ResearchPDF Available

Five-Item Guilt Proneness Scale (GP-5)

Authors:

Abstract and Figures

Cohen, T. R., Kim, Y., & Panter, A. T. (2014). The five-item guilt proneness scale (GP-5). Carnegie Mellon University, Pittsburgh, PA.
Content may be subject to copyright.
Five-Item Guilt proneness Scale 1
The Five-Item Guilt Proneness Scale (GP-5)
Taya R. Cohen1, Yeonjeong Kim1, & A. T. Panter2
1Carnegie Mellon University
2University of North Carolina at Chapel Hill
2014
Citation:
Cohen, T. R., Kim, Y., & Panter, A. T. (2014). The five-item guilt proneness scale (GP-5). Carnegie Mellon
University, Pittsburgh, PA. doi: http://dx.doi.org/10.17605/OSF.IO/G389S
The first published use of the GP-5 scale was in Study 3 of: Cohen, T. R., Panter, A. T., Turan, N., Morse, L. A., &
Kim, Y. (2014). Moral character in the workplace. Journal of Personality and Social Psychology, 107(5), 943-963.
doi: http://dx.doi.org/10.1037/a0037245
The GP-5 was used with a sample of law enforcement job applicants in: Cohen, T. R., Kim, Y., Jordan, K. P., &
Panter, A. T. (2016). Guilt-proneness is a marker of integrity and employment suitability. Personality and Individual
Differences, 92, 109-112. doi: http://dx.doi.org/10.1016/j.paid.2015.12.026
The first four items in the GP-5 scale were originally published in: Cohen, T. R., Wolf, S. T., Panter, A. T., & Insko,
C. A. (2011). Introducing the GASP scale: A new measure of guilt and shame proneness. Journal of Personality and
Social Psychology, 100(5), 947-966. doi: http://dx.doi.org/10.1037/a0022641
Five-Item Guilt proneness Scale 2
Guilt Proneness Measurement Studies:
Development of the GP-5
Study 1
In Study 1 we experimentally varied the labels for the response options in the GASP
scale. The purpose of manipulating the labels was to test whether small revisions could improve
the measurement properties of the instrument. In addition to investigating the effect of different
rating-scale labels, we also tested whether the measurement of guilt proneness could be
improved by adding an additional item. Although the entire 16-item GASP scale was
administered in this study, we limit our focus to the guilt proneness subscale (i.e., guilt-negative-
behavior-evaluation) given its status as particularly important element of moral character (Cohen
et al., 2011, 2013a, 2013b, 2014), along with our goal of creating a brief, targeted assessment of
this personality trait.
A critical property of any scale is that it has broad, meaningful content coverage rather
than overly narrow or redundant content (Furr, 2011; McCrae et al., 2011; Schmitt, 1996).
Because various types of transgressions are included in the GASP (i.e., receiving too much
change; secretly committing a felony; covering up a wine spill with a chair; and lying to people),
we can be reasonably sure that the four-item measure assesses a more general trait rather than an
overly narrow attribute limited to a particular type of transgression. Broad content coverage,
however, can have an adverse impact on internal consistency reliability (i.e., coefficient alpha).
Indeed, while some investigations find a high alpha coefficient for the GASP guilt proneness
scale (e.g., α =.82 in Schaumberg & Flynn, 2011, Study 2), others find alphas that are on the
lower side of acceptable (e.g., αs =.69, .70, and .71 in Cohen et al., 2011). As a brief scenario-
based measure, some degree of unreliability in the alpha coefficient is expected because each
Five-Item Guilt proneness Scale 3
item has unique variance due to idiosyncrasies of the particular transgression. The differences
among the transgression scenarios, however, are critical for ensuring that the underlying
theoretical construct is sampled broadly. One way to address these dual concerns of sampling the
construct broadly yet maintaining a high alpha level is to increase the length of the scale, as
alpha tends to increase with additional items (Furr, 2011; McCrae et al., 2011; Schmitt, 1996).
However, the brevity of the instrument is an important strength of the measure because short
scales can be included in a variety of assessment contexts and are less likely to burden
respondents. To balance these concerns, we tested the viability of including one additional item
to improve the measurement of guilt proneness.
Method
Participants
Participants were recruited from three online research participation pools: (1) a paid pool
maintained by Amazon Mechanical Turk (MTurk participants received $0.25 in MTurk credit);
(2) a paid pool maintained by the Center for Behavioral and Decision Research (CBDR) at
Carnegie Mellon University (CBDR participants were entered into a lottery for $25 giftcards);
and (3) an undergraduate student pool maintained by Carnegie Mellon University (students
received class credit).
We surveyed 835 participants, but only 737 (88.3%) passed the two attentiveness checks
that were embedded at random points in the survey. The attention checks were simple face-valid
items that asked participants to check a box indicating that they were paying attention: “It is
important for our research that you read each question carefully. To let us know that you are
paying attention right now, please select box number 2 [box number 4]”. After excluding the
Five-Item Guilt proneness Scale 4
participants who failed these checks (14.2% MTurk, 10.4% CBDR, 13.4% students), we were
left with a final sample of 737 adults (n = 194 MTurk, n = 459 CBDR, n = 84 students).
Participants lived in 41 U.S. states, with the majority being from Pennsylvania (58.2%),
and the others spread relatively evenly across the other states. The average age was 28.3 years
(range = 18 to 74). Women comprised 64.8% of the sample. The majority of participants (60.7%)
self-identified as White, 20.9% as Asian, 6.8% as Black, 2.9% as Hispanic, and 8.7% as other or
multi-racial. In response to a question about employment, 45.2% reported being full-time
students, 24.7% reported full-time employment, 12.5% reported part-time employment, 12.6%
reported being unemployed, 2.3% reported being retired, and 2.7% other.
