Compassion in Engineering Education: Validation of the Compassionate Engagement and Action Scales (CEAS) and Conceptual Insights

Abstract

This study validates the Compassionate Engagement and Action Scales for Self and Others (CEAS) for use with undergraduate engineering students in the United States. Compassion, defined as sensitivity to suffering in oneself and others coupled with a commitment to alleviate and prevent it, is increasingly recognized as a vital socio-emotional competency in professional education. Using a cross-sectional survey design, 434 engineering undergraduates completed the CEAS instrument. In addition, students responded to open-ended questions about their definition of compassion and “others” as well as a validated engineering identity scale. Structural equation modeling supported the hypothesized three-flow, two-component structure of compassion, with excellent fit indices (CFI = 0.980, RMSEA = 0.037) and generally strong factor loadings. Reliability was high for most subscales (α = 0.716–0.762), though self-compassion engagement showed lower internal consistency (α = 0.614). Divergent validity was confirmed through weak correlations with engineering identity dimensions. Qualitative salience and thematic analysis revealed that participants most frequently associated compassion with empathy, kindness, caring, and understanding and defined “others” mainly as friends, family, and classmates, with high-compassion scorers being more compassion oriented and including broader social circles. Findings support the CEAS’s structural validity and utility in engineering education while highlighting opportunities to strengthen self-compassion engagement to enhance well-being, ethical reasoning, and socially responsible practice among future engineers.

1. Introduction

This study validates the Compassionate Engagement and Action Scales for Self and Others (CEAS) within the context of undergraduate engineering education in the United States, with the aim of enabling educators and researchers to position compassion as a professional competency.

The engineering profession is widely recognized as scientific in nature. Consequently, engineering education has traditionally emphasized engineering science alongside the mathematics and basic sciences that underpin it (Akera, 2017). However, engineering also has a long-standing recognition that while grounded in scientific knowledge, it is not solely a scientific pursuit; it is inherently a social endeavor (Sørensen, 2009). Engineers work within teams, serve communities, and produce technologies that have profound social implications. Therefore, they must be prepared to address not only the question, “Will this work?”, but also, “Will this have a positive overall impact?”

The need for engineering education to address social concerns is not new. The Grinter Report—a 1955 national review of engineering education commissioned by the American Society for Engineering Education—assigned equal importance to the social and technical dimensions of the profession. It emphasized that engineering programs should cultivate “the development of both a personal philosophy which will ensure satisfaction in the pursuit of a productive life and a sense of moral and ethical values consistent with the career of a professional engineer” (Harris et al., 1994, p. 76).

Moreover, ABET, the accreditor for engineering programs in the United States and in several other countries, currently requires programs to assess the students’ ability to “recognize ethical and professional responsibilities in engineering situations and make informed judgments, which must consider the impact of engineering solutions in global, economic, environmental, and societal contexts” (ABET, 2025). While this outcome supports the development of ethical awareness, it does not explicitly assess the students’ ability to translate that awareness into action in real-world contexts.

Traditional approaches in engineering ethics emphasize reflective consideration of technology’s societal impacts, while more recent approaches have incorporated empathy as a cognitive precursor to ethical behavior. Compassion extends these perspectives by adding a motivational dimension that propels ethically informed action aimed at mitigating harm. Scholars such as Williams (2008) and Berne (2018) argue that compassion can be intentionally cultivated as a professional competency, supporting care-centered engineering education and advancing goals of social justice, sustainability, and human well-being.

To advocate for the intentional integration of compassion as a core element of engineering education, it is necessary to establish robust means of assessing it. Currently, no validated tools exist for measuring compassion within the engineering context. To address this gap, the present study undertakes the validation of the CEAS for use with engineering students. While the CEAS has been validated in other undergraduate populations, this represents its first validation in engineering education.

1.1. Compassion in Engineering

In this study, we defined compassion following Gilbert et al. (2017) as a “sensitivity to suffering in oneself and others, combined with a commitment to alleviate and prevent that suffering” (p. 1). This definition emphasizes two core components: (1) the recognition of suffering and (2) the motivation to take action to alleviate it. This dual aspect distinguishes compassion from constructs such as empathy or sympathy, which share overlapping affective and cognitive features but differ in scope, motivation, and behavioral orientation.

Sympathy, often referred to as empathic concern, involves feelings of care or sorrow for another’s suffering without necessarily sharing or deeply understanding the other’s internal state (Batson, 2009). Sympathy is primarily other-oriented and affective, but it does not require perspective-taking or the motivation to act beyond offering comfort or expressing concern.

Empathy, in contrast, is a multifaceted construct encompassing both affective and cognitive dimensions (Batson, 2009; Hess & Fila, 2016). Affective empathy involves sharing or mirroring another’s emotional state (e.g., empathic distress or empathic concern), while cognitive empathy entails perspective-taking—imagining another person’s experiences either as if one were in their place (imagine-self) or as they themselves experience it (imagine-other) (Hess, 2024). In engineering education, empathy has been promoted as critical for user-centered design, ethical reasoning, and innovation (Hess et al., 2015; Walther et al., 2020). However, empathy alone does not inherently include the motivation to take action; it may inform ethical reasoning but stops short of compelling behavior change.

Compassion extends beyond sympathy and empathy by adding an action-oriented dimension. While empathy can provide the cognitive and affective understanding of another’s situation, and sympathy can foster concern, compassion incorporates the deliberate intention to act to alleviate suffering (Gilbert et al., 2017). A similar distinction has been articulated in engineering education research, where ethics has been framed as awareness and professional responsibility as its active enactment (Bielefeldt, 2018). While professional responsibility and compassion emerge from different theoretical traditions—normative ethics versus socio-emotional competencies—the parallel highlights a broader pattern in engineering formation: the movement from passive recognition to action-oriented engagement. This motivational force positions compassion as particularly relevant for engineering ethics, where the challenge often lies not only in recognizing harm, but in taking responsible, proactive steps to prevent or mitigate it. In this sense, compassion can be understood as a moral emotion—one that integrates awareness, affect, and motivation to support post-conventional moral reasoning and prosocial behavior (Dunivan et al., 2024).

