Best practices from a cohort model for underrepresented PhD students in engineering and physics

1 Introduction

1.1 Background and need

Graduating additional, diverse PhD students is important for fostering innovation, as well as potentially providing opportunities to the graduates for leadership in industrial and academic settings. Studies show that companies with higher numbers of PhD scientists in leadership generate superior innovative output and more efficient monitoring, playing a crucial role in growth and research outputs (Afcha et al., 2023; Alam, et al., 2025; Halse and Mowbray, 2011; Swift, 2018). Similarly, diversity in organizations fosters innovation by bringing together people of different cultures and genders with different experiences and approaches to problem solving, showing the ability to interact and make connections previously unrelated (Daily and Eugene, 2013; Okatta et al., 2024). Taken together, diverse PhD graduates could foster increased innovation and excellence. 2021 numbers from the National Science Foundation (NSF) show that 14.8%, 8.2%, and 0.3% of science, engineering, technology and math (STEM) workers are Hispanic, Black, and Native American, respectively, even though these groups together represented about 35% of the US population combined (National Science Board and National Science Foundation, 2024). Thus, these groups of individuals represent partially untapped resources for maintaining a talented and innovative workforce, and equitable representation is desirable for the benefit of the graduates themselves (Benish, 2018; Burt et al., 2023; Main et al., 2022; McNeely and Fealing, 2018).

Of course, graduating more diverse PhD students depends not only on enrollments but also on retention or persistence. There have been many studies on the persistence of STEM students at the undergraduate level (e.g., Anderson et al., 2019; Davis and Bergstrom, 2023; Geisinger and Raman, 2013; Hansen et al., 2024 Oliviera et al., 2025; Seymour and Hunter, 2019), with factors including grades and conceptual understanding, economics, self-efficacy and self-confidence, high-school preparation, interest and career goals, classroom and academic climate, race and gender, and sense of belonging. Possible interventions to improve retention include academic communities or cohorts, co-curricular activities such as research, and mentorship (Anderson et al., 2019; Davis and Bergstrom, 2023; Dennehy and Dasgupta, 2017; Geisinger and Raman, 2013; Hansen et al., 2024; Oliviera et al., 2025).

When it comes to mentorship in the context of STEM, it has also been widely considered a useful tool for pursuing a variety of outcomes. At the undergraduate level, the impact of mentorship on STEM students has been widely researched; for example, studies show that quality mentorship, including near-peer mentorship, has potential to increase student sense of belonging, better their grades, and, importantly, increase retention rates (Dennehy and Dasgupta, 2017; Ravishankar et al., 2024; Wilson et al., 2012). Interestingly, some studies have even demonstrated benefits for the mentors themselves, including better communication skills, ability to collaborate, problem solving, networking, positive relationship development, and the rewarding feeling of giving back (Ravishankar et al., 2024; Spaulding et al., 2020). This attitude that mentors take on is also called the protégé effect (Gates, 2019).

Moreover, specific interventions and studies on the impact of mentorship on underrepresented minorities (URMs) in STEM, including students who identify as Black, African American, Hispanic, Latino American, Native American, Alaska Native, Native Hawaiian, or Pacific Islander have shown that multiple mentors, including models incorporating faculty and near-peer mentorship, have a positive impact on supporting URMs and their retention (Estrada et al., 2016; Kuchynka et al., 2025; Lisberg and Woods, 2018; National Academies of Sciences, Engineering, and Medicine, 2019; Venegas-Muggli et al., 2023). Namely, mentorship of URMs significantly contributes to integration of students into STEM communities, a positive indicator for retention and persistence (Brobst et al., 2026; Markle et al., 2022; Trujillo et al., 2015).

Despite support in the literature for increasing the number of URM PhD graduates, as well as evidence for mentorship and cohort programs increasing retention rates for URMs at the undergraduate level, there are few studies that have examined the efficacy of mentorship and cohort programs at the graduate level, particularly for STEM URMs. The existing studies to date have not fully evaluated the relationship of the mentorship programs to retention and often focused on mentorship by faculty advisors, rather than near-peer mentorship (Council of Graduate Schools, 2007). Studies that focused on near-peer mentorship of graduate students have not suggested how academic institutions can promote positive near-peer-mentoring relationships in graduate education (Lorenzetti et al., 2019). Although there have been fewer studies at the graduate level, they show that near-peer mentorship is promising for retention: Graduate students who receive high levels of near-peer mentorship experience further academic and professional socialization, ongoing personal and professional development, steady and highly autonomous motivation for their studies and, thus, lower intentions to leave (Liu et al., 2022; Lorenzetti et al., 2019; Peña-Lima et al., 2026; Swift, 2018).

Attrition rates of PhD students of all backgrounds are notoriously high (van Rooij et al., 2021). A longitudinal study from 1992 to 2007 conducted by the Council of Graduate Schools and involving 29 major U.S. and Canadian universities found that, on average, only 48.5% of engineering students and 39.3% of math and physical science students completed a PhD degree within six years of entering a PhD program, rising to 63.6% of engineering students and 54.7% of math and physical science students completing the PhD within 10 years (Council of Graduate Schools, 2007). Of more direct relevance to the current work is that the same study found that 5.5%, 14.5% and 20.2% of engineering students left their PhD programs within one, two or three years, respectively; for math and physical science students, these attrition rates are 8.2%, 17.8%, and 22.1%, respectively. While there has not been a nationwide longitudinal study on attrition published since then, recent data from the University of Colorado are examined herein. For students in PhD programs in physics, engineering and computer science at CU, 65% of students who entered from 2008 to 2017 completed PhD degrees within eight years. The graduation rate for URMs is lower: 53% of PhD URM students graduated from CU within the same time frame.

