ORIGINAL RESEARCH

published: 03 December 2020

doi: 10.3389/fpsyg.2020.595651

Psychometric Evaluation of the

Czech Version of Group
Cohesiveness Scale (GCS) in a
Clinical Sample: A Two-Dimensional
Model
Adam Klocek* † , Tomáš Řiháček* † and Hynek Cígler †
Department of Psychology, Faculty of Social Studies, Brno, Czechia

Edited by:
The Group Cohesiveness Scale (GCS, 7 items) measures patient-rated group
Susana Sanduvete-Chaves, cohesiveness. The English version of the scale has demonstrated good psychometric
Seville University, Spain
properties. This study describes the validation of the Czech version of the GCS.
Reviewed by:
A total of 369 patients participated in the study. Unlike the original study, the
Sai-fu Fung,
City University of Hong Kong, ordinal confirmatory factor analysis (CFA) supported a two-dimensional solution
Hong Kong (RMSEA = 0.075; TLI = 0.986). The analysis demonstrated the existence of two
Giorgia Molinengo,
University of Turin, Italy
moderately to highly associated (r = 0.79) domains of group cohesiveness—affective
*Correspondence:
and behavioral. The two-dimensional model was invariant across genders, age,
Adam Klocek education, and time (retest after 6 weeks) up to factor means level. Internal consistency
[email protected];
reached satisfactory values for both domains (affective, ω = 0.86; behavioral, ω = 0.81).
[email protected]
Tomáš Řiháček In terms of convergent validity, only weak association was found between the GCS
[email protected] domains and the group working alliance measured by the Group Outcome Rating Scale
† ORCID:
(GSRS). This is the first revision of the factor structure of the GCS in the European
Adam Klocek
orcid.org/0000-0002-0797-4890 context. The scale showed that the Czech version of the GCS is a valid and reliable brief
Tomáš Řiháček tool for measuring both aspects of group cohesiveness.
orcid.org/0000-0001-5893-9289
Hynek Cígler Keywords: confirmatory factor analysis, group cohesion, Group Cohesiveness Scale, Czech validation study,
orcid.org/0000-0001-9959-6227 affective and behavioral group cohesion

Specialty section:
This article was submitted to INTRODUCTION
Quantitative Psychology
and Measurement, Group cohesion is one of the elemental group phenomena that allows other therapeutic processes to
a section of the journal occur within the group therapy framework. It is defined as the ability of the members of a group to
Frontiers in Psychology tolerate negative emotions and self-disclosure (Wongpakaran et al., 2013). Group cohesion partially
Received: 17 August 2020 overlaps with other group phenomena, such as the working alliance and empathy (Johnson et al.,
Accepted: 11 November 2020 2005). Group cohesion is conceptually akin to the working alliance in individual therapy. Although
Published: 03 December 2020
it is primarily based on the relationships among the group members, it can also be extended to the
Citation: relationship with the therapists (Budman et al., 1989). Group cohesion is also related to empathy
Klocek A, Řiháček T and Cígler H because a cohesive group demands that its members have an understanding of others’ feelings and
(2020) Psychometric Evaluation of the
experiences and can effectively express this understanding (Roark and Sharah, 1989).
Czech Version of Group
Cohesiveness Scale (GCS) in a
Until recently, group phenomena and processes were measured by measures such as the
Clinical Sample: A Two-Dimensional Group Climate Questionnaire (MacKenzie, 1983), the Therapeutic Factors Inventory (Lese and
Model. Front. Psychol. 11:595651. MacNair-Semands, 2000), and the Working Alliance Inventory (Horvath and Greenberg, 1989).
doi: 10.3389/fpsyg.2020.595651 However, these scales were too lengthy to be used in routine care or rapid hospital environments

Frontiers in Psychology | www.frontiersin.org 1 December 2020 | Volume 11 | Article 595651

