higher
values indicated higher levels of trust.
For the assessment of the con-
trol mechanism, we adopted the mea-
surement construct from Müller and
Lecoeuvre (2014). The questions
addressed organizations’ preference
for following the rules versus creat-
ing expected outcomes; preference for
more formal, tight or more informal,
loose control; adherence to job descrip-
tions; the need for individuals to com-
ply with their job descriptions or decide
on their own appropriate on-the-job
behavior; the role of support institu-
tions (such as PMOs) being process
or results oriented; and compliance
expectations by the organization, such
as prioritizing methodology compliance
over an individual’s own experience.
Higher scores indicated a preference for
outcome control and lower scores for
Corporate governance was assessed
using the 12 dimensions listed in the
literature review above, with low fulfillment of the good corporate governance
principles on the low end of the scale
and high fulfillment on the high end.
Validity and Reliability
Cronbach Alpha values above 0.6 indicated sufficient reliability for an exploratory study like this (Hair et al., 2014).
Validity was achieved through use of
published and tested measurement
constructs for ethical issues, TO governance control, people, and system
trust. Questions for corporate governance were newly developed, but based
on existing literature as well as their
validity tested during the questionnaire
pilot test. No issues were raised during
the pilot test and only minor grammatical changes were made. Answers to the
pilot test were not included in the final
Following suggestions by Hair et al.
(2014), we chose Partial Least Squares
Structural Equation Modeling (PLS-SEM)
over covariance-based SEM because the
nature of the study is exploratory and
aims for the highest predictability of a
single dependent latent construct, that
Issues. The underlying structural model
defines TO governance with its trust and
control constructs as exogenous latent
constructs; the model defines corporate
governance as an endogenous mediator
For the measurement model, we
designed all measurement items as
reflective items, assuming that they are
caused by their respective latent construct (Hair et al., 2014).
For the quantitative analysis, we
started with unrotated factor analysis
to test for internal consistency of the
construct measures and a Haman test
for assessing Common Methods Bias
(Podsakoff & Organ, 1986). All but one
item loaded high (.0.5) on their respective construct. This indicates no issues
with internal consistency and common methods bias; then we developed
the PLS-SEM model using SmartPLS.
Missing values were less than 5%, thus
acceptable for the chosen technique
(Hair et al., 2014).
For the analysis we followed Hair et al.
(2014) and report below in line with
As suggested by Hair et al., we
excluded indicator items with loadings
below 0.7, except in cases in which
the item was close to the threshold
and important for the reliability of the
construct. The final set of indicators is
shown in Figure 1 and described in the
Appendix (see Table 4 in Appendix).
SmartPLS does not require normal distribution of the data, but we checked
skewness and kurtosis nonetheless and
found all constructs to be within the
range of 61 standard deviation, thus
Table 2 shows the details and results of
the hypothesis tests. The initial analysis of the path model called for an
assessment of the main effect of trust
and control of the TO’s governance on
ethical issues. The path from control
to ethical issues is positively correlated
and significant at a t-value of 2.084 (at
5% significance level). This supports
Hypothesis 1: Higher levels of outcome