Shadow education, pandemic style: Social class, race, and supplemental education during Covid-19

Abstract

Research on shadow education—i.e., one-on-one or group learning intended to supplement children’s experiences in school—has documented persistent social class and racial/ethnic inequalities. Yet, as with many things during the Covid-19 pandemic, the nature of shadow education changed dramatically. Much supplemental education shifted online, potentially increasing accessibility; many universities became testoptional, potentially reducing the demand for the shadow education industry; and a new form of shadow education—learning pods—emerged to take pandemic schooling from a more individual to a more collective experience. In this article, we use data from a sample of U.S. parents of K-12 students stratified by race/ethnicity (N = 1911) to assess social class and racial/ethnic inequalities in shadow education in 2020–21, the first full academic year of the Covid-19 pandemic. We are also the first scholars, to our knowledge, to assess high-quality data on the use of learning pods. We find that during the pandemic, African American and South Asian students were more likely than White student to use test preparation services and online supplemental education, and that African Americans, East Asians and Latinx were more likely to utilize private tutoring. We find few disparities by family income, however, thus supporting the idea that some forms of shadow education have become more accessible than they once were. Regarding learning pods, we find that pods were most common among African American families and families with parents who were less educated and worked fulltime. Thus, most learning pods were not a means of “opportunity hoarding,” as some scholars originally feared, but instead provided sorely needed childcare and support during a time of social turbulence.

1. Introduction

The move to remote learning among many schools during the Covid-19 pandemic led to strong concerns of children “falling behind” or “losing a year” in school. Although many people quite rightly asked “falling behind whom?,” others, especially parents, remained concerned about children’s ability to learn. Many parents, anxious about the uncertainty and dissatisfied with schools’ responses to the pandemic, took their children’s education into their own hands. Much recent research has been dedicated to estimating the impact of the pandemic on social class and racial/ethnic inequalities in learning, particularly inequalities between school districts and between families in terms of school resources and support for online schooling (Bailey et al., 2021; Domina et al., 2021). Yet, another frequently-mentioned source of potential inequality—but one that perhaps has received less attention in the research literature (Bailey et al., 2021)—is in families’ ability to afford and support supplementary instruction. One way that many parents tried to prevent learning loss during the pandemic was with commercial learning centers, test preparation services, and tutors—i.e., activities that comprise shadow education—to enhance what was being learned (or not learned) in school.

Prior research on shadow education in the U.S. has focused on SAT/ACT preparation and has found persistent social class inequalities, particularly when it comes to commercial test preparation services, private tutoring, and other costly forms of shadow education (Buchmann et al., 2010a; Byun & Park, 2012; Park et al., 2016). In addition, research generally finds that African Americans and Asian Americans are more likely than similar-income Whites to use shadow education in preparing for the SAT (Alon, 2010, Buchmann et al., 2010aaa, Byun and Park, 2012). Much of the most influential research on shadow education relied on data from the National Education Longitudinal Study of 1988 (NELS-88), a cohort of students who were in the eighth grade in 1988 (long before the advent of the internet), thus making an update warranted at this point. In their influential piece, Buchmann et al. (2010b) encouraged scholars to “explore the utility of the shadow education concept while remaining sensitive to the particulars of time and place” (489). In addition to technological innovation that introduced online forms of supplemental learning, the Covid-19 pandemic presents a particularly intriguing case in this regard because the pandemic altered not only formal education, but also shadow education, in several ways. First, when schools shifted online, so too did many test preparation classes and tutoring centers. Commercial tutoring companies such as Kumon and Sylvan Learning Center, which offer live, group tutoring sessions, expanded their virtual options. Some reduced their fees in order to increase accessibility and affordability. Khan Academy, which already provided free online recorded lessons, expanded their suite of course offerings to include more subjects and grade levels. Second, as some universities adopted SAT/ACT-optional policies, demand for other forms of shadow education beyond test preparation may have increased. Third, a new form of shadow education—learning pods—also emerged during the pandemic, and along with pods came great concern over their role in the reproduction of inequality.

Given these changes in the demand for shadow education (as well as, potentially, changes in families’ reasons for using shadow education), in this article, we assess the extent to which documented social class and racial/ethnic differences in the use of shadow education persisted during the Covid-19 pandemic. We also make the case for learning pods as an emerging form of shadow education and assess demographic predictors of their use. Our data come from an original survey we conducted with 1918 parents of K-12 students in the U.S. We focus in particular on the use of test preparation services, supplemental online education (including both services that charge and those that are free),1 private tutoring (paid and free), and learning pods during the 2020–21 school year. These data are critical for helping scholars assess these potential sources of educational inequality, while also demonstrating how parents are thinking about both formal and shadow education in this period of educational uncertainty.

2. Background

2.1. The rise and diversification of shadow education

As Park et al. (2016) illustrate, most students in the U.S. participate in organized, academically oriented learning activities “beyond the schools walls” to supplement their formal education. Stevenson & Baker, (1992) coined the term “shadow education” when they referred to such “educational activities that occur outside formal schooling and are designed to enhance the student’s formal school career” (1639). Shadow education works alongside the “regular” education system (i.e., schools), often taking the form of one-on-one tutoring, small group tutoring, or lectures. Research shows that shadow education most often emerges in educational contexts that use formal examinations, and when there are strong linkages between early educational outcomes and future opportunities (Stevenson & Baker, 1992). For example, Zhang (2020) describes how investment in shadow education in Shanghai is a product of educational expansion and increased competition for higher education, in conjunction with educational reforms to reduce study burden in schools. As the responsibility for test preparation gradually moved from schools to families, middle-class families turned to shadow education when they did not have the economic and social capital needed to ensure access to elite colleges.

In the years since its inception, the prevalence of shadow education has grown rapidly worldwide (Aurini, 2004, Baker et al., 2001, Bray, 2021, Byun et al., 2018, Entrich, 2020), including in the U.S. (Park et al., 2016). Given the importance of the SAT/ACT for college admission in the U.S. (although with the movement toward test-optional policies, this may become less so), shadow education often takes the form of test preparation in learning centers, private tutoring, and even after-school classes offered in public schools (Buchmann et al., 2010aaa, Byun and Park, 2012, Park et al., 2016). As admission to selective colleges and universities has become more competitive, shadow education has become a key component of students’ strategies to maximize their chances of admission (Bound et al., 2009). Shadow education also has expanded in the U.S. in recent years through private learning centers, such as Kumon and Sylvan Learning Center (Aurini, 2004). Indeed, between 1997 and 2016, the number of private learning centers in the U.S. grew from approximately 3000 to almost 10,000 (Kim, Hassel, & Gilliam, 2022).

