Collaboration is defined as a “coordinated, synchronous activity that is the result of a continued attempt to construct and maintain a shared conception of a problem” (Roschelle and Teasley, 1995, p. 70). CPS is often dichotomized into the social domain of collaboration skills and the cognitive domain of problem-solving skills. The problem-solving process involves cognitive abilities including task regulation, perspective-taking, and reflecting. The collaboration process is defined as the activity of working together as a team instead of individuals to achieve a common goal (Care et al., 2015). The process of collaboration can be further divided into three main aspects: participation, perspective-taking, and social regulation. Specifically, participation is defined as an individual’s willingness to offer and share information and thoughts, and the extent of their involvement as part of a team. Perspective-taking skills, as another dimension of exchange information, refers to an individual’s ability to internalize others’ opinions and make adaptations or adjustments for their purposes (Dehler et al., 2011).

In the Framework for Teachable Collaborative Problem Solving Skills, one of the key factors that ensures a successful collaboration is shared mental representation, that is, the participant’s mental representation should be aligned to other group members’ mental spaces within the same group (Hesse et al., 2015). Many studies have shown that participants usually demonstrate a higher performance of collaboration or collaborative problem solving when higher levels of shared mental representations are achieved (Klimoski and Mohammed, 1994). This type of alignment could take effect in each step of collaboration including goal setting, plan execution, communication, and progress evaluation. In this sequence of joint actions, participant’s awareness, or willingness to collaborate, plays a crucial role in positively fostering the successful fulfillment of the CPS task.

Given the importance of endorsement with collaboration to CPS skills, many theoretical, as well as empirical studies, were dedicated to exploring factors that influence individual’s attitudes towards collaboration in different domains (e.g., Bollen et al., 2019; Cradock-Henry et al., 2017; Mattessich and Monsey, 1992). Among many causes, it is found that culture plays an important role in shaping individuals’ awareness of collaborations and cognitive development through social interactions (Patel et al., 2012). Culture can be referred to as national or regional culture, which serves as a vital factor that shapes the collective programming of human minds and behavior patterns, and often is represented by a set of values or beliefs about the willingness to communicate, openness to knowledge sharing, and trustfulness of interpersonal interaction (Hofstede, 2011).

To date, several studies were conducted to define culture and to investigate how culture steers individuals’ social and cognitive behaviors at large. Hofstede’s multi-dimensional model (Hofstede, 2011), a seminal framework in cross-cultural psychology, assesses national cultures across six dimensions: Power Distance, Uncertainty Avoidance, Individualism versus Collectivism, Masculinity versus Femininity, Long Term Orientation, and Indulgence versus Restraint. Originating from an extensive survey spanning over 50 countries and 100,000 participants, the model interprets cultural differences and their impact on societal behaviors and values. It postulates that these dimensions shape not just ideologies but also the collective behaviors within a society, making human actions predictable to an extent.

The impact of these cultural dimensions, particularly Individualism versus Collectivism, has been profound on collaborative practices and problem-solving, as subsequent studies indicate the profound influence of a society’s orientation on individuals’ approach to group tasks (Hofstede, 2011). These studies reveal that individualists prioritize personal goals and task achievement, often above group harmony, while collectivists tend to value group cohesion and interpersonal relationships, potentially at the expense of task outcomes. This cultural orientation influences not only personal relationships and work dynamics but also educational approaches, particularly in collaborative learning settings (e.g., Arieli and Sagiv, 2018; Arpaci et al., 2018; Zhang and Yin, 2019).

Hofstede’s studies utilized the World Values Survey to measure cultural values across 76 countries, introducing the Individualism vs. Collectivism (IDV) index. They found a link between a country’s wealth, geographical location, and its level of individualism, with richer Western countries tending towards individualism and Eastern countries towards collectivism. The U.S., Australia, the UK, and Canada scored high on individualism, while East Asian countries like Korea and China scored low. Subsequent research has continued to explore these cultural dimensions in representative nations. Chen (2000) conducted a comparative study and claimed that Chinese children grow up in a collectivist culture, and thus their psychological and cognitive development is different from children from individualist countries such as the United States. In terms of behavioral representations, Asian societies often display shyness, self-restraint, and high levels of socialization, whereas North American societies encourage self-assertion and personal independence. Oyserman et al. (2002) meta-analysis found that Canada and the U.S. are among the world’s most individualistic societies, valuing individual rights over social ties. However, compared to Japan and Korea, these differences in individualism versus collectivism are not as pronounced in certain behaviors. Only China distinctly favored collectivism, which has implications for understanding social behaviors in collaboration and conflict resolution across cultures.

