Introduction
Despite significant efforts to improve student support and retention, attrition remains a major problem in engineering programs nationwide. While efforts to improve retention have historically focused on academic preparation, significant research suggests that non-academic factors, including peer relationships, department culture, and students’ sense of belonging play a major role in determining whether students stay in a program.
As part of an ongoing NSF-sponsored RED project, our computer engineering (CPE) department is engaged in a multi-year effort to identify and dismantle barriers to student success. Our project is particularly focused on increasing department inclusion and a shared sense of identity among students from all backgrounds. As part of this project, we have developed and administered a department climate survey to better understand how students experience our program. The survey includes sub-instruments designed to measure student experiences with interacting with the department, experiences interacting with faculty, perceived quality of peer interactions, CPE identity, and intent to persist in the major. Prior analyses of this work focused on documenting students’ areas of concern, including challenges related to workload (Danowitz & Slivovsky, 2025b), and competition among peers (Danowitz & Slivovsky, 2025a).
Building on this foundation, the present study focuses on the interactions between student experiences. Specifically, we address the following research questions:
RQ1: Which aspects of students’ experiences (including departmental, faculty, and peer interactions, and professional identity) are the strongest predictors of academic anxiety and intent to persist?
RQ2: How do students describe and contextualize the stressors and barriers that influence their persistence in the program?
Using data collected from students over three survey administrations from 2024–2025, we conduct a regression analysis to examine which of our measured aspects of student experience are the most significant predictors of student anxiety and self-reported intent to persist. To better provide context for our work, we also include a triangulating qualitative analysis of open-ended survey responses about what students view as their most significant stressors and how barriers to success show up in student’s lived experiences in the department.
By incorporating both quantitative and qualitative analysis of our longitudinal data set, this paper aims to provide insight on the types of challenges that most limit persistence in engineering. The findings will be used in our work to shape future interventions, and, we hope, will help to spark a conversation in the broader engineering education community about how department culture and peer relationships shape student outcomes.
Background
High rates of attrition, especially for students from historically excluded identity groups, have long been a problem for engineering and computing programs (Astin & Astin, 1992; Bahnson & Berdanier, 2023; National Science Board, 2016). Historically, efforts to address problems of retention have focused on issues of academic preparation and curricular reform (Adelman, 2006; Bradley & Bradley, 2006; Katz et al., 2006; National Academy of Engineering, 2005; Weston et al., 2019). While these approaches have had some success, a growing body of research indicates that academic factors alone are insufficient to explain patterns of persistence in engineering.
Recent scholarship has emphasized the role of students’ lived experiences within engineering programs. Factors such as department climate, peer interactions, professor interactions, and sense of belonging in the field have been shown to play a central role in shaping student persistence in STEM-fields generally (Carlone & Johnson, 2007; Godwin et al., 2016; Lord et al., 2011; Tonso, 2006), and in computing majors in particular (Lehman et al., 2023). This line of research aligns closely with psychological and educational frameworks of motivation (Edward & Ryan, 1985; Tinto, 1975), and suggests that attention to social and affective dimensions of engineering education hold greater promise for addressing attrition than academic factors alone.
One factor that has been shown to be impactful for persistence in engineering is a student’s sense of belonging. Research has documented how issues of exclusion, including microaggressions and other unwelcoming aspects of department and campus culture, can have deeply negative effects on students from historically excluded identity groups and for first-generation and socio-economically disadvantaged students (Blue et al., 2018; Lee et al., 2020; True-Funk et al., 2021). Conversely, a strong sense of belonging has been shown to have positive effects on student persistence and success within STEM majors (Brawner et al., 2020; Marra et al., 2012; Wilson et al., 2008) and computing majors specifically (Achenbach et al., 2018).
While department culture and student-faculty interactions help shape belonging, peer relationships and the quality of peer-peer interactions also play a role in encouraging student retention and success. Environments marked with excessive peer competition can lead to increased disengagement among students (Faulkner, 2000). Strong peer-peer ties and social interactions have been associated with increased student motivation and other positive measures of persistence (Christe, 2015; Kittur et al., 2021). It is important to note, however, that not all research has found a statistically significant link between peer interactions and student success (Dika & Lim, 2012).
Another factor that has been linked to student success and persistence is disciplinary identity. Students who view engineering work as appealing, and who identify with their engineering discipline tend to be more successful and show stronger motivation to complete their programs than students who do not (Rodriguez et al., 2018). Indeed, disciplinary identity is arguably closely related to the type of goal-setting framed as a precondition for retention and success in Tinto’s student-centered model (2017). The influence of identity on persistence is not uniform across groups, however, and prior research suggests that identity is a weaker predictor of persistence in engineering for women than for men (Atwood et al., 2022; Godwin et al., 2016).