Procedure
The “personality and social behavior” online study took approximately 15 minutes to
complete. Participants were assured that “all information collected in this survey is kept
completely confidential and secure, and only the research team will have access to the data.” The
experiment began with the 16-item GASP scale (Cohen et al., 2011); the items were presented in
a randomized order for each participant. Each participant was randomly assigned to one of three
experimental conditions. The difference between these conditions was the labels used in the
GASP’s seven-point rating scale. The first condition (“original”) used the GASP’s original
labels: very unlikely (1), unlikely (2), slightly unlikely (3), about 50% likely (4), slightly likely
(5), likely (6), very likely (7). The second condition (“extremely”) used alternate labels:
extremely unlikely (1), very unlikely (2), unlikely (3), about 50% likely (4), likely (5), very
likely (6), extremely likely (7); as did the third condition (“not at all”): not at all likely (1),
slightly likely (2), somewhat likely (3), moderately likely (4), quite a bit likely (5), very likely
(6), extremely likely (7).
Five-Item Guilt proneness Scale 5
For 370 of the 737 participants, an additional guilt proneness item was embedded in the
GASP. This additional item was added after the study was already in progress, so the 367
participants who completed the study prior to its inclusion have missing data on the new item.
The new item is:
“Out of frustration, you break the photocopier at work. Nobody is around and you leave
without telling anyone. What is the likelihood you would feel bad about the way you
acted?”
Following the GASP, participants completed an attentiveness check, and then the
HEXACO-60 Personality Inventory (Ashton & Lee, 2007, 2009; Lee & Ashton, 2012). The
HEXACO-60 measures the six major dimensions of personality. Each of the six broad
dimensions contains four facets: Honesty-Humility (sincerity, fairness, greed-avoidance,
modesty); Emotionality (fearfulness, anxiety, dependence, sentimentality); Extraversion (social
self-esteem, social boldness, sociability, liveliness); Agreeableness (forgiveness, gentleness,
flexibility, patience); Conscientiousness (organization, diligence, perfectionism, prudence);
Openness to Experience (aesthetic appreciation, inquisitiveness, creativity, unconventionality).
Participants indicated their agreement with the 60 HEXACO items (presented in a randomized
order for each participant) with a five-point rating scale: Strongly Disagree (1), Disagree (2),
Neutral (neither agree nor disagree) (3), Agree (4), Strongly Agree (5). Composite scores were
created by averaging the items in each factor, after reverse-keying the appropriate items.
After the HEXACO-60, participants answered a question about their employment status.
Those who indicated full-time or part-time employment (n = 274) answered additional questions
about their job and work behaviors; the remainder of the participants (n = 463) skipped the job
section, and continued on to the final section, which assessed demographic information. The
Five-Item Guilt proneness Scale 6
study ended with a box for comments. Due to too few respondents in each condition completing
the job section, we do not discuss those variables further.
Results
Classical Test Theory Analysis
In accordance with a CTT approach to scale construction, we compared the coefficient
alphas of the three versions of the guilt proneness scale. Due to the missing data on the new guilt
proneness item, we examined alpha coefficients with and without the new item. Alpha was
highest in the second condition (“extremely”): four-item α = .73, N = 255; five-item α = .79, N =
127. It was lowest in the third condition (“not at all”): four-item α = .65, N = 233; five-item α =
.74, N = 118. The original condition was somewhat similar to the extremely condition: four-item
α = .71, N = 249; five-item α = .79, N = 125.
We averaged the five guilt proneness items to form a composite guilt proneness score,
and then compared the descriptive statistics for the three conditions. Table 1 shows the results for
the participants who answered all five items. The mean score in the first (“original”) condition
was significantly higher than the third (“not at all”) condition, t(241) = 2.70, p = .007, and
marginally higher than the second (“extremely”) condition, t(250) = 1.94, p = .053. The
variances did not significantly differ, as evidenced by the nearly identical standard deviations
and standard errors, and non-significant Levene’s tests for equality of variances (Fs < 1).
Factor Analysis
Next, we examined the dimensionality of the items via confirmatory factor analysis
(CFA). We computed a multigroup CFA to test for differences in the factor loadings between the
three experimental conditions. A one-factor solution was specified. We estimated the models in
MPlus 6.0 (Muthén & Muthén, 1998-2011) and used the “missing” option to account for the
Five-Item Guilt proneness Scale 7
missing data. WLSMV estimation was used because of the ordinal nature of the response
options. To set the scale of the latent factor, its variance was set to 1 in each group. The five
factor loadings and the item thresholds were freely estimated. The mean of the guilt proneness
factor was set to 0 in the original condition, and estimated in the others.
Model fit was excellent: RMSEA = .036 (90% C.I. = 0.006, 0.056); CFI = 0.989; TLI =
0.995; χ2(df = 63, free parameters = 57) = 83.20, p = .045. The factor loadings for each condition
are presented in Table 2; all were strong and statistically significant (p < .001). Of particular
interest, the new guilt proneness item about breaking the copy machine had factor loadings as
high as or higher than the four previously validated guilt proneness items.
Item Response Theory Analysis
IRT provides item and scale level information, as well as information about response
categories. We examined the item and scale curves to evaluate the measurement properties of the
GP-5. Our models used a two-parameter graded response model (Samejima, 1969), which
estimates discrimination and difficulty parameters. The item parameter estimates from the IRT
analysis are shown in Table 3. The a column shows that the “extremely” condition produced the
highest discrimination parameter for three of the five items. For example, the discrimination
parameter for the new item (“break the copier”) was 2.88 in the “extremely” condition, but only
2.22 in the “original” condition, and 2.41 in the “not at all” condition.
The IRT analysis provided item characteristic curves (ICCs) for each item in each of the
three conditions. In general, the “extremely” condition showed psychometrically more desirable
ICCs than the “original” or “not at all” conditions, as illustrated by Figure 1. As shown, the
“extremely” condition had relatively sharper bell-shaped curves for each rating scale option
without much overlap across options. Figure 2 shows the total information curve (TIC) for the
Five-Item Guilt proneness Scale 8
guilt proneness scale for each of the three conditions. As shown, the “extremely” condition
provided the highest level of information and thus lowest measurement errors among the three
conditions across all latent trait levels. Thus, the IRT analyses, like the CTT and CFA analyses,
suggest that the rating scale labels in the “extremely” condition are better than those in the
“original’ and “not at all” conditions.