Shadow education also has become more varied and accessible with the growth of the internet and other technology. Even before the onset of the Covid-19 pandemic, new technologies led to new modes of tutoring and other private instruction (Aurini, 2012, Bray, 2010; Byun & Baker, 2015). Prominent examples include Khan Academy, which is fully online and free to use, and online versions of Kumon and Sylvan Learning Center, which require a paid subscription. By most accounts, the pandemic expedited the proliferation of online supplemental education, as teachers and families had unprecedented anxiety about learning loss and demand for learning support. In April 2020, for example, Google searches for online learning resources had already doubled compared to pre-Covid levels (Bacher-Hicks et al., 2021). Between March and May of 2020, Khan Academy use tripled; teacher registrations were six times that of 2019 and parent registrations were 20 times that of 2019 (Khan Academy, 2021). Similarly, Outschool’s enrollment increased from 80,000 to 300,000 (Horn, 2021). While some of these increases are likely attributable to teachers using Khan Academy or Outschool lessons in lieu of in-person learning (to be clear—something that we would not classify as shadow education because teachers were assigning the content to the whole class, thus making it traditional education), families also used these resources to supplement assigned work. For these reasons, in addition to examining test preparation, in this article we also assess data on supplemental online commercial learning (both paid and free) and private tutoring in general (not just for test preparation) among school-aged children of all grade levels.

2.2. Social class disparities in shadow education

Much prior research on shadow education has focused on social class as a primary dimension of inequality. This research considers multiple mechanisms through which social class inequalities in shadow education operate; for example, many forms of shadow education are expensive, demand ample time and support, and require the social and cultural capital needed to know about these resources in the first place (Baker, 2020, Bray, 2021, Park et al., 2016). As a result, prior studies in the U.S. and other countries have found a consistent socioeconomic gap in shadow education participation, particularly forms that are most costly (and arguably higher quality; Bray, 2021; Entrich, 2020; Jansen et al., 2021; Park et al., 2016). Using data from the U.S., Buchmann et al. (2010a) find that students from higher-income families are more likely to take a private course and more likely to hire a private tutor than their lower-income peers. Relatedly, Kim, Hassel, & Gilliam, (2022) find that private tutoring centers are more likely to be located and experience growth in areas with high household income and parental education. Publicly provided forms of tutoring, such as those offered by public schools, typically are targeted toward lower-SES families. Although this has the benefit of making test preparation and other shadow education more accessible, these programs often end up serving students who are relatively advantaged in terms of access to transportation and other resources (Park et al., 2016).

Some emerging research suggests that the pandemic has contributed to widening socioeconomic inequality, as schools have relied heavily on parents with different resources, schedules, and skills to assist children. For example, parental education was a strong and consistent predictor of children’s submitting assignments on time and logging in for remote instruction during the pandemic (Domina et al. 2021) More specific to shadow education, the Google search study we mentioned earlier found that searches for online learning resources increased most dramatically in high-SES areas (Bacher-Hicks et al., 2021), although more research is needed to assess actual rates of shadow education by SES.

We also have good reason to suspect, however, that social class inequalities in shadow education may have been mitigated during the pandemic. Online options are much more affordable than they once were. Khan Academy offers online courses in a variety of subjects for free, and Outschool also began to offer free and discounted online courses. Additionally, many school districts provided students with tablets, laptops, mobile internet, and other technology to make learning more accessible. Asynchronous online courses may also be more accessible than traditional commercial classes because transportation is not needed and work can be completed any time.2 These circumstances collectively suggest that although higher-SES students may have been more likely to engage in shadow education before the pandemic, these gaps may have become smaller or even closed in the wake of Covid-19.

2.3. Racial/ethnic disparities in shadow education

In addition to family background differences in the use of shadow education, research also points to differences across race/ethnicity. Much research suggests that non-White families in the U.S. are more likely to use shadow education than White families, net of SES. Buchmann et al. (2010a), for example, find that after accounting for SES, African American, Asian American, and Latinx youth are more likely than Whites to participate in shadow education, including the more costly forms such as commercial test preparation and private tutoring. African Americans are more likely than Whites to participate in more intensive and more costly types of shadow education, such as high school-based test preparation, commercial test preparation, and private tutoring (also see Byun & Park, 2012)—although the African American advantage in private shadow education may be most pronounced among high-SES students with low test scores (Alon, 2010). These findings echo research showing that, given the same level of income, African American parents tend to invest more money (often much more) in their children’s education than White parents (Quadlin & Conwell, 2021; Quadlin & Powell, 2022; Steelman & Powell, 1991). Additionally, Latinx families are more likely than White families to hire a private tutor (net of SES), either alone or in combination with other forms of test preparation (Buchmann et al., 2010aaa, Byun and Park, 2012).

Research on supplemental education often has focused on Asians and Asian Americans, given strong traditions of investment in education (including shadow education) among these groups (see, e.g., Lee & Zhou, 2015; Yin, 2017). Asian Americans are more likely than Whites to participate in shadow education, particularly in the form of a private course (Buchmann et al., 2010a; Byun & Park, 2012; Park et al., 2016). Some of this work distinguishes between East Asians and “other” Asians, finding that East Asian students are more likely to take commercial test preparation courses compared to other racial/ethnic groups, including “other” Asians (Byun & Park, 2012). East Asians also are more likely than White and Latinx students to receive one-on-one tutoring (Byun & Park, 2012).

Some scholars have posited that investment in shadow education should be considered a form of “concerted cultivation,” especially among Asian Americans (Dhingra, 2020, Jansen et al., 2021). Lareau (2003) originally described concerted cultivation as a style of parenting characterized by (among other things) enrollment in structured activities, such as music or sports. Dhingra (2020) argues, however, that for Asian Americans who lack familiarity with, or otherwise are excluded from, these types of activities, parents see shadow education as a way for their children to stand out. This is true especially for groups who come from countries with intensive or high-stakes educational testing and longer histories of shadow education (Dhingra, 2020). Some Asian American groups, such as Chinese, Korean, and Vietnamese Americans, have established shadow education offerings within their co-ethnic communities, which may provide families with accessible and affordable options (Zhou & Kim, 2006; Zhou & Li, 2003). Relatedly, private tutoring centers tend to be concentrated in areas with relatively large immigrant and Asian American populations (Kim, Hassel, & Gilliam, 2022).

A complementary line of research contends that racial/ethnic differences in shadow education arise because racially minorized groups are aware of the history of racial/ethnic disadvantages in American education—and on the SAT in particular—and thus they take on extra preparation in order to counter this inequality (Buchmann et al., 2010a). We likewise suggest that minorized groups, particularly African American families, may seek alternative forms of education to help counter disadvantages they encounter in public schools. Scholars recently have noted an increase in homeschooling among African Americans, specifically due to concerns about the quality of education provided in public schools as well as the need to provide a curriculum that depicts Black culture in a positive light (Mazama & Lundy, 2013). This trend may have been exacerbated during the pandemic, when schools, especially those that were under-resourced, were unable to meet many families’ needs. One recent study showed that in the fall of 2020, African American and Latinx students were more likely than White students to attend school remotely rather than in-person (Camp and Zamorro, 2022), so perhaps these families sought out shadow education to help support their children. Based on these trends, we might expect that racial/ethnic trends in shadow education during the pandemic would mimic trends from before the pandemic (i.e., more shadow education among non-White groups than among Whites, net of SES).