Thus, there is a need to improve the retention and success of STEM PhD students, especially from underrepresented populations. One programmatic framework created for such a task is the Louis Stokes Alliances for Minority Participation (LSAMP) Bridge to the Doctorate (BD) Program. The National Science Foundation (NSF) began supporting LSAMP programs across the United States in the 1990s, with the mission of increasing the number of STEM baccalaureate and graduate degrees awarded to persons from LSAMP populations (Gates et al., 2022). LSAMP populations include groups who are underrepresented in STEM: Black and African Americans, Hispanic and Latino Americans, Native Americans, Alaska Natives, Native Hawaiians, and Pacific Islanders. LSAMP Bridge to the Doctorate (LSAMP-BD) programs are a sub-type of LSAMP activities and focus on graduate degree access and retention, with grant recipients designing their own unique programs that typically include goals involving recruitment, academic and professional skills development, and PhD enrollment and graduation of underrepresented STEM students. Despite the existence of LSAMP BD since 2003, studies of the program's efficacy are limited (e.g., Gloster et al., 2023; Preuss et al., 2020). Although detailed accounts of LSAMP-BD programming structures exist, such as those at Texas A&M University and the North Carolina Agricultural and Technical State University, near-peer mentorship is not often implemented or studied, and overall program results (such as retention and graduate rates) are also limited (Gloster et al., 2023; Preuss et al., 2020). There remains a need to better understand how near-peer mentoring and cohort-based programs might contribute to improved retention at the graduate level. As a step toward meeting these needs, the authors describe and analyze student satisfaction of a structured near-peer-mentorship model piloted along with professional and social activities within a graduate cohort program. Surveys, group interviews, and retention data of participants, as well as suggestions for scaling the model and program, are discussed. Best practices that evolved through student feedback and continuous improvement are also described.

1.2 Description of the program and participants

The current study examines a near-peer-mentorship program and its associated cohort-based community of URM STEM doctoral students participating in the LSAMP BD program at the University of Colorado Boulder (CU). The LSAMP-BD program at CU was funded by NSF during 2023–2025 with financial support (stipends, tuition, and fees) for two-year fellowships for a cohort of 12 graduate students in engineering (9), computer science (1) and physics (2). LSAMP BD at CU matriculated its first cohort of twelve fellows in fall 2023. All 12 students are from LSAMP/URM populations (Table 1), which historically have had higher attrition rates as noted above, and several of these students had background deficiencies (e.g., five students were undergraduates at non-PhD schools with few research opportunities, and five students had undergraduate GPAs of 3.5 or less) – as was the intention for a “bridge” program. A second cohort, comprised of eight LSAMP-BD affiliates (without fellowship funding from NSF but with teaching or research support from CU), was added to the program in fall 2024. Because the affiliates were not funded by NSF, it was not required that the students be from LSAMP populations, although seven of the eight affiliates in the 2024 cohort identified as URMs (Table 1). Another cohort of ten affiliates was added to the program in fall 2025. Due to changes in university procedures, official race/ethnicity data were not collected for the 2025 cohort nor used for student selection – instead, students were recommended by their departments for the program based on perceptions of which entering students would most benefit from a supportive community, including mentoring, networking, skills for success in graduate school and leadership, and funding opportunities for conference travel. Nevertheless, it was a diverse group, including (unofficially) eight people of color and six URMs.

Table 1. Demographic data of participants – this table gives the breakdown of the first two cohorts of LSAMP-BD students at CU by gender and race or ethnicity, based on enrollment data.

Program leadership obtained an Impact Grant from CU for programmatic support, which has included community-building and professional-development opportunities through several types of events: Details of these activities are provided as supplemental material. As described in the next paragraph, mentoring has also been a key aspect of the LSAMP-BD program.

• Orientation/Planning Meeting – Once per semester, a planning meeting is held to welcome new students (fall semester), discuss program goals and plans for the semester, and provide a mini-workshop on mentoring.
• Networking Dinners – Once per semester, a networking dinner is held with industry (spring semesters) or faculty (fall semesters) representatives, using a “speed-dating” format where students meet with several different representatives during the evening.
• Workshops – Several times per semester in the first year but then decreased to typically once per semester based on student feedback, workshops or seminars were held on subjects such as communications, time management, financial planning, microaggressions, and DEI in STEM, often with guest speakers.
• Outreach – Starting in spring 2025, the program has partnered with a public charter high school in Denver, CO to offer a day-long science and engineering program each semester for students, led by LSAMP-BD students. The school's population is primarily URM students. The school is a Title 1 STEM school with a mission of achieving 100% college-acceptance rates. Because of its geographic location, the students do not interact often with college students, particularly diverse graduate students conducting research, like the LSAMP students.
• Retreats – The program has held a summer retreat each year at a conference center or large cabin in the mountains, with community- and skill-building activities plus a guest speaker or workshop.