In engineering education, most ethics-related interventions have focused on empathy development (e.g., Fertig and Kumpaty (2022); Hess et al. (2014, 2015); Surma-aho et al. (2018)). These studies provide a valuable foundation but tend to conceptualize compassion as either equivalent to empathy or as a subset of emotional engagement in moral decision-making. For example, in research using the Engineering and Science Issues Test (Borenstein et al., 2010), compassion was subsumed under broader affective responses within moral reasoning tasks rather than measured as a distinct construct. This approach overlooks compassion’s unique role as a driver of ethically informed action.

1.2. Previous Work in Assessing Compassion

A number of validated instruments have been developed to assess compassion across different fields including psychology, education, healthcare, and sociology. Although each tool provides valuable insights into specific aspects of compassion, none has been specifically validated for engineering education.

In psychology and education, the Self-Compassion Scale (Neff, 2016) measures how individuals respond to themselves in times of failure or suffering, and the Self-Compassion Scale-Short Form (SCS-SF) (Raes et al., 2011) provides a briefer version with the same purpose. Both instruments have been widely applied with higher education students and show solid psychometric properties in clinical and educational settings. Also in these fields, the Compassion Scale (Pommier et al., 2020) measures compassion directed toward others, with a focus on emotional and behavioral reactions to others’ suffering; it was originally validated in general populations but has also been used to study prosocial tendencies among college students.

In healthcare, compassion is often understood in relation to caregiving. The Compassion Scale-Medical (Martins et al., 2013) was created to adapt compassion measurement to medical environments. The Compassionate Care Assessment Tool (Burnell & Agan, 2013) and the Schwartz Center Compassionate Care Scale (Lown et al., 2015) both evaluate compassion from the patient’s perspective, highlighting the relational and behavioral aspects of care in clinical practice. The Relational Compassion Scale (RCS) (Hacker, 2008) expands this focus by examining mutual compassion exchanged between people, and has been applied both in health education and with general populations.

From sociology, the Compassionate Love Scale (Sprecher & Fehr, 2005) assesses altruistic concern and care for others in situations of suffering. This instrument has been mainly used in the study of social relationships and spirituality, providing a broader conceptualization of compassion that contrasts with more clinical or intrapersonal measures. Finally, the Santa Clara Brief Compassion Scale (Hwang et al., 2008) offers a short tool rooted in religious and spiritual traditions, emphasizing compassionate attitudes toward others in the general population.

Despite the availability of these instruments, a conceptual gap remains: most existing measures tend to capture compassion as either a personal attitude, a relational disposition, or a clinical behavior, without integrating a broader theoretical account of its underlying motivational and evolutionary functions. To address this limitation, the Compassionate Engagement and Action Scales for Self and Others (CEAS) (Gilbert et al., 2017) were developed. Rooted in an evolutionary model of caring motives, these scales conceptualize compassion as both a single emotional reaction and a complex, adaptive process involving multiple competencies. Specifically, compassion is defined as a “sensitivity to suffering in self and others, with a commitment to try to alleviate and prevent it” (p. 1), which is operationalized through two core competencies: motivated attention to suffering (engagement) and the enactment of helpful responses (action). These competencies operate within three distinct but interconnected domains, or “flows” of compassion: toward others, from others, and toward oneself. This two-part structure is consistent with foundational perspectives in Buddhist traditions, which emphasize awareness of suffering and the commitment to alleviate it (Harvey, 2000) as well as with Western philosophical accounts of moral motivation (Eisenberg, 2000). The CEAS therefore assesses the competencies of engagement and action within each flow of compassion, supporting an understanding of compassion as a socially intelligent skillset with implications for mental health, relationships, and well-being.

1.3. Psychometric Properties of the Compassionate Engagement and Action Scales for Self and Others (CEAS)

Based on the comprehensive foundations of compassion as an evolutionary competence, the Compassionate Engagement and Action Scales for Self and Others (CEAS) measure an individual’s ability to engage with and act compassionately toward themselves, others, and receive compassion from others. The construct is based on the idea that compassion involves sensitivity to suffering and a commitment to alleviating it, encompassing emotional engagement and proactive behavior. The CEAS assesses self-compassion (SC), compassion to others (CTO), and compassion from others (CFO), each subdivided into engagement (-E) and action (-A) components.

The CEAS has been employed in multiple studies to examine the relationship between compassion and psychological well-being. However, only three papers have psychometrically assessed the validity evidence of the instrument in different contexts (Gilbert et al., 2017; Lindsey et al., 2022; Murfield et al., 2021). Internal consistency has been supported, with Cronbach’s alphas ranging from 0.70 to 0.96 across subscales. Lindsey et al. (2022) reported reliability coefficients between 0.73 and 0.94, Murfield et al. (2021) found values between 0.70 and 0.96, and Gilbert et al. (2017) reported internal consistency scores from 0.77 to 0.92 across the self-compassion, compassion to others, and compassion from others subscales.

Confirmatory factor analysis has demonstrated adequate to good model fit. Lindsey et al. (2022) found acceptable fit indices across subscales, with the comparative fit index (CFI) values ranging from 0.89 to 0.96 and root mean square error of approximation (RMSEA) between 0.03 and 0.07. Murfield et al. (2021) reported strong model fit for all subscales, with CFI values of 0.97 to 0.98 and RMSEA between 0.07 and 0.08. Gilbert et al. (2017) similarly found good fit indices, with CFI values between 0.95 and 0.98 and RMSEA ranging from 0.073 to 0.098.

Convergent validity has been demonstrated through significant correlations with well-being and affect. Lindsey et al. (2022) reported positive associations between self-compassion and positive affect (PA Active: 0.37, PA Safe: 0.46) and negative associations with distress (DASS-21 total score: –0.35). Gilbert et al. (2017) found that self-compassion was positively correlated with well-being (β = 0.21, p < 0.001). Measurement invariance has been assessed across different populations, with Gilbert et al. (2017) confirming the cross-cultural applicability of the CEAS.

The CEAS have been widely applied in psychological research across diverse populations to investigate the role of compassion in well-being, emotional regulation, and psychological distress (Gilbert et al., 2025) In the engineering education context, cultivating compassion may enhance ethical reasoning, social responsibility, and human-centered design practices; however, the survey has not yet been validated for this field.

Establishing its reliability and validity for undergraduate engineering students will equip researchers and educators with a robust instrument to measure compassion, evaluate educational interventions, track student development longitudinally, and embed compassion into broader conversations about engineering ethics.

This study was guided by the following research questions:

• General question:
• What evidence of validity and reliability does the Compassionate Engagement and Action Scales (CEAS) provide when applied to undergraduate engineering students, considering both construct validity and response process validity?