Finally, in the past few years, there has been an explosion of research examining the effects of mental health in engineering programs. Studies have shown that engineering students are significantly more likely to suffer from mental health issues than the general population (Danowitz & Beddoes, 2022), and that these issues are often correlated student experiences in engineering program (Beddoes & Danowitz, 2022; Jensen & Cross, 2021). While mental health issues like anxiety, depression, and high levels of stress have been associated with negative academic and well-being outcomes, their relationship to persistence is less straightforward, particularly when considered alongside factors such as belonging and identity. Within the broader landscape of mental health, academic anxiety represents a context-specific construct that captures students’ responses to academic demands and performance expectations (Pekrun, 2006). Research has shown that higher academic anxiety is associated with negative academic outcomes, including reduced persistence in STEM fields and reduced well-being (England et al., 2019; Pekrun et al., 2017). Examining academic anxiety alongside other social factors may clarify the conditions under which mental health-related factors influence student persistence in engineering.
The goal of this work is to explore how these various factors interact to influence student success in a computer engineering program. Specifically, we explore relationships among academic anxiety, CPE identity, peer and faculty interactions and intent to persist to graduation.
Methods
Data were collected using a pre-existing survey instrument designed to capture various aspects of the student experience within a department (Danowitz & Slivovsky, 2025a, 2025b). The survey is made up of 39 questions across 5 subscales as summarized in Table 1. The sub-instruments themselves are pre-existing scales designed to measure a single aspect of student experience; as a result, each sub-instrument contains a different number of items. Our instrument is designed to measure respondent perceptions of department belonging, respondent experience with faculty interactions, quality of peer-peer interactions in the department, the extent to which respondents identify as a computer engineer, and the extent to which respondents experience academic anxiety. Internal consistency of the multi-item survey scales was assessed using Cronbach’s alpha based on the most recent survey administration, and values were acceptable to strong across all scales (α = 0.83–0.90). As the survey items are unchanged from previous works, we do not reproduce them here. The responses to the survey are used to answer our research question RQ1: Which aspects of students’ experiences (departmental, faculty, and peer interactions) are the strongest predictors of academic anxiety and intent to persist?
In addition to the scales, the survey included a 4-point Likert question asking students about their commitment and intent to persist to graduation in CPE. The survey also contained two open-ended response questions designed to elicit students’ biggest stressors in CPE and suggestions for what the department could do to improve the student experience. Specifically
-
What would you describe as your challenges or stressors being a CPE student?
-
What, if anything, do you think is holding you back?
The open-ended questions were optional, and 163 students elected to respond. Responses from the surveys were analyzed to triangulate the findings of barriers to success and to explore if any findings emerged beyond the survey measures (Borrego et al., 2009). The responses were emergently structurally coded for themes of responses and grouped based on commonalities (Saldaña, 2012). Answers were coded with multiple themes, if present. Twenty-four initial themes were generated. The number of responses for each theme were frequency counted to capture both the commonalities and range of experiences. The emergent themes were then associated with the quantitative scales. The initial coding was done by the second author, then the first author reviewed the entirety of the codebook and coded data to provide feedback on the any unclear codes and overall thematic analysis.
With IRB approval, data was collected in Spring 2024, Winter–Spring 2025, and Fall 2025. Students were recruited from within the computer engineering department via email sent to the department and through in-class announcements in major courses from willing CPE faculty. Students were incentivized to participate through a gift-card raffle.
To ensure data integrity and student privacy, all waves of the survey were administered by Redwood Consulting Collective. Redwood Consulting Collective hosted the instrument on their Qualtrics platform and cleaned and anonymized the results before providing them to the survey authors. Since student participation was anonymous, there is no way of knowing how many respondents participated in more than one data collection. Since the survey was sent to the entire computer engineering department, however, it is likely that some participants are represented in more than one wave.
Number of survey responses per wave and respondent demographics are summarized in Table 2. This represented department response rates in the range of 15–30%, and the sample population for each was roughly in-line with the overall demographics of our program. Wave 2 saw a particularly high response rate as we were able to recruit significant numbers of faculty members to mention the survey in their courses. Faculty participation in the first and third waves was less robust, however, leading to smaller response rates by comparison.
