Online mentoring for girls in secondary education to increase participation rates of women in STEM: A long‐term follow‐up study on later university major and career choices

An important first step in talent development in science, technology, engineering, and mathematics (STEM) is getting individuals excited about STEM. Females, in particular, are underrepresented in many STEM fields. Since girls’ interest in STEM declines in adolescence, interventions should begin in secondary education at the latest. One appropriate intervention is (online) mentoring. Although its short‐term effectiveness has been demonstrated for proximal outcomes during secondary education (e.g., positive changes in elective intentions in STEM), studies of the long‐term effectiveness of STEM mentoring provided during secondary education—especially for real‐life choices of university STEM majors and professions—are lacking. In our study, we examine females’ real‐life decisions about university majors and entering professions made years after they had participated in an online mentoring program (CyberMentor) during secondary education. The program's proximal positive influence on girls’ elective intentions in STEM and certainty about career plans during secondary education had previously been demonstrated in several studies with pre–post‐test waitlist control group designs. Specifically, we compared the choices that former mentees (n = 410) made about university majors and entering professions several years after program participation with (1) females of their age cohort and (2) females of a group of girls comparably interested in STEM who had signed up for the program but then not participated (n = 71). Further, we examined the explanatory contribution to these later career‐path‐relevant, real‐life choices based on (1) mentees’ baseline conditions prior to entering the program (e.g., elective intentions in STEM), (2) successful 1‐year program participation, and (3) multiyear program participation. Findings indicate positive long‐term effects of the program in all areas investigated.