But at the same time, we also might expect to see fewer racial/ethnic differences in shadow education during the pandemic than in years prior. Just as formal schooling largely moved online during the pandemic, shadow education also moved online, and lack of adequate internet access during the pandemic was greatest among African Americans (Friedman et al., 2021). This might have been further exacerbated with income loss given that African American workers, as well as Latinx workers, were more likely to lose their jobs during the pandemic than White workers (Couch et al., 2020, Parolin, 2021; Vargas & Sanchez, 2020). Accordingly, if online shadow education became less accessible for non-White families, racial/ethnic differences in its use may have shifted compared to pre-pandemic times. In addition, many at school-resources were not available due to safety precautions or staff shortages, and teachers often were not able to support children as well as they had prior to the pandemic. Thus, perhaps more White families looked to shadow education to supplement their children’s formal schooling. In either case, we would expect to see fewer racial/ethnic differences in shadow education during this recent period, compared to what we have seen in prior research.

2.4. The case of learning pods

As a final focus of the analyses, we consider a unique form of shadow education that emerged during the pandemic—learning pods. Pods are not a new phenomenon; modern homeschooling emerged in the 1970s and has increased rapidly in prevalence since then (Mazama and Lundy, 2013; Stevens, 2003), and self-directed micro-schools have existed since at least 2009 (Horn, 2015). As soon as it became clear that schools would be fully or partially online—or, the opposite, that some schools were not going to have online options despite public health uncertainty—parents began to establish their own home-based schooling options. By August 2020, the “Pandemic Pods” Facebook group had nearly 40,000 members (Horn, 2021). Some parents who worried about the quality of remote teaching pulled their children out of local schools and hired private tutors. Others who worried about the potential dangers of sending their children to school in-person did the same. These “pandemic pods” or “micro-schools” raised concerns among scholars (Horn, 2021), who suggested that these very expensive options would not only take funds away from local schools, but also potentially provide already-privileged children with better educational opportunities, thus exacerbating inequality and contributing to racial and ethnic segregation—akin to White flight from public to private schools.

In the end, few parents un-enrolled their children from school (Jochim & Poon, 2022), but many participated in “supplemental learning pods,” in which children remained enrolled in school but formed small groups of children that learned remotely together. Parents cited several reasons for forming these pods. A 2020 survey of families and educators who participated in pods found that the most-cited reason for the pods was socialization, and the second-most-cited reason was child supervision (Jochim & Poon, 2022)—both of which are non-academic in nature. Other commonly cited reasons were to support children’s emotional needs, improve engagement, provide individualized instruction, and prevent children from falling behind. Regardless of their reasons, most pods hired pod facilitators who had a professional background in education and supported children with learning (Jochim & Poon, 2022). Thus, regardless of the primary intended purpose from the parent’s perspective (e.g., childcare, socialization, emotional needs), most children in pods were receiving academic support. In this sense, we contend, pods can be considered a form of shadow education. Many pod facilitators were, quite literally, sitting off-camera in the shadows, teaching children in-between classes that were being held online.

Although many commentators assumed there would be social class and racial/ethnic inequalities in the use of pods, little empirical research has examined whether this was the case. Regarding social class in particular, the cost of pods varied widely. As we just discussed, pods frequently hired instructors with a professional background in education (especially in high-income areas). Yet, other pods cost the equivalent of a part-time babysitter—still costly, but not as much—and others relied on family members to supervise the pods. As a response to early concerns over equity, some school districts and community-based organizations also established pod-like learning environments where children could go to learn remotely at little to no cost (Kim, Hassel, & Gilliam, 2022). Even in the case of district-organized pods (which one might argue is not shadow education because they are coordinated through the school), there were still pod facilitators present that provided additional help beyond the children’s teachers, helping to manage schedules, facilitate communication between teachers and families, and advocate for students. The few nationally representative surveys of pod use that have been conducted have yielded mixed findings regarding social class and which families were more and less likely to use pods (Jochim & Poon, 2022). Even less is known about racial/ethnic differences in pod use, but because African American children were the most likely to attend school remotely (Camp and Zamorro, 2022), it may be that they were also the most likely to participate in a learning pod. Thus, in this article, in addition to extending prior research on shadow education to include supplemental online learning, we also examine the extent to which pod users mirror the social class and racial/ethnic differences we find for other forms of shadow education.

3. Data & methods

3.1. Data

We collected original survey data through Qualtrics (N = 1918), a web-based company that offers survey design tools and access to research participants for market research and academic purposes. We used screening questions at the beginning of the survey to limit the sample in two main ways. First, respondents had to have at least one child who was enrolled in school (or homeschooled) in grades K-12. If respondents had more than one child in grades K-12, we randomized whether they were asked to answer questions about their youngest or oldest child. Second, due to our substantive interest in racial/ethnic differences in shadow education, we created a quota such that the sample was evenly divided between respondents who identified as Asian, Black, Latinx, and White (discussed further below). Quota sampling methods are common in educational surveys when the goal is to uncover broad racial/ethnic patterns in student experiences and outcomes (see, e.g., Massey et al., 2003). Although the data are not nationally representative samples of Asian, Black, Latinx, and White parents, our sample is comparable to national estimates in terms of income and other characteristics (see Table 1 for recent national data on parents of school-aged children; National Center for Education Statistics, 2019). Notably, this sample is more educated than the broader population of parents of school-aged children, and Asians are slightly more of a positive outlier than we might expect in terms of income, although research shows that Asian Americans on average are relatively advantaged across multiple socioeconomic dimensions (Lee & Kye, 2016; Sakamoto et al., 2009).

Table 1.

Descriptive statistics for shadow education and other variables of interest.