Klocek et al. Czech Psychometric Evaluation of GCS

(compared to research) and were not directly focused on group According to Carron (1982), group cohesion is a “dynamic
cohesion. Therefore, the Group Cohesiveness Scale (GCS1 ) was process that is reflected in the tendency for a group to stick
developed (Wongpakaran et al., 2013). together [emphasis added] and remain united in the pursuit of
The GCS (Wongpakaran et al., 2013; see Table 1) was its goals and objectives. [emphasis added]” (p. 124). Similarly,
created from an original pool of 40 items and reduced to seven Mudrack (1989) divided group cohesion into attraction-to-group
items representing two domains: cohesion and engagement. (affective component) and commitment to the group task
The former domain was represented by two items from the (behavioral component).
Therapeutic Factors Inventory, while the latter was represented Originally, the GCS was standardized in the Thai language
by five items from the Group Climate Questionnaire. However, (Wongpakaran et al., 2013) in a clinical sample of 96 patients
since both domains were similar in content, Wongpakaran (56% women) with a mean age of 28.22 (SD = 6.84). Patients
et al. (2013) considered them to be representations of the were hospitalized for up to 2 weeks. A principal component
unidimensional group cohesiveness construct. analysis revealed a unidimensional factor structure (57.2% of
Alternatively, Wongpakaran et al. (2013) suggested that the explained variance). Based on a confirmatory factor analysis
GCS items can be differentiated into the affective (items 1, 2, (CFA) conducted on the same dataset, the authors claimed
and 3) and behavioral (items 4, 5, 6, and 7) components of that the unidimensional model had moderately acceptable fit
group cohesiveness. They argued that these components might be despite unsatisfactory RMSEA values (χ2 (14) = 32.29; CFI = 0.94;
related to each other in a fashion similar to the unidimensional TLI = 0.90; SRMR = 0.04; RMSEA = 0.12).
construct of depression, in which the feeling of sadness is Although Wongpakaran et al. (2013) tried to fit a two-
functionally different from a behavioral lack of interest, yet both dimensional model (i.e., cohesion and engagement), they did
components measure the same latent construct of depression not report the results, arguing that the two dimensions were
(Wongpakaran et al., 2013). too strongly correlated to be set apart (r = 0.83). Instead, they
The distinction between the affective and behavioral fine-tuned the unidimensional model based on modification
components is consistent with the theoretical literature. indices by allowing residual correlations between pairs of items
(items 1 and 2; items 2 and 3), reaching an excellent fit
1
The GCS used in this study is unrelated to the Harvard Community Health [χ2 (12) = 12.41; CFI = 0.99; TLI = 0.99; SRMR = 0.04;
Plan Group Cohesiveness Scale (which is also referred to as GCS in the literature;
Budman et al., 1993).
RMSEA = 0.02]. Arguably, by allowing the residual correlations,
the authors developed a model that was very similar to (but
less parsimonious than) the suggested two-factor model with the
TABLE 1 | Group Cohesiveness Scale (Wongpakaran et al., 2013). affective and behavioral factors. Therefore, we found it desirable
to formally test this alternative two-factor model as well. In terms
Item Item wording (Czech in italics) Cohesiveness (C) Affective (A) or
no. or engagement behavioral (B)
of convergent validity, the GCS was correlated to the Group
(E) domain domain Benefit Questionnaire (r = 0.71, p < 0.001) and to the Cohesion
to Therapist Scale (r = 0.77, p < 0.001) in the original study.
1 I feel accepted by the group. C A The GCS is a relatively new measure that has been employed
(Cítím se být skupinou přijímaný/á.)
in a limited number of studies thus far. Psychometric information
2 In my group, we trust each other. C A
(Ve skupině si vzájemně důvěřujeme.)
about the GCS is rather scarce and often unsatisfactory given
3 The members like and care about E A
small sample sizes. Poyner-Del Vento et al. (2018) used the GCS
each other. as a measure of group cohesion in a pilot study in a sample of
(Členové skupiny se mají rádi a seven female military veterans. They found that removing item
vzájemně jim na sobě záleží.) 6 (“The members appear to do things the way they think will be
4 The members try to understand why E B acceptable to the group”) increased the internal consistency of the
they do the things they do; they try to
scale from α = 0.72–0.90. Tulin et al. (2018) used the GCS to
reason it out.
(Členové se snaží porozumět tomu, measure group cohesion in a sample of 109 students with internal
proč dělají věci, které dělají; snaží se consistency of α = 0.90. In another sample of 22 students, Ashby
na to přijít.) et al. (2018) found a mean interitem correlation of r = 0.43. This
5 The members feel a sense of E B limited evidence does not allow us to thoroughly evaluate the
participation. GCS, and the applicability of the measure in Western culture
(Členové skupiny cítí, že se podílejí na
is still missing.
chodu skupiny.)
6 The members appear to do things E B
This study aimed to validate the Czech version of the
the way they think will be acceptable GCS using the ordinal CFA paradigm. Four models were
to the group. tested, including the unidimensional model (model 1), the
(Vypadá to, že členové dělají věci unidimensional model with residual covariances between items
způsobem, o němž si myslí, že bude 1 and 2 and items 2 and 3 allowed (model 2), a two-factor model
pro skupinu přijatelný.)
with the factors of cohesion (items 1 and 2 originally extracted
7 The members reveal sensitive E B
personal information or feelings.
from the Therapeutic Factors Inventory) and engagement (items
(Členové si sdělují citlivé osobní 3–7 originally extracted from the Group Climate Questionnaire)
informace a pocity.) (model 3), and a two-factor model with affective (items 1–3)

Frontiers in Psychology | www.frontiersin.org 2 December 2020 | Volume 11 | Article 595651

