Research has found that access to K-12 CS learning opportunities is unequally available in the US by race, ethnicity, socio-economic status, and urbanicity (i.e., population density) [16]. These patterns of disparate access mirror patterns of underrepresentation of racial/ethnic groups in CS majors and CS careers [47]. Equitable access to CS learning opportunities is imperative because CS is a relatively new eld [20,21,35] with an enormous impact that is not equitable across society [10,22,32]. Students could benet from CS learning opportunities to understand and shape the computational world around them [31].
Scholars have sought to understand K-12 CS inequity using the CAPE Framework [23,46]. The CAPE Framework focuses on how inequality can be present in the Capacity of a district to oer CS, the Access students have to CS courses, the Participation of students in CS courses, and the Experiences of students in CS courses. Our research focuses on CS teachers as they are central to understanding equity and inequity in Capacity, Access, Participation, and Experiences. CS teacher isolation can limit the Access to CS and negatively impact the Experience of both students and teachers. We focus specically on the reported problem of CS teacher isolation (i.e., being the only CS teacher at their school) [15] due to its importance in improving CS education.
In the United States, isolation was identied as one of the ve main challenges for CS teachers, with 55% of CS teachers reporting being the only CS teacher at their school in 2013 [15]. However, survey self-reports may be inaccurate or non-representative and therefore may not reect the true rates of CS teacher isolation at the time of that study. Additionally, current rates of CS teacher isolation and how they have changed as CS education has expanded in the subsequent decade are unknown. Accurately measuring rates of CS teacher isolation is important because teacher isolation can negatively impact teachers and their students. In contrast to isolated CS teachers, non-isolated CS teachers have the potential to collaborate, and the resulting teacher social networks can benet students' exam scores [39,41]. Having opportunities to collaborate with and learn from other CS teachers at their school may address some of the insucient training for CS teachers [18], many of whom are credentialed in other disciplines [12]. Additionally, non-isolated CS teachers indicate a larger capacity to allow more students to participate in CS coursework.
Our work is inspired by analyses from physics education because K-12 physics education suers from similar problems of inequitable access [28] and additional challenges caused by isolated teachers. In particular, Kelly and Sheppard [29] found that small schools of a few hundred students were less likely to have physics courses. These small schools were created as part of the Small Schools Movement and were meant to improve the experience of students from underrepresented groups. Previous work has focused on relationships between student demographics and CS access [44,45] but has not considered school size. Some districts in California participated in the Small Schools Movement as well [14], which created the possibility that disparities in access to CS could be attributed to more students in underrepresented groups being in these smaller schools. Ultimately, larger schools may have more capacity to offer CS courses or other electives due to being able to hire more teachers.
Our work explores how school characteristics, such as school size and urbanicity, impact a school's capacity to oer CS. We also investigate whether these school characteristics impact whether the school has an isolated CS teacher, further exacerbating issues of equity and access. We use data from California because we can access fourteen years of teacher, course, and student data from over 1,000 districts, over 10,000 schools, and over six million students. Additionally, dierent regions in California vary in their demographics, nancial resources, and population density.
Motivated by the goal of understanding and addressing current patterns of inequity, our research extends prior work to document the current state of CS teacher isolation and takes into account the impact of school size to address the following research questions:
RQ1: How do rates of CS teacher isolation vary over time and compare to other subjects? RQ2: To what extent do school size, urbanicity, or student demographics aect the likelihood that a school:
• (a) does not oer CS? • (b) has only a single (i.e., isolated) CS teacher? • (c) has more than one (i.e., non-isolated) CS teacher?
We nd that teachers in CS experience isolation at a higher rate than almost all other subjects. We provide a more comprehensive view of teacher isolation than is possible from survey data. We also illustrate dierences between middle school and high school teachers. While prior research has documented teacher isolation in other subjects in other states [34], these rates of isolation had previously not been demonstrated in CS. Our ndings illustrate that despite recent improvements, California shows a high rate of CS teacher isolation. Based upon previous work highlighting the benets of social networks [7,38] and the negative impacts of teacher isolation [8], supporting CS teachers who experience isolation may improve their work experience and lead to better student outcomes. Our ndings support previous work documenting disparities in CS education access and additionally demonstrate that these access disparities persist even when accounting for school size. We suggest that when expanding the number of CS teachers in a school or district is not possible, support could be oered through targeted professional development funding and additional resources.