The initial cohort of 2023 fellows were mentored by 2nd- and 3rd-year PhD students during the 2023–24 academic year, but it was relatively unstructured and survey data on its effectiveness were not collected. A structured near-peer-mentorship program was piloted in the 2024–25 academic year, with eight of the students in the 2023 cohort serving as mentors for the eight affiliates who entered the program in fall 2024. Subsequently, six of the 2023 fellows and six of the 2024 affiliates were matched with 2025 affiliates as mentors or co-mentors. All returning LSAMP students who volunteered to mentor were paired with a mentee or co-mentor and mentee. They were paired based on department or program, or the closest content area to their department/program if there was not an exact match. The structured near-peer-mentorship program has the following features:

• Participants attend a near-peer-mentorship training during orientation in the first week of the fall semester, which includes slides on key mentoring concepts and interactive scenarios facilitated by the Program Director.
• Second-year or third-year students (mentors) are paired with first-year students (mentees), most often but not always within the same discipline. Mentors are asked to prepare discussion items for each meeting, and mentees are asked to prepare questions for mentors. Suggested topics include tips for courses, PhD project selection procedures, communicating with professors, PhD preliminary exam preparation, graduate fellowship opportunities, work-life balance, time management, and career interests and opportunities.
• Mentor-mentee pairs are expected to meet at least four times per semester, including two or more times in person. They are asked to set the next meeting during the current meeting and discuss content/goals of the next meetings. On-campus meal cards are provided as an incentive for in-person meetings.
• Each program participant completes a survey at the end of each semester, including Fall 2024, Spring 2025, and Fall 2025 to date. Once each pair has met at least four times and each participant has completed the survey to document and evaluate these meetings, students receive stipends for their participation: $300 for mentors and $100 for mentees.

1.3 Aim of the study

The primary aim of this study is to analyze student satisfaction and outcomes for the graduate cohort model and then to use this information to improve the program and determine best practices. In developing the program, informal student feedback from the first semester led to the hypothesis that community building and relationships would be valued more than workshops by the students. Although CU's LSAMP-BD program has had a relatively short duration and modest size, its experience, outcomes, and lessons learned may be of value in designing other cohort-based programs for promoting retention and success of URM PhD students.

2 Methods

As described below, student satisfaction with the program, as well as a self-analysis of sense of community and skills progression, was assessed through surveys and group interviews (formative assessment). Student retention and outcomes were assessed through University of Colorado enrollment and graduation data (summative assessment). The study took place at the University of Colorado Boulder during fall 2024 – spring 2026.

2.1 Formative assessment

Survey results at the end of each semester are the primary data source for the analysis of student evaluation of the program. Student satisfaction with the mentoring program as well as the other aspects of the LSAMP-BD program were assessed in these surveys. The mentorship surveys asked participants how many times per semester they met with their mentor/mentee, which topics were discussed (out of the list of suggestions above), and how valuable they found the mentorship opportunity that semester on a Likert scale from 1 (low) to 5 (high). Students were also given the opportunity to add any additional comments on their experience and were asked to write what they gained most from the experience. The event surveys asked students to rate how effective they found each event on a 5-point Likert scale and gave opportunities for participants to add their comments. The retreat surveys asked how much they enjoyed different aspects of the retreat (activities, amenities, food, etc.) and how much they enjoyed the retreat overall on a 5-point Likert scale. Survey examples are provided in Supplemental Material. Individual meetings between the participants and either the Program Director or Program Coordinator (called a “1:1 check-in”) were held each semester. While informal, notes were taken on these meetings by program leadership, especially regarding suggestions for programmatic changes. As such, individual check-ins and survey responses provided feedback for continuous program improvement, as well as assessing student satisfaction and perceived skills progression.

Additionally, formative assessment was performed by the program's NSF-supported external evaluator, Dr. Sarah Hug. Dr. Hug conducted anonymous online surveys with the LSAMP-BD participants in April 2024 and April 2025. She also hosted group interviews online in December 2023 and December 2024 and in-person in March 2024 and April 2025. She aggregated and summarized the results of these surveys and interviews and submitted reports to the LSAMP-BD program leadership at CU in May 2024 and May 2025. Each survey asked students for their level of agreement on a 7-point Likert scale for a variety of statements. Question themes included peer support, faculty mentor support, skill progression, and belonging at CU and LSAMP. For the skill progression questions, students were asked to rate their agreement to statements such as, “I know how to manage my time effectively as a graduate student” and “I know how to give an ‘elevator speech’ describing my academic interests.” For the belonging questions, students were asked to rate their agreement to statements such as, “I feel a sense of belonging to my academic major's community” and “I feel comfortable in my LSAMP at CU Boulder community.” The surveys also included yes or no questions regarding the academic activities the students participated in the prior academic year (e.g., “In the past year, I have participated in a research project as a part of coursework” and “In the past year, I have prepared a poster for a conference”). In the group interviews, students answered questions related to the benefits of LSAMP, improvements to the program over the course of the prior year, graduate students’ needs not yet addressed, and proposed improvements for the future. More details on the survey questions used by the external evaluator are included in the Supplemental Material.

Finally, this study also looked at the 2021 CU Campus Culture Survey, which evaluated the satisfaction of students in their studies at CU (University of Colorado Boulder, 2022). The results of these surveys function as the control group, as it surveyed nearly 18,000 CU community members, including a 48% response rate of graduate students on campus, regarding student sense of belonging, connection and community, and more. Survey questions by the external evaluator mirrored some of the campus questions, for comparison purposes.

2.2 Summative assessment

Summative assessment was performed using University of Colorado enrollment data. Internally, individual enrollment and graduation data regarding the LSAMP-BD fellows and affiliates are accessed using the CU Campus Solutions database. These data are then compared to university-wide and nation-wide data obtained from the CU Office of Information Technology Data and Analytics Division and the Council of Graduate Schools, respectively (Council of Graduate Schools, 2007; Office of Information Technology, University of Colorado Boulder, 2025). This paper analyzes enrollment data from the 2023 fellow cohort and 2024 affiliate cohort, as year-to-year persistence data were not available for the 2025 affiliate cohort at the time that this report was prepared. Note that the CU Data and Analytics Division includes on its website reports available to the public and reports available only to CU personnel, but the current study required customized reports that the authors requested.