• Specific questions:
  • What are the psychometric properties of the CEAS in this population including factor structure and internal consistency?
  • How do engineering students’ word associations and definitions of compassion and “others” provide response process evidence that supports or challenges the CEAS construct?

2. Materials and Methods

2.1. Instrument’s Characteristics

The CEAS (Gilbert et al., 2017) measures compassion toward oneself (SC) and others (CTO),as well as compassion received from others (CFO), with each orientation divided into engagement (-E) and action (-A) subscales. Items are rated on a 10-point Likert scale from 1 (never) to 10 (always). Engagement items capture competencies such as sensitivity to distress and empathy, while action items assess attention to helpful responses and the implementation of supportive behaviors. Prior to the present study, the instrument underwent a content validation process in which three expert scholars evaluated its phrasing and contextual adequacy for this population (Vargas-Ordonez et al., 2024). Their feedback underscored the need to clarify the intended meaning of “others” and prompted refinements to ensure clarity and consistency. These insights directly informed the design of the current study, particularly the inclusion of open-ended questions to capture how engineering students themselves interpret compassion and define “others”.

2.1.1. Open-Ended Questions

Two qualitative prompts were added to the instrument to illuminate how engineering students conceptualize key aspects of the survey. The first used a freelisting approach (Quinlan, 2019) asking participants to “Please list ten words, in order of significance, that best describe what compassion means to you. Separate each word with a comma.” This method was chosen to elicit a broad range of responses and assess the relative cultural salience of each term within an engineering education context. The second was an open-ended question—“When responding to the previous set of questions, who did you consider as the ‘Others’ in your answers?”—designed to capture the participants’ reference groups without constraining their responses to predetermined categories.

2.1.2. Engineering Identity Scale

To obtain evidence of divergent validity, engineering identity was measured using Godwin’s (2016) Engineering Identity Scale: (1) recognition with four items such as “My peers see me as an engineer” (M = 4.23, SD = 1.34, range: 0.00–6.00, Cronbach’s α = 0.77); (2) interest with three items such as “I find fulfillment in doing engineering” (M = 4.61, SD = 1.37, range: 0.30–6.00, Cronbach’s α = 0.89); and (3) performance/competence with six items (M = 4.09, SD = 1.17, range: 0.80–6.00, Cronbach’s α = 0.88). Responses were given on a 7-point scale (0 = strongly disagree, 6 = strongly agree).

2.2. Procedure

This study employed a cross-sectional, non-experimental survey design. Following institutional review board approval (IRB# 2024-140), data collection was conducted entirely online during the Fall 2024 and Spring 2025 semesters. The research team distributed mass email invitations directly and also contacted faculty with access to undergraduate engineering program listservs, which included all students formally enrolled in engineering programs at participating institutions across the U.S. The recruitment email linked to an informed consent form; consent was required to proceed. Participation was voluntary and anonymous, and to encourage participation, respondents were given the option to enter a raffle for one of fifty USD 10 gift cards, with email addresses collected through a separate mechanism to preserve anonymity. Surveys were administered in a single browser session without time limits, and the target sample size was set at 384 students based on conventional sample size estimation for a population proportion with a 95% confidence level and 5% margin of error as well as the rule of securing at least ten responses per item of the psychometric instrument under validation.

2.3. Participants

A total of 569 individuals accessed the online survey, with 434 participants completing the relevant sections and advancing in the form. Approximately 67.74% identified as male (n = 294), 27.42% as female (n = 119), 3.23% as non-binary (n = 14), and 1.61% did not report gender (n = 7). Ages ranged from 18 to 78 years (M = 20.66, SD = 3.67), with a median age of 20 years. All participants were undergraduate engineering students enrolled in 108 different colleges across the United States.

The most frequent majors were Mechanical Engineering (33.18%, n = 144), First Year Engineering (6.68%, n = 29), Industrial Engineering (4.84%, n = 21), Computer Engineering (5.07%, n = 22), Civil Engineering (5.53%, n = 24), and Chemical Engineering (4.61%, n = 20), with 38.71% (n = 168) representing 83 other engineering programs. Regarding academic standing, 37.56% were freshmen (n = 163), 24.42% sophomores (n = 106), 23.04% juniors (n = 100), 13.36% seniors (n = 58), and 1.61% did not report their academic year (n = 7).

2.4. Data Analysis

2.4.1. Psychometric Analysis

For quantitative analysis, after data cleaning and coding, psychometric analyses were conducted using structural equation modeling (SEM) to evaluate the internal structure and construct validity of the CEAS. A second-order measurement model tested three latent factors SC, CTO, and CFO each with -E and -A subscales. Model fit was evaluated using the comparative fit index (CFI), Tucker–Lewis index (TLI), root mean square error of approximation (RMSEA), standardized root mean square residual (SRMR), and goodness-of-fit index (GFI), with thresholds based on Hu and Bentler (1999).

Internal consistency was assessed using the Cronbach’s alpha for each subscale. Pearson correlations tested the convergent and discriminant validity with the Engineering Identity Scale. Missing data were handled using pairwise deletion under the assumption of missing-at-random. Analyses were conducted in JASP 0.18.3.0 (JASP Team, 2024).

2.4.2. Representational Content of Compassion—Freelisting

A total of 338 undergraduate engineering students completed the freelisting task in the online survey. Participants were instructed as follows: “Please list ten words, in order of significance, that best describe what compassion means to you. Separate each word with a comma.” Responses were analyzed using the AnthroTools a simple package for use in R version 4.5.1 for cultural domain analysis (Purzycki & Jamieson-Lane, 2017), treating each participant’s list as an ordered set. Data preparation involved light normalization (converting to lowercase and removing punctuation) and harmonizing obvious lexical variants and synonyms through a codebook (e.g., support, supporting, supportive, supportiveness → support). Three salience measures were computed: mean salience (average salience among participants who mentioned the item), sum salience (total salience across all participants), and Smith’s salience (a standardized index integrating frequency of mention and average rank position (Smith & Bogartti, 1998)). To explore potential differences in conceptualization, we analyzed the salience of the whole population, and the lowest quartile (LQ; bottom 0–25%) and highest quartile (HQ; top 75–100%) of the CEAS composite score.