Prior research has looked at who is taking computer science at a high school level and have highlighted disparities in access for Black and Hispanic students when compared to white students [44,45]. There is further evidence of disparities in terms of access within underrepresented populations when considering intersectionality [45]. Additionally, there is research that has looked to determine school characteristics that might aid in expanding access or help further explain disparities, such as course modality and urbanicity [17]. However, there is no research to the authors' knowledge that considers access to CS courses while accounting for student demographics, school size, and urbanicity simultaneously.
We use the CAPE Framework [23] to help us understand factors that impact equality in K-12 CS education. Notably, the CAPE framework expands beyond typical measures of Access and textit-Participation to consider the underlying Capacity within educational systems to teach CS as well as student Experience . Some of the components of capacity are having the resources, teachers, and the facilities to teach CS. Teacher isolation both aects the access to CS, as one CS teacher may not be enough to provide all students in a school the opportunity to take a course, and experience, as it can aect both teacher and subsequently student experience.
Teacher isolation has the potential to negatively impact teachers in a variety of ways. Researchers have classied teacher isolation along dierent dimensions: geographic isolation (i.e., how geographically distant teachers are from other teachers or resources), intellectual isolation (i.e., how there is a lack of access to teachers they could collaborate with), and social isolation (i.e., how distant they feel to the community outside of work) [4]. Prior work has focused primarily on geographic and intellectual isolation as it has the potential to cause teachers to burn out, leave their current school, or exit the profession entirely [4,8,9]. Geographic and intellectual isolation are also the types of teacher isolation where policymakers have the greatest opportunity for impact; social isolation is more dicult to legislate solutions for. This paper focuses on intellectual isolation, particularly on lacking access to teachers to collaborate with on the same subject.
Research has found that teacher social networks can lead to an increase in teacher self-ecacy [40,42] and retention [7,38], and positively inuence student performance [2]. For example, professional development provided by the Exploring Computer Science (ECS) program specically for the US-based high school course ECS after "[r]recognizing that teachers who teach CS are often isolated within their schools without organized academic departments of colleagues" [36, p. 352]. Teachers that attended this professional development reported "increased understanding, condence, and application of inquiry and equity-based teaching practice" [36, p. 356]. Understanding CS teachers' intellectual isolation and the corresponding challenges of not having a social network is important. While some of these CS teachers might have experience teaching other subjects, many could benet from resources and other teaching support specic to CS teaching.
Formal social networks can foster communities to improve teachers' condence in their curriculum and their self-ecacy [40,42]. Bandura argued that successful learning environments rest heavily on teacher's self-ecacy [6]. Researchers examined 20 middle school math departments across two districts and found that more connections in a teacher social network were associated with higher collective ecacy among school sta [11]. However, researchers have found that social connections are not always enough and that providing curriculum strategies and specic guidance have the most positive impact on instructional practices and self-ecacy [42].
Teachers report that collaborations and professional learning communities (PLCs) were important for their learning and their decision to stay in the profession [7]. In the U.S. from 2007 to 2012, around 17% of new teachers left teaching every year [25]. However, new teachers appear to be less likely to leave their schools where there is high engagement among teachers [37]. Reducing teacher turnover is important because it is expensive to recruit and train new hire teachers. These costs are more recurrent for lower-resource schools where turnover is more common [3]. Teacher retention is also a particular concern for increasing the number of teachers of color as data suggests that non-white teachers have higher turnover rates than white teachers [1]. Workforce morale is also inuenced by teacher turnover as being in a school with high turnover rate is reported by teachers to be stressful [27].
Researchers found that teachers who interact with higher performing peers improve their own students' performance [26], and that the addition of a higher performing teacher to a school improves the performance of all teachers in it [41]. One explanation as to why higher performing teachers seem to benet all teachers is that higher performing teachers are more likely to initiate pedagogical discussions which in turn improved their instructional practices [39]. However, teacher interactions seem to inuence teaching practices more when these are from the same school, with across school teacher interactions providing less curriculum support [5].
In summary, teacher social networks appear to benet teacher self-ecacy, teacher retention, and student achievement. Collectively, teacher isolation can impact student experience. Additionally, teacher isolation can impact capacity and access within schools because fewer teachers in a school reduces the ability of the school to oer sucient CS courses for all students. There is limited research looking to measure teacher isolation with longitudinal, statewide data. Analyzing the extent of CS teacher isolation is relevant to be able to better support them and to improve access to social networks and their benets.