	Full sample	Asian	Black	Latinx	White	National estimate
Shadow education outcomes
From Buchmann et al. (2010) – grades 6–12
None	0.49	0.46	0.43*	0.52	0.54	–
Books/videos/software	0.16	0.16	0.16	0.14	0.16	–
High school course	0.05	0.05	0.04	0.05	0.07	–
Private course	0.15	0.16	0.19*	0.14	0.11	–
Private tutor	0.16	0.18	0.17	0.16	0.12	–
From Grodsky (2010) – grades 6–12
None	0.49	0.46	0.43*	0.52	0.54	–
Public	0.21	0.21	0.21	0.19	0.23	–
Private	0.07	0.08	0.08	0.06	0.05	–
Both public and private	0.24	0.25	0.29*	0.24	0.18	–
Online education in 2020–21 – all grades
None	0.66	0.64	0.63	0.68	0.68	–
Free	0.21	0.24*	0.23*	0.21*	0.15	–
Paid	0.14	0.12	0.14	0.11*	0.17	–
Tutoring in 2020–21 – all grades
None	0.82	0.83*	0.78*	0.81*	0.87	–
Free	0.06	0.04	0.10*	0.08*	0.03	–
Paid	0.12	0.13	0.12	0.11	0.10	–
Learning pod in 2020–21 – all grades	0.19	0.13	0.25*	0.21	0.17	–
Background variables and controls
Parent’s racial background
East Asian	0.09	0.35	–	–	–	0.02
Filipino	0.05	0.21	–	–	–	0.01
South Asian	0.07	0.27	–	–	–	0.03
Other Asian	0.04	0.16	–	–	–	0.01
Black	0.25	–	1.00	–	–	0.11
Latinx	0.25	–	–	1.00	–	0.21
White	0.25	–	–	–	1.00	0.72
Family income (categorical, in tens of thousands)	6.79	9.33*	5.26*	5.90*	6.70	7.36
	(5.13)	(5.60)	(4.54)	(4.53)	(4.84)
Parent education
High school or less	0.21	0.08*	0.25	0.29*	0.21	0.32
AA/Some college	0.35	0.18*	0.44	0.35*	0.43	0.33
BA or more	0.44	0.74*	0.31	0.36	0.36	0.38
Both parents (or single parent) work(s) full time	0.42	0.40	0.47	0.39	0.42	–
Child’s gender identity
Male	0.49	0.49	0.48	0.49	0.52	–
Female	0.50	0.50	0.52	0.51	0.47	–
Non-binary or other	0.01	0.00	0.00	0.01	0.01	–
Child is in grades 6–12	0.41	0.43	0.40	0.37	0.43	–
R parent is married	0.64	0.79*	0.50*	0.59*	0.69	0.70
Number of children in household	1.98	1.76*	2.04	2.02	2.08	1.95
	(1.12)	(.94)	(1.23)	(1.14)	(1.14)
Urbanicity
Urban	0.38	0.35*	0.50*	0.45*	0.23	–
Suburban	0.45	0.58*	0.36*	0.40*	0.48	–
Rural	0.17	0.07*	0.15*	0.16*	0.29	–
Resides in the South	0.43	0.28	0.59	0.37	0.46	–
Prior achievement (standardized to 4-point scale)	3.18	3.31*	3.17	3.10	3.13	–
	(.76)	(.73)	(.71)	(.80)	(.80)
Child has ever been enrolled in accelerated classes	0.48	0.53*	0.50*	0.47	0.43	–
Child has been assessed as having a disability	0.32	0.22*	0.35	0.34	0.37	–
Child has an IEP	.19	0.11*	0.22	0.17*	0.24	–
Parent expects child to attend college	0.85	0.90*	0.86*	0.86*	0.77	–
Number of academic topics discussed with child	2.59	2.68	2.61	2.47	2.60	–
	(1.44)	(1.49)	(1.42)	(1.43)	(1.43)
School type in 2020–21
Public	0.75	0.78	0.77	0.73	0.74	–
Private	0.08	0.10	0.05	0.08	0.08	–
Charter	0.07	0.05	0.09*	0.07	0.04	–
Child is homeschooled	0.10	0.06*	0.09*	0.11	0.14	–
Child’s school was in-person all year in 2020–21	0.21	0.18*	0.23	0.20	0.24	–
N/n	1918	477	487	479	475

Source: Authors’ original data collected through Qualtrics panels; National estimates are taken from the NCES EDGE dataset (2015–19).

Note: National estimates of marital status are based on mothers of school-aged children.

^*p < .05 for comparison to White respondents (two-tailed tests). More detailed p-values are included in the main text where applicable.

3.2. Dependent variables

3.2.1. SAT/ACT preparation

Respondents who had children enrolled in grades 6–12 (n = 791) were asked if their child planned to take the SAT/ACT and, if so, whether they participated in any of the following test preparation activities during 2020–2021: studied from test preparation books/videos/software; enrolled in a special course in school; enrolled in an in-person commercial test preparation service; enrolled in an online commercial test preparation service; or received private one-on-one tutoring. Although some prior studies have assessed SAT/ACT preparation among high school students only (e.g., Buchmann et al., 2010a), we broadened this population to include middle schoolers because recent research suggests that for some students (especially those in higher-SES groups), the college admissions and preparation process begins much earlier than it once did (see, e.g., Pappano, 2015). Indeed, in our sample, rates of test preparation among middle schoolers (47% had any form of test preparation, not shown) were only slightly lower than among high schoolers (58%, not shown), although we replicate these analyses among high schoolers only (discussed in the results) to maintain consistency with prior research.3

We use these data to create two separate but related measures of test preparation. The first draws from Buchmann et al. (2010a), where test preparation is measured with a categorical variable indicating the highest level of test preparation (0 = no test prep; 1 = books/videos/software; 2 = course offered through the child’s school; 3 = private course; 4 = private tutoring).4 The second measure draws from Grodsky (2010) study, in which he distinguished between test preparation that occurs in the private sector and public instruction offered through schools. This second measure of shadow education is a categorical variable indicating private, public, or both types of test preparation. This variable is coded as: 0 = no test prep; 1 = public only (books/videos/software and/or a course offered through school); 2 = private only (commercial test prep and/or tutor, but not books/videos/software or school course); and 3 = both public and private.

Table 1 shows descriptive statistics for our sample, with the top portion of the table focusing on test preparation among children in grades 6–12. As we see in the first column, about half of respondents reported that their child participated in some form of preparation for the SAT/ACT exams. This is lower than that reported by Buchmann et al. (2010a) (at 73% of their sample), but we suspect this is because their sample was restricted to high school seniors who already took or were planning to take the SAT or ACT.5 In addition, Buchmann et al. (2010a) found that the most common highest level of test preparation was books/videos/software; this is also one of the top two categories in our sample, with the other being private tutoring. In terms of private versus public forms of test preparation, about one-quarter of families used both public and private, followed by public only. Use of only private test preparation was rare, at less than 10% of the sample.