Klocek et al. Czech Psychometric Evaluation of GCS

and behavioral factors (items 4–7) (model 4). Furthermore, Instruments

to assess the convergent validity, we used the Group Session Group Cohesiveness Scale (GCS)
Rating Scale (GSRS, Quirk et al., 2013), a measure of the group The seven items of the GCS are scored on a Likert scale from
working alliance, as a comparison. Although group cohesion 1 (strongly disagree) to 5 (strongly agree). None of the items
and group working alliance are distinct constructs, we expected is negatively worded. A higher score indicates higher perceived
the GCS scores to be related to the GSRS scores because both group cohesion. In the original study, the GCS yielded an average
instruments measure non-specific group-based relational factors score of 4.73 out of 5 (SD = 0.62), the internal consistency of the
of the therapeutic process. whole scale was α = 0.87, and the item-total correlations ranged
from 0.497 to 0.752.
The scale was translated into Czech from the English
MATERIALS AND METHODS version. Five native Czech speakers (a psychology student, two
psychologists, and two laypeople) created five independent Czech
Sample and Procedure translations. A group of three people (the two psychologists and
The sample included patients from seven clinical sites in the psychology student) then discussed all the translations and
the Czech Republic who provided informed consent to consolidated them into a single version. Third, this version was
participate in research tracking the mechanisms of change during back-translated into English by a bilingual, native English speaker
psychotherapy from January 2018 to December 2019. All patients and compared to the original English version. Fourth, the final
underwent group therapy lasting from 4 to 12 weeks (depending Czech version was field-tested with five respondents to check the
on the site, median of 6 weeks). Data were collected on a comprehensibility of the items.
paper-and-pencil form on a weekly basis during the whole
treatment. Participants completed a battery of questionnaires Group Session Rating Scale (GSRS)
regarding demographic variables, several outcome variables and The GSRS (Quirk et al., 2013) is a measure of the working
several mechanisms of change, including group cohesion and alliance in group psychotherapy. It includes four 10-cm-long
working alliance. The study was approved by the Research Ethics visual analog scales, each framed by a verbal anchor on both
Committee of Masaryk University (Ref. No. EKV-2017-029-R1). ends. The continuous dimension of each item is framed by
In this study, the dataset used to validate the GCS included bipolar points, and participants rate the group working alliance
data from the second week of therapy (i.e., the first measurement by making a mark on each scale. The response is measured as
of the group cohesion). Out of 448 patients who provided their the length of the line from the left-hand side to the mark in
baseline data, 380 patients (85%) participated in the second week millimeters. The range of the total score, computed as the sum
of treatment. Out of 380 participants, 11 were characterized of all items, can thus reach values between 0 and 400. A higher
by missing data regarding the GCS, resulting in a total sample score indicates a better perceived working alliance. The scale
size of N = 369 patients. Differences between participants with was reported to be unidimensional, and the internal consistency
missing data (n = 80) and the final sample (n = 369) in the ranged from α = 0.86 to 0.90 in the original study.
demographics and clinical diagnosis data were investigated using
t-tests and χ2 -tests. Data Analysis
Software and General Settings
Group Therapy The statistical procedures were performed using statistical
The treatment was integrative with major psychodynamic and software R, version 4.0.2 (R Core Team, 2020). The significance
minor humanistic and experiential aspects, supplemented with level was set at p < 0.05.
art, physical activity, music, ergo-, drama-, physio-, and biblio-
therapy, relaxation and cognitive training, and community Factorial Validity
meetings2 . Five sites were characterized by a frequency of five The factor structure was estimated through ordinal CFA using
sessions of psychotherapy per week. The remaining two sites had the lavaan package (Rosseel, 2012). The ordinal factor analysis is
three and four sessions per week, respectively. A session of group equivalent to the two-parameter logistic graded response model
therapy lasted 90 min3 . in item response theory. Hence, this approach is not as vulnerable
The sample comprised small closed groups of inpatients to the violation of assumptions as the standard factor analysis
within four clinical sites and small open groups of outpatients (Raykov and Marcoulides, 2011). Each item has five parameters
in a program with a daycare basis within three clinical sites. (one slope and four thresholds between all neighboring response
Twenty-five (16 female) therapists participated in this research options). All five models were estimated using the stochastic
(M age = 44.13 years, SDage = 10.29). They were trained in the weighted least squares means and variance adjusted estimator
psychodynamic or psychoanalytic approach (n = 15), gestalt method (WLSMV), which seems to perform well with ordered
(n = 4), person-centered approach (n = 3), integrative approach categorical data (Raykov and Marcoulides, 2011). The fit indices
(n = 2) or Daseinanalysis (n = 1). Their experience fluctuated employed in this study included χ2 , χ2 /df, root mean square
between 1 and 25 years (M = 12.21, SD = 7.30). error of approximation (RMSEA), Tucker-Lewis index (TLI),
comparative fit index (CFI), and standardized root mean residual
2
The supplemental therapeutic techniques and session differed by site. (SRMR). According to Hu and Bentler’s (1999) and Hooper et al.
3
One site was characterized by the session length of 75 min. (2008) evaluation criteria, the χ2 /df should not exceed 3, the

Frontiers in Psychology | www.frontiersin.org 3 December 2020 | Volume 11 | Article 595651