Item response theory (IRT) models are commonly adopted tools to analyze survey responses. Traditional IRT models are widely used for item parameter estimation but often criticized for failing to incorporate additional variables that are likely to influence responses to the items. Unlike traditional IRT models, explanatory IRT models allow researchers to use both item-level (i.e., item properties) and person-level (i.e., student characteristics) when estimating item parameters and latent trait levels (De Boeck and Wilson, 2004). EIRM has originated from the generalized linear mixed modeling (GLMM) framework in which items (i.e., repeated measures) are assumed to be nested within students (i.e., random effects) within a multilevel structure. This approach enables the use of explanatory variables (i.e., covariates) at the item and person levels to predict student responses in the same model. To date, many studies have applied EIRM to a wide range of areas, including differential item functioning (French and Finch, 2010), contextual effect detection (Randall et al., 2011), and proficiency scaling (Hartig et al., 2012). However, these studies employed EIRM to model the effects of item-level and person-level variables using dichotomous item responses (i.e., correct or incorrect). To date, only several studies have utilized the EIRM approach to model polytomous responses, survey items based on a Likert scale.

Tuerlinckx and Wang (2004) first attempted to analyze a publicly-available verbal aggression dataset using two polytomous IRT models including the Rating Scale Model (RSM; Andrich, 1978) and Partial Credit Model (PCM; Masters, 1982) with item-level and person-level covariates. In their study, the explanatory form of PCM (EPCM) can be formulated as the log-odds of choosing response j over (j − 1) on item i for person n:

where θn represents the latent trait of person n, Znij is a matrix of fixed and random effects related to the latent trait θn, δi is the threshold parameter between the first (j = 0) and second (j = 1) response categories for item i, Xnij′ is the matrix of fixed and random effects related to individual items, and τij is the matrix of the distances between the first threshold parameter and the other threshold parameters for item i. Tuerlinckx and Wang (2004) model allowed the estimation of not only item threshold parameters but also the effects of covariates related to items and persons. However, the model could only estimate the effects of item-related covariates on the first threshold parameter. That is, the model contains no explanatory variables to describe the common variability beyond the initial threshold parameter.

Stanke and Bulut (2019) used the same verbal aggression dataset and estimated four polytomous IRT models including RSM, PCM, EPCM, and cross-classified PCM with a new parameterization that enables the estimation of the effects of explanatory variables on both the first threshold parameter and the subsequent threshold parameters. That is, the model can account for the variability in all of the threshold parameters based on item-level covariates. In Equation 2, the log-odds of choosing response j over (j − 1) for item i for person n can be calculated as:

where Wnij′ is a matrix of fixed and random effects related to the threshold parameters and the remaining elements are the same as those defined for Equation 1. Compared to Tuerlinckx and Wang (2004) approach, this method is more flexible in terms of explaining how all threshold parameters differ based on item-level covariates. In this study, Stanke and Bulut (2019) approach was employed to examine the impact of cultural orientations on attitudes towards collaboration. Students’ responses to the survey items in the Collaboration Scale of PISA 2015 were analyzed based on item-level (item type: valuing relationships vs. teamwork) and person-level (country type: individualism vs. collectivism) covariates.

To sum, the impact of culture on individuals’ attitudes towards collaboration revealed by previous studies suggests that the cultural factor provides the fundamental contexts for arousing awareness of collaboration and promoting individuals’ psychological engagement in collaborative tasks (Hofstede, 2011). Therefore, in the present study, we make two hypotheses: (1) students from collectivist cultures will have a higher endorsement of collaboration compared with students from individualistic cultures and (2) higher endorsement of collaboration will lead to better success in the CPS assessment.

The sample of the current study consists of students from five countries that participated in PISA 2015: China (n = 9,841), Japan (n = 6,647), Korea (n = 5,581), Canada (n = 20,058), and the United States (n = 5,712).

In total, 47,839 students took both the CPS assessment and the Collaboration Scale in the selected countries. The demographic information of the students and their CPS assessment scores from the five countries reported by OECD (2017) are shown in Table 1. To prepare the dataset, we first detected the missing values and found that the variables Country ID and Gender contained no missing data, whereas students’ responses to Item 1—Item 8 contained missing responses. We calculated that the proportion of missing data was lower than 5% (about 2%). Therefore, we deleted the samples with missing data and conducted further analyses.

PISA is a triennial assessment of competencies and attitudes that aims to measure “how well 15-year-old students approaching the end of compulsory schooling are prepared to meet the challenges of today’s knowledge societies” (OECD, 2017). Covering nearly half a million students in 2015, PISA uses a two-hour test on core subjects and a questionnaire suite. It has probed problem-solving since 2003, with a shift to computer-based assessment in 2012, defining it as cognitive processing to resolve non-obvious problems. In 2015, PISA expanded to include collaborative problem-solving (CPS), assessing individuals’ capacity to pool knowledge and efforts in a group problem-solving context. It also introduced the Collaboration Scale, assessing students’ attitudes towards valuing relationships and teamwork.