2.4.3. “Others” Question—Open

We analyzed the responses to the open-ended question, “List up to ten words you associate with the concept of compassion” using conventional (inductive) content analysis as described by Hsieh and Shannon (2005). The first author identified preliminary categories directly emerging from the participants’ responses. Based on these initial codes, the third and fourth authors collaboratively reviewed the data in three iterative rounds. Through discussion and consensus, a shared framework of nine content categories was established: family, partner, friends, colleagues, housemates, social community, strangers, anyone, and self. This framework was then applied to the lowest quartile (LQ; bottom 0–25%) and highest quartile (HQ; top 75–100%) of the CEAS composite score to conduct a comparative content analysis (Melnyk et al., 2021). Frequencies for each category were contrasted between LQ and HQ, and a comparative reading of the responses was conducted to identify additional divergences in how categories were expressed.

3. Results

3.1. Psychometric Properties

The statistical analysis of the CEAS instrument via SEM provided robust evidence supporting the hypothesized hierarchical measurement model of compassion: SC, CTO, CFO with subdomains of -E and -A. The initial model (Model 1, full item set) demonstrated excellent fit across multiple widely accepted indices: the comparative fit index (CFI = 0.980), Tucker–Lewis index (TLI = 0.987), non-normed fit index (NNFI = 0.987), and normed fit index (NFI = 0.973) all exceeded the recommended threshold of 0.95, indicating a close fit between the hypothesized and observed covariance structures. The root mean square error of approximation (RMSEA = 0.037; 90% CI [0.033, 0.041]) was well below the 0.05 cutoff, suggesting an excellent approximation to the population model, and the standardized root mean square residual (SRMR = 0.058) fell below the conventional 0.08 criterion. The goodness-of-fit index (GFI = 0.982) further reinforced the appropriateness of the specified model.

A second model (Model 2) was estimated with the removal of Item 6, “I reflect on and make sense of my feelings of distress”, which exhibited low explained variance (R² = 0.145) and a moderate standardized loading in Model 1. Model 2 retained similarly strong fit indices—CFI = 0.981, TLI = 0.987, RMSEA = 0.037, SRMR = 0.059—with negligible change relative to Model 1 (ΔCFI = +0.001), indicating that SC-E Item 6 was not critical to overall model fit. Notably, the removal of this item slightly increased the R² of the first-order factor SC engagement (from 0.638 to 0.772), suggesting improved internal consistency within this subscale. However, the exclusion also reduced the conceptual coverage of the reflective dimension of distress processing, a theoretically relevant aspect of the construct. Therefore, Model 1 was retained.

As shown in

Table 1. Explained variance (R²) for the second-order factor on CEAS in Model 1.

Second-Order Factor	R2	First-Order Factor	R2
Self-Compassion (SC)	0.64	Engagement with own distress (-E)	0.84
		Actions toward own distress (-A)	0.57
Compassion toward Others (CTO)	0.57	Engagement with others’ distress (-E)	0.92
		Actions toward others’ distress (-A)	0.90
Compassion from Others (CFO)	0.47	Engagement from others toward my distress (-E)	0.90
		Actions from others toward my distress (-A)	0.93

, Model 1 indicated that the second-order latent variables—SC (R² = 0.64), CTO (R² = 0.57), and CFO (R² = 0.47)—explained substantial proportions of variance in their respective first-order factors. SC predicted SC-E (R² = 0.84) and SC A (R² = 0.57), CTO predicted CTO-E (R² = 0.92) and CTO A (R² = 0.90), and CFO predicted CFO-E (R² = 0.90) and CFO A (R² = 0.93). These high coefficients indicate that the model meaningfully captured the intended multidimensional structure.

All standardized factor loadings in both models were statistically significant (p < 0.001). In Model 1, the loadings ranged from 0.514 to 0.798 for SC indicators, 0.603 to 0.748 for CTO, and 0.518 to 0.691 for CFO. Second-order factor loadings onto the higher-order compassion factor were also strong, with SC (0.798), CTO (0.753), and CFO (0.686) each contributing significantly (see

Table 2. Standardized factor loadings for the second-order factor on CEAS.

1st-Order Factor	Item	Std. Loading
Self-Compassion—Engagement (SC-E)	I notice and am sensitive to my distressed feelings when they arise in me.	0.471
	I am motivated to engage and work with my distress when it arises.	0.585
	I am emotionally moved by my distressed feelings or situations.	0.479
	I reflect on and make sense of my feelings of distress.	0.381
	I avoid thinking about my distress, try to distract myself, and put it out of my mind.	0.562
	I tolerate the various feelings that are part of my distress.	0.417
Self-Compassion—Actions (SC-A)	I am accepting, non-critical, and non-judgmental of my feelings of distress.	0.636
	I think about and come up with helpful ways to cope with my distress.	0.673
	I take the actions and do the things that will be helpful to me.	0.584
	I create inner feelings of support, helpfulness, and encouragement.	0.604
Compassion toward Others—Engagement (CTO-E)	I am motivated to engage and work with other peoples’ distress when it arises.	0.660
	I notice and am sensitive to distress in others when it arises.	0.603
	I am emotionally moved by expressions of distress in others.	0.568
	I reflect on and make sense of other people’s distress.	0.488
	I avoid thinking about other peoples’ distress, try to distract myself, and put it out of my mind.	0.685
	I tolerate the various feelings that are part of other people’s distress.	0.549
Compassion toward Others—Actions (CTO-A)	I am accepting, non-critical, and non-judgmental of other people’s distress.	0.689
	I direct attention to what is likely to be helpful to others.	0.634
	I take the actions and do the things that will be helpful to others.	0.690
	I express feelings of support, helpfulness, and encouragement to others.	0.748
Compassion from Others—Engagement (CFO-E)	Others tolerate my various feelings that are part of my distress.	0.629
	Others notice and are sensitive to my distressed feelings when they arise in me.	0.518
	Others avoid thinking about my distress, try to distract themselves, and put it out of their mind.	0.514
	Others are emotionally moved by my distressed feelings.	0.488
	Others are actively motivated to engage and work with my distress when it arises.	0.657
	Others reflect on and make sense of my feelings of distress.	0.546
Compassion from Others—Actions (CFO-A)	Others are accepting, non-critical, and non-judgmental of my feelings of distress.	0.618
	Others think about and come up with helpful ways for me to cope with my distress.	0.670
	Others take the actions and do the things that will be helpful to me.	0.683
	Others treat me with feelings of support, helpfulness, and encouragement.	0.691

).