Klocek et al. Czech Psychometric Evaluation of GCS

RMSEA should optimally be below 0.05, but values up to 0.10 are Descriptive Characteristics
still considered to indicate a satisfactory fit. The SRMR should not The total sample included 369 patients (73.7% females). Their
exceed 0.08. Optimally, the TLI and CFI should be above 0.95; nationality included Czech (95%), Slovak (2%), and others (3%).
nevertheless, values above 0.90 are still considered to indicate a The patients’ ages ranged from 18 to 71 years (M age = 39.6,
satisfactory fit. SD = 11.1). Psychiatric diagnoses were represented as follows:
Within models with more than one dimension (models 3 and F4x (n = 261), F3x (n = 69), F6x (n = 53), F5x (n = 8), and
4), factors were allowed to be correlated. Since model 3 contained F1x (n = 7). Several participants possessed multiple diagnoses
a factor represented only by two items, these items were (n = 33), mainly a combination of F4x and F6x (n = 13), F3x and
constrained to load equally on their factor. Otherwise, models F4x (n = 9), and F3x and F6x (n = 7). The remaining demographic
were identified by standardizing the latent variable. The internal variables are reported in Table 2.
consistency was estimated using bootstrapped Cronbach’s alpha The mean scores for each GCS item, the GCS total score,
and McDonald’s omega coefficients (McDonald, 1999). In terms and the GSRS total score, as well as the internal consistency of
of convergent validity, the association between the GCS and the the unidimensional model, are reported in Table 3. The average
GSRS was tested on the level of latent scores. total score was 3.7 (SD = 0.69). Corrected item-total correlations
ranged from 0.49 to 0.75.
Measurement Invariance
The invariance was tested with regard to age, gender, Factor Structure
education, and time. Measurement invariance was assessed First, the assumptions of factor analysis were tested. The data did
by testing differences between nested models with continually not show multivariate normality, and the standardized residuals
increasing constraints: configural, metric (factor loadings), scalar were positively skewed. Homoscedasticity was not observed.
(intercepts), strict (residuals), and factor means. Age groups After the preliminary data analyses, an ordinal factor analysis
were created by dividing the sample according to a median was employed to estimate the fit of the factor models using
split. Gender invariance was assessed between male and female these skewed non-linear data. The RMSEA of the null model was
participants. Education invariance was assessed between higher 0.398. This value is above 0.148; thus, the TLI fit index could be
(university, high technical school) and lower education (primary interpreted (Kenny et al., 2015).
and secondary school with or without graduation) levels. Time
invariance was assessed between the second and sixth weeks of
group therapy (the sixth week was chosen pragmatically because TABLE 2 | Descriptive characteristics of the sample (N = 369).
in most sites, the therapy lasted only 6 weeks). We used four
different fit indices to test the invariance, namely, 1χ2 , 1CFI, Variable Categories n Percent
1SRMR, and 1RMSEA. We employed “theta” parametrization Gender Female 272 74%
and invariance guidelines with regard to ordinal data according Male 90 24%
to Wu and Estabrook (2016). Two groups are considered to be Missing 7 2%
invariant if the item parameters (i.e., factor loadings, thresholds, Household In partnership 189 51%
intercepts, residuals, and factor means) are similar across groups. Single 71 19%
Items 3 and 7 demonstrated missing response frequency at With parents 39 11%
response option 1 (i.e., 1 or “strongly disagree”). The remaining Other 62 17%
items demonstrated near-to-missing response frequency (0.01) Missing 8 2%
at response option 1. Response option 2 (unnamed) was also Marital status Single 178 48%
very seldom selected by the participants in all items. Therefore, Married 114 31%
all items were recoded into three categories (i.e., responses Divorced 67 18%
from 1 to 3 were recoded as a single category, representing Widowed 2 1%
a low level of group cohesion) for the purpose of testing the Missing 8 2%
measurement invariance. Education Primary school 17 5%
Secondary school 180 49%
High technical school 22 6%
RESULTS University 141 38%
Missing 9 2%
Missing Data Occupation Employee 163 44%
No significant differences between the final sample (N = 369) Unemployed 53 14%
and the respondents with missing responses or respondents Invalidity pension 35 10%
not participating in the study at the second week (n = 80, Entrepreneur 23 6%
who were the remaining part of the initial sample of 449 Student 20 6%
participants) were found for the mean age, gender, education, Maternity leave 7 2%
and psychiatric diagnosis. The pattern of missingness could be Retirement 4 1%
considered missing at random. Therefore, only complete cases Other 15 4%
were included in the analyses. Missing 49 13%

Frontiers in Psychology | www.frontiersin.org 4 December 2020 | Volume 11 | Article 595651

Klocek et al. Czech Psychometric Evaluation of GCS

TABLE 3 | Descriptive characteristics of scales (N = 369).

Item M SD Range (min-max) Skewness Kurtosis Corrected item-total correlation Cronbach’s alpha if item deleted

GCS 1 3.70 0.91 4 (1 − 5) −0.01 −0.64 0.63 0.85

GCS 2 3.72 0.94 4 (1 − 5) −0.04 −0.87 0.75 0.84
GCS 3 3.51 0.92 3 (2 − 5) 0.21 −0.85 0.69 0.85
GCS 4 3.81 0.94 4 (1 − 5) −0.27 −0.64 0.69 0.85
GCS 5 3.68 0.86 4 (1 − 5) −0.01 −0.53 0.69 0.85
GCS 6 3.57 0.93 4 (1 − 5) 0.00 −0.44 0.59 0.86
GCS 7 4.00 0.93 3 (2 − 5) −0.36 −0.97 0.49 0.87

M SD Range Skewness Kurtosis McDonald’s omega Cronbach’s alpha

GCS total 25.99 4.82 22 (13 − 35) 0.20 −0.83 0.91 0.87 [0.85–0.89]
GSRS 290.82 75.20 368 (32 − 400) −0.66 0.09 0.83 0.82 [0.79–0.85]