INTRODUCTION
To meet the challenges of our time-the COVID-19 pandemic or climate change, for example-and realize positive visions for a common future, outstanding scientists are crucial. Talent development toward eminence and innovation has been the subject of research for decades.
Bloom's seminal interview study of 120 individuals who had achieved world-class levels of performance in various domains has made an important contribution to a better understanding of what eminence and innovation entail. 1 He identified three stages of talent development. Stage 1 is about interest development. Individuals fall in love with a subject, an idea, or a discipline. Stage 2 is about skill acquisition. Individuals develop technical mastery through deliberate practice. continue to play an important role in practice. 3 In science, technology, engineering, and mathematics (STEM), the first stage of talent development poses a particular problem. This can, for example, be seen in the chronic shortages of skilled professionals. 4 Attracting talented girls and women to STEM has proven particularly challenging. Although the situation at this initial stage of talent development has improved in recent years, women are still less likely to opt for STEM majors and professions in many countries, especially in disciplines such as computer science and engineering. [5][6][7] For this reason, extensive efforts have been made in recent decades to improve the situation. In this context, it has proven important to start interventions as early as possible, at the latest early on in secondary education, as girls' interest in STEM subjects declines substantially during adolescence, 8 and decisions related to future career choices are made during this period. 9,10 As crucial as it is that interventions to attract talented girls to STEM careers start early, it is difficult to verify the long-term effectiveness of such measures. Definitive real-life choices for or against long-term engagement with STEM domains-in the sense of decisions about university majors and about careers-often take place years after an intervention targeting girls. Research studies often rely on self-reports completed after early interventions, for example, about girls' elective intentions or certainty about future career plans, and assume that any such proximal improvements in STEM-related outlooks or preferences presage definitive later choices about university majors and about careers. In more rigorous studies, researchers specify developmental trajectories based on participants' self-reports made before, during, and after an intervention and compare these with responses provided by suitable control groups (i.e., groups of individuals with similar initial characteristics but who did not receive the intervention). 11 However, follow-up studies on later real-life choices are hard to find. 12,13 The aim of our study was, therefore, to investigate whether a measure that has proven effective for getting girls interested in STEM subjects early on-namely, online mentoring being offered to girls enrolled in secondary education-also increases the rates at which par-ticipants make STEM career choices after high school (i.e., majoring in a STEM subject at university and/or entering a STEM profession).
We examined this for CyberMentor, a Germany-wide online mentoring program for girls enrolled in university-track secondary education.
CyberMentor has been scientifically evaluated and extensively studied for nearly two decades in terms of its short-term effectiveness and determinants of success. [14][15][16] Attracting girls in secondary education to STEM via mentoring Mentoring is a relatively stable dyadic relationship between one or more experienced individuals (mentors) and one or more less experienced individuals (mentees) characterized by mutual trust, goodwill, and the shared goal of mentees' advancement and growth. 17 Mentoring is important for youth career guidance 13 and for Bloom's Stage 1 of talent development 1 in three ways. First, it provides mentees with a connection to adults who probe and cultivate mentees' interests and help them achieve their goals. Second, mentors serve as role models who share their own experiences in identifying (career) interests and their career paths with their mentees. For career guidance and the promotion of girls' STEM interests, female role models who are themselves studying a STEM subject or working professionally in a STEM field are particularly effective. 18 Third, mentors act as advocates for their mentees, giving them access to other individuals and institutions to explore and deepen their career interests. 19 Online mentoring is a specific form of mentoring in which inter- In the context of STEM promotion for girls, online mentoring has proven to be particularly successful for at least four reasons. First, online mentoring makes it easier to find suitable female role models as mentors. The rationale is that while women who are themselves studying a STEM subject or working in a STEM field act as particularly suitable role models (e.g., Ref. 18), it is often difficult to find them in the mentees' immediate vicinity due to the low participation rates of women in STEM. Online mentoring facilitates matching mentees with suitable mentors due to its spatial and temporal flexibility. Moreover, the online format makes frequent communication (ideally on a weekly basis) easier, which is essential for successful mentoring. 22 Second, by using an appropriate platform, online mentoring enables mentees to network with other females interested in STEM. Networking with other mentors prevents counterproductive subtyping processes, 23 that is, girls recognize that their individual mentors are not an exception (i.e., a subtype), but that their mentors are among many women successful in STEM. This can help to reduce the stereotype that STEM is a typically male domain. [24][25][26] In addition, by networking with other female mentors, mentees learn about different STEM career paths and discuss a wide variety of STEM-related topics, thereby deepening their interests. Networking with other mentees in an online mentoring program can make mentees aware that many other girls are also interested in STEM, which is usually not the case in their immediate social environments. 15 Moreover, the other mentees in a program provide same-age role models. Same-age role models have been shown to be particularly effective for changing girls' perceptions of STEM subjects as unfeminine. 27 Furthermore, studies indicate that peer support plays an important role in girls' willingness to stay in STEM fields. 28 Third, online mentoring enables the establishment of an optimal learning environment for girls in STEM. 17 For example, it is possible to host topic chats on STEM or provide low-threshold access to STEM lectures, discussions, and Q-and-A sessions. Mentees and mentors can use collaboration tools to work jointly on STEM projects or manuscripts, or, with proper facilitation, to organize their own symposia. 3 Finally, some aspects that predict successful mentoring 29 are particularly well realized in online mentoring. For example, it is more feasible for employed mentors to participate in training sessions essential to mentoring success before and during a mentoring program 22 if these are offered via instructional videos or online. 30 The continuous support of participants by trained program staff (e.g., regular check-ins with mentees and mentors) is an important criterion for successful mentoring 31 and easier to implement online.

Research on mentoring during secondary education and females' later STEM-related choices
There is evidence that mentors have an influence on mentees' career choices in STEM (i.e., majoring in a STEM subject at university or entering a STEM profession). For example, in a retrospective survey of 1425 female graduates of selective science, high schools in the United States 32 found that having a teacher as a mentor during high school correlated with university STEM major choices and degrees in STEM.
However, studies examining associations between participation in formal STEM mentoring programs during secondary education and later career choices are lacking. 13,33 To date, evaluation studies of such programs have mainly examined more proximal program effects on precursors of later choices of STEM majors or careers (e.g., elective intentions in STEM, certainty about career plans, or career interests; for an overview, see Ref. 13). In the following, we describe the results of two evaluation studies reporting proximal beneficial effects of STEM mentoring programs offered during secondary education.
In their evaluation study of the Spanish Inspira STEAM program, which aims to increase girls' participation in STEM, 34  The results of these studies are encouraging. However, they focus on the early prerequisites of later career choices. Whether these promising changes in elective intentions in STEM, in certainty about career plans, and for other relevant constructs-observed while girls were in secondary education and participating in online and in-person mentoring programs-actually lead to real-life decisions, often made years later, to major in a STEM subject at university and/or to choose a STEM career has not yet been investigated. 13,33 Although studies with adults suggest that online and in-person mentoring do indeed influence career-path-relevant, real-life choices, 36,37 in these studies, the career choices occurred immediately or very shortly after participation. In the case of online and in-person mentor-