Convergent and Discriminant Validity Analysis
We tested for validity evidence of the five-item guilt proneness scale by examining its
correlations with the six HEXCO factors. Consistent with past research and the notion that guilt
proneness is a moral character trait (Cohen et al., 2011, 2013b, 2014), we predicted a large
positive correlation between the guilt proneness scale and Honesty-Humility and a moderate
positive correlation between guilt proneness and Conscientiousness. We also expected a
moderate positive correlation between guilt proneness and Agreeableness in light of past
empirical results and the idea that guilt proneness and Agreeableness both capture a concern for
other people (Cohen et al., 2011, 2013b, 2014). Small non-significant correlations were expected
between guilt proneness and Emotionality, Extraversion, and Openness to Experience. Table 1
shows these correlations for the participants who answered all five guilt proneness items.
The correlations in each of the three conditions provide evidence of convergent and
discriminant validity of the five-item guilt proneness scale. In support of guilt proneness’ status
as a moral character trait, the scale was significantly correlated with Honesty-Humility and
Conscientiousness—traits that are also indicative of moral character (Cohen et al., 2013b, 2014;
Lee & Ashton, 2012). The correlations with Agreeableness were significant in the first two
conditions, but not the third. As expected, Extraversion was not correlated with guilt proneness
in any of the conditions. Unexpectedly, significant correlations emerged with Emotionality and
Five-Item Guilt proneness Scale 9
with Openness to Experience in the second (“extremely”) condition. The subsequent studies also
included the HEXACO-60, allowing us to test to reliability of the unexpected correlations.
Discussion
The results of Study 1 suggest that the guilt proneness subscale of the GASP can be
improved by the inclusion of the new fifth item and by altering the response labels so that the
anchors are extremely unlikely/extremely likely rather than very unlikely/very likely or not at all
likely/extremely likely. These changes made the scale more reliable, as evidenced by
improvements in coefficient alpha. Furthermore, changing the anchors in the rating scale to
extremely unlikely/extremely likely versus very unlikely/very likely lowered the mean score of
the scale and did not alter its variance. This result is important because participants tend to use
the upper-end of the response scale when responding to guilt proneness items. Moreover, IRT
analyses also favored the extremely unlikely/extremely likely anchors. The version of the
measure with these response labels provided more information and less error across all trait
levels. Finally, the CFA indicated that the five items have a unidimensional structure despite
broadly sampling the construct of guilt proneness with different types of transgressions and
responses. Each of the five items loaded highly on the latent guilt proneness factor. Overall, then,
Study 1 indicated that small changes to the scale labels improved the quality of the guilt
proneness measure, as did the inclusion of an additional item.
Study 2
Study 1 offered evidence supporting a revised guilt proneness scale. In Study 2, we tested
this measure—the Five-Item Guilt Proneness Scale (GP-5)—in a new sample. Rather than
recruiting participants from online subject pools as we did in Study 1, we instead surveyed
graduate business students a top school in the United States. This sample is a good proxy for the
Five-Item Guilt proneness Scale 10
population managers and human resource professionals would evaluate when making personnel
selection and promotion decisions.
Method
Students in a required Masters of Business Administration (MBA) class on ethics at the
Tepper School of Business at Carnegie Mellon University completed an online survey the week
before the course began. They were assured that their responses were confidential, and only the
197 students who consented to have their anonymous data used for research purposes were
included in the study (two students declined the use of their data).
The survey began with the GP-5 and the HEXACO-60 (Ashton & Lee, 2009), with the
order of the scales and the order of the items within each scale randomized for each participant.
After those measures, participants responded to additional questions for class purposes not
related to this research. The GP-5 used the items with the anchors extremely unlikely/extremely
likely (identical to condition two from Study 1).
Results & Discussion
Table 4 presents descriptive statistics for the GP-5, and bivariate correlations of the scale
with the six HEXACO scales. As in Study 1, the internal consistency was good (α = .81). The
mean score of 5.41 out of 7 was higher than in Study 1 (4.93), possibly because of differences in
the sampling methods. As expected, the GP-5 was strongly correlated with Honesty-Humility,
moderately correlated with Conscientiousness, and uncorrelated with Emotionality, Extraversion,
and Openness to Experience. The GP-5 was uncorrelated with Agreeableness, which is
consistent with Study 1, but contrary to our expectations based on prior research (Cohen et al.,
2011, 2013b).
Five-Item Guilt proneness Scale 11
We conducted a one-factor CFA of the five guilt proneness items (see Table 5). The CFI
and TLI indicated very good fit, but the RMSEA and chi-square were higher than expected.
Nonetheless, the factor loadings were all high and statistically significant (ps < .001), consistent
with the results from Study 1. Again, the new guilt proneness item about breaking the copy
machine had factor loadings as high as the four previously validated guilt proneness items.
Overall, the results of Study 2 are consistent with Study 1, and provide further evidence of the
reliability and validity of the GP-5 scale.
Five-Item Guilt proneness Scale 12
Study 3
In our final study, we sought to determine whether the GP-5 scale could be improved by
varying the number of options in the rating scale. We used an experimental design to compare a
seven-point, five-point, and four-point response scale. Underutilized categories are problematic
because it means that the rating scale does not differentiate respondents as intended. On the one
hand, more response options provide a finer-grained analysis that helps to differentiate
individuals, assuming the respondents are not confused by small differences between adjacent
categories. On the other hand, too many response options could cause respondents with the same
latent trait level to differentially endorse rating categories resulting in less reliable scale scores.
Thus, it is important to determine the optimal number of rating points to best differentiate
individuals without resulting in confounding endorsement.
In addition to examining the optimal number of response options, this study also allowed
us to investigate convergent and discriminant validity further. Specifically, we measured the
HEXACO dimensions, as well as empathic concern and perspective taking, which are both moral
character traits and expected to show moderately high correlations with the GP-5, consistent with
previous research (Cohen et al., 2011, 2014). Study 4 also included measures of personal distress
and self-esteem, to provide evidence of discriminant validity. These patterns have been
documented in prior research with the GASP (Cohen et al., 2011) and TOSCA-3 (Tangney &
Dearing, 2002), so we expected to replicate them here with the GP-5.