Second, four different factor solutions were tested for fit and values, and the fit even increased with more restricted models.
compared (see Table 4). We concluded that the best fit was We could, therefore, consider the model invariant in time on the
obtained by model 4, a two-factor solution with the affective configural, metric, scalar, and factor mean levels. The final model
and behavioral factors (see Table 5 and Figure 1). Model 4 fit was non-invariant only on the level of residual variances between
the data significantly better than did model 1 [unidimensional; the second and sixth weeks of measurement.
1χ2 (1) = 87.66, p < 0.0001] and model 3 [two-factor with the
cohesion and engagement factors; 1χ2 (2) = 104.31, p < 0.0001]. Reliability and Convergent Validity
Furthermore, the fit of model 4 did not significantly differ The internal consistency of the final model was ω = 0.86 for the
from that of model 2 [unidimensional with residual correlations; affective and ω = 0.81 for the behavioral domains (see Table 5).
1χ2 (1) = 2.35, p > 0.10]. However, model 4 can be considered Additionally, the internal consistency of the general factor in
superior in terms of parsimony as well as theoretical justification. model 1 was ω = 0.91. None of the GCS items would increase
While the affective factor represents the same underlying the internal consistency when dropped.
structure as the empirically derived residual correlations in Thirteen participants had missing data on the GSRS scale,
model 2, it explains the item interrelationships more efficiently resulting in 367 patients. With respect to the final two-factor
and is consistent with theoretical expectations (Carron, 1982; model with affective and behavioral dimensions (model 4),
Mudrack, 1989). the affective domain was correlated more strongly with the
GSRS (r = 0.449, p < 0.05) than the behavioral domain was
Measurement Invariance (r = 0.290, p < 0.05). Additionally, a small to moderate positive
Measurement invariance was assessed for model 4 with respect correlation between the latent constructs of the unidimensional
to age, gender, and education (see Table 6). Several patients GCS (Model 1) and GSRS scales was found (r = 0.394, p < 0.05).
were lost due to missing responses on the demographic variables,
namely, age (n = 9), gender (n = 7), and education (n = 9).
Measurement invariance between the younger (n = 185) and DISCUSSION
older (n = 175) cohorts was reached on the configural, metric,
scalar, factor mean, and residual levels. Even though the χ2 -test The present study described the validation of the Czech
was significant on the scalar and residual invariance level, other version of the Group Cohesiveness Scale (GCS). The average
1fit indices showed desirable values. Measurement invariance item scores and reliability were compatible with those of the
between women (n = 90) and men (n = 272) was reached on original Thai version (Wongpakaran et al., 2013). However,
the configural, metric, scalar, and factor mean levels. Genders we concluded that, based on a CFA, the most preferable
were not invariant only on the level of residual variances. model was a two-factor solution with the correlated affective
Measurement invariance between lower (n = 197) and higher and behavioral domains (model 4). This solution is more
education levels (n = 163) was reached on the configural level. parsimonious than the fine-tuned unidimensional solution
Even though the χ2 -test was significant on both metric and scalar (model 2) suggested by Wongpakaran et al. (2013).
invariance levels, other 1fit indices showed desirable values, The final model demonstrated excellent fit and was
and the fit even increased with more restricted models. We invariant across age groups, genders, education levels, and
could, therefore, consider the model invariant between education time. The Czech version did not even show any problematic
levels on the configural, metric, scalar, factor mean, and residual functioning of item 6 as presented in the English translation by
variance levels. Measurement invariance between the second Poyner-Del Vento et al. (2018). Theoretically, group cohesion is
(n = 369) and sixth weeks (n = 273) was reached on the configural related to the working alliance (Johnson et al., 2005). However, in
level. Even though the χ2 -test was significant on both the metric our study, we found only small to medium correlations between
and scalar invariance levels, other 1fit indices showed desirable the GCS subscales and the GSRS. This finding was unexpected,

Frontiers in Psychology | www.frontiersin.org 5 December 2020 | Volume 11 | Article 595651

Klocek et al. Czech Psychometric Evaluation of GCS

TABLE 4 | Fit indices of the tested models (N = 369).

Model χ2 df χ2 /df TLI CFI SRMR RMSEA

Model 1 249.67*** 14 17.8 0.975 0.983 0.072 0.156[0.133;0.180]

(unidimensional)
Model 2 78.97*** 12 6.6 0.994 0.997 0.041 0.076[0.050;0.105]
(unidimensional, Item 1 ∼∼ Item 2, Item 2 ∼∼ Item 3)
Model 3} 238.95*** 14 17.1 0.975 0.983 0.071 0.158[0.135;0.182]
(two-factor: cohesion and engagement)
Model 4}} 79.71*** 13 6.1 0.994 0.986 0.040 0.075[0.049;0.102]
(two-factor: affective and behavioral)

***p < 0.001.

}Correlation between the cohesion and engagement latent factors was r = 0.88.
}}Correlation between the affective and behavioral latent factors was r = 0.79.