3.2.2. Supplemental online education

We also asked all respondents whether their child used an online education program outside of assigned schoolwork (meaning that the teacher did not assign the online work as part of a formal lesson or for homework), and whether they paid for this online instruction. Our dependent variable is a 3-category measure of online supplemental education indicating whether the child did not use any online instruction, used free online instruction, or used paid online instruction. As shown in Table 1, slightly more than one-third of students received some form of online instruction outside of assigned homework. However, most of this instruction was free. Fourteen percent of the sample did not pay for online instruction, and 21% of the sample paid for online instruction.6

3.2.3. Tutoring

We asked all parents whether their child received any tutoring outside of regular school hours and, if so, whether they paid for the tutoring. Our dependent variable is a 3-category measure of tutoring indicating whether the child did not have a tutor, had a tutor but did not pay,7 or had a paid tutor. In our sample, 12% hired a paid tutor, and 6% had a free tutor, such as a volunteer tutor through an after-school program.8

3.2.4. Learning pods

As a final dependent variable of special interest during the pandemic, we consider students’ enrollment in learning pods. Following prior commentary and research in this area, we defined pods in our survey as “small groups of children (typically 2–10 children) from different families, who are homeschooled or attend school remotely, gathering to learn together.” Our dependent variable is a dichotomous measure of whether children were enrolled in a pod. As shown in Table 1, 19% of families in our sample participated in a pod for at least part of the 2020–2021 school year.9

3.3. Key independent variables

3.3.1. Race/ethnicity

Due to our theoretical and substantive interest in racial/ethnic differences in shadow education, we constructed a quota sample with equal representation (i.e., 25% each) of parents who identify as Asian, Black, Latinx, and White.10 Those who identified as Asian were asked to select from a list of common Asian ethnic groups. We combined this information to create a seven-category variable indicating whether the respondent was East Asian, Filipino, South Asian, Other Asian, Black, Latinx, or White. As shown in Table 1, 9% of parents were East Asian, 4% were Filipino, 7% were South Asian, and 4% were in “Other Asian” groups.

3.3.2. Socioeconomic status

Family socioeconomic status is indicated by two variables: household income and parents’ education. We defined household income as total income from all sources prior to taxes and deductions in 2020 and is reported in tens of thousands. Parents’ education is a categorical measure of the highest level of education the parent has attained (or the parent’s partner, if the partner is present in the household and attained more education than the respondent parent).

3.4. Control variables

In multivariable analyses, the models include additional controls to account for factors aside from race and socioeconomic status that may be associated with test preparation and other forms of shadow education. Parents’ employment status is likely to predict shadow education, particularly pod use, because parents who work full-time may have less time available to assist children with homework and/or remote schooling. In our models, parents’ employment status is measured as a dichotomous variable indicating whether both parents in a two-parent household work full-time (or whether a single parent works full-time).

We also control for child’s gender (girl, boy, non-binary); child’s grade level (K-5th, 6–12th); respondent parent’s marital status; number of children in the household; urbanicity; and region of the U.S. In addition, we include a host of control variables related to students’ academic performance and experiences in school. Prior achievement measures the student’s achievement prior to the start of the pandemic and is measured on a 0–4 scale. For students in grades 6–12, the scale corresponds to their GPA, and for students in K-5, we had parents estimate their child’s proficiency level and we recoded this to a numerical scale. “Accelerated learning” is an indicator for whether the student has ever been enrolled in a “gifted and talented” program or honors, advanced placement (AP), or accelerated learning classes. We also control for parents’ educational expectations, which is measured with a dichotomous variable indicating whether the parent expects their child to attend college. A continuous measure of number of parent-child discussions about academic topics ranges from 0 to 6 (e.g., discussions about grades, college). We also include indicators for whether the student has been formally diagnosed with a learning and/or physical disability, and whether the student has an IEP (i.e., an Individualized Education Plan). Finally, we control for school type (public, private, charter, homeschool) and school modality (remote/hybrid/mixed, entirely in-person) during 2020–2021, the first full academic year of the Covid-19 pandemic.

3.5. Analytic strategy

The findings are presented in three main parts, all of which focus on racial/ethnic and social class disparities in shadow education. First, following Buchmann et al. (2010a), we examine test preparation among 6th-12th grade students by gauging their highest level of test preparation, using multinomial logistic regression. We also examine test preparation with a related measure proposed by Grodsky (2010), as we discussed earlier, also using multinomial logistic regression. We then turn to examining two additional forms of shadow education—online supplemental education and tutoring—and their use during the 2020–21 school year, using the full K-12 sample. Here we distinguish between paid and free shadow education, recognizing that students may access different types of shadow education, as well as to distinguish between more public versus private forms, depending on their race/ethnicity and SES. In both cases we use multinomial logistic regressions. Finally, we use logistic regression to predict students’ participation in a learning pod in 2020–21. For all analyses, we show average marginal effects (AMEs), which are readily interpretable as changes in predicted probabilities given a one-unit change in the independent variable.

4. Results

4.1. Bivariate results

In addition to pooled descriptive statistics, Table 1 also includes descriptive statistics for each of the four main racial/ethnic groups in our sample (for parsimony, Asian is shown as one group here). As we see in columns 2–5, in sheer point-estimate terms, Whites were the most likely to report that they used no shadow education for test preparation (significantly more so than Black students; p < .05). White students also tended not to use the more intensive, and more costly, forms of shadow education, as reflected in their low point-estimates for these measures. Black students were more likely than White students to take a private course (p < .05), and Black students also were more likely than White students to use both public and private forms of test preparation (p < .05).

Turning to children of all grade levels, we find that while White students tended to use paid online education more than other groups (significantly more so than Latinx students; p < .05), White students were the least likely to use free online education—significantly less than Asian students (p < .001), Black students (p < .01), and Latinx students (p < .05).11 White youth also were the least likely to use any form of tutoring, as compared to Asian students (p < .05), Black students (p < .001), and Latinx students (p < .01). The most notable difference is in the use of free tutoring, such as tutoring offered through after-school programs or public libraries. Black students and Latinx students both were more likely to use free tutoring than White students (both p < .001), while we find no significant racial differences in paid tutoring.

Finally turning to pods, we find that Black students were the most likely to use learning pods during the pandemic. About 25% of Black respondents reported that their child was enrolled in a learning pod, versus about 17% of White respondents (p < .01). Latinx respondents (21%) also were marginally more likely than White respondents to use pods (p = .09). These findings run counter to the popular early narrative that pods were being used primarily by privileged White families to secure additional educational advantages during the pandemic. Yet, these initial patterns also are in line with findings from other scholars (Camp and Zamorro, 2022) as well as our data (see Table 1 as well as Appendix C in the online supplement) showing that White students were the most likely to attend school in-person all year rather than remotely. Because many White students attended school in-person during the pandemic, they would have no need for learning pods. As shown in Appendix D in the online supplement, most families we surveyed did not unenroll their children from school in order to form micro-schools, as some commentators originally feared. Most families’ pods were organized through their school district, but White families were more likely than other groups to participate in pods that they organized themselves or with other parents. In addition, Latinx families were less likely than other groups to hire someone to facilitate their child’s pod.

4.2. Multivariable results

4.2.1. SAT/ACT preparation

We next turn to examining race/ethnicity, income, and other predictors of shadow education, incorporating the suite of controls we described earlier in the Data & Methods section. We begin with test preparation, given that much prior research on shadow education has focused specifically on preparation for the SAT/ACT and other high-stakes exams. Table 2 shows average marginal effects of race/ethnicity and income on two related conceptions of test preparation. The first assesses the highest level of preparation students used to prepare for the SAT/ACT (versus no preparation). The second distinguishes between “public,” “private,” and “both public and private” forms of test preparation (versus no preparation). Recall that these analyses rely on students in grades 6–12, rather than grades 9–12 as in prior research, because test preparation tends to begin earlier than it once did. We replicated these models using students in grades 9–12 and results were highly consistent with what is shown here; see Appendix A in the online supplement.