The Collaboration Scale measures students’ self-report levels of “establishing and maintaining shared understanding,” “taking appropriate action to solve the problem,” and “establishing and maintaining team organization” (OECD, 2017, p. 44). There were eight items from two domains of collaboration which measures students’ attitudes towards valuing relationships and measuring students’ attitudes towards teamwork, which are presented in Table 2. In the present study, we calculated the reliability indices (Cronbach’s Alpha) of the two scales among the selected countries and revealed that the reliability for scale of attitudes towards valuing relationships (Item1—Item4) is 0.7, and the reliability for scale of attitudes towards valuing teamwork (Item5—Item8) is 0.8, indicating satisfactory internal reliabilities of the two scales. Moreover, we also examined the construct validity of the two scales using Confirmatory Factor Analysis. The model yielded satisfactory model fit indices, with CFI = 0.94, TLI = 0.92, AGFI = 0.96, RMSEA = 0.08, and SRMR = 0.08. The standardized factor loadings of the eight items in the two scales are presented in Table 2.

To address the research questions, the effects of two explanatory variables were analyzed using EPCM. The item-level covariate item type is a two-way factor that explains whether an item measures attitudes towards valuing relationships or teamwork. Two sets of person-level covariates were used to code the countries based on the cultural orientations represented in Hofstede’s culture dimensions and their country identities. More specifically, country type refers to whether the selected country has an Eastern collectivist culture or a Western individualist culture; and country ID represents the identity of the selected country.

In PCM, the locations of the first threshold parameters “Strongly Disagree/Disagree” can be interpreted as item easiness. That is, the higher the estimate is, the easier it is to endorse the item for students. Generally, the easiness of all thresholds of the eight items are within [−3, 3], suggesting the items in the two scales are set appropriately with no invalid items. Table 4 presents the estimates of the first threshold parameters (δ) and the distances between the first threshold and the subsequent parameters (τ) for all items in the Collaboration Scale. Results show that students from the selected countries are least likely to choose to Disagree over Strongly Disagree with the statement “I prefer working as part of a team to working alone” (Item1:1.94 logits), followed by “I find that teamwork raises my own efficiency” (Item7:2.25 logits) and “I enjoy cooperating with peers” (Item8:2.34 logits) among all items based on the Strongly Disagree/Disagree threshold. In the contrast, students are most likely to choose to Disagree over Strongly Disagree with the statement “I am a good listener” (Item2:2.90 logits), followed by “I take into account what others are interested in” (Item4:2.88 logits) and “I enjoy considering different perspectives” (Item6:2.85 logits).

The distances to Disagree/Agree categories for the eight items vary from 1.06 to 2.25 logits, which shows the eight items within collaboration scales have different distances from Disagree to Agree. Item 1 (1.06 logits) and Item 7 (1.12 logits) have the smallest distances from Disagree to Agree, indicating it is more difficult for students to change their attitudes from Disagree to Agree on the statements “I prefer working as part of a team to working alone” and “I find that teamwork raises my own efficiency.” However, Item 3 and Item 8 have the larger distances between Disagree and Agree categories, indicating that it is easier to raise students’ levels of agreement on the statement of “I enjoy seeing my classmates be successful” and “I enjoy cooperating with peers”.

It is noted that the distances between Agree and Strongly Agree categories are all negative values ranging from −1.63 to −0.81. PCM does not require strict ordering of response categories and thus it allows researchers to estimate threshold parameters that may not follow the same order as the response categories. The negative threshold parameters suggest that the thresholds are not increasingly ordered as the response categories, suggesting it is harder for students to choose Strongly Agree category to Agree category compared with other thresholds. Additionally, the results from PCM demonstrate that students generally hold positive attitudes towards collaboration, it is easier for them to choose the response category Strongly Disagree over Disagree compared with the other thresholds.

Table 5 summarizes the results estimated by the five explanatory models using EPCMs. As explained earlier, the explanatory models included different item-level covariates, person-level covariates, and their interactions. Relative fit indices are presented at the bottom of Table 5 to facilitate the comparison of non-nested models.

Collaborative problem solving requires cognitive skills and non-cognitive social skills, where attitudes towards collaboration shaped by cultural orientations play a significant role. The present study shed light on understanding the relationship between psychological attributes (i.e., endorsement with collaboration) and performance on collaborative problem solving. Findings revealed that attitudes towards collaboration is not determined by national culture. Further, CPS is not simply determined by psychological characteristics like attitudes. Future research is necessary to understand and address the pedagogical and instructional gaps in CPS training among different countries, after accounting for the effects of confounding factors such as cultural orientations.

Some limitations were identified. First, the present study selected limited countries to represent the Western and Eastern cultures. Future research could be conducted among more countries and regions for global comparisons. Second, the sample selection and data analyses in the present study used country as the minimum unit, whereas overlooking the internal cultural variation within large countries like China, the US, and Canada. Further inspections could be conducted to investigate the differences of attitudes towards collaborations within countries.