since the GSRS measures patients’ relationships not only with (2013), even though these authors did not report fit indices
the therapists/group leaders but also with other members of for this model. Group cohesiveness has been recognized as a
the group; therefore, there is an apparent overlap in what the multidimensional construct several times in the past (Hogg,
instrument is expected to measure. Although the affective 1993). Mudrack’s (1989) definition of group cohesion as a
domain was more promising than the behavioral domain in combination of attraction-to-group and commitment to the
terms of convergent validity, overall, the convergent validity of group task provides a solid rationale for the differentiation of
the GCS was not particularly supported in this study. group cohesion into the affective and behavioral domain. The
former is associated with the attraction to the group or its
members and by collectively sharing positive, as well as negative,
Theoretical Support for Two-Dimensional
emotional experiences (Barsade and Knight, 2015). The latter,
Group Cohesion on the other hand, is associated with a commitment to the
The GCS was conceived as a unidimensional construct by group (Mudrack, 1989) that may be manifested, for instance, by
Wongpakaran et al. (2013). However, the unidimensional model following group rules or giving gifts to other members (Lawler
(model 1) demonstrated an acceptable fit neither in their et al., 2000). Another literature supporting the two-dimensional
study nor in ours. Although the large correlation between the model was Carron et al. (1985) who defined the individual
affective and behavioral factors may be interpreted in favor group factor (commitment to other members of group) and task-
of the unidimensionality of the scale, the two dimensions social factor (interest in the goals of the group). Cota et al.
are still independent to some degree and represent different (1995) in their review of group cohesion structure discussed both
phenomena conceptually. Theoretical support for the two-factor unidimensionality and multidimensionality resulting in favoring
model with the affective and behavioral domains can already the multidimensional perspective (normative and behavioral
be found in the standardization study by Wongpakaran et al. components are divided and considered primary components
of group cohesion). Kipnes et al. (2002) tested group cohesion
dimensionality using two different instruments and claimed that
TABLE 5 | Standardized regression weights (factor loadings) and errors (N = 369). cohesion is a multidimensional construct and offer a hierarchical
structure [first order factors will be (1) bond to individual
Model 1 Model 4
members and (2) level of trust and encouragement of the
λ h2 λF1 λF2 h2
group as a whole].
In summary, given the high internal consistency of the
Item 1 0.75 0.56 0.78 – 0.61 unidimensional solution and the large correlation between the
Item 2 0.89 0.80 0.94 – 0.87 affective and behavioral dimensions, the GCS may be used as
Item 3 0.82 0.67 0.85 – 0.73 an essentially unidimensional measure of group cohesiveness.
Item 4 0.79 0.63 – 0.84 0.71 However, it should be done with caution and with the awareness
Item 5 0.83 0.69 – 0.87 0.75 of the fact that group cohesiveness may be, in fact, composed of
Item 6 0.69 0.47 – 0.72 0.52 different and partially independent phenomena.
Item 7 0.56 0.31 – 0.58 0.34
McDonald’s omega 0.910 0.860 0.811
Raykov’s omega 0.879 0.850 0.797
Similarities and Differences Between the
Cronbach’s alpha 0.896 0.884 0.828 Thai and Czech Versions
R2 61.9% 32.7% 35.6% The Czech version of the GCS demonstrated some features
similar to those of the Thai version. Both versions were
R2 , explained variance of the model;λ, factor loadings; h2 , communality;
characterized by similar values of item-total correlations
F1, affective domain; F2, behavioral domain.
Correlation between F1 and F2 in model 4: r = 0.785. McDonald’s omega total for and internal consistency. Item loadings in terms of the
model 4 = 0.894. unidimensional model were very similar for both versions as

Frontiers in Psychology | www.frontiersin.org 6 December 2020 | Volume 11 | Article 595651

Klocek et al. Czech Psychometric Evaluation of GCS

FIGURE 1 | Factor structure of Group Cohesiveness Scale.

well. The GCS scores were relatively skewed in both studies. Second, 67 patients dropped out of the study by the second week
Patients tended to perceive their groups as rather cohesive in (i.e., the time when the first measurement of group cohesion took
both cultures. Based on these similarities, we can argue that both place). Although there were no significant differences between
versions are comparable. those who dropped out and those who continued with the
However, certain differences between the Czech and Thai treatment, this number of participants could have changed some
versions can be found. The two-dimensional solution as the subtle structures within the data. Third, two models yielded a
best fitting solution is different from the original unidimensional satisfactory fit. The selection of the final model, even though
solution. This may be attributed to cultural differences. theoretically anchored, is always relatively arbitrary in such cases.
Furthermore, the mean total score of the unidimensional model Moreover, none of the models fulfilled the criteria for a good
was higher in the Thai version (4.7) than in the Czech version fit regarding the χ2 /df fit index. However, the chi-square test
(3.7). Therefore, Thai participants might perceive therapeutic of model fit (and its derivatives) are sample size sensitive and
groups as generally more cohesive than Czech participants do or could lead to the rejection of factor model even when residual
might be less willing to report a lack of cohesion. variances are negligible. Fourth, the final two-factor model was
invariant across age cohorts, genders, education levels, and time.
Limitations Nevertheless, response options 1, 2, and 3 were clustered into a
First, the sample was relatively heterogeneous and did not single response option because of missing response patterns in
represent both genders equally (70% were female). Although the data. This reduction of thresholds might have distorted our
this corresponds to the fact that most psychotherapy clients are conclusions about the invariance. This response pattern might be
women, future studies may investigate male groups to explore explained by the tendency of group members to perceive their
possible differences in the factor structure of group cohesion. group likewise; hence, their responses to the measurement tool

Frontiers in Psychology | www.frontiersin.org 7 December 2020 | Volume 11 | Article 595651

Klocek et al. Czech Psychometric Evaluation of GCS

TABLE 6 | Measurement invariance for Model 4 across age, gender, and education level.