Research questions
In our study, we, therefore, investigated whether girls who participated in the online mentoring program CyberMentor for at least 1 year during secondary education were more likely to make a STEM career choice (majoring in a STEM subject or entering a STEM profession) later on. To do this, we compared the STEM career choices of former mentees with different control groups. We formulated three research questions. In the following, we state each research question and provide additional rationale for Research Questions 2 and 3.
Research Question 1: A few years after participating in the program, are former CyberMentor participants significantly more likely to make a career choice in STEM than women of the same age cohort?
Girls who enroll in longer-term STEM programs, such as CyberMentor, differ from girls who do not enroll in such programs on several characteristics. 15,38 For example, they exhibit significantly greater interest in STEM and have better grades in STEM subjects. Therefore, to determine whether subsequent career choices can actually be attributed to program participation, it is not sufficient to compare the proportion of participants' career choices to those of the same age cohort. Rather, it is important to compare the proportion of participants' career choices with those of comparable nonparticipating females of the same age who were similarly interested in STEM at the time of program participation. Only in this way can it be determined whether participation in the mentoring program is (partly) responsible for subsequent career choices in STEM, or whether the above-average interest would have led such girls to make these choices even without having participated in such a mentoring program.
Research Question 2: A few years after participating in the program, are former CyberMentor participants significantly more likely to make a STEM career choice than females who had also originally enrolled in the program (i.e., girls of the same age who were similarly interested in STEM at the time) but then did not participate?
Not only does STEM interest differ among girls who enroll in long-term STEM programs. Often, these girls are characterized by greater elective intentions in STEM and less certainty about career plans before program participation. 15

CyberMentor as a research setting
CyberMentor is a Germany-wide online mentoring program founded in 2005. Its goal is to inspire girls for STEM and to contribute to an increased participation of women in STEM in the long term. The program takes place on a members-only online platform that was planned and programmed by the project team based on the mentoring goals and the underlying mentoring concept. Mentees are enrolled in grades 5−13 of university-track secondary education in Germany. a Each mentee is mentored for at least 1 year by a mentor, a woman who is majoring in a STEM subject or working in a STEM field. Up to 800 mentees and up to 800 mentors participate in the program annually.
The mentees and mentors communicate with one another on a weekly basis for at least half an hour via emails, instant messages, and forum posts. The program is free of charge for the students, and the mentors volunteer their time.
In CyberMentor, dyads are matched based on the mentees' STEM interests and the mentors' STEM fields, as well as shared personal interests. Mentors act as successful STEM role models, provide insight into their careers and everyday work, discuss STEM-related topics as well as personal issues, and provide support for mentees' STEM projects. To prevent subtyping processes and to provide both role models that are professionally successful in STEM (mentors) and same-age role models who are also interested in STEM (mentees), the platform offers extensive networking opportunities with other mentors and mentees.
The mentoring year is divided into four phases of equal length. In the first quarter, the focus is on getting to know one another and learning more about STEM majors and professions. In addition, mentees and mentors jointly investigate where STEM plays a role in everyday life.
The results of these discussions are made available to the entire online mentoring community in STEM wikis written to stimulate platformwide discussions about STEM in everyday contexts and illustrate that STEM plays a role in diverse everyday settings that are often not apparent at first glance. In the second quarter, two mentoring dyads with similar STEM interests (e.g., computer science) collaborate on a project in their STEM field (e.g., programming an app). In the third quarter, several mentoring dyads from different STEM fields work together on an interdisciplinary project (e.g., researching and writing a plan for surviving on Mars). The last quarter is dedicated to a review of and reflection on the first three quarters of the mentoring year. For example, participants write articles for the monthly program magazine, CyberNews, and present their most interesting projects from the mentoring year.
a In most German federal states, secondary education starts in fifth grade and runs through twelfth and in some federal states the thirteenth grade. For Research Question 2, we compared the percentage of the 410 former mentees who had later made a STEM career choice with the percentage of STEM career choices in the group of the 71 women who had applied for the program during secondary school but never participated. As the two groups were not comparable concerning age and STEM interest when entering the program, propensity score matching was used to make the groups as comparable as possible (described in detail in the section on data analysis), which resulted in a matched sample of 265 former mentees who were then compared to the 71 women who had not participated in the program but had originally applied for participation. b Although our response rate is appropriate for our sample size, 53 selection bias could still be a problem. Therefore, we analyzed differences between respondents and nonrespondents for age, elective intentions in STEM, and certainty about career plans at the beginning of the mentoring period. While there were significant differences, the differences were small. Cohen's ds were 0.13 for both age (respondents were slightly older) and certainty about career plans (lower values for respondents) and 0.27 for elective intentions in STEM (higher values for respondents), which indicates low selection bias. c While some of our former mentees reported only their recent job in the follow-up, it is often possible to deduce what they studied at university. When analyzing a reduced sample of former mentees who were still studying, the differences compared to the age cohort were even slightly larger.