2.3 Statistical analysis

Statistical analysis was conducted on summative outcome data and on a limited number of comparable data for the formative outcomes. Statistical methods were used to compare LSAMP-BD enrollment data with general population data, as well as the general URM population data. Enrollment data are reported as the number of students who exhibited certain outcomes in a given year. Because the frequency of one event (attrition of LSAMP students) is less than five, a Fisher Exact test was used to compare outcomes (Finney, 1948). Using this method, the p-value is calculated from Equation 1:

p=(a+b)!(c+d)!(a+c)!(b+d)!a!b!c!d!n!,(1)

where arepresents the number of LSAMP-BD students being retained, brepresents the number of the control population being retained, crepresents the number of LSAMP-BD students leaving the PhD program, d represents the number from the control population leaving the PhD program, and nrepresents the number total number of LSAMP-BD students and the control-population students combined. Note, for example, that a value of p=0.1 implies 90% confidence that the difference in proportions is statistically significant.

For formative data, such as Likert-scale averages, comparisons between two populations (e.g., different years or sample groups) were made using standard statistical analysis (Milton and Arnold, 1995). A two-sample, two-sided t-test (Equation 2) was used for comparing the mean values of two samples:where x¯1 and x¯2 are the measured means of the two samples, n1 and n2 are the sample sizes, s1 and s2are the standard deviations (or, more accurately, s12 and s22are the variances), μ1−μ2is the difference in the true means, and tα/2 is the t statistic or value with significance α and ν degrees of freedom. Degrees of freedom are calculated with Equation 3:Note, for example, that a value of α=0.1 implies 90% confidence that the difference in the true means is within the confidence interval given by evaluating the right-hand side of Equation 2. When one sample is large (such as campus data vs. LSAMP data), n2≫ n1, these equations simplify to Equation 4:

μ1−μ2=x¯1−x¯2±tα/2s12/n1+s22/n2,(2)

ν=(s12/n1+s22/n2)2/((s12/n1)2/(n1−1)+(s22/n2)2/(n2−1)).(3)

μ1−μ2=x¯1−x¯2±tα/2s12/n1,ν=n1−1.(4)

3 Results

3.1 Formative outcomes

Mentorship in Action - In the first year of the program at CU, there was not a structured near-peer-mentorship program in place where senior LSAMP students mentored junior LSAMP students. First-year LSAMP students were instead paired with second- or third-year PhD students in their fields, but the mentors and mentees were not specifically trained in mentorship and expectations, nor were data collected. In Dr. Hug's 2024 report, students noted this lack of guidance, without necessarily attributing it to a lack of a structured near-peer-mentorship program. One student said, “…especially my first semester, I had no idea what I should have been doing in terms of research. Should I be looking to write a literature review at this point? Should I be trying to talk to different people about just like…Maybe to have some suggestions.” Another said “I think some [faculty] advisors, are too laid back. And some are too rushed in things. And so, some will add too much pressure, and you’ll feel like you’re drowning. But others, they don’t add enough pressure, and then you’re like three years in, and you’re like, “oh, what did I do?” kind of energy. I think if you have a general template, you can see where your advisor fits in that, and see maybe, “okay, do I need to push myself early on because they won’t do it?” so I kind of gauge how much you need to advocate for yourself early on.” Students were receiving mentorship from their faculty mentors, but those relationships did not provide guidance for the more nuanced experiences in graduate school, such as how to navigate different lab cultures and social norms, manage time and balance work with life, and handle a variety of coursework.

Thus, one of the largest programmatic changes starting in the 2024–25 academic year was to add an affiliate class and introduce a structured, “pay-it-forward” near-peer-mentorship program, as described in the Methods section. Qualitatively, the LSAMP near-peer-mentorship program was received by students in an overwhelmingly positive manner. When students were asked in the post-semester mentorship surveys for comments on their experience and to describe what they gained the most from it, mentors and mentees used words such as, “insightful”, “enlightening”, “inspiring” and “rewarding”. Participants gained the most from receiving help navigating classes and graduate student life, finding shared experiences, getting their mentor/mentee's perspective, becoming friends, and feeling validated. In Dr. Hug's 2025 report, one student is quoted saying, “Adding the mentees was fantastic! They don’t really cost me anything. I am going to the meeting regardless and they add a ton of new perspective! I enjoy mentoring and getting a meal card makes it super easy to meet.” Dr. Hug summarized with, “Having mentees has been a positive experience for the second-year students. They report that helping novice graduate students helps them see how they have developed over time and makes them more reflective. They enjoy “paying it forward” and feel that the role of mentor will help them in the future, particularly for those intending to become academics who will build and mentor their own science labs.” Studies show that mentorship of URMs significantly contributes to the integration of students into the STEM community, which is why this mentorship program was added (Markle et al., 2022). A result that was not anticipated is how beneficial the mentorship program would be for the mentors themselves. In fact, when arranging the mentorship program, the program leadership assumed the mentees would greatly benefit from the program, and the mentors would either marginally benefit from the program or feel it was an added time barrier. Instead, the combined average score from mentors in Table 1 is 4.68 out of 5, whereas that for mentees is 4.43 out of 5, indicating that the mentors valued the opportunity even more than the mentees. This is likely due to the protégé effect, where mentors are highly impacted because of their own role as a former protégé, in addition to the satisfaction gained via “learning by teaching” (Gates, 2019). The survey responses showed that 9% of the respondents met more than the required four times in a semester and, anecdotally, program leadership is aware of several mentor-mentee pairs who have become friends and continued to socialize after the formal mentor-mentee assignment ended.