The authors completed data analysis using the open-source statistical language R (R Core Team, 2025), and the RStudio integrated development environment (RStudio Team, 2019). For each subscale in Table 1, subitem scores were averaged together to create a composite score variable. Composite scores were only computed for respondents who answered all questions in each sub-instrument. Composite variables for belongingness, peer relationships, and CPE identity and the data collection wave (1–3) were used as explanatory variables in a linear regression to see how they affect academic anxiety and reported commitment to graduate in CPE. Since peer relationship data was only collected in waves 2 and 3, regressions were run without peer relationship data on waves 1–3 and rerun with peer relationship data for only waves 2–3.
AI Usage
ChatGPT in a limited, assistive capacity during this project. The specific model available through the ChatGPT interface at the time of analysis was GPT-5.2.
During the draft planning phase, it was used to generate a preliminary writing schedule. During background research, it was used alongside traditional search tools like Google Scholar to surface potentially relevant background sources; all sources were independently reviewed, selected, and synthesized by the authors. During analysis, it was used only for discrete R syntax questions (e.g. combining data frames by column name), like consulting documentation or online forums; all analytic decisions, model specifications, and interpretations were determined by the authors. Finally, ChatGPT was used to suggest minor edits to grammar, usage, and phrasing. All aspects of study design, analysis, and interpretation were conducted and verified by the authors.
Limitations
There are some limitations of the study that may limit the generalizability of the results. First, our survey instrument captures student-reported intent to persist, which is commonly used as an indicator of persistence-related attitudes and decision-making. However, because the survey was only administered to students still enrolled in the major at the time of data collection, the study does not capture the perspectives of students who have already left the program. As a result, the findings may not fully reflect the experiences or factors influencing students who ultimately do not persist.
Second, the peer interaction scale was only added after the first data collection, limiting the analysis of its effects to two data sets rather than three.
Third, the study may be subject to self-selection bias. Although recruitment efforts included in-class announcements and modest incentives (e.g. gift card raffles) to encourage broad participation, it is possible that individuals with strong opinions about the department climate and culture were more likely to respond.
Fourth, the open-ended survey questions followed the Likert-scale items, which may have introduced a priming effect in participant responses: prior exposure to structured scales could have influenced how students framed or emphasized their qualitative responses. For this reason, the qualitative findings are used to clarify the quantitative responses, rather than as a fully independent data source.
Also, although gender was collected, this variable was not included in the primary analysis. Participation from some demographic groups was limited, and disaggregating the data—particularly at the intersection of gender and race—would have resulted in groups too small to support meaningful statistical analysis. Including these variables under such conditions risked producing unstable or potentially misleading results. We acknowledge that gender identity likely plays an important role in shaping students’ experiences and persistence, and future work will explore these relationships to better understand how these factors interact with departmental climate and student outcomes.
Finally, it is important to note that data collection took place at a single public primarily undergraduate institution. As a result, the findings may not generalize to computer engineering programs with substantially different student demographics, institutional structures, resource levels, or departmental cultures. This limitation is particularly relevant when interpreting the observed racial and ethnic differences, as these patterns may be influenced by characteristics specific to our institution and should therefore be validated through future studies conducted across a broader range of institutions.
Results
Linear regressions were used to determine how the different measures in the survey instrument contributed to students’ academic anxiety and intent to persist. Since peer interaction data was only collected for two of the three years, regressions were run both on three-year data without factoring in peer relationship data, and on wave 2 and 3 data with the peer relationship score included. The results of these analyses are shown in Table 3 and Table 4. The adjusted values range from 0.27 to 0.34. Only participants who completely answered each relevant subsection were included in each regression as noted by N in Table 3 and Table 4. As a result, roughly of wave responses were not used in each analysis due to missing data. There are also slight discrepancies in N for regressions involving academic anxiety and intent to persist due to a small number of students not completing all the relevant subinstruments used in both regressions. The authors consider this to be in line with previous experiences.
Data wave (data collection time) was used as a control variable across all regressions and was not found be to be significant.
CPE professional identity was a statistically significant predictor across all four regressions. It is negatively correlated with academic anxiety and positively correlated with student intent to persist to graduation in computer engineering.
In the waves 2 and 3 regression, positive peer interactions were negatively correlated with intent to persist. Conversely, in the context of the “All waves” dataset, academic anxiety was positively correlated with persistence. Academic anxiety is not significant for just waves 2 and 3, however, and it is unclear if this is simply because that dataset has fewer responses and more explanatory variables, or if another factor is at play.