Parameter estimates further affirmed the internal consistency and construct validity of the measurement model. All loadings of the observed variables on latent constructs were statistically significant (p < 0.001), with standardized loadings ranging from 0.514 to 0.798 for SC indicators, from 0.603 to 0.748 for CTO, and from 0.518 to 0.691 for CFO. These robust loadings confirm that the indicators reliably reflect the latent variables. Furthermore, the second-order factor loadings demonstrated strong coherence within the overall compassion construct. For instance, SC (standardized estimate = 0.363, p = 0.005), CTO (0.433, p < 0.001), and CFO (0.529, p < 0.001) all significantly contributed to the higher-order compassion latent variable, thus empirically supporting the hierarchical structure of compassion as a multidimensional construct.

Fit comparison based on the Akaike information criterion (AIC) and Bayesian information criterion (BIC) showed minimal differences between models (Model 1: AIC = 419, BIC = 911.812; Model 2: AIC = 419, BIC = 911.812), underscoring their statistical equivalence in terms of parsimony. Given the excellent fit of both models, the small magnitude of improvement in Model 2, and the theoretical importance of retaining content breadth for the SC-E subscale, Model 1 was selected as the final model for subsequent analyses. Internal consistency reliability was evaluated using Cronbach’s alpha (α) with pairwise complete cases, as shown in

Table 3. Cronbach’s alpha reliability estimates and 95% confidence intervals.

Scale	Cronbach’s α	95% CI Lower	95% CI Upper
Self-compassion engagement (SC-E)	0.614	0.554	0.667
Self-compassion actions (SC-A)	0.701	0.652	0.744
Compassion toward others engagement (CTO-E)	0.745	0.705	0.780
Compassion toward others’ actions (CTO-A)	0.762	0.723	0.796
Compassion from others’ engagement (CFO-E)	0.716	0.672	0.756
Compassion from others’ actions (CFO-A)	0.749	0.708	0.785

. The SC-E demonstrated relatively low internal consistency (α = 0.614, 95% CI [0.554, 0.667]), suggesting that the items may not be strongly cohesive in measuring the construct. In contrast, the SC-A scale showed acceptable reliability (α = 0.701, 95% CI [0.652, 0.744]), supporting its use in psychological assessment.

Dimensions related to the CTO-E (α = 0.745, 95% CI [0.705, 0.780]) and CTO A (α = 0.762, 95% CI [0.723, 0.796]) subscales indicated good internal consistency, suggesting that these items consistently captured their respective constructs. Similarly, the CFO-E (α = 0.716, 95% CI [0.672, 0.756]) and CFO A (α = 0.749, 95% CI [0.708, 0.785]) subscales also demonstrated adequate to good reliability.

To assess the construct validity, bivariate Pearson correlations were computed between the CEAS subdimensions and the engineering identity dimensions: recognition, interest, and performance/competence. Given the theoretical distinction between compassion and engineering identity, low correlations were hypothesized, supporting divergent validity.

As expected, the results demonstrated generally weak to modest correlations, confirming the discriminant nature of the constructs (see

Table 4. Mean, SD, and correlation between CEAS dimensions and engineering identity measurement.

Variable	Mean	SD	1	2	3	4	5	6	7	8	9
1. SC-E	6.04	1.37	—
2. SC-A	6.18	1.66	0.48 *	—
3. CTO-E	6.36	1.56	0.43 *	0.26 *	—
4. CTO-A	6.65	1.72	0.37 *	0.30 *	0.68 *	—
5. CFO-E	5.65	1.46	0.32 *	0.33 *	0.37 *	0.30 *	—
6. CFO-A	6.03	1.66	0.29 *	0.35 *	0.35 *	0.36 *	0.66 *	—
7. Recognition	4.23	1.34	0.29 *	0.28 *	0.28 *	0.34 *	0.23 *	0.24 *	—
8. Interest	4.61	1.37	0.21 *	0.21 *	0.28 *	0.39 *	0.03	0.11 *	0.48 *	—
9. Performance/competence	4.09	1.17	0.27 *	0.30 *	0.22 *	0.34 *	0.16	0.18 *	0.55 *	0.63 *	—

* p < 0.05.

). The correlation between SC-E and recognition was r = 0.29, p < 0.001, and between SC-A and recognition, r = 0.28, p < 0.001, both in the low range. Similarly, correlations with interest were even lower, ranging from r = 0.03 (ns) to r = 0.21, p < 0.001 across the CEAS subscales. Notably, performance/competence showed a modest association with SC-A (r = 0.30, p < 0.001) and recognition (r = 0.55, p < 0.001), though these values remained within acceptable bounds for demonstrating convergent independence rather than overlap.

Intercorrelations among CEAS subdimensions were moderate to high, particularly within the same construct domains (e.g., SC-E and SC-A, r = 0.48, p < 0.001), supporting the convergent validity of the CEAS structure itself. In contrast, the overall pattern of correlations between the CEAS subscales and engineering identity dimensions supports divergent validity, with no evidence of shared method variance or conceptual redundancy between these instruments.

3.2. Representational Content of Compassion

The freelisting analysis on compassion revealed the frequencies of associations and cultural salience of the construct. Smith’s salience (S) was selected for ranking items, as it enables cross-study comparisons and accounts for the frequency of mention and the average position of each item in the list. Items were sorted from most to least salient based on Smith’s S.

3.2.1. Overall Salience Patterns

Table 5. Salience of compassion-related terms for the full sample.

Item	Frequency (%)	Mean Rank	Mean Salience	Smith’s Salience (S)
Empathy	71.30	0.78	188.11	0.56
Kindness	73.08	0.74	178.34	0.53
Caring	68.64	0.68	153.32	0.45
Understanding	65.98	0.69	152.17	0.45
Love	40.24	0.59	79.66	0.24
Sympathy	35.80	0.61	73.47	0.22
Patience	25.74	0.57	49.48	0.15
Support	27.51	0.51	46.58	0.14

Note: Smith’s salience was interpreted relative to items within the same dataset rather than against fixed cutoffs.

presents the aggregated results for the full sample. Empathy emerged as the most salient item (S = 0.56), followed closely by kindness (S = 0.53). These findings indicate the centrality of affective resonance and prosocial disposition in the participants’ conceptualizations of compassion, thereby offering content-based validity evidence that the CEAS items align with the students’ spontaneous definitions of the construct. The higher relative salience of empathy compared with other terms suggests that students conceptualize compassion primarily as an affective experience, rather than as a catalyst for action following emotional resonance. The prominence of kindness, along with caring (S = 0.45) and understanding (S = 0.45), points to the integration of volitional and cognitive dimensions alongside affective resonance.