Finally, this study included a social desirability inventory. The developers of the GASP
reported a null correlation between guilt proneness and the self-deception aspect of social
desirability but they did not include an impression management measure in their scale
development work (Cohen et al., 2011). A high correlation between guilt proneness and
Five-Item Guilt proneness Scale 13
impression management would indicate that people high in guilt proneness are concerned about
other people’s perceptions of them. However, correlations with impression management do not
necessarily provide evidence of artificial response style variance. Rather, they are likely
indicative of substantive personality variance, at least to some extent, given that self-ratings and
observer ratings of impression management are correlated (r = .45 in Lee et al., 2003). We
expected a moderately high correlation between impression management and guilt proneness
given that both constructs reflect a concern for other people. Furthermore, other moral character
traits tend to correlate highly with impression management (e.g., the correlation of impression
management with Honesty-Humility has been shown to be substantial: r = .50 in Lee, Gizzarone,
& Ashton, 2003).
Method
This study had a design similar to Study 1, with exceptions related to the specifics of the
experimental conditions and the measures included in the survey.
Participants
Participants were recruited from the same three pools as in Study 1. In this study, the
MTurk participants were paid $0.50. After excluding the participants who failed the two
attentiveness checks (8.7% MTurk, 15.2% CBDR, 12.7% students), we were left with a final
sample of 699 adults (n = 295 MTurk, n = 273 CBDR, n = 131 students).
Participants lived in 47 U.S. states, with the majority being from Pennsylvania (51.9%),
and the others spread relatively evenly across the other states. Women comprised 63.2% of the
sample. The majority of participants (63.8%) self-identified as White, 19.6% as Asian, 7.0% as
Black, 2.9% as Hispanic, and 6.7% as other or multi-racial. In response to a question about
employment, 36.2% reported being full-time students, 24.3% reported full-time employment,
Five-Item Guilt proneness Scale 14
17.3% reported part-time employment, 14.0% reported being unemployed, 4.0% reported being
retired, and 4.1% other.
Procedure
The study began with the GP-5 items interspersed with the 12 other items from the GASP
(Cohen et al., 2011) presented in a randomized order for each participant. Each participant was
randomly assigned to one of three conditions. The difference between these conditions was the
number of response options in the rating scale. The first condition used a seven-point scale:
extremely unlikely (1), very unlikely (2), unlikely (3), about 50% likely (4), likely (5), very
likely (6), extremely likely (7). The second condition used a five-point scale: extremely unlikely
(1), unlikely (2), about 50% likely (3), likely (4), extremely likely (5); and the third used a four-
point scale: extremely unlikely (1), unlikely (2), likely (3), extremely likely (4).
Next, participants completed an attentiveness check, and then the HEXACO-60
Personality Inventory (Ashton & Lee, 2009). Then, participants completed the empathic concern
(EC), perspective taking (PT), and personal distress (PD) scales from the Interpersonal
Reactivity Index (IRI) (Davis, 1983), followed by the Rosenberg Self-Esteem scale (Rosenberg,
1965). The 21 IRI items were presented in a randomized order for each participant. Participants
indicated how well each item described them using a five-point scale: does not describe me well
(1), (2), describes me moderately well (3), (4), describes me very well (5). The 10 RSE items
were also presented in a randomized order for each participant. Participants indicated their
agreement with each statement: strongly disagree (1), disagree (2), agree (3), strongly agree (4).
Composite scores were created by averaging the items in each scale, after reverse-keying the
appropriate items.
Five-Item Guilt proneness Scale 15
After completing the IRI and RSE scales, participants completed a 40-item inventory of
socially-desirable responding: the Balanced Inventory of Desirable Responding (BIDR) Version
6 (Paulhus, 1991, p. 40). The BIDR was added while data collection was already in progress
(after 87 participants had completed the study), so analyses with the BIDR have missing data.
Participants indicated how much they agreed with each of the 40 BIDR statements (randomized
for each participant) using a seven-point scale: not true (1), (2), (3), somewhat true (4), (5), (6),
very true (7). The BIDR contains two 20-item subscales: self-deception and impression
management. The inventory is scored differently than other measures: Participants are given a
point for extreme responses (6 or 7), after half of the items have been reversed. Thus, total scores
for self-deception and impression management range from 0 to 20, with subjects giving
excessively desirable responses receiving high scores. After the BIDR, there was a demographics
section and the study ended with a box for comments.
Results
Table 6 presents descriptive statistics and bivariate correlations with the other scales, and
Table 7 presents the correlations among the various inventories included in the study.
As shown, the alphas in the seven-point-scale condition and the five-point-scale condition were
good (αs =.79, .80), whereas the alpha in the four-point-scale condition was lower (.71). The
correlations with the theoretically-related and unrelated measures were as predicted in the seven-
point-scale and five-point-scale conditions, but were weaker in the four-point-scale condition.
Specifically, in the seven-point-scale and five-point-scale conditions, we observed moderate to
high correlations with Honesty-Humility, Conscientiousness, Agreeableness, empathic concern,
perspective taking, and impression management. The other measures had weaker or non-
significant relationships, thus providing evidence of both convergent and discriminant validity.
Five-Item Guilt proneness Scale 16
These results suggest that the four-point-scale does not function as well as the versions with five-
or seven-point response scales.
For each of the three conditions, we conducted a one-factor CFA of the five items. As
shown in Table 8, model fit was good in all the models, as indicated by the high CFI and TLI
statistics, and high factor loadings. RMSEA and chi-square values were higher than expected; as
in Study 2, it is unclear why those indexes performed differently than the others. The factor
loadings in the four-point scale condition were lower than in the other two conditions, providing
further evidence that the five-point and seven-point scales perform better than the four-point
scale.
Next, we conducted an IRT analysis to determine how many response options are
optimal. A two-parameter graded response IRT model was fitted to the data in each of the three
conditions. Table 9 shows the results. It was found that the four-point rating scale was not as
good as the five- or seven-point rating scales, in that four out of five of items had lower
discrimination parameters (a) when the four-point response scale was used. The differences in
the discrimination parameters between the five-point and seven-point conditions were relatively
small compared to the difference between the four-point condition and the others.
