How To (and How Not To) Assess the Integrity of Managers (19 pages)
violations are discovered only after the fact. Therefore, ratings of an integrity competency are
unlikely to identify managers who have yet to be caught in the act.
Our last point is that integrity ratings are heavily skewed; nearly all managers receive high
ratings for integrity—and this does not seem plausible. For example, the CEO of a major corporation
with whom we worked believed that integrity was a simple matter: you either have it or you don’t.
He instituted a performance review that ended with a single dichotomous rating, “This manager
conducts business with ethics and integrity.” One year every one of his top 400-plus managers
received a “Yes” rating. Nonetheless, several of these managers were subsequently indicted for an
accounting scandal, bribing a government official, and conspiracy.
We conducted an empirical study of ratings on a competency-based integrity scale that is repre-
sentative of those commonly used in practice, but with psychometric properties that are perhaps
better than is typical. The study had two goals. First, we tested our expectation that few managers
would be rated as lacking integrity. Second, to the extent that ratings of integrity failed to identify
individuals at the low end of the continuum, we expected that they would not predict managerial
The data included subordinates’ ratings of 672 directors and vice presidents
(VPs) employed by a Fortune 500 technology firm. We only included managers and raters based in
the United States to control for cross-cultural effects. On average, each manager was rated by five
subordinates. For security reasons, precise demographic data were not made available with the
research database; however, we are very familiar with the organization and estimate that approxi-
mately 75% of managers and 65% of raters were middle-aged white males with a college education.
The company has a long tradition of promoting from within and we estimate that most of this sample
had worked in the company for five years or more.
Subordinate ratings were collected using a proprietary competency
survey developed for the host company. The ratings were gathered for the sole and stated purpose
of developmental feedback and all raters were assured that their ratings would be anonymous.
Further, they were told that their ratings would be averaged together with other subordinates’
ratings, and that those averages would only be reported if three or more subordinate surveys were
available. The organization had used this confidentiality policy with similar feedback surveys for
several years, and employees were generally familiar with the rules.
The survey included 23 items intended to measure five competencies. Those competencies (and
sample items) included Vision (“Comes up with a vision for the future”), Execution (“Keeps the
organization focused on executing the plan”), Managerial Courage (“Raises tough issues, even if it
makes people uncomfortable”), Building Talent (“Develops talent by coaching people”), and
Integrity (the three items were “Conducts business with integrity,” “Tells the truth,” and “Puts the
organization’s interests ahead of his or her personal ambitions”). Raters used a five-point response
scale, ranging from “0
ϭ ineffective” to “4 ϭ extremely effective.”
We evaluated the adequacy of the measurement model implied by the competency survey using
confirmatory factor analysis (CFA). First, we tested how well the rating data fit a five-correlated
factors model with five items each for Vision, Execution, Managerial Courage, and Building Talent
and the three items for Integrity. The results indicated that the data fit this model well according to
Hu and Bentler’s (1999) multiple criteria for judging model fit,
ϭ 1786.42, p Ͻ .001;
comparative fit index
ϭ .94, root mean square residual ϭ .03, and root mean square error of
ϭ .06. In fact, the data fit the five correlated-factors model much better than a
one-factor model representing a single, overall “good– bad leader” or “halo” dimension, providing
support for the validity of the measurement model representing five distinct, but empirically related,
Full results from the CFAs for the five- and one-factor models are available from the first author upon
KAISER AND HOGAN