The relative salience of understanding (S = 0.45) may indicate cognitive competence in interpreting emotional states without judgment, skills essential for avoiding defensive or avoidant responses that could inhibit compassionate action. Love (S = 0.24) and sympathy (S = 0.22), while positive affective states, appeared to be less directly tied to the regulation of suffering, a key function of compassion within Gilbert’s framework. However, the salience of love warrants caution, as the term may encompass a wide range of emotional and relational meanings, some of which may indeed play a regulatory role. Its multifaceted nature makes it difficult to draw clear inferences without further context. Patience (S = 0.15) and support (S = 0.14) showed the lowest salience, suggesting that sustained compassionate responses and distress tolerance were less emphasized in the participants’ conceptualizations.

3.2.2. Quartile Comparisons

To explore whether compassion is conceptualized differently among students with higher versus lower compassion scores, we examined the salience patterns separately for the highest and lowest quartiles of the CEAS total scores.

In HQ, caring held the highest salience (S = 0.50), followed closely by empathy (S = 0.49) and kindness (S = 0.48). This profile suggests that high-scoring students placed similar emphasis on relational warmth (caring) and affective resonance (empathy), indicating a balanced integration of motivational and emotional components (

Table 6. Salience of compassion-related terms for the highest quartile of CEAS scores (HQ).

Item	Frequency (%)	Mean Rank	Mean Salience	Smith’s Salience (S)
Caring	73.85	0.70	32.80	0.50
Empathy	61.54	0.80	31.95	0.49
Kindness	66.15	0.76	31.16	0.48
Understanding	64.62	0.65	27.39	0.42
Love	36.92	0.60	14.49	0.22
Sympathy	23.08	0.67	10.10	0.16
Patience	27.69	0.54	9.68	0.15
Support	27.69	0.48	8.20	0.13

).

In LQ, empathy (S = 0.54) and kindness (S = 0.51) were most salient, with caring ranked fourth (S = 0.41). This ordering suggests that lower-scoring students foreground affective resonance and general prosociality over relational warmth (

Table 7. Salience of compassion-related terms for the lowest quartile of CEAS scores (LQ).

Item	Frequency (%)	Mean Rank	Mean Salience	Smith’s Salience (S)
Empathy	66.99	0.81	55.94	0.54
Kindness	72.82	0.72	52.48	0.51
Understanding	72.82	0.69	49.13	0.48
Caring	68.93	0.62	41.74	0.41
Love	36.89	0.63	23.29	0.23
Sympathy	33.01	0.63	21.35	0.21
Patience	27.18	0.55	15.53	0.15
Support	26.21	0.55	14.79	0.14

).

3.2.3. Comparative Patterns

Figure 1 compares the Smith’s S rankings for LQ and HQ, showing convergence on the top four items—empathy, kindness, caring, and understanding—but with meaningful shifts in their relative order. While both quartiles highlighted these terms, the reversal of caring and empathy between groups suggests different emphases: low-compassion (LQ) responses prioritized affective resonance (empathy), whereas high-compassion (HQ) responses elevated relational engagement (caring) to nearly equal salience. This balance in HQ responses may reflect a more integrated understanding of compassion as involving emotional attunement and sustained interpersonal involvement, supporting response-process validity by showing how students at different score levels interpret the construct in ways consistent with or diverging from the theoretical model.

Items such as love, sympathy, patience, and support remained low in salience across both quartiles, underscoring their peripheral role in conceptualizations of compassion regardless of overall compassion level.

3.3. Definitions and Relational Constructions of “Others”

Table 8. Classification of “others” for LQ, HQ, and the total sample.

Category	LQ (n)	LQ (%)	HQ (n)	HQ (%)	Total (n)	Total (%)
Friends	31	58.5	83	79.8	243	76.9
Family	20	37.7	54	51.9	144	45.6
Colleagues	20	37.7	49	47.1	137	43.4
Social community	5	9.4	13	12.5	34	10.8
Strangers	6	11.3	9	8.7	29	9.2
Anyone	6	11.3	6	5.8	21	6.6
Partner	1	1.9	4	3.8	16	5.1
Housemates	1	2.0	6	6.0	15	5.0
Self	1	1.9	1	1.0	3	0.9
Variety (M, SD)	0.84	(1.15)	2.04	(1.13)	1.46	(1.24)

presents the classification of “others” referenced by participants in their open-ended responses, disaggregated by quartile of CEAS scores. Categories included diverse social referents. Frequencies and proportions were calculated to identify patterns across the lowest (LQ) and highest (HQ) quartiles as well as the total sample. This analysis captured how participants relationally situated compassion, highlighting common and peripheral categories of “others” and offering insight into the social scope of compassionate engagement.

Across the sample, friends (77%) and family (46%) were the most frequently identified categories of “others”, followed by colleagues (43%). Less common references included partners (5%), housemates (5%), social community (11%), strangers (9%), and anyone (7%). Mentions of self as the “other” were rare (1%). Participants in the lowest quartile (LQ) of the CEAS composite score named, on average, fewer relationship categories (M = 0.84, SD = 1.15) than those in the highest quartile (HQ; M = 2.04, SD = 1.13).

In LQ, responses frequently centered on close ties, often restricted to friends (“My close friends”; “Friends, mostly”) and family (“Family and friends”; “The others in my answers were my parents, my close friends [both from college and home]”). Some LQ participants included colleagues or teammates (“Team members/colleagues”; “People who I have worked with either in groups or individually”), but their accounts often listed categories without elaboration (“coworkers, students, family, friends”). A number of LQ responses suggested a more distant or even withdrawn stance such as “Anyone who would share or express feeling with me”, or “I have no friends, so me myself and I”.