The data used is from the California Department of Education. California has detailed, privacy-compliant, non-traceable data that includes the courses taught in schools (see Section 3.1.2), numeric identiers for who teaches them (see Section 3.1.3), and details on school size, urbanicity, and students' demographics (see Section 3.1.1). There is annual data available from Fall 2003 until Fall 2018 with the exception of Fall 2010. We use California data in large part because it is public data with a fourteen year time frame. As of the 2019-2020 school year, in California there are 1029 districts and 10,545 schools serving over 6 million students. To add context to our results, the number of high schools oering CS has doubled in the fteen years from 2003 until 2018, reaching 1120 schools in 2018.
The dataset provides the total enrollment per school, which refers to how many students are registered at that school, and is what we refer to as school size. The dataset also provides the percentage of the enrollment of students of various demographics by school and district. Specically, there is both the percentage based on race/ethnicity as well as percentage of students receiving free or reduced-price lunch (FRPL). 1 The percentage of students receiving FRPL is used in the literature as a common, but coarse, proxy for estimating the percentage of students from low-income households 1 Students are eligible to receive lunch prepared at their school for free or at a reduced price when their household income is at or below the eligibility guidelines provided by the US Department of Agriculture. [30]. The dataset provides the urbanicity of the school districts. Urbanicity is divided into four categories from least to most densely populated: rural, town, suburban, and urban [24]. This is calculated by the average population density of the district.
Classification. California provides course data per school. At the middle school and high school levels, California provides the names of courses oered in each school in each year. California does not provide a specic classication for CS courses and so a category was created based on previous work [12]. This category was created by classifying course codes based on their title and course description. In our analysis, the subjects considered are CS, math, English Language Arts (ELA), art, and science; we further split art and science. Science teachers were grouped by ve class categories: general science, life science, earth and space science, physics, and chemistry. These were chosen as they are the categories that are involved in California's science graduation requirements for middle schools and high schools [33]. Art was classied under the ve categories specied by California: Music, Theater, Visual Arts, Media Arts, and Dance. Math and ELA teachers require a single certication to teach any courses from their subject. In contrast, science and art teachers teach substantially dierent courses with varied certication requirements, and were given a narrower denition of isolation than math or ELA teachers.
Each teacher has a unique identier across their courses from each year, but not across years. This means that we can identify the courses a teacher taught within a year but cannot track a teacher over multiple years. We classify a teacher with their given subject by what courses are associated with each unique teacher identier. For example, we considered the number of CS teachers in a school to be the number of unique teacher identiers to be associated with at least one CS course in a given year.
3.1.4 Defining Peers and Isolation. We consider a teacher isolated in their school if they had no peers at their school teaching the same subject. Since some teachers teach multiple subjects, we classied teachers by the subjects they teach, such that a teacher might be considered isolated in some subjects but not others.
To answer RQ1, we focused on comparing rates of isolation across time and subject. To answer RQ2, we use the following three outcomes (i) there being no CS teacher at the school, (ii) there being an isolated CS teacher, and (iii) there being more than one CS teacher at the school. We predict our three outcomes across ve distinct linear regression models, with outcomes represented below by $DC2><4 BC for school s and year t. These outcomes were all coded to be binary. That is, schools were coded as 1 for our third outcome if they had more than one CS teacher at their school, and coded 0 otherwise. Our resulting linear probability model is presented below:
We consider school size because of how school size can serve as a capacity and access constraint to course oerings. California has extremely varied school sizes; to account for potentially diminishing returns to size we use school size as a quadratic term, seen in the variable B@A (B2⌘>>;(8I4). There are three variables relating to student demographics with %<8=>A8C~B C , referring to the percentage of students at a school s in year t that are neither white nor Asian, % '%! referring to students receiving free or reduced priced lunch (as a proxy for socioeconomic status), and %⇢!! referring to English Language Learners. These were included due to there being evidence of disparities in CS access across these dierent groups.
We also consider urbanicity represented in our model by urban and rural_town. We chose suburban as the reference group as it is the most represented in the dataset. Urbanicity is considered in some of our models as schools in rural and urban areas have been shown to oer less CS than their suburban counterparts [16]. We use year xed eects (i.e., a dummy variable per year) to account for changes all schools might experience each year, such as statewide regulatory changes. School xed eects are used to isolate changes within schools. We cluster standard errors on schools, to account for a lack of independence between observations of the same school in multiple years.