Program Satisfaction Surveys - Three satisfaction surveys were conducted each semester: one for mentorship, one for the retreat, and one for all other programmatic events. Table 2 shows student satisfaction ratings of the mentorship experience. Surveys were deployed at the end of Fall 2024, Spring 2025, and Fall 2025 and the question asked was, “How valuable did you find the mentorship opportunity this semester?” The data are disaggregated by semester and whether the student's role was the mentor or mentee. Table 3 shows student satisfaction by event type. A question was asked regarding each individual event in Fall 2024, Spring 2025, and Fall 2025 in the format, “How effective did you find the [insert event name]?” The questions regarding the retreats were phrased as, “How much did you enjoy the retreat overall?” The different adjectives used in the questions (valuable for mentorship, effective for individual events, enjoyable for retreats) for the different activities may have affected student responses but were chosen due to the different natures of these activities. For example, the retreat was likely rated as highly “enjoyable” because it included an all-inclusive trip to the beautiful Rocky Mountains. Students participated in professional-development activities, but a large portion of the retreat was dedicated to socializing and community building. Students rated their answers on a 5-point Likert scale. The events were then grouped together by event type, and an average rating was taken for each event. There is high satisfaction shown for all LSAMP-BD activities (average scores of 4.42–4.80), not including the workshops (average score of 3.76), which represent the largest category and included six events ranging from DEI workshops to writing workshops to group interviews. For the workshops, a repeated qualitative comment from students is that they wanted workshops in which they were active participants, as many of the workshops primarily consisted of presentations.

Table 2. LSAMP mentorship satisfaction survey results – this table includes average responses to the survey question “how valuable did you find the mentorship opportunity this semester?” on a 5-point Likert scale. Mentee data includes all first-year LSAMP affiliates in the specified semester and mentor data includes 2nd and 3rd year LSAMP students who volunteered to participate as a mentor in the specified semester. Here, n_23, n₂₄ and n₂₅ are the numbers of students who entered in the fall 2023, fall 2024 and fall 2025 semesters, respectively, and participated in mentorship in the specified semester. The average ratings are listed respectively with n-values.

Table 3. LSAMP events and retreats satisfaction survey results – this table includes average responses to each specified question on a 5-pt Likert scale, disaggregated by cohort. Data includes ratings from the 2023, 2024, and 2025 cohorts, where n_23, n₂₄ and n₂₅ are the numbers of students who entered in the fall 2023, fall 2024 and fall 2025 semesters, respectively, and attended the rated event. Average ratings are listed respectively with n-values. F24 signifies an event which occurred in Fall 2024, S25 signifies an event which occurred in Spring 2025 and F25 signifies an event which occurred in Fall 2025. .

Community and Belonging - As mentioned in the literature review, a sense of belonging and community is a favorable indicator for STEM retention and persistence (Kuchynka et al., 2022; Liu et al., 2022; Markle et al., 2022; Ravishankar et al., 2024). Through Dr. Sarah Hug's surveys and focus groups, she found that the LSAMP-BD students at CU have felt a sense of belonging within the program in both years evaluated. Table 4 outlines her data. In her 2024 report, she found that 81% of participants agreed to the statement “I feel like I am part of the community.” Here, “agreed” represents the three highest choices on a 7-point Likert scale, namely “highly agreed”, “agreed” and “somewhat agreed”. She quotes one student saying, “I think that the most important thing that I got out of, first and second semester, and I think a lot of you would probably say the same thing, is the community was wonderful. All the community-building aspects brought a lot…it was very welcoming to have people who are in the same position as me. I don't have a lot of friends in the PhD program. And being able to share this experience was important to me.” In her 2025 report, she found that 93% of participants agreed to the statement “I feel like I am part of the community.” When asked why they joined the program, one of the 2024 affiliates said, “I was excited to build and foster a community of inclusive professionalism as we embarked on various activities that allowed me to learn and grow with others on things regarding my PhD and further down the line.”

Table 4. Community and belonging in LSAMP and CU – this table includes responses to surveys conducted by NSF external evaluator, Dr. Sarah Hug. Columns under “2024” include data collected from the 2023 cohort, presented in an April 2024 report. Columns under “2025” include data collected from the 2023 and 2024 cohort, presented in an April 2025 report. Data reported in the April 2025 report was not disaggregated by cohort. Data denoted with * are data from the 2021 CU Campus Culture survey, which asked the same questions for engineering and physics students. The significance value α for comparing the two populations for each question is also provided.