In contrast, HQ responses typically combined multiple relational categories and offered more detailed descriptions. Friends and family remained common referents (“My closest friends here at school and my boyfriend”; “My family and friends who confide in me or come to me for guidance when they are distressed”), but these were frequently extended to classmates, coworkers, roommates, and teammates (“Classmates and teammates, people I work with”; “My peers, teammates, friends, family, co-workers, roommates”). Several participants in HQ explicitly included broader social circles or strangers (“Friends, families, peers, and strangers seeking support”; “Strangers looking downcast [stressed]”), and some reflected on their own role or perspective in these relationships (“I put myself in the same place and thought how I would feel when another person is doing the same for me”).

Taken together, the quartile comparison shows that while both groups most often defined “others” through friendships and family ties, HQ participants reported a wider range of relationship categories and provided richer narrative detail. In contrast, LQ responses more often relied on brief or general listings, occasionally expressing detachment or self-reference. These relational patterns provide response-process validity evidence by clarifying which categories of ‘others’ the students most often considered when responding to the CEAS items.

4. Discussion

The present study validated the Compassionate Engagement and Action Scales—Self and Others (CEAS) in a sample of U.S. undergraduate engineering students, examining its psychometric properties and the ways participants conceptualized compassion and defined “others” within it. The construct was approached as multidimensional, reflecting contemporary evolutionary and motivational models in which compassion involves sensitivity to suffering, whether in oneself, in others, or received from others, paired with the commitment to alleviate and prevent it. Across these three flows, engagement entails emotional and cognitive attunement to suffering, while action represents the behavioral enactment of alleviating it. In the context of engineering education, where ethical reasoning and social responsibility are gaining prominence, compassion emerges as a socio-emotional competence that complements technical expertise and anchors professional decision-making in concern for human well-being.

4.1. Psychometric Validation of the CEAS

The hypothesized three-flow, two-component CEAS structure received empirical support with excellent fit indices (CFI = 0.980; RMSEA = 0.037), confirming that the model captured the latent structure of compassion in this population. Most items loaded strongly on their intended factors, and the pattern of intercorrelations between subscales was consistent with theoretical expectations. However, several self-compassion engagement items fell below the 0.40 loading threshold, most notably “I reflect on and make sense of my feelings of distress” (0.381) and “I tolerate the various feelings that are part of my distress” (0.417). These weaker indicators suggest that among engineering students, engagement with one’s own suffering was less consistently developed, potentially reflecting educational norms that privilege external problem-solving and task performance over introspection and emotional processing. Convergent validity was supported through moderate correlations among the CEAS subscales, while divergent validity emerged from the weak associations with engineering identity dimensions, indicating that compassion, as measured here, was distinct from professional self-concept.

The quantitative evidence revealed an asymmetry across compassion flows. Reliability coefficients were stronger for compassion toward others and compassion from others (α = 0.716–0.762), pointing to the salience and social reinforcement of interpersonal compassion in this group. In contrast, self-compassion engagement demonstrated a lower internal consistency (α = 0.614), suggesting less cohesive measurement and possibly weaker underlying skill development. This imbalance aligns with theoretical propositions that self-directed compassion often encounters more barriers, both culturally and individually, than outward compassion (Neff, 2023). In professional training environments, particularly in engineering, action-oriented coping strategies may be encouraged more than sustained emotional engagement with one’s own distress. This could explain why action subscales in self-compassion performed better than the engagement subscales. The most fragile competencies clustered around tolerating and cognitively processing distress, processes that compassion theory sees as necessary for transforming affective resonance into sustained compassionate action (Gilbert et al., 2025). Conversely, the strongest loadings were observed in action items for compassion toward and from others, reinforcing the role of prosocial behavioral norms in shaping compassion profiles.

These patterns are partly consistent with other research on the self-compassion engagement domain of Gilbert’s Compassionate Engagement and Action Scales (CEAS), which has, in some studies, shown conceptual or psychometric challenges, while in others has demonstrated solid performance. Gilbert et al. (2017) noted that self-criticism and other negative affective states can interfere with self-compassionate engagement, suggesting that individual emotional profiles may shape the measurement outcomes. Liu et al. (2025), studying clinical nurses, found that emotional regulation mechanisms linked to self-compassion enhanced work engagement, indicating that this domain can function as a strength in certain contexts. Román-Calderón et al. (2024) reported variations in the dimensionality of the self-compassion engagement subscale, which they interpreted as a need for closer theoretical and contextual examination. Collectively, these studies suggest that the functioning of this domain may depend on factors such as occupational demands, cultural context, and individual differences, rather than reflecting an inherent weakness of the scale. The present findings fit within this nuanced landscape, showing that while self-compassion engagement may require targeted support in populations such as engineering students, it remains a viable and valuable component of the CEAS that can be strengthened through educational and training interventions.

4.2. Conceptualizations of Compassion and Definitions of “Others”

The qualitative findings contribute to the validity argument by showing how students interpret the construct of compassion and identify the referents of “others” when responding to the CEAS. The most salient descriptors of compassion were empathy, kindness, caring, and understanding, which indicates that the participants primarily conceptualized compassion as an emotionally resonant and interpersonally warm experience. Empathy emerged as the central descriptor, particularly among students with lower CEAS scores, suggesting that many equate compassion with emotional responsiveness rather than with deliberate or sustained action. Kindness and caring reflected goodwill but often lacked reference to effortful engagement, while understanding pointed to emerging cognitive skills such as non-judgmental interpretation of emotional states, which support prosocial engagement (Decety & Jackson, 2006). In this sense, the evidence of content validity shows that the participants’ spontaneous definitions aligned with the engagement dimension of the CEAS but underemphasized its action-oriented component.

This imbalance is important in light of theoretical models that define compassion as both sensitivity to suffering and a commitment to alleviate it (Gilbert, 2010). While students recognized the affective resonance of compassion, traits associated with sustained action, such as patience or support, appeared marginal. This pattern reflects broader tendencies in engineering and STEM contexts where emotional labor and care are often undervalued (Cech, 2014). Students with higher compassion scores did show a more balanced profile, emphasizing both caring and empathy, which suggests an implicit integration of emotional attunement and motivational stance. This resonates with Strauss et al. (2016), who argue that definitions of compassion typically include the recognition of suffering and motivation to respond. Nevertheless, for the broader sample, compassion was more often framed as a feeling than as an action, which highlights a conceptual gap and supports the need for educational interventions that cultivate not only awareness, but also regulatory and behavioral capacities for sustained engagement.