Model 0 includes only student demographics and year xed effects, and provides an initial estimate of how student demographics within a school predict whether CS is oered and if CS teachers are isolated or not. Model 0 does not include school factors or school xed eects. Model 1 additionally includes school size because smaller schools may have less capacity to oer additional courses. Model 2 is Model 1 with the addition of school xed eects. Model 3 adds urbanicity and removes school size and school xed eects because we only want to consider the eects of urbanicity without considering other school factors to better understand the relationship between our school factors. Finally, Model 4 uses all variables from the previous models except for school xed eects due to concerns of collinearity between all of our school level factors.
While our dataset covers a 14 year period and allows us to consider time-varying school characteristics, there is still the possibility that there are other factors unaccounted for relating to teacher isolation and course oerings. The fact that our dataset ends in 2018-2019 is because later years of data are not available. Additionally, data from 2020 and 2021 would have been challenging to interpret due to the eects of Covid-19. Our analysis also only considers access to one or more CS courses, and not the participation or experience within them. This is also data strictly from California and the results, with the exception of perhaps school size, cannot necessarily be generalized, as many states approach CS course oerings dierently. For example, some states have CS as a graduation requirement.
The rst author is a PhD student, who is half Latina and half Middle Eastern. Both her parents are educators, in particular, her mother is a science teacher at a public school and her mother actively participates in and has provided professional development. Seeing the impact her mother has had on other teachers and students has inuenced the rst author's interest in supporting and documenting teachers' experiences. Growing up in an area where access to CS education is currently limited, also inuenced the author's particular interest in identifying disparities in access to CS. The second author is a professor who is a white woman who participated in compulsory CS education learning opportunities that were integrated within her public school in California. This pre-college background was essential in allowing her to pursue CS in higher education. In her research, she has participated in several activities to support CSteacher collaboration and sees how such collaboration can benet teachers and students. The percentage of isolated CS teachers has decreased over time. The percentage of high school CS teachers isolated at their school, shown in Figure 1, has decreased from 75% in 2004 to 52% in 2019. 2Figure 2 shows that while the proportion of middle school CS teachers isolated at their school started to decrease in 2014, it has plateaued at around 80%, which is higher than it was for high school teachers in 2004.
As shown in Figures 1 and 2, we found that high school and middle school CS teachers experience a higher rate of isolation than teachers of most other subjects. Math and English teachers in both high schools and middle schools experience isolation at a drastically lower rate than CS teachers, with less than 5% isolated at their school. high school Physics teachers experience a rate of isolation similar to CS teachers. In 2019, approximately 50% of high school CS and Physics teachers were isolated. In middle school, General Sciences is the subset in science with the highest rates of isolation but those teachers still experience isolation at less than half the rate of CS teachers. Within the arts, Theatre and Dance teachers experience higher rates of isolation than high school and middle school CS teachers, and Visual Arts, Media Arts, and Music teachers experience isolation at a lower rate. This is the rst study to capture these high rates of CS teacher isolation without the use of self-reported surveys.
RQ2: Do school size, urbanicity, and student demographics aect the likelihood that a CS teacher is isolated at their school or that the school does not oer CS?
Table 1 shows the relationships between our ve models and our three outcomes. To facilitate the explanation of the models, they will be grouped in this result section by independent variables.
show disparities in access. Model 0 shows that schools with higher representations of students who are ELLs (%ELL) and who receive FRPL (%FRPL) are less likely to oer CS. These results are fairly consistent across all models. In models that control for school size (i.e., Models 1, 2, and 4) our estimates shrink slightly. Model 4 shows that a 10 percentage point increase in representation of ELLs or students receiving FRPL is associated with a 1.8 or 0.6 percentage point increase, respectively, in the likelihood of CS not being oered.
In some of the models that predict an isolated or non-isolated teacher, the coecient for %FRPL is not signicant. This may be in part due to the eects of the Community Eligibility Provision [43]. Starting in 2010, this provision allows schools that already have 40% of their students receiving free lunch [43] to classify all students as eligible, making %FRPL an even coarser proxy to estimate how many students come from low-income households [30].