At the academic-major and university levels, participants showed a stronger sense of belonging in 2025 than in 2024. In Dr. Hug's 2024 report, 70% of participants agreed with the statement “I feel a sense of belonging to my academic major's community.” In 2025, for which the report included both the 2023 and 2024 cohorts but not disaggregated, 92% of participants agreed. In 2024, only 50% of students agreed with the statement “I feel a sense of belonging to my university.” In 2025, 93% of participants agreed. The increase in academic major belonging is not statistically significant (α=0.4), but the larger increase in university belonging is significant at approximately 95% confidence and could be attributed to various factors. One such factor is that, in 2025, some of the participants surveyed were second-year students, and the longevity of their stay at CU could have increased the feeling of belonging to their major and university as they found their place. Additionally, the program was simply larger in its second year, increasing opportunities for participants to find like-minded peers and make friends. Importantly, a key difference between LSAMP BD in its first and second years is that the structured near-peer-mentorship program was added in the second year. The addition of the affiliate class meant that the LSAMP-BD program brought a higher likelihood of there being another student in the same major within the program. Indeed, most of the mentor-mentee pairs were same-major pairings, which may have increased the sense of belonging in an academic major for mentors and mentees alike.

The above results can be compared to the most recent Campus Culture Survey, administered in 2021. Over 3,350 graduate students across the campus responded to the survey, including 1,100 from the College of Engineering and Applied Science (CEAS) and 175 students from the Department of Physics. In the Campus Culture Survey, 80% of CEAS students and 71% of physics students agreed with the statement, “I have a sense of community at CU” (including “strongly agree”, “agree” and “somewhat agree”). When disaggregated into racial/ethnic groups, 77% of CEAS URMs and 62% of physics URMs responded that they agreed with the same statement. Similarly, 80% of CEAS students and 77% of physics students agreed with the statement, “I have a sense of community in my department/graduate program.” 75% of CEAS URMs and 67% of physics URMs responded that they agreed with the same statement. Thus, a higher percentage of LSAMP program participants in 2025 expressed feelings of belonging than both the general CEAS and physics PhD population and the URM CEAS and physics PhD population in 2021. As noted in Table 4, the differences in the mean values between the 2025 LSAMP survey and the 2021 campus survey general population (weighted average of CEAS and physics) are significant at the 97% and 98.5% confidence levels for the sense of belonging to the academic major and university, respectively. Note that some lower campus-wide responses may have been impacted by isolation stemming from the COVID-19 pandemic. Note also that 10 of the 12 students in the first cohort and 7 of the 8 students in the second cohort were enrolled in CEAS, and the remaining students were enrolled in physics.

Skills Progression - A long-term goal of retaining diverse graduate students is to increase the number of diverse PhD graduates in the workforce, thus fostering innovation. It is important, then, that students graduate with the skillsets necessary for successful careers in engineering, computer science and physics fields. Dr. Hug surveyed how LSAMP BD at CU may have contributed to the development of these skillsets. Her data are outlined in Table 5. Some of these skillsets can be explicitly taught through a workshop such as applying for fellowships, identifying microaggressions, giving an elevator speech, and writing a research abstract. Indeed, since the inception of the program, at least one workshop has been dedicated to each of these topics.

Table 5. Skills progression through LSAMP – this table includes responses to surveys conducted by NSF external evaluator, Dr. Sarah Hug. Columns under “2024” include data collected from the 2023 cohort, presented in an April 2024 report. Columns under “2025” include data collected from the 2023 and 2024 cohort, presented in an April 2025 report. Data reported in the April 2025 report was not disaggregated by cohort.

Some skills, however, are taught more implicitly over time, and our team hypothesizes that a near-peer-mentorship relationship is an effective tool in this regard. Related skills statements include “I know how to manage my time effectively as a graduate student,” “I know how to balance work and life as a graduate student,” and “I know how to give helpful feedback to my peers.” On average, 84% of LSAMP-BD mentorship program participants discussed “Work/Life Balance” as a topic during their mentorship meetings during a semester and 63% of mentorship program participants discussed “Time Management” as a topic during their mentorship meetings in a semester. 79% of mentorship program participants discussed “Communicating with Professors” during their mentorship meetings, which ties back to a need discussed in the subsection on Mentorship in Action. Although some increases in these skillsets could be attributed to the 2nd-year students having spent more time as graduate students and thus learning these skills through experience, any 1st-year student that noted possessing these skills could have learned them from their mentorship relationship.

3.2 Summative outcomes

Enrollment and graduation data at CU are tracked by the Data Analytics Division of the Office of Information Technology. For students who entered PhD programs in physics, engineering and computer science at CU between 2008 and 2017, 65% graduated within 8 years. Only 53% of PhD URM students from these disciplines graduated within the same time frame. For this study, the year-to-year enrollment was analyzed, as the students in the LSAMP-BD program at CU are only in their first, second, or third years, and most PhD students in engineering, computer science and physics take four to eight years to complete their degrees.

Table 6 shows the year-to-year enrollment data for students entering PhD programs in engineering, computer science, or physics in fall 2023 or 2024. The LSAMP-BD program is compared to the general CU population, as well as the URM population at CU. A “Stayer” is a student who continues to be enrolled in a PhD program in engineering, computer science, or physics. A “Leaver” is a student who left the program or university for any reason, with or without an MS degree, but without a PhD degree. Of note is 11 of the 12 students in the original cohort were admitted directly into PhD programs and were all still enrolled at the time this report was written (their sixth semester) at CU. One student was admitted into an MS rather than PhD program, and this student received an MS degree and took a position in industry after two years. For simplicity's sake, this student is included in Table 6 as entering a PhD program, as there was intention to complete a PhD. None of the original 12 students had an MS degree when they enrolled at CU. For the second cohort, seven of the eight students were admitted directly into PhD programs, and one of them already had an MS degree. The eighth student was admitted into an MS program but has since transferred to a PhD program (with much help and advice for doing so from her peer mentor!). This student also is included in Table 6 as entering a PhD program, as they transferred to the PhD program within one semester of starting at CU.