χ2 (df) TLI RMSEA SRMR 1χ2 1df 1TLI 1RMSEA 1SRMR

Age
Configural 58.11 (26) 0.988 0.083 0.042 – – – – –
Metric (loadings, free var, free means) 65.18 (31) 0.989 0.078 0.042 5.76 5 0.001 −0.005 0.00
Scalar (loadings, intercepts, free var, 79.00 (36) 0.988 0.082 0.043 11.92* 5 −0.001 0.003 0.00
free means)
Factor means 61.05 (38) 0.994 0.058 0.043 0.48 2 0.006 −0.023 0.00
Residuals 101.3 (43) 0.987 0.087 0.052 18.97** 7 −0.002 0.005 0.01
Gender
Configural 52.60 (26) 0.991 0.075 0.041 – – – – –
Metric (loadings, free var, free means) 58.77 (31) 0.992 0.071 0.041 5.65 5 0.001 −0.005 0.000
Scalar (loadings, intercepts, free var, 65.51 (36) 0.992 0.067 0.041 6.58 5 0.001 −0.003 0.000
free means)
Factor means 59.98 (38) 0.995 0.057 0.041 2.11 2 0.002 −0.011 0.000
Residuals 90.90 (43) 0.990 0.079 0.055 16.74* 7 −0.003 0.011 0.014
Education
Configural 64.19 (26) 0.985 0.091 0.046 – – – – –
Metric (loadings, free var, free means) 76.73 (31) 0.985 0.091 0.047 11.54* 5 0.000 0.000 0.001
Scalar (loadings, intercepts, free var, 88.38 (36) 0.985 0.090 0.047 10.52* 5 0.000 −0.001 0.001
free means)
Factor means 76.33 (38) 0.990 0.075 0.047 2.35 2 0.005 −0.015 0.000
Residuals 96.26 (43) 0.987 0.083 0.056 11.71 7 0.002 −0.007 0.009
Time (comparing week 2 and week 6)
Configural 66.49 (26) 0.994 0.070 0.029 – – – – –
Metric (loadings, free var, free means) 71.35 (31) 0.995 0.064 0.029 3.18 5 0.001 −0.006 0.000
Scalar (loadings, intercepts, free var, 85.44 (36) 0.995 0.066 0.029 11.84* 5 0.000 0.002 0.000
free means)
Factor means 71.66 (38) 0.997 0.053 0.029 2.88 2 0.002 −0.013 0.000
Residuals 120.7 (43) 0.993 0.075 0.037 25.92*** 7 −0.002 0.010 0.007

*p < 0.05, **p < 0.01, ***p < 0.001.

Values in bold are the best fitting values when all invariance models within the same groups are compared. The factor mean level and residual level were both compared
with the scalar level.

or to particular items could be limited to a very homogenous to AK, [email protected]; https://www.researchgate.

response style (Evans and Jarvis, 1980). net/publication/344571321_GCS_full_dataset.

CONCLUSION ETHICS STATEMENT

The Czech version of the GCS is a reliable and psychometrically The studies involving human participants were reviewed
valid tool for the measurement of the affective and behavioral and approved by the Research Ethics Committee of
domains of group cohesiveness. Thanks to its brevity, the scale Masaryk University (Ref. No. EKV-2017-029-R1). The
is useful in the rapid hospital or therapeutic environment. As patients/participants provided their written informed consent to
far as we know, this is the first psychometric validation of participate in this study.
the GCS in Western culture and the Caucasian population. In
this study, we revised the originally proposed unidimensional AUTHOR CONTRIBUTIONS
factor structure (Wongpakaran et al., 2013) and found support
for the existence of the affective and behavioral domain AK: conceptualization, theoretical literature search, analysis,
of group cohesion. writing, and reviewing. TŘ: conceptualization and reviewing. HC:
analysis and reviewing. All authors contributed to the article and
approved the submitted version.
DATA AVAILABILITY STATEMENT
The data analyzed in this study is subject to the following FUNDING
licenses/restrictions: The datasets analyzed during the current
study are available from the corresponding author on reasonable This research was funded by the Czech Ministry of Education,
request. Requests to access these datasets should be directed Youth, and Sports, internal grant MUNI/A/0993/2019.

Frontiers in Psychology | www.frontiersin.org 8 December 2020 | Volume 11 | Article 595651