7.2.2 Race/Ethnicity results show disparities in access when controlling for school size. In Model 0, we see that the representation of students from minority groups is a positive predictor of CS being oered. This is while controlling for the representation of ELLs and students receiving FRPL. In Model 0, a 10 percentage point increase in the percentage of students from minority groups is associated with a 1.1% increase in the likelihood that CS is oered. However, once we control for school school size in the subsequent models (1,2,4), the sign is reversed. We see a positive relationship between %minority and there being no CS oered at the school, and a negative relationship with there being one or more CS teachers. Based on Model 4, which includes student demographics and school factors, a 10 percentage point increase in representation from students from minority groups is associated with a 0.6 percentage point increase in the likelihood of there being no CS oered at the school.
School size is a positive predictor for CS. Our Model 1 adds school size. Changes between Model 0 and Model 1 show dierences in sign for %minority that are discussed above. Additionally, across all models with school size (1,2,4), school size is a positive predictor for a school oering CS, as well as having more than one CS teacher. Model 4 shows that every increase of 100 students is associated with a 2 percentage point decrease in there being no CS oered and a 0.8 percentage point increase in there being more than one CS teacher. This arms that larger schools are more likely to oer CS, however, it should be noted that the results signal a diminishing return to additional enrollment.
Rural schools are less likely to oer CS. Model 3 adds urbanicity through our rural_town and urban variables. The variable rural_town shows a positive relationship with there being no CS oered at a school and a negative relationship with there being more than one CS teacher. This implies that schools in rural and town areas are less likely to oer CS than schools in suburban areas. In particular, schools in rural and town areas have a 3.6 percentage point increase in the likelihood there is no CS compared to schools in suburban areas. When we compare Model 3 and 4, where Model 4 includes school size, we see the sign reversed for rural/town. This can imply that among schools of similar sizes, rural and town schools are more likely to oer CS than urban schools.
The results from our rst research question arm that CS teachers are largely isolated at their school and experience isolation at higher rates than teachers of most other subjects. 30% of CS teachers in the last year of our dataset were also isolated in their district. These results indicate capacity issues at a district level, and an access and experience issue at a district and school level. Capacity can in part be addressed through policy changes such as requiring every school to oer CS and then oering dedicated funding for such initiatives. Additionally, while this does not fully address issues caused by isolation, providing teachers with resources and support may improve student experience.
Reducing teacher isolation through providing more avenues for teacher social networks has been shown to improve work experience [7] and can lead to better student outcomes [2]. When hiring new sta is not possible, districts and states can attempt to compensate for isolation at a school by providing professional development and CS teacher communities. There is evidence that supports that professional development positively impacts student performance [2,13], including when teachers are tasked with teaching a new curriculum [19]. To provide professional development equitably, they can provide incentives for teachers to attend these. Such incentives or virtual options may be important for teachers far from the meeting location. School size is a positive predictor of there being one or more CS teachers but is not frequently considered in analyses of this kind. School size is an understandable capacity and access constraint that can potentially be addressed by policy change. However, inequity in CS access by race/ethnicity remains even after controlling for urbanicity, students' average socioeconomic status, and school size. This indicates that dierences in access cannot be fully explained by factors such as the Small Schools Movement. This supports existing eorts to address inequity by race/ethnicity and previous research that has highlighted disparities by race/ethnicity.
Despite teacher isolation being reported as one of the biggest challenges for CS teachers in the US, the extent of CS teacher isolation has not been documented beyond teachers' self-report. Using data from California, we determine how CS teacher isolation has varied over time and how this variation compares to teachers in other subjects. We use linear regression to determine what factors aect the likelihood of CS being oered or there being an isolated CS teacher. We nd that CS teachers do experience a higher rate of isolation when compared to other subjects. We also nd that school size is a positive predictor of the presence or quantity of CS teachers. However, even when controlling for school size, schools with higher representation of students from groups underrepresented in computing are less likely to oer CS at all. These ndings suggest that policymakers should support eorts to reduce the impact of teacher isolation, such as professional development. It also illustrates that more work is needed to provide students equitable access to CS.
RESPECT 2024, May 16-17, 2024, Atlanta, GA, USA Mariam Saar Perez and Colleen M. Lewis
Although not the focus of our analysis, we also looked at isolation at a district level and found that
30% of CS teachers were isolated in their district.