STEM interests
Participants indicated with "yes" or "no" whether they were interested in each of the following six areas: mathematics, computer science, biology, chemistry, physics, and technology. It was possible to select "yes" or "no" more than once.

Length of program participation
We counted the number of years the mentees participated in the program.

STEM career choice
Depending on which applied to them, participants either indicated which major they were currently studying at university, or they indicated their current profession via an open-response item. Using the classification of the National Pact for Women in STEM Professions, 39 we coded whether their major or profession was in a STEM field (1) or not (0).

STEMM career choice
We additionally coded a broader STEM variable, called STEMM. It refers to science, technology, engineering, mathematics, and also medical sciences.

Computer science and engineering career choice
As the participation of women in STEM and STEMM in Germany is not generally low in all domains but is especially low in computer science and engineering, we also coded choices in these fields separately.

Data analysis
For answering Research Question 1, simple descriptive statistics were used. For properly assessing the treatment effect for Research Question 2, we first had to check the balance of our two groups concerning relevant pretreatment covariates. We considered age, the CyberMentor cohort (cohort 1, 2, or 3), and the pretreatment STEM interests from the application questionnaire described in the methods section (i.e., in mathematics, computer science, biology, chemistry, physics, and technology) as relevant variables. Information about these variables was available for all participants as the corresponding questions were part of the program application.
There was a significant age difference as well as marginally significant differences concerning interest in chemistry and interest in technology. However, even nonsignificant differences should be reduced as much as possible. 40 Therefore, we used propensity score matching based on all noted pretreatment variables using the program PS Matching 3.0.4 to achieve balance concerning these variables. 41 More specifically, we used nearest-neighbor matching without replacement and a ratio of 1:4. The 1:4 ratio means that we found four matches for each person in our control group, drawing from the larger treatment group. There are many possible procedures for matching, but as the goal of each matching procedure is to achieve balance, a procedure that achieves balance can be deemed suitable. 40 Our analyses for Research Question 3 are based on the latent growth-curve approach. In the latent growth-curve approach-which is situated in the framework of structural equation modeling 42

Estimation of the models
The analyses were conducted with Mplus 8. 43

Research Question 1
As can be seen in Table 1

Research Question 2
As can be seen in Table 2, in the treatment and the control groups, similar means and standard deviations concerning age and STEM interests were achieved using the nearest-neighbor propensity score matching procedure described in the prior section on data analyses. The matching procedure also resulted in a similar composition of the dif-

Research Question 3
Next, we tested within our sample of former mentees which variables predict later STEMM career choices. The structural equation models showed very good model fit according to every index we examined (see Table 3).
In  Note. Beta indicates the standardized probit regression weights. LL and UL indicate the lower and upper limits of the 95% confidence interval, respectively.
The predictors explained 16.4% of the variance in the outcome, and the baseline value of elective intentions in STEM was clearly the best predictor (see Table 4, step 1).
In step 2 (Model 2), we examined how much the prediction improved when adding the changes (slopes) in elective intentions in STEM and certainty about career plans during the first mentoring year as further predictors. The explained variance almost doubled to 31.0%, and both slope variables significantly contributed to this increase (see Table 4, In the next step, we went beyond predictors from the first mentoring year and added the overall length of program participation (the overall number of years the mentees participated in the program) as another predictor. While the overall length of program participation was a significant predictor, it only improved the explained amount of variance to 32.6% (Model 3, see Table 4, step 3). Women's overall participation rates-at all stages of talent development and all levels of seniority-are significantly lower than men's. [5][6][7] This situation has led to extensive attempts in recent decades to inspire females to pursue STEM and to increase females' participation in STEM majors and careers. Because girls' interest in STEM declines markedly during adolescence 8  With our study, we endeavored to make a contribution to addressing this research gap. Specifically, we investigated the real-life choices of STEM majors and careers made by women who had-years before while enrolled in secondary education-participated in the Germanywide online mentoring program CyberMentor. We chose this program because online mentoring is a particularly promising intervention for engaging girls in STEM. [14][15][16] Furthermore, the short-term effectiveness of CyberMentor has been demonstrated in numerous evaluation studies that meet quality standards. 33 Girls who participated in the program had more positive developmental trajectories in terms of elective intentions in STEM and certainty about career plans than did girls in waitlist control groups who were comparably interested in STEM. 15 Thus, the program was particularly well suited to investigating the significance of successful STEM promotion provided to girls during secondary education for their later real-life STEM choices.