Table 6. Year-to-year enrollment data in engineering, computer science and physics PhD programs at CU for 2023 and 2024 entering students – this table includes fall enrollment data collected from the office of data analytics at CU Boulder. LSAMP population data includes LSAMP fellows (2023 cohort) and affiliates (2024 cohort), the general population data includes students entering PhD programs in engineering, computer science, and physics in fall of the specified year, and the URM population data includes Black, African American, Hispanic, Latino American, Native American, Alaska Native, Native Hawaiian, and Pacific Islander students entering PhD programs in engineering, computer science, and physics at CU in the fall of the specified year. A “Stayer” is a student who continues to be enrolled in a PhD program in engineering, computer science, or physics while a “Leaver” is a student who left the program or university for any reason, with or without an MS degree, but without a PhD degree. LSAMP students were included in both the general population and URM population numbers.

Students in the LSAMP-BD program were retained at a higher rate (100%) after one year in the PhD program than the general population (96.5%) or the URM population (95.5%). A Fisher's Exact test was conducted to compare each iteration of the data (2023 LSAMP versus 2023 general population, 2024 LSAMP vs. 2024 URM population, etc.). None of these differences is statistically significant at greater than 60% confidence, primarily due to the small population size of the LSAMP-BD program.

Table 7 compares LSAMP-BD attrition data with national attrition data collected by the Council of Graduate Schools (CGS, 2007). Data for the PhD Completion Project were collected from 29 universities across the US and Canada on students entering PhD programs from 1992 to 1995. The attrition rates listed are for degree programs most similar to those in LSAMP BD. Engineering includes biomedical engineering, chemical engineering, civil engineering, electrical engineering and mechanical engineering. Mathematics and physical sciences include chemistry, computer and information sciences, mathematics, physics and astronomy. The attrition rates for Year 2 are cumulative rates.

Table 7. Year-to-year attrition data in engineering, computer science and physics phD programs in LSAMP-BD at CU and nationally – this table includes national attrition data collected by the Council of Graduate Schools (CGS) through the PhD cCompletion Project from 29 universities across the US and Canada on students entering PhD programs from 1992 to 1995. Attrition rates listed are for degree programs most similar to those in LSAMP BD. Engineering includes biomedical engineering, chemical engineering, civil engineering, electrical engineering and mechanical engineering. Mathematics and physical sciences include chemistry, computer and information sciences, mathematics, physics and astronomy. The attrition rates for Year 2 are cumulative rates.

Students in the LSAMP-BD program left their PhD programs at a lower rate (0.0% after one year and 8.3% after two years) than national engineering averages (5.5% after one year and 14.5% after two years). The national averages for mathematics and physical sciences, which include computer science and physics, have even higher attrition rates. The differences between attrition rates after one or two years for LSAMP and national engineering data are not significant at greater than 30% confidence, due to the small sample size. Similarly, the differences between attrition rates after one or two years for LSAMP and national mathematics and physical sciences are not significant at greater than 40% confidence, due to the small sample size. One possible source of error in comparing these data sets is the difference in entering years. Graduation rates for PhD programs have increased in the last 10 years (Lanier et al., 2024). Thus, comparing 1992–1995 national data with 2023–2024 LSAMP-BD data may introduce errors in comparison. More recent longitudinal research from Ohio demonstrates that attrition from STEM doctoral programs occurs disproportionately during the first several years of enrollment (Bostwick and Weinberg, 2022), but comparable contemporary national data reporting cumulative first- and second-year attrition by STEM discipline were not identified. Nevertheless, program leadership is encouraged to see that the LSAMP-BD students, which arguably represent an at-risk population, have exhibited early retention data that are comparable to or above national averages.

4 Discussion

4.1 Programmatic changes over time

LSAMP BD at CU strives for continuous improvement, which is why satisfaction surveys, individual check-ins, and group interviews have been deployed. It is important to note that the data in Tables 2–5 were collected after at least the first semester of the program, and that the results show changes in averages of the satisfaction ratings over the course of the program. Thus, programmatic events and operations are ever-changing in response to student feedback, and satisfaction has not always been so high. The most pertinent feedback from the feedback mechanisms above and their subsequent changes are outlined in Table 8. Key takeaways include that the number of group events should be about one per month (representing a balance between building community and the academic demands on the students) and that they should be interactive in nature versus didactic presentations.

Table 8. Changes made in response to student feedback – this table includes changes that were implemented over time as results from either ongoing or one-time feedback, as specified in the evidence column.

4.2 Relationship to other LSAMP programs and prior studies

The Texas A&M University System (TAMUS) was awarded LSAMP-BD funding for a total of 13 two-year awards from 2003 to 2020. Through continuous improvement and refinement, the TAMUS program developed a streamlined, replicable and participant-informed model for LSAMP-BD institutions across the US (Preuss et al., 2020). The program served over 90 underrepresented graduate students studying over 32 distinct STEM disciplines. The primary component of TAMUS programming included 12–15 seminars per year on topics ranging from stress management to scientific writing. Although skills development was the program's main focus, it also included a near-peer mentorship program where former LSAMP-BD fellows who were still enrolled in TAMUS mentored incoming BD fellows, similar to the program discussed in this paper but without quantitative assessment reported for the mentoring program. Details on the mentorship program structure and requirements are not included in public reporting. While not tabulating year-to-year retention that can be compared to the current work, the TAMUS LSAMP-BD program reports that 41 of 72 (57%) of its fellows entering during 2003–2011 had completed PhD degrees by 2020, that 18 of 36 (50%) entering during 2012–2014 had PhD degrees completed or still in progress by 2020, and that “nearly all of the students in the most recent cohorts are continuing study.” NSF final reports of other LSAMP-BD programs have shown student enrollment in a PhD program after the two years of fellowship support ranging from less than half to all the students in a cohort (e.g., Gillman and Read de Alaniz, 2021; Hammersley et al., 2020; Hammersley et al., 2022; Harris et al., 2023 Hudson et al., 2022; Kirksey and Irvin, 2020; Pastides et al., 2023). The holistic model proposed by North Carolina Agricultural and Technical State University is also based on skills development, using a framework of social cognitive career theory (Gloster et al., 2023), but this program was in the initial stages at the time of the report, without outcomes evaluation. The LSAMP-BD program reported in the current work uses a model of community and belonging, with skills development a smaller but important component. With the end of LSAMP-BD funding from NSF in 2025, however, there is a need for new models that leverage funding from alternate sources