Item information and test total information depends on the number of scale points, thus
they cannot be used for the comparison of different number of rating points. By contrast, ICCs
provide information about which rating options best discriminate respondents with different
latent traits without confusing adjacent categories. Accordingly, the results of ICC comparisons
are presented in Figure 3.
As shown, in the seven-point condition, the curves from each category were overlapping
and the probabilities for middle range categories were not as high as in the five-point condition.
Five-Item Guilt proneness Scale 17
Specifically, in the seven-point rating condition, items 1, 3, and 4 had underutilization problems
for some rating scale options. Specifically, the second and third categories did not fully function
in detecting differences according to latent scores. For example, for item 1 (too much change),
the probability of selecting the first category was higher than that of second or third category for
individuals whose latent trait was lower than approximately -1 standard deviation. That is,
category 2 and 3 did not work properly for distinguishing different latent scores. Similarly, for
the new item (break the copier), it was found that the fourth category was not fully utilized in the
seven point condition. The probability of selecting this category was not highest at any level of
the latent score. Thus, the ICCs suggest that the seven-point rating scale did not fully function for
detecting different latent trait levels of guilt proneness.
In contrast to the results for the four-point and seven-point rating scales, the five-point
rating scale resulted in better utilization of all the scale categories. All five categories showed
distinctive non-overlapping probability patterns. For example, for item 1 (too much change), the
probability of selecting the first category was highest for individuals who had latent trait scores
of less than -2. For individuals with latent trait scores between -2 and -1.2, the probability of
selecting the second category was the highest. The probability of marking the third category was
highest between -1.2 and -0.30. Individuals whose latent traits were located between -0.30 and
0.70 would be expected to select the fourth category, and individuals whose latent guilt
proneness scores were greater than 0.70 would be expected to mark the fifth category.
Discussion
We systematically examined the number of response options in the GP-5 using a
randomized experimental design and a variety of statistical methods. Our conclusion is that a
five-point rating scale is best, with the following labels for the response options: 1 = Extremely
Five-Item Guilt proneness Scale 18
Unlikely; 2 = Unlikely; 3 = About 50% Likely; 4 = Likely; 5 = Extremely Likely. The GP-5 with
the five-point rating scale had moderately strong correlations with four key moral character traits
identified by prior research (Cohen et al., 2014): Honesty-Humility, Conscientiousness, empathic
concern, and perspective taking. It was also correlated to a lesser extent with Agreeableness and
Emotionality. In support of its discriminant validity, the GP-5 was uncorrelated with
Extraversion, Openness to Experience, self-esteem, and personal distress.
In regards to social desirability, we found that the GP-5 with the five-point rating scale
was uncorrelated with self-deception, which suggests that people are aware of their own level of
guilt proneness and able to report on it at least somewhat accurately. This conclusion is
consistent with prior empirical evidence of high self-other agreement for guilt proneness among
friends and coworkers who know each other well (Cohen et al., 2013b). The GP-5 was
moderately correlated with the impression management aspect of social desirability, which we
hypothesized based on the idea that both guilt proneness and impression management reflect an
orientation toward other people. Consistent with this interpretation, empathic concern and
perspective taking were also moderately to highly correlated with impression management, as
were Honesty-Humility, Agreeableness, and Conscientiousness.
Five-Item Guilt proneness Scale 19
Table 1
Study 1: Descriptive Statistics and Bivariate Correlations for the Five-Item Guilt Proneness
Scale (GP-5)
Condition 1 Condition 2 Condition 3
Anchors very unlikely (1),
very likely (7)
extremely unlikely (1),
extremely likely (7)
not at all likely (1),
extremely likely (7)
N 125 127 118
Alpha 0.79 0.79 0.74
Mean 5.23 4.93 4.79
Standard Deviation 1.24 1.24 1.29
Standard Error of Mean 0.11 0.11 0.12
Skew -0.61 -0.33 -0.58
Range 1.00 to 7.00 1.00 to 7.00 1.00 to 7.00
Percentiles: 25th 4.40 4.00 4.00
Percentiles: 50th 5.40 4.80 5.00
Percentiles: 75th 6.20 6.00 5.80
Correlations
Honesty-Humility .46** .48** .44**
Emotionality .22* .36** .36**
Extraversion .08 -.05 .06
Agreeableness .19* .18* .10
Conscientiousness .23* .37** .33**
Openness to Experience .03 .24** .18
Note. Composite guilt proneness scores were created by averaging the five items.