In accounting for race and ethnicity, identifying as White non-Latina/o/x was significantly correlated with lower academic anxiety, while identifying as Asian was correlated with lower intent to persist. It is important to note that while both academic anxiety and CPE identity were positively associated with intent to persist for “All waves” (Table 4), CPE identity was shown to be negatively correlated with academic anxiety overall (Table 3).
Qualitative Results
The emergent qualitative themes were frequency counted and the most common codes that aligned with the quantitative survey categories are shown in Table 5. These codes triangulate what stands out most to students as challenges they were facing.
Some qualitative codes were aligned with a specific survey question. Other responses integrated multiple questions, for example the response “What is holding me back is burnout, the sting of staring at a screen for 8+ hours a day, and having an echo chamber in my head that makes me feel like I am always falling behind, not performing well enough, and that my teachers think I am the dumbest person in the world” is aligned with measures: “I feel supported by faculty in CPE,” “the professors in CPE are sensitive to the ability levels of all students,” “I often worry that my best is not as good as expected in school,” and “I am less confident about school than my classmates.” This answer was coded as imposter syndrome, which aligns with the quantitative scales for academic anxiety and CPE identity dimensions. Associations between qualitative codes and quantitative are highlighted in Table 5.
Discussion
Overall, the quantitative models explain a meaningful proportion of variance in both academic anxiety and intent persist (adjusted between 0.27 and 0.34). While a substantial portion of the variation remains unexplained, our results are consistent with prior work in engineering education where outcomes like persistence are shaped by a complex range of social and academic factors.
The quantitative data analysis illustrates several key points about persistence and student experience in our CPE program. First, our regressions indicate that higher levels of professional identity are associated with increased intent to persist and decreased academic anxiety. This finding is perhaps unsurprising considering existing research that describes professional identity as emergent from the totality of student experiences at an institution including coursework and social interactions (Mann et al., 2008; McCall et al., 2021). The high correlation between CPE professional identity and both academic anxiety and intent to persist indicates that interventions should be made to encourage professional identity as early as possible.
Prior research has indicated the importance of both encouraging identity through teamwork and accomplishment as well as affirming the value of students’ social identities as contributing to their success as engineers, specifically through coursework that emphasizes socio-technical dimensions of engineering (Hadis, 2005; McAlister et al., 2026). The department in this study is making strides to address this by strengthening a mentoring program for first-year students and introducing sociotechnical engineering lessons into key required courses (Hoang et al., 2026).
Additionally, our regressions indicate a strong relationship between peer interactions and professional identity when examining factors influencing intent to persist. Although peer interactions exhibit a negative coefficient in models that include professional identity (b = −0.29, p < .001; Table 4), this result appears to reflect a suppression effect rather than a direct negative relationship. Specifically, peer interactions were not a statistically significant predictor of intent to persist when we ran the model without professional identity, suggesting that their explanatory power is largely shared with professional identity. When both variables are included, professional identity emerges as a strong positive predictor, while the residual variance in peer interactions is associated with a negative coefficient. This pattern is consistent with the interpretation that peer interactions primarily contribute to persistence indirectly through their role in shaping professional identity, rather than functioning as an independent predictor. Supporting this interpretation, qualitative responses highlight a lack of community and camaraderie among CPE majors as a key concern: “Lacking a sense of belonging and community has been one of the major things holding me back. I ended up making friends with others outside my major and have very few friends within the major.” Therefore, it makes sense that the lack of a strong CPE community among students or struggling to fit in with peers would impact student’s professional identity formation. Unfortunately, our current dataset does not speak to the likelihood of either of these potential explanations.
Next, academic anxiety showed a small positive association with intent to persist when controlling for professional identity and related factors. However, when professional identity was removed from the model, anxiety was negatively associated with persistence and no longer statistically significant. Consistent with this pattern, professional identity was strongly negatively associated with academic anxiety (Table 3). Together, these results indicate a suppression effect, in which anxiety captures variance associated with lower identity and with a residual component positively related to persistence. In this context, professional identity appears to reduce academic anxiety, while the remaining component of anxiety—net of identity—may reflect heightened investment or perceived stakes associated with persistence.
Overall, the effect of academic anxiety was modest and substantially smaller than that of professional identity. Prior research has documented negative associations between anxiety and persistence (England et al., 2019), consistent with the negative bivariate relationship observed here. Our qualitative findings further suggest that anxiety may hinder the development of professional identity (e.g., “[a challenge is] thinking of myself as a student that has the capability to be in CPE. I need the confidence for me to succeed”). Given the strong link between professional identity and persistence, these findings underscore the importance of addressing academic anxiety and related mental health challenges in efforts to support student success.