Markle et al. (2022). gave nine recommendations for supporting historically underrepresented groups in STEM higher education through mentorship. One such recommendation emphasized the importance of academic institutions partnering with LSAMP programs to “develop successful mentoring approaches on a larger scale,” as LSAMP programs often only have funds to support a small cohort each year. While near-peer-mentorship programs are commonly deployed and show promising results for retention at the undergraduate level, they are not as often deployed or studied at the graduate level. This paper responds directly to the above needs by demonstrating how supplemental funding and program design enabled CU to expand its cohort-based community and mentoring impact beyond the initial NSF-funded LSAMP-BD cohort comprised of 12 students. Based on student feedback, LSAMP-BD at CU implemented a scalable, pay-it-forward near-peer-mentorship program in which a cohort of affiliated PhD students was added to the program each year, with program support funded by a supplemental Impact Grant from CU. The cost of adding these students each year is relatively small, as events for the original fellows were planned regardless, and the supplemental stipends that are provided for mentorship are small, yet meaningful to graduate students (the much larger costs of tuition and living expenses of the affiliates are covered through standard research and teaching assistantships from their home departments).

4.3 Recommendations for model replication and future research

The authors of the current study expect the model presented to be replicable at other LSAMP-BD institutions or any PhD institution that implements cohort-type programs for graduate-student success. The model is reiterated in the recommendations below: Since research consistently shows a strong link between mentorship and retention (Lorenzetti et al., 2019; Swift, 2018), this model has broad implications for supporting diverse graduate students across STEM disciplines. LSAMP-BD leadership at CU hopes to continue this model for several years, even without continued NSF LSAMP-BD funding. As such, additional data will be collected for participant satisfaction, enrollment and degree completion, allowing for additional program refinement and analysis. Finally, PhD institutions should pool more recent attrition and completion data to analyze trends and share best practices, particularly for retaining diverse PhD students, so that they can foster innovation and growth (Daily and Eugene, 2013; Halse and Mowbray, 2011; Swift, 2018).

• Using LSAMP-BD (or another mechanism) funding as leverage, seek supplemental institutional support for programmatic activities that build community. Even without outside funding, consider departmental funding and/or campus-wide funding of these activities.
• Invite incoming diverse students as affiliates in the second and subsequent years of the program. The number of new students should not supersede the number of current students each year, as each incoming affiliate will need a mentor.
• Pair similar-discipline, second-year students (mentors) with first-year students (mentees). Provide mentorship training and incentivize meetings with dining-hall meal cards and supplemental stipends upon completion of meetings.
• Provide programming throughout the year that supports student success in graduate programs: workshops and networking events for professional development, outreach events for STEM-identity reinforcement and encouragement of goodwill, and retreats and social events for community building and belonging. Events should be highly participatory for students, not didactic.
• Regularly seek and implement student feedback and improve/repeat in subsequent years.

5 Conclusions

The survey results for events and retreats (Table 3) show that student satisfaction is highest for the retreats, followed by outreach, orientation/planning, networking, and finally workshops well behind the others. These results support the hypothesis (made at the end of the first semester based on informal student feedback, before any surveys were done) that community building and relationships would be valued more than workshops by the students. The results have also been used for continuous program improvement and identification of best practices. The retreats provide the greatest and most extended opportunities for building community and relationships (see retreat agendas in the supplemental material). The outreach events are planned and put on by the LSAMP-BD students, and so also have a high level of community building. The orientation and planning meetings provide the students with the opportunity to create a shared vision for the program, thus building ownership/belonging among the participants. The networking events are also very relational, but they usually involve short-term interactions between students and academic or industry professionals at a single event and not the establishment of longer-term, peer relationships. Although the workshops have received lower student satisfaction ratings, their content is associated with the progression in student skills evident in Table 5.

The survey results for the near-peer-mentorship program (Table 2) also show high satisfaction by the mentees and especially the mentors. Near-peer mentoring has further supported the building of relationships within the LSAMP-BD community and may have contributed to the enhanced sense of belonging, both over time (Table 4) and relative to campus-wide results (subsection on Community and Belonging in the Results section). Although the cohorts in the current study are small and in the early stages of PhD studies, their enrollment data relative to campus (Table 6) and national (Table 7) results show promise for the efficacy of cohort-based and near-peer-mentorship programs at the graduate level in relation to community building and retention of diverse students.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/feduc.2026.1799508/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Alexander Gates, Rutgers University, Newark, United States

Reviewed by: Ken Teter, University of Central Florida, United States

Kathleen Grogan, University of Cincinnati, United States