Table 2.

Race, income, and other demographic predictors of test preparation among students in grades 6–12 (N = 609).

	Outcomes from Buchmann et al. (2010)				Outcomes from Grodsky (2010)
	Books/Videos/ Software (versus None)	High School Course (versus None)	Private Course (versus None)	Private Tutor (versus None)	Public (versus None)	Private (versus None)	Public & Private (versus None)
East Asian	-0.020	-0.011	0.051	0.061	-0.033	0.067	0.053
	(0.054)	(0.038)	(0.051)	(0.052)	(0.062)	(0.042)	(0.057)
Filipino	0.130	-0.038	0.015	0.042	0.082	-0.018	0.089
	(0.087)	(0.038)	(0.062)	(0.070)	(0.090)	(0.031)	(0.082)
South Asian	0.116	0.017	-0.006	0.140+	0.128	-0.044*	0.179*
	(0.092)	(0.065)	(0.061)	(0.082)	(0.099)	(0.018)	(0.087)
Other Asian	-0.036	0.052	-0.048	0.005	-0.010	0.082	-0.099
	(0.073)	(0.077)	(0.055)	(0.069)	(0.093)	(0.069)	(0.063)
Black	-0.008	-0.037	0.090*	0.043	-0.046	0.035	0.100*
	(0.042)	(0.027)	(0.045)	(0.040)	(0.047)	(0.030)	(0.048)
Latinx	-0.013	-0.021	0.004	0.035	-0.037	0.005	0.044
	(0.042)	(0.029)	(0.039)	(0.040)	(0.048)	(0.025)	(0.046)
Household income	-0.003	-0.004	-0.001	0.001	-0.006	-0.003	0.003
	(0.004)	(0.003)	(0.003)	(0.003)	(0.004)	(0.003)	(0.004)
Parent attained AA/Some college	-0.007	-0.009	-0.050	0.002	-0.015	-0.040	-0.008
	(0.045)	(0.026)	(0.044)	(0.040)	(0.049)	(0.028)	(0.048)
Parent attained BA or more	-0.057	-0.006	-0.033	0.055	-0.054	-0.003	0.020
	(0.049)	(0.030)	(0.048)	(0.045)	(0.054)	(0.034)	(0.053)
Parent(s) work full-time	0.063+	-0.006	0.042	0.021	0.055	0.031	0.033
	(0.033)	(0.020)	(0.030)	(0.031)	(0.036)	(0.022)	(0.035)
Child is male	0.039	-0.001	-0.021	0.002	0.036	0.030	-0.044
	(0.030)	(0.019)	(0.028)	(0.029)	(0.033)	(0.021)	(0.033)
Child is non-binary	0.029	0.088	0.144	0.013	0.143	-0.050***	0.177
	(0.154)	(0.141)	(0.178)	(0.155)	(0.189)	(0.012)	(0.180)
Parent is married	-0.029	0.021	-0.012	-0.007	-0.013	0.011	-0.026
	(0.032)	(0.021)	(0.032)	(0.033)	(0.036)	(0.022)	(0.036)
Number of children	-0.009	-0.015	0.002	0.011	-0.023	-0.007	0.019
	(0.014)	(0.010)	(0.013)	(0.013)	(0.016)	(0.010)	(0.015)
Suburban	-0.039	-0.045*	0.027	-0.110**	-0.086*	0.015	-0.094*
	(0.034)	(0.023)	(0.033)	(0.034)	(0.038)	(0.023)	(0.038)
Rural	0.029	-0.038	-0.044	-0.100*	-0.013	-0.027	-0.115*
	(0.048)	(0.028)	(0.038)	(0.044)	(0.052)	(0.025)	(0.048)
South	0.019	-0.002	-0.040	0.015	0.016	-0.029	0.005
	(0.032)	(0.019)	(0.029)	(0.031)	(0.035)	(0.020)	(0.035)
Prior achievement	0.008	-0.013	0.031	0.002	-0.008	-0.003	0.032
	(0.026)	(0.014)	(0.026)	(0.024)	(0.028)	(0.017)	(0.028)
Accelerated classes	-0.006	0.015	0.106**	0.048	0.008	-0.013	0.171***
	(0.034)	(0.023)	(0.038)	(0.036)	(0.039)	(0.022)	(0.042)
Disability	-0.034	0.032	0.105**	0.081*	0.005	0.071**	0.112**
	(0.038)	(0.021)	(0.033)	(0.033)	(0.041)	(0.022)	(0.038)
IEP	-.010	0.035	-0.003	0.077*	0.047	-0.052+	0.113*
	(0.051)	(0.023)	(0.041)	(0.039)	(0.051)	(0.031)	(0.046)
Expect college	-0.033	0.058	0.087	0.059	0.013	-0.002	0.166*
	(0.047)	(0.039)	(0.059)	(0.058)	(0.055)	(0.031)	(0.070)
Discuss academics with child	0.025**	0.009	-0.011	0.005	0.034***	0.000	-0.005
	(0.009)	(0.006)	(0.009)	(0.009)	(0.010)	(0.006)	(0.010)
Private school	-0.018	-0.040+	0.017	0.104+	-0.064	-0.048*	0.167**
	(0.058)	(0.020)	(0.051)	(0.058)	(0.058)	(0.020)	(0.064)
Charter school	-0.056	0.005	0.022	0.080	-0.045	0.032	0.060
	(0.049)	(0.035)	(0.054)	(0.061)	(0.059)	(0.043)	(0.063)
School fully in-person in 20–21	-0.065	0.010	0.001	0.013	-0.046	0.017	-0.007
	(0.041)	(0.022)	(0.035)	(0.035)	(0.043)	(0.024)	(0.040)

+ p < .10, * p < .05, ** p < .01, *** p < .001 (two-tailed tests).

Source: Authors’ original data collected through Qualtrics panels.

Note: Multinomial logistic regressions; average marginal effects (AMEs) shown. Sample is restricted to students in grades 6–12 who are not homeschooled. Omitted categories are White; parent attained high school diploma or less; parent(s) not working full-time; child is female; respondent parent is not married; urban residence; does not reside in the South; no enrollment in accelerated classes; no disability; no IEP; parent does not expect child to attend college, doesn’t know, or is unsure; public school; school not fully in-person.