Another major result from the quantitative analysis is that a sense of belonging from faculty was not significantly correlated with either intent to persist or academic anxiety. This result was unexpected given the important role faculty can play in working with individual students as mentors, advocates, and role-models (Main et al., 2020; Vogt, 2008). Though some relationships take these positive forms, faculty interactions can still be a barrier to students: some students qualitatively reported experiencing negative interactions with faculty “Getting some professors to empathize with my situation has been a struggle. I have always wanted to learn computer engineering, but some professors and classmates make me feel lazy and less capable because I can’t think and grind the ‘INSTITUTION’ way.” Several other respondents mentioned that they are too intimidated to approach professors. As one student reported: “I feel like office hours are quite daunting to me. I don’t like the idea of 1 on 1 time with a professor as I don’t like asking for help.” If students are too afraid to approach professors either out of anxiety or from bad experiences, it is possible that some students are simply devaluing close professor interactions as part of their overall educational experience. It is also likely that some of the positive and negative effects of belongingness from faculty are moderated by professional identity: if students have strong feelings about their identity as a computer engineer, welcoming or unwelcoming interactions with individual faculty may not have as large a direct impact on a student’s academic anxiety or persistence.
Regardless of the reason, for a primarily undergraduate program like ours, the fact that faculty interactions did not rate as a statistically significant factor in predicting academic anxiety or persistence is a concern. Future study will be needed to determine the level of interaction between students and faculty, and interventions could be developed to foster deeper mentoring relationships between faculty members and students.
Finally, results indicated differences in academic anxiety and intent to persist across racial and ethnic groups. Identifying as White was associated with lower academic anxiety. Identifying as Asian was associated with lower intent to persist relative to the reference group, even after accounting for professional identity, academic anxiety, and related factors. This finding contrasts with prior research showing strong retention rates among Asian students (Lord et al., 2009), although it could be a result of added pressure and stress on Asian American students as a result of model minority stereotypes that are prevalent in engineering and other STEM fields (Ma, 2010; Trytten et al., 2012). While the present study does not identify the mechanisms underlying these differences, the gap in intent to persist among racial and ethnic groups calls for further study and intervention development to improve persistence for students from different groups.
Conclusions
As retention continues to be a problem for engineering programs, it is critical that engineering educators take a holistic approach to identifying the different barriers and challenges that prevent students from succeeding. This study represents our attempt at such an analysis for a computer engineering department at a public primarily undergraduate institution. For this project, we developed an instrument to measure several dimensions of student-department experience including peer-peer relationships, departmental belonging, sense of belonging from faculty, academic anxiety, and computer engineering professional identity. We then used linear regressions to explore which of the measured factors correlate significantly with academic anxiety and self-reported intent to persist.
The results showed that CPE professional identity was strongly correlated with both variables: positive correlation for intent to persist and negative correlation for academic anxiety. The results also showed that identifying as White was statistically correlated with lower academic anxiety while identifying as Asian was statistically correlated with a lower intent to persist in Computer engineering. To help contextualize our results, we conducted a thematic analysis on student responses to questions about biggest stressors they face in CPE. Peer relationships, imposter syndrome, and relationship with faculty were among the most frequent survey-measure aligned stressors cited by students.
A key benefit of collecting the qualitative responses was that the data captured responses that were beyond the survey measures. While they were beyond the scope of this paper, additional relevant topics addressed by respondents include struggles with course learning, issues with work-life balance, and institutional difficulties such as challenges with enrollment. To begin to address these challenges, the department has created an internal tutoring center. The CPE student organization has been growing in terms of recruitment and funding, which strengthens institutional knowledge among students and is aiding peer-peer relationships.
Additional interventions to resolve student reported barriers will be a topic for future work, however. Future work will also explore qualitative responses relating to different segments of the respondent population such as supports for neurodiverse students, improving transfer student experiences, and addressing mental health concerns. Further study is also needed to understand how different demographic variables, such as gender and first-generation student status affect persistence and academic anxiety. With a better understanding of the barriers faced by different groups, interventions can be targeted to provide support for all students in the program.
Acknowledgements
This material is based upon work supported by the National Science Foundation under Grant No. IUSE/PFE:RED 2234256. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. We would also like to thank Redwood Consulting Collective for their work in adapting and administering the survey, and our student respondents for completing it.