Klocek et al. Czech Psychometric Evaluation of GCS

REFERENCES Lawler, E. J., Thye, S. R., and Yoon, J. (2000). Emotion and group cohesion in
productive exchange. Am. J. Sociol. 106, 616–657. doi: 10.1086/318965
Ashby, K. M., Collins, D. L., Helms, J. E., and Manlove, J. (2018). Let’s talk about Lese, K. P., and MacNair-Semands, R. R. (2000). The therapeutic factors inventory:
race: evaluating a college interracial discussion group on race. J. Multicult. development of a scale. Group 24, 303–317. doi: 10.1023/A:1026616626780
Counsel. Dev. 46, 97–114. doi: 10.1002/jmcd.12095 MacKenzie, K. R. (1983). “The clinical application of group measure,” in Advances
Barsade, S. G., and Knight, A. P. (2015). Group affect. Annu. Rev. Organ. in Group Psychotherapy: Integrating Research and Practice, eds R. R. Dies, and
Psychol. Organ. Behav. 2, 21–46. doi: 10.1146/annurev-orgpsych-032414-11 K. R. MacKenzie (New York: International Universities Press), 159–170.
1316 Mudrack, P. E. (1989). Defining group cohesiveness: a legacy of confusion? Small
Budman, S. H., Soldz, S., Demby, A., Davis, M., and Merry, J. (1993). What is Group Behav. 20, 37–49. doi: 10.1177/104649648902000103
cohesiveness? An empirical examination. Small Group Res. 24, 199–216. doi: Poyner-Del Vento, P., Goy, E., Baddeley, J., and Libet, J. (2018). The caregivers’
10.1177/1046496493242003 attachment and relationship education class: a new and promising group
Budman, S. H., Soldz, S., Demby, A., Feldstein, M., Springer, T., and Davis, M. S. therapy for caregivers of individuals with Parkinson’s disease. J. Couple
(1989). Cohesion, alliance and outcome in group psychotherapy. Psychiatry 52, Relationsh. Ther. 17, 97–113. doi: 10.1080/15332691.2017.1341356
339–350. doi: 10.1080/00332747.1989.11024456 Quirk, K., Miller, S., Duncan, B., and Owen, J. (2013). Group session rating scale:
Carron, A. V. (1982). Cohesiveness in sport groups: interpretations and preliminary psychometrics in substance abuse group interventions. Counsel.
considerations. J. Sport Psychol. 4, 123–138. Psychother. Res. 13, 194–200. doi: 10.1080/14733145.2012.744425
Carron, A. V., Widmeyer, W. N., and Brawley, L. R. (1985). The development R Core Team (2020). R: A Language and Environment for Statistical Computing.
of an instrument to assess cohesion in sport teams: the group environment Vienna: R Foundation for Statistical Computing.
questionnaire. J. Sport Exerc. Psychol. 7, 244–266. doi: 10.1123/jsp.7.3.244 Raykov, T., and Marcoulides, G. A. (2011). Introduction to Psychometric Theory.
Cota, A. A., Evans, C. R., Dion, K. L., Kilik, L., and Longman, R. S. (1995). The London: Routledge.
structure of group cohesion. Pers. Soc. Psychol. Bull. 21, 572–580. doi: 10.1177/ Roark, A. E., and Sharah, H. S. (1989). Factors related to group cohesiveness. Small
0146167295216003 Group Behav. 20, 62–69. doi: 10.1177/104649648902000105
Evans, N. J., and Jarvis, P. A. (1980). Group cohesion: a review and reevaluation. Rosseel, Y. (2012). Lavaan: an R package for structural equation modeling
Small Group Behav. 11, 359–370. doi: 10.1177/104649648001100401 and more. Version 0.5–12 (BETA). J. Stat. Softw. 48, 1–36. doi: 10.1002/
Hogg, M. A. (1993). Group cohesiveness: a critical review and some new directions. 9781119579038.ch1
Eur. Rev. Soc. Psychol. 4, 85–111. doi: 10.1080/14792779343000031 Tulin, M., Pollet, T. V., and Lehmann-Willenbrock, N. (2018). Perceived group
Hooper, D., Coughlan, J., and Mullen, M. R. (2008). Structural equation modelling: cohesion versus actual social structure: a study using social network analysis
guidelines for determining model fit. Electron. J. Bus. Res. Methods 6, 53–60. of egocentric Facebook networks. Soc. Sci. Res. 74, 161–175. doi: 10.1016/j.
Horvath, A. O., and Greenberg, L. S. (1989). Development and validation of the ssresearch.2018.04.004
working alliance inventory. J. Counsel. Psychol. 36:223. doi: 10.1037/0022-0167. Wongpakaran, T., Wongpakaran, N., Intachote-Sakamoto, R., and Boripuntakul,
36.2.223 T. (2013). The Group Cohesiveness Scale (GCS) for psychiatric inpatients.
Hu, L. T., and Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance Perspect. Psychiatr. Care 49, 58–64. doi: 10.1111/j.1744-6163.2012.00342.x
structure analysis: conventional criteria versus new alternatives. Struct. Equ. Wu, H., and Estabrook, R. (2016). Identification of confirmatory factor analysis
Model. Multidiscip. J. 6, 1–55. doi: 10.1080/10705519909540118 models of different levels of invariance for ordered categorical outcomes.
Johnson, J. E., Burlingame, G. M., Olsen, J. A., Davies, D. R., and Gleave, Psychometrika 81, 1014–1045. doi: 10.1007/s11336-016-9506-0
R. L. (2005). Group climate, cohesion, alliance, and empathy in group
psychotherapy: multilevel structural equation models. J. Couns. Psychol. 52, Conflict of Interest: The authors declare that the research was conducted in the
310–321. doi: 10.1037/0022-0167.52.3.310 absence of any commercial or financial relationships that could be construed as a
McDonald, R. P. (1999). Test Theory: A Unified Treatment. Lawrence Erlbaum potential conflict of interest.
Associate.
Kenny, D. A., Kaniskan, B., and McCoach, D. B. (2015). The performance of Copyright © 2020 Klocek, Řiháček and Cígler. This is an open-access article
RMSEA in models with small degrees of freedom. Sociol. Methods Res. 44, distributed under the terms of the Creative Commons Attribution License (CC BY).
486–507. doi: 10.1177/0049124114543236 The use, distribution or reproduction in other forums is permitted, provided the
Kipnes, D. R., Piper, W. E., and Joyce, A. S. (2002). Cohesion and outcome in short- original author(s) and the copyright owner(s) are credited and that the original
term psychodynamic groups for complicated grief. Int. J. Group Psychother. 52, publication in this journal is cited, in accordance with accepted academic practice. No
483–509. doi: 10.1521/ijgp.52.4.483.45525 use, distribution or reproduction is permitted which does not comply with these terms.

Frontiers in Psychology | www.frontiersin.org 9 December 2020 | Volume 11 | Article 595651