In the first section of the table, we see that after controlling for factors such as parents’ employment status, urbanicity, accelerated learning, and disability status, the predicted probability of using private courses to prepare for the SAT was 9 points higher for African Americans compared to Whites (p < .05), which is consistent with Buchmann et al.’s (2010a) findings. In addition, South Asians were more likely than Whites to use a private tutor—their predicted probability of doing so was 14 points higher than that of Whites (p < .10). Contrary to earlier research on shadow education, however, household income was not significantly related to students’ highest level of test preparation. Buchmann et al. (2010a) found that students from families with higher incomes were more likely to use the most intensive and most costly forms of test preparation (i.e., private courses and private tutors). Yet, we did not find this to be the case. We defined private courses in our study to include both in-person and online courses, which is consistent with the current state of shadow education but is necessarily broader than what scholars would have encountered in earlier cohorts. As we discussed earlier, commercial test preparation companies expanded in recent years to provide online versions of their courses; these courses are often asynchronous and not capped by a physical class size, and thus tend to be relatively affordable. Additionally, as we show next, many students who used tutors to assist with their schooling (not just SAT/ACT preparation) used free tutors, and this may have been the case for test preparation as well.

Although we found relatively few differences by race/ethnicity and income, some other notable findings emerged with regard to the controls. The strongest predictors of test preparation were indicators of accelerated learning and disability status. Both groups of students were more likely to use more intensive forms of test preparation. Specifically, the predicted probability for students in accelerated courses was 11 points higher than for students who had never been enrolled in such courses (p < .01). Students with a diagnosed disability were also more likely to take a private course (11 points higher; p < .01) and use a private tutor (8 points higher; p < .05) than students without a disability.12 Finally, students who lived in urban areas and those who attended private schools used higher levels of test preparation than their relevant counterparts.13

The second section of the table uses a slightly different conception of test preparation but yields similar results with respect to race/ethnicity and income. When private versus public forms of shadow education are distinguished, we see that South Asians were less likely than Whites to use only private test preparation, but were more likely than Whites to use both public and private (both p < .05). Specifically, their predicted probability of using only public forms of test preparation was 4 points lower than that of Whites, but their predicted probability of using both public and private test preparation services was much higher (18 points) than that of White students. African Americans’ probability of using a combination of public and private forms of test preparation was 10 points higher than that of Whites (p < .05). As before, household income was not associated with test preparation, whether public or private. We also found that accelerated learning students, students with a diagnosed disability, and students with an IEP were more likely to use both public and private test preparation (all at least p < .05).14 Students who lived in urban areas and those who attended private schools tended to use both public and private forms of test preparation.

4.2.2. Supplemental online education and private tutoring

We next expand our analysis beyond test preparation to examine two additional forms of shadow education—supplemental online education and private tutoring—and their use during the Covid-19 pandemic. The first section of Table 3 shows AMEs for free (column 1) and paid (column 2) online education, which are derived from a multinomial logistic regression model. As shown in the first column, South Asians (p < .01), and African Americans (p < .05) were more likely than Whites to use free online education during the pandemic. Compared to White students, the predicted probability of using free online education was 13 points higher for South Asians and 6 points higher for African Americans. In the second column, however, we see that these groups were no more likely than Whites to use paid online education. Latinx students were less likely than White students to use paid online education: their predicted probability of using paid online education was 5 points lower than that of Whites (p < .10). As before, household income was unrelated to the use of online education, either paid or free. Younger children also were more likely than older children to use both free and paid online education (both at least p < .05), suggesting that these resources were most often developed for and/or used by students in elementary school.

Table 3.

Race, income, and other demographic predictors of online education and tutoring (N = 1657).

	Online education		Tutoring
	Free (versus None)	Paid (versus None)	Free (versus None)	Paid (versus None)
East Asian	0.067	-0.016	0.057*	0.042
	(0.041)	(0.031)	(0.027)	(0.028)
Filipino	0.010	-0.037	0.019	-0.021
	(0.045)	(0.036)	(0.024)	(0.030)
South Asian	0.127**	-0.031	0.027	0.042
	(0.049)	(0.032)	(0.025)	(0.030)
Other Asian	0.063	-0.032	0.011	0.062
	(0.052)	(0.041)	(0.024)	(0.042)
Black	0.059*	-0.016	0.066***	0.032
	(0.029)	(0.025)	(0.015)	(0.021)
Latinx	0.035	-0.045+	0.057***	0.034
	(0.028)	(0.023)	(0.015)	(0.021)
Household income	-0.002	0.001	-0.004*	0.005**
	(0.002)	(0.002)	(0.002)	(0.001)
Parent AA/Some coll	-0.008	-0.054*	0.009	-0.010
	(0.028)	(0.024)	(0.015)	(0.020)
Parent BA+	0.015	-0.009	0.012	0.054*
	(0.031)	(0.027)	(0.017)	(0.023)
Parent(s) work FT	-0.014	0.029+	0.001	0.013
	(0.021)	(0.016)	(0.013)	(0.016)
Child is male	-0.009	0.002	0.009	-0.005
	(0.020)	(0.015)	(0.012)	(0.015)
Child is non-binary	0.160	0.007	-0.058***	-0.121***
	(0.162)	(0.108)	(0.008)	(0.010)
Child in grades 6–12	-0.039+	-0.061***	0.004	-0.036*
	(0.022)	(0.017)	(0.013)	(0.016)
Parent is married	-0.031	0.013	-0.005	-0.024
	(0.023)	(0.018)	(0.013)	(0.017)
Number of children	0.002	0.012+	0.001	0.008
	(0.009)	(0.007)	(0.005)	(0.007)
Suburban	0.022	-0.046*	-0.020	-0.038*
	(0.022)	(0.018)	(0.013)	(0.017)
Rural	-0.002	-0.103***	0.001	-0.079***
	(0.030)	(0.020)	(0.019)	(0.021)
South	0.004	-0.013	0.016	0.010
	(0.021)	(0.016)	(0.012)	(0.016)
Prior achievement	-0.002	0.026*	-0.007	0.004
	(0.015)	(0.012)	(0.008)	(0.011)
Accelerated classes	0.017	0.077***	0.002	0.104***
	(0.022)	(0.017)	(0.013)	(0.017)
Disability	0.028	0.032+	-0.002	0.056**
	(0.025)	(0.019)	(0.015)	(0.018)
IEP	.004	0.074***	0.040*	0.066***
	(0.030)	(0.021)	(0.016)	(0.020)
Expect college	0.093*	0.120**	0.008	0.044
	(0.036)	(0.038)	(0.018)	(0.028)
Discuss w/ child	0.011	-0.002	0.010*	0.000
	(0.007)	(0.006)	(0.004)	(0.005)
Private school	-0.047	0.112***	-0.002	0.147***
	(0.032)	(0.032)	(0.022)	(0.033)
Charter school	0.087*	0.007	0.017	0.016
	(0.042)	(0.027)	(0.024)	(0.027)
School in-person	-0.049+	0.037*	0.023+	0.030+
	(0.025)	(0.017)	(0.013)	(0.017)
N	1657		1686

+ p < .10, * p < .05, ** p < .01, *** p < .001 (two-tailed tests).