*p < .05, ** p < .001
Five-Item Guilt proneness Scale 20
Table 2
Study 1: Factor Loadings for the Five-Item guilt proneness Scale (GP-5)
Condition 1
Condition 2 Condition 3
Anchors very unlikely (1),
very likely (7)
extremely unlikely (1),
extremely likely (7)
not at all likely (1),
extremely likely (7)
N 249 255 233
Factor mean 0.00 -0.15 -0.34
Factor loadings
(with standard errors)
(1) too much change 0.66 (0.04)** 0.49 (0.06)** 0.41 (0.07)**
(2) secret felony
0.72 (0.04)** 0.76 (0.09)** 0.71 (0.09)**
(3) cover wine spill 0.58 (0.05)** 0.55 (0.07)** 0.57 (0.08)**
(4) tell lies 0.78 (0.03)** 0.66 (0.07)** 0.65 (0.08)**
(5) break the copier 0.78 (0.05)** 0.76 (0.11)** 0.72 (0.11)**
Note. *p < .05, ** p < .001
Five-Item Guilt proneness Scale 21
Table 3
Study 1: Item Parameters from Item Response Theory (IRT) Analyses
Item Condition Item parameters
a b1 b2 b3 b4 b5 b6
(1) too much change original 1.45 -1.84 -1.18 -0.77 -0.17 0.33 1.10
extremely 1.70 -2.30 -1.55 -0.90 -0.33 0.21 1.06
not at all 1.13 -2.47 -1.60 -0.80 -0.14 0.39 1.47
(2) secret felony original 1.93 -2.77 -1.96 -1.61 -1.04 -0.52 0.24
extremely 2.06 -2.29 -1.96 -1.43 -0.86 -0.44 0.28
not at all 1.92 -2.80 -1.96 -1.39 -0.87 -0.46 0.21
(3) cover wine spill original 1.36 -3.31 -2.53 -1.88 -1.24 -0.46 0.63
extremely 1.40 -3.15 -2.64 -1.92 -1.26 -0.43 0.62
not at all 1.47 -3.28 -2.40 -1.71 -1.11 -0.53 0.30
(4) tell lies original 2.44 -2.23 -1.40 -1.03 -0.49 0.03 1.05
extremely 2.41 -2.55 -1.62 -0.94 -0.32 0.20 1.06
not at all 1.73 -2.44 -1.64 -1.00 -0.35 0.38 1.12
(5) break the copier original 2.22 -2.35 -1.74 -1.32 -0.69 -0.06 0.83
extremely 2.88 -2.12 -1.68 -1.27 -0.70 -0.19 0.70
not at all 2.41 -2.56 -1.75 -1.29 -0.72 -0.26 0.51
Five-Item Guilt proneness Scale 22
Table 4
Study 2: Descriptive Statistics for the Five-Item Guilt Proneness Scale (GP-5) and Bivariate
Correlations with HEXACO Personal Inventory
Study 2 (N = 197)
Alpha .81
Mean 5.41
Standard Deviation 1.13
Standard Error of Mean 0.08
Skew -0.75
Range 1.40 to 7.00
Percentiles: 25
th
4.60
Percentiles: 50th 5.60
Percentiles: 75
th
6.20
Correlations
Honesty-Humility
.49**
Emotionality -.01
Extraversion .11
Agreeableness .07
Conscientiousness .29*
Openness to Experience .11
Note. Composite guilt proneness scores were created by averaging the five items.
*p < .05, ** p < .001
Five-Item Guilt proneness Scale 23
Table 5
Study 2: Factor Loadings for the Five-Item Guilt Proneness Scale (GP-5)
Model Fit Study 2 (N = 197)
χ2(df = 5) 25.95**
RMSEA (90% C.I.) .146 (.094, .204)
CFI .975
TLI .950
Item
Factor loadings (with standard
errors)
(1) too much change 0.72 (0.04)**
(2) secret felony
0.71 (0.04)**
(3) cover wine spill 0.68 (0.04)**
(4) tell lies 0.80 (0.04)**
(5) break the copier 0.76 (0.04)**
Note. *p < .05, ** p < .001
Five-Item Guilt proneness Scale 24
Table 6
Study 3: Descriptive Statistics and Bivariate Correlations for the Five-Item Guilt Proneness
Scale (GP-5)
Seven-Point Rating
Scale
Five-Point Rating
Scale
Four-Point Rating
Scale
N 232 232 235
Alpha .79 .80 .71
Mean 5.31 3.82 3.11
Standard Deviation 1.17 0.86 0.58
Standard Error of Mean 0.08 0.06 0.04
Skew -0.77 -0.60 -0.47
Range 1.20 to 7.00 1.40 to 5.00 1.20 to 4.00
Percentiles: 25
th
4.60 3.20 2.80
Percentiles: 50th 5.60 4.00 3.20
Percentiles: 75
th
6.20 4.40 3.60
Correlations
Honesty-Humility .54** .49** .58*
Emotionality .13* .26** .14*
Extraversion .06 .04 .08
Agreeableness .32** .21* .12
Conscientiousness .29** .31** .33**
Openness to Experience .10 .06 .10
Empathic Concern .41** .34** .29**
Perspective Taking .34** .30* .15*
Personal Distress -.02 -.09 -.03
Self-Esteem .19** .05 .11
Self-Deception .19** .09 .15*
Impression Management .55** .37** .50**
Note. Composite guilt proneness scores were created by averaging the five items.
*p < .05, ** p < .001
Five-Item Guilt proneness Scale 25
Table 7
Study 3: Bivariate Correlations
M SD 1 2 3 4 5 6 7 8 9 10 11 12
1. Honesty-Humility 3.33
0.65
(.75)
2. Emotionality 3.24
0.64
.03 (.78)
3. Extraversion 3.25
0.69
-.01 -.18 (.82)
4. Agreeableness 3.14
0.63
.31 -.12 .16 (.79)
5. Conscientiousness 3.59
0.60
.29 .05 .18 .12 (.78)
6. Openness to Experience 3.60
0.65
.14 -.03 .13 .10 .16 (.77)
7. Empathic Concern 3.74
0.72
.31 .37 .17 .29 .20 .30 (.82)
8. Perspective Taking 3.52
0.72
.25 .04 .20 .50 .23 .30 .48 (.82)
9. Personal Distress 2.57
0.76
-.13 .50 -.37 -.19 -.32 -.20 -.03 -.23 (.82)
10. Self-Esteem 3.01
0.59
.17 -.24 .65 .25 .35 .12 .16 .23 -.42 (.91)
11. Self-Deception 4.69
3.61
.23 -.20 .38 .19 .38 .23 .18 .23 -.41 .52 (.77)
12. Impression Management
5.44
4.03
.53 .02 .18 .35 .38 .10 .30 .29 -.21 .33 .61 (.82)
Note. N = 699 (N = 611 for the self-deception and impression management correlations). Bivariate correlations are presented with
coefficient alphas on the diagonal. Correlations greater than |.08| are significant at p < .05. Correlations greater than |.12| are
significant at p < .001.