The students’ definitions of “others” further provide evidence of response-process validity. Across the sample, friends and family were the most common categories, reflecting the central role of intimate ties in emerging adulthood (Arnett, 2000). However, differences appeared between quartiles of CEAS total score. Low scorers tended to mention one or two categories, often in general or impersonal terms such as “people I work with” and sometimes conveyed emotional distance or withdrawal. High scorers, in contrast, included a greater variety of categories such as classmates, roommates, coworkers, and even strangers, often elaborating their answers with richer narrative detail. These responses reveal a broader moral imagination (Singer, 2011) and a greater capacity for perspective-taking (Zaki, 2014), suggesting that higher levels of compassion are associated with more flexible and inclusive social schemas.

Taken together, these findings show that while the CEAS resonates with how students define compassion and conceive “others”, their interpretations also revealed limitations. Compassion is often reduced to emotional resonance without sustained action, and “others” are frequently defined within narrow relational circles. By uncovering these tendencies, the qualitative analysis not only complements the psychometric evidence, but also demonstrates how engineering students make sense of the instrument, thereby reinforcing the overall validity of the CEAS in this context.

4.3. Interpretive Synthesis of CEAS Performance and Student Perspectives

The combined analyses show that the CEAS captured compassion as a multidimensional construct among undergraduate engineering students, but with important asymmetries. Psychometric results confirmed the three-flow, two-component structure with strong validity and reliability, but the weakest performance appeared in the self-compassion engagement domain. This pattern is illuminated by the qualitative findings, which revealed that students seldom described compassion in terms of enduring presence with their own distress. Instead, their definitions emphasized empathy, kindness, and caring toward others, while action-oriented or self-regulatory competencies such as patience and support remained marginal. The limited salience of these terms helps explain why items targeting tolerance and cognitive processing of personal distress showed low consistency: the skills required to engage with one’s own suffering are not central to how students define compassion.

At the same time, the robustness of the scales for compassion to and from others resonated with the strong presence of relational descriptors in the students’ accounts. Friends and family were the most common categories of “others”, and the emphasis on interpersonal warmth aligned with the reliable measurement of outward-facing flows. High-scoring students went further, extending their circle of concern to include broader communities and even strangers, offering evidence that the CEAS meaningfully differentiates levels of perspective-taking and social imagination. The convergence between quantitative reliability in others-directed subscales and qualitative emphasis on relational closeness indicates that compassion is more consistently expressed outward than inward.

This synthesis suggests that the psychometric asymmetry of the CEAS—particularly the lower cohesion of the self-compassion engagement (SC-E) subscale—can be interpreted from different angles. Since the instrument underwent content validation by three engineering education experts, we do not attribute the result to linguistic or contextual misunderstanding. Instead, the low internal consistency of SC-E may reflect deeper cultural features of engineering education. Prior research has documented a persistent devaluation of emotional expression, self-reflection, and compassion in engineering contexts, which are often perceived as incompatible with the technical and rational ideals of the field (Cech, 2014; Riley, 2008; Leydens et al., 2012). While socio-emotional skills like teamwork and leadership are promoted, other competencies—such as emotional self-awareness, self-directed reflection, and compassionate self-regulation—tend to be underdeveloped or neglected (Walther et al., 2017; Hess & Fila, 2016). These findings suggest that the lower cohesion of SC-E may reflect not only measurement variance, but also structural and cultural limitations in how engineering students are socialized into their emotional lives. Further empirical work is needed to examine how interventions aimed at fostering emotional reflexivity might strengthen intrapersonal compassion among engineering students

The integration of quantitative and qualitative strands therefore strengthens the overall validity argument. The psychometric model demonstrates structural soundness, while the students’ interpretations provide content and response-process evidence that clarify why certain domains perform unevenly. Together, they highlight both the utility of the CEAS in this context and the need for pedagogical efforts that cultivate self-compassion and action-oriented skills alongside relational attunement.

5. Conclusions

This study provides empirical support for the CEAS as a valid and multidimensional instrument for assessing compassion among undergraduate engineering students. The combination of psychometric and qualitative analyses revealed consistent asymmetries between self-directed and other-directed compassion, emphasizing the need to strengthen action-oriented and self-compassionate competencies. These findings underscore both the utility of the instrument and the relevance of compassion as a socio-emotional capacity within engineering education.

Compassion in this population appeared most robust when directed toward close relational networks and least developed when directed inward, suggesting that self-compassion is under-cultivated. This imbalance carries ethical and psychological consequences, particularly in a field characterized by high cognitive demands and norms that value distance and efficiency over vulnerability and care. Educational interventions that foster reflection, peer support, and mindfulness may offer practical pathways to cultivate balanced compassionate engagement. Without such initiatives, engineering ethics education risks reinforcing a narrow conception of care that excludes the self and limits the students’ capacity for broader ethical action.

While the focus of this study was on undergraduate engineering students, the convergence of weak item performance, subscale reliability patterns, and qualitative salience data offers a coherent and multidimensional picture of how compassion operates in this sample. At the same time, several limitations should be noted. The cross-sectional design prevents causal inference, reliance on self-report introduces possible social desirability bias, and the sample was predominantly male and drawn from a single disciplinary context. These factors restrict the generalizability of the findings.

Future research should examine the development of compassion flows across time, disciplines, and cultural settings. Longitudinal designs could trace developmental trajectories, particularly in relation to interventions that target emotional regulation or distress tolerance. Comparative studies are also needed to determine whether the self–other asymmetries observed in this study are distinctive of engineering education or whether they resemble patterns found in other professional fields such as the social sciences, education, or healthcare. Clarifying these disciplinary differences would strengthen the understanding of how professional training shapes compassion. Future work could also establish normative benchmarks and cutoff scores for engineering and other professional populations, providing empirically grounded criteria to interpret variation in compassion profiles. Finally, it would be valuable to examine how compassion interacts with ethical reasoning and professional judgment, further clarifying its role in translating ethical awareness into engineering practice.

Finally, further psychometric work using approaches such as item response theory, cross-cultural validation, and abbreviated forms may enhance the scale’s precision and adaptability across populations. In conclusion, this study contributes to the recognition of compassion as both a psychological capacity and an ethical competence in higher education. By integrating structural validation with the students’ lived conceptualizations, it offers a multidimensional tool for understanding how compassion is interpreted and directed in an engineering context. Ultimately, these findings establish a foundation for cultivating more balanced and sustained compassion in future engineers, equipping them to respond ethically to the complex human challenges embedded in their professional practice.