DISCUSSION
In a first step, we investigated whether years after program participation of former CyberMentor participants were more likely to choose STEM majors and professions than females of the same age cohort. The results of these analyses were encouraging. Former female CyberMentor participants were more than twice as likely to choose STEM majors (51.2% vs. 23.9%). When medical science-a STEM field in which women are actually overrepresented in many countries-was included in the analysis, former mentees were still almost twice as likely to choose STEM majors as females in their age cohort (61.7% vs. 31.1%). Interestingly, however, this means the ratio between the groups became somewhat smaller when medical science was included.
The real-life choices for computer science and engineering, that is, the two STEM fields that are by far the least likely to be chosen by females in Germany, 45 confirm this tendency of higher ratios in fields with fewer women. While only 10.6% of females in the age cohort chose these fields, more than twice as many of CyberMentor's former mentees did, namely, 24.9%. Thus, these findings suggest that there was an even stronger increase in percentages in STEM fields in which there are few females. One reason for this could be that a disproportionately high number of mentors in the CyberMentor program came from these fields. 51 As CyberMentor not only offers one-on-one mentoring, but also facilitates networking with numerous other mentors and mentees on the platform, it is conceivable that the particularly large number of role models from the computer science and engineering fields were partly responsible for the frequent choices of these STEM fields that former participants went on to make after high school. Our study shows that online mentoring can make an important contribution to Stage 1 talent development according to Bloom 1,2especially when it comes to inspiring females to enter a STEM field and motivate them to stay in the STEM talent development pipeline.
To our knowledge, this is the first study to systematically examine whether successful participation in a mentoring program offered to girls enrolled in secondary education can positively influence real-life STEM choices made years after participation. While studies have examined the impact of (online) mentoring on real-life choices of STEM majors and careers, 32 they were either retrospective surveys that did not address formal (online) mentoring programs or were studies with adults. 36,37 Studies with students in secondary education have so far primarily examined the short-term effectiveness of formal (online) mentoring programs (e.g., influences on STEM interests and elective intentions in STEM). 13 Moreover, many of these studies evince methodological shortcomings (e.g., inappropriate or wholly lacking control groups, reliance on postprogram satisfaction surveys), making even conclusions about the short-term effectiveness of the programs difficult in some cases. 13 Our evaluation study, which included appropriately designed randomized waitlist control groups and followed up with former participants years later, provides preliminary evidence that a formal online mentoring program that had been demonstrated to be effective in the short term did indeed also contribute to females' later real-life STEM choices and, therefore, also possesses long-term effectiveness. However, our study also has its own limitations. These should be considered when interpreting our findings and can inform future research.

Limitations and future research directions
A first limitation is the generalizability of our results. CyberMentor is a Germany-wide online mentoring program that has been continuously optimized and adapted to the needs of participants via an extensive program of accompanying research. 52   practice, important quality standards of mentoring programs often go unaddressed or are insufficiently attended to. 11,55,56 In the case of mentoring programs in secondary education aiming to increase the labor-force participation rates of females and other underrepresented groups for the long term, a crucial weakness has been a lack of evaluation studies that considered the long-term effects of such programs to understand whether and, if so, to which extent such programs are effective for achieving such long-term goals. This

ACKNOWLEDGMENTS
We thank Dr. Daniel Balestrini for his feedback on our manuscript and assistance with language revisions and Manuel Hopp for his assistance with data collection.
Open access funding enabled and organized by Projekt DEAL.