Five-Item Guilt proneness Scale 26
Table 8
Study 3: Factor Loadings for the Five-Item Guilt Proneness Scale (GP-5) in each Condition
Seven-Point Rating
Scale
Five-Point Rating
Scale
Four-Point Rating
Scale
N 232 232 235
Model Fit
χ2(df = 5) 29.05 ** 19.86* 6.67
RMSEA (90% C.I.) .144 (.096, .197) .113 (.064, .167) .038 (.000, .104)
CFI .974 .981 .995
TLI .947 .962 .991
Factor loadings
(with standard errors)
(1) too much change .69 (.04)** .68 (.04)** .62 (.05)**
(2) secret felony .78 (.03)** .76 (.04)** .63 (.06)**
(3) cover wine spill .57 (.05)** .68 ( .04)** .63 (.06)**
(4) tell lies .75 (.03)** .74 (.04)** .66 (.05)**
(5) break the copier .77 (.03)** .79 (.04)** .70 (.05)**
Note. *p < .05, ** p < .001
Five-Item Guilt proneness Scale 27
Table 9
Study 3: Item Parameters from Item Response Theory (IRT) Analyses
Item Condition Item parameters
a b1 b2 b3 b4 b5 b6
(1) too much change 4 points 1.38 -2.07 -0.42 0.92
5 points 1.68 -1.66 -0.88 -0.15 0.71
7 points 1.74 -1.89 -1.39 -0.9 -0.28 0.32 0.97
(2) secret felony 4 points 1.45 -2.74 -1.54 -0.05
5 points 2.24 -2.3 -1.61 -0.85 -0.00
7 points 2.32 -2.98 -1.94 -1.51 -0.93 -0.53 0.39
(3) cover wine spill 4 points 1.48 -3.1 -1.71 0.03
5 points 1.71 -2.92 -1.73 -0.95 0.26
7 points 1.26 -3.41 -3.05 -2.48 -1.59 -0.68 0.66
(4) tell lies 4 points 1.56 -2.68 -0.9 0.95
5 points 1.94 -2.13 -1.24 -0.27 0.83
7 points 2.16 -2.33 -1.77 -1.28 -0.59 0.18 1.08
(5) break the copier 4 points 1.79 -2.69 -1.12 0.21
5 points 2.27 -2.25 -1.41 -0.66 0.48
7 points 2.39 -3.14 -2.32 -1.44 -0.9 -0.20 0.55
Five-Item Guilt proneness Scale 28
Figure 1. Study 1: Item Characteristics Curve (ICC) comparison in the three conditions for item
4: You lie to people but they never find out about it. What is the likelihood that you would feel
terrible about the lies you told?
Five-Item Guilt proneness Scale 29
Figure 2. Study 1: Total Information Curves in the three conditions.
Five-Item Guilt proneness Scale 30
Figure 3. Study 3: Item Characteristics Curve (ICC) comparison in the three conditions for all items
Five-Item Guilt proneness Scale 31
Five-Item Guilt Proneness Scale (GP-5)
Instructions: In this questionnaire, you will read about five situations that people could
encounter in day-to-day life, followed by reactions to those situations. As you read each
scenario, try to imagine yourself in that situation. Then indicate the likelihood that you
would react in the way described.
1 2 3 4 5
Extremely
Unlikely
Unlikely About
50%
Likely
Likely Extremely
Likely
(1) After realizing you have received too much change at a store, you decide to keep it
because the salesclerk doesn't notice. What is the likelihood that you would feel
uncomfortable about keeping the money?
(2) You secretly commit a felony. What is the likelihood that you would feel remorse
about breaking the law?
(3) At a coworker’s housewarming party, you spill red wine on their new cream-colored
carpet. You cover the stain with a chair so that nobody notices your mess. What is
the likelihood that you would feel that the way you acted was pathetic?
(4) You lie to people but they never find out about it. What is the likelihood that you
would feel terrible about the lies you told?
(5) Out of frustration, you break the photocopier at work. Nobody is around and you
leave without telling anyone. What is the likelihood you would feel bad about the way
you acted?
SCORING: The scale is scored by averaging the 5 items. Higher scores indicate more guilt proneness.
... We also asked participants to respond to three items about the extent to which they were influenced by self-interest: "How happy I would feel about keeping all the money", "How important it is for me to earn the largest possible bonus", and "How excited I would feel to earn $1.50" (α = .65). 5 To mask the purpose of the study, we also asked participants to imagine that they were Player 1 and asked a similar set of questions (whether their decision as Player 1 would be influenced by their sense of responsibility, anticipated guilt, warm glow, and self-interest). We had no hypotheses pertaining to these questions and did not analyze those data. ...
Article
Full-text available
Existing trust research has disproportionately focused on what makes people more or less trusting, and has largely ignored the question of what makes people more or less trustworthy. In this investigation, we deepen our understanding of trustworthiness. Across six studies using economic games that measure trustworthy behavior and survey items that measure trustworthy intentions, we explore the personality traits that predict trustworthiness. We demonstrate that guilt-proneness predicts trustworthiness better than a variety of other personality measures, and we identify sense of interpersonal responsibility as the underlying mechanism by both measuring it and manipulating it directly. People who are high in guilt-proneness are more likely to be trustworthy than are individuals who are low in guilt-proneness, but they are not universally more generous. We demonstrate that people high in guilt-proneness are more likely to behave in interpersonally sensitive ways when they are more responsible for others’ outcomes. We also explore potential interventions to increase trustworthiness. Our findings fill a significant gap in the trust literature by building a foundation for investigating trustworthiness, by identifying a trait predictor of trustworthy intentions and behavior, and by providing practical advice for deciding in whom we should place our trust.
Chapter
Signages that elicit guilt is an effective strategy in influencing segregation behaviour especially if it includes a statement of action and an appropriate picture to illustrate the consequence of bad behavior. This paper examines the direct effect of signage design factors: type of guilt statement and pictorial component to guilt. Clarity of purpose was considered as a mediating variable between the design factors and guilt, while guilt proneness was considered as moderating variable between guilt and segregation behavior. The evaluation of the model was done through a Structural Equation Model (SEM) with a data set count of 404. There was no significant effect in the use of actual picture and cartoon in promoting the clarity of purpose of a signage. The mediating effect of clarity of purpose was established between the design factors and guilt, while guilt proneness also moderated the relationship between guilt and segregation behavior.
Preprint
In this article, we discuss the role of moral character in negotiation and identify open questions and promising directions for future scholars to explore. We advance research in this area by introducing a dyadic model of moral character in negotiation, which highlights the joint influence of each party’s moral character on negotiation attitudes, motives, and behaviors. We discuss the implications of our model and conclude that personality science, and especially the study of moral character, has great potential to enhance research and practice in negotiations. Our hope is that this work will accelerate theoretical development and empirical studies that address the question of how moral character influences negotiation processes and outcomes—from pre-negotiation (e.g., planning, selecting negotiating partners) to actual bargaining (e.g., bargaining tactics, concessions) and finally, post-negotiation (e.g., deal implementation, long-term consequences, relationship building and maintenance, reputations)—and provide a springboard for future studies on this topic.
Preprint
Existing trust research has disproportionately focused on what makes people more or less trusting, and has largely ignored the question of what makes people more or less trustworthy. In this investigation, we deepen our understanding of trustworthiness. Across six studies using economic games that measure trustworthy behavior and survey items that measure trustworthy intentions, we explore the personality traits that predict trustworthiness. We demonstrate that guilt-proneness predicts trustworthiness better than a variety of other personality measures, and we identify sense of interpersonal responsibility as the underlying mechanism by both measuring it and manipulating it directly. People who are high in guilt-proneness are more likely to be trustworthy than are individuals who are low in guilt-proneness, but they are not universally more generous. We demonstrate that people high in guilt-proneness are more likely to behave in interpersonally sensitive ways when they are more responsible for others’ outcomes. We also explore potential interventions to increase trustworthiness. Our findings fill a significant gap in the trust literature by building a foundation for investigating trustworthiness, by identifying a trait predictor of trustworthy intentions and behavior, and by providing practical advice for deciding in whom we should place our trust
Preprint
Full-text available
From the time we are children, we are taught to say “thank you” and “I’m sorry.” These communications are central to many social interactions, and the failure to say them often leads to conflict in relationships. Research has documented that, alongside the impact they can have on relationships, apologies and thanks can also impact material outcomes as small as restaurant tips and as significant as settlements of medical malpractice lawsuits. But how can such “cheap talk” carry so much value? In this paper, we propose a theory that explains why these communications are not costless and draws connections between four forms of communication that have not previously been connected: thanking, apologizing, bragging, and blaming. All four of these communications relay information about credit or blame for a positive or negative outcome, and thus introduce image-based costs and benefits for both the communicator and the recipient of communication: Each of the four communications involves a tradeoff between appearing competent and appearing warm. By formalizing these social psychological insights with a cognitive approach to modeling communication, and by applying game theoretic analysis, we offer new insights and predictions about social communication. In the first experiment, we test the main assumptions of the model (i.e., that these communications produce a tradeoff between perceived competence and warmth). In a second experiment, we test several of the model’s novel predictions about strategic communication. We end by discussing extended predictions and applications of the model as examples of future directions for research on the topic.
Article
Full-text available
In this article, we discuss the role of moral character in negotiation and identify open questions and promising directions for future scholars to explore. We advance research in this area by introducing a dyadic model of moral character in negotiation that highlights the joint influence of each party’s moral character on negotiation attitudes, motives, and behaviors. We discuss the implications of our model and conclude that personality science—especially the study of moral character—has great potential to enhance research and practice in negotiations. Our hope is that this work will accelerate theoretical development and empirical studies that address the question of how moral character influences negotiation processes and outcomes—from pre-negotiation (e.g., planning, selecting negotiating partners) to actual bargaining (e.g., bargaining tactics, concessions) to post-negotiation (e.g., deal implementation, long-term consequences, relationship building and maintenance, reputations)—and provide a springboard for future studies on this topic.
Chapter
Full-text available
Growth mindset has gained popularity in the fields of psychology and education, yet there is surprisingly little research on the psychometric properties of the Growth Mindset Scale. This research presents an item response theory analysis of the Growth Mindset Scale when used among college students in the United States. Growth Mindset is the belief that success comes through hard work and effort rather than fixed intelligence. Having a growth mindset is believed to be important for academic success among historically marginalized groups; therefore it is important to know if the Growth Mindset Scale functions well among first generation college students. The sample consists of 1260 individuals who completed the Growth Mindset Scale on one of 5 surveys. The Growth Mindset Scale consists of 8 items, with responses ranging from strongly disagree (1) to strongly agree (5). IRT analysis is used to assess item fit, scale dimensionality, local dependence, and differential item functioning (DIF). Due to local dependence within the 8-item scale, the final IRT model fit 4 items to a unidimensional model. The 4-item scale did not exhibit any local dependence or DIF among known groups within the sample. The 4-item scale also had high marginal reliability (0.90) and high total information. Cronbach’s alpha for the 4-item scale was α = 0.89. Discussion of the local dependence issues within the 8-item scale is provided.
Article
Full-text available
Guilt-proneness (GP) is an individual difference characterized by a tendency to feel bad about committing transgressions. We investigated how law enforcement job candidates' guilt-proneness relates to their employment suitability, history of employment and legal problems, and counterproductive tendencies. By demonstrating relationships between GP and variables important for personnel selection and organizational functioning, this work highlights the potential utility of measuring this trait in applied settings where researchers and/or practitioners wish to gauge the integrity of respondents with a personality-based test. In light of its ability to predict employment suitability and counterproductive tendencies, the five-item guilt-proneness scale (GP-5) may prove to be a useful measure for pre-employment integrity assessment for public safety occupations, as well as other occupations where honesty and accountability are everyday concerns.
Conference Paper
Full-text available
The concept of grit is of interest in the field of education, particularly as it pertains to persistence to a 4-year college degree. This study offers an IRT analysis of the Grit Scale when used among first-generation college students (FGCSs) as well as recent first-generation college graduates and non-FGCS recent graduates. The Grit Scale was included in surveys administered as part of an array of other research projects within The Finish Line Project—a US Department of Education First in the World grant-funded project that seeks to improve FGCS access to, persistence in, and completion of postsecondary education through rigorous research into various programs and supports for FGCSs. The reliability and validity of the Grit Scale have not yet been analyzed for use with FGCS or overall with students at large, research universities. By comparing enrolled students and recent graduates, the psychometric analysis in this study offers insight into the measurement of student grit for use in program development and policy-making to improve student retention. Item response theory (IRT) analyses, analysis of differential item functioning (DIF), reliability analyses, convergent and discriminant validity analyses, and known groups validity analyses were used to examine the Grit Scale.
ResearchGate has not been able to resolve any references for this publication.