Source: Authors’ original data collected through Qualtrics panels.

Note: Multinomial logistic regressions; average marginal effects (AMEs) shown. Sample is restricted to students who are not homeschooled. Omitted categories are White; parent attained high school diploma or less; parent(s) not working full-time; child is female; respondent parent is not married; urban residence; does not reside in the South; no enrollment in accelerated classes; no disability; no IEP; parent does not expect child to attend college, doesn’t know, or is unsure; public school; school not fully in-person.

The second section of Table 3 shows AMEs for free (column 3) and paid (column 4) tutoring, which are again derived from a multinomial logistic regression model. Compared to White students, the predicted probability of using free tutoring (but not paid tutoring) was 6 points higher among East Asians (p < .05), 7 points higher among African Americans (p < .001), and 6 points higher among Latinx students (p < .001). In addition, household income was a significant predictor of tutoring during the pandemic. Children from families with higher household incomes were less likely to use free tutoring than their lower-income peers (p < .05), but they were more likely to use paid tutors (p < .01). These effects are substantively minor, however, especially compared to the effects of race/ethnicity; a $10,000 increase in household income was associated with only a 0.4-point decrease in the probability of using free tutoring, and a 0.5-point increase the probability of using paid tutoring. Parental education also was a significant, positive predictor of paid tutoring. Students whose parents had at least a bachelor’s degree were about 5 points more likely to use a paid tutor, relative to those whose parents had less than a bachelor’s degree.

In addition, younger students, urban students, and those who attended private schools were more likely than their relevant counterparts to use paid tutors. Students with an IEP were more likely to use both free and paid tutors than those without an IEP, and students enrolled in accelerated courses were more likely to use paid tutoring, as were students with a disability. Although these patterns run counter to the common belief that tutoring is remedial, prior research on shadow education shows that in many parts of the world, private tutoring is most common among families who want to maintain or further enhance their children’s position in a competitive society (Bray, 2021). Our findings likewise indicate that families may use private tutoring not only if their children are struggling in school, but also if they want to further enhance their children’s progress.

In summary, we found racial/ethnic differences in shadow education during the first year of the pandemic that were broadly consistent with prior research on shadow education (Alon, 2010, Buchmann et al., 2010aaa). Asians and African Americans were the most likely to use test preparation services, particularly commercial learning centers and private tutoring, as well as other types of shadow education such as free online courses and free tutoring. Contrary to prior research, however, we found little evidence of a social class gradient in shadow education during the pandemic; household income and parental education were associated only with paid tutoring. This pattern is consistent with the idea that shadow education has become broader and more accessible in recent years, with lower-SES families having more opportunities for shadow education than in previous cohorts. Across all outcomes, the most consistent predictors of shadow education were measures of coursetaking and disabilities. Students enrolled in accelerated learning, and students with a disability and/or an IEP, were more likely to participate in more intensive forms of test preparation, as well as other shadow education such as online courses and tutoring.

4.2.3. Learning pods

As we discussed earlier, learning pods were a unique form of shadow education that emerged during the pandemic to support children’s experiences with remote schooling. Table 4 shows AMEs for learning pod attendance, derived from a logistic regression. Our findings point to some similarities between learning pods and other forms of shadow education with respect to demographic correlates, but we also found some key differences that are important for explaining the purpose and potential impact of pods.

Table 4.

Race, income, and other demographic predictors of pod attendance (N = 1894).

East Asian	-0.059+
	(0.031)
Filipino	-0.028
	(0.040)
South Asian	0.011
	(0.039)
Other Asian	-0.027
	(0.045)
Black	0.069**
	(0.026)
Latinx	0.031
	(0.025)
Household income	0.003
	(0.002)
Parent attained AA/Some college	-0.056*
	(0.026)
Parent attained BA or more	-0.076**
	(0.029)
Parent(s) work full-time	0.048*
	(0.019)
Child is male	-0.001
	(0.017)
Child is non-binary	-0.109
	(0.076)
Child is in grades 6–12	-0.076***
	(0.019)
Parent is married	-0.003
	(0.019)
Number of children	0.007
	(0.008)
Suburban	-0.047*
	(0.020)
Rural	-0.064*
	(0.026)
South	-0.026
	(0.018)
Prior achievement	0.022+
	(0.013)
Accelerated classes	0.092***
	(0.019)
Disability	0.024
	(0.021)
IEP	.126***
	(0.023)
Expect college	0.051+
	(0.027)
Discuss academics with child	-0.006
	(0.007)
Private school	0.048
	(0.035)
Charter school	-0.005
	(0.032)
Homeschool	0.052
	(0.032)
School fully in-person in 20–21	0.017
	(0.021)

+ p < .10, * p < .05, ** p < .01, *** p < .001 (two-tailed tests).

Source: Authors’ original data collected through Qualtrics panels.

Note: Logistic regression; average marginal effects (AMEs) shown. Sample is restricted to students who are not homeschooled. Omitted categories are White; parent attained high school diploma or less; parent(s) not working full-time; child is female; respondent parent is not married; urban residence; does not reside in the South; no enrollment in accelerated classes; no disability; no IEP; parent does not expect child to attend college, doesn’t know, or is unsure; public school; school not fully in-person.

Generally speaking, we found that children who were enrolled in pods were disproportionately African American; had parents who were less educated and worked full-time; were in elementary school; and lived in urban areas. These patterns run contrary to the narrative that pods were a method of opportunity hoarding during the pandemic, and instead suggest that pods were a lifeline for parents who needed to work in-person and could not supervise their young children during the school day. Specifically, as with test preparation, online education, and tutoring, the predicted probability of pod attendance among African American students was 7 points higher than that of White students (p < .01). East Asians were marginally less likely, by 6 points, than Whites to use pods (p < .10). Parents’ education was negatively associated with pods use: compared to parents who had a high school diploma or less, the probability of putting their child in a pod was 6 points lower among parents who completed some college, and 8 points lower among parents who completed a bachelor’s degree or more. Parents who worked full-time also tended to enroll their children in pods, as did parents of younger students (i.e., those in grades K-5), and those who lived in urban areas. To reiterate, these findings provide further evidence that learning pods were not concentrated among higher-income, privileged families—something that concerned scholars and other commentators in the early stages of the pandemic. Instead, these findings, in combination with the descriptive statistics we discussed earlier, suggest that pods were perhaps less about hoarding opportunities and more about providing sorely needed childcare and supervision.

That said, we also found that children who had accelerated learning opportunities and those with stronger prior achievement were more likely to be enrolled in pods, which was consistent with our findings for test preparation, online education, and tutoring. This was also the case for students with an IEP. These patterns suggest that, to some extent, learning pods—like other forms of shadow education—may have been an important resource for families who were concerned about possible learning loss and other challenges that emerged due to remote schooling.15

5. Discussion

The purpose of this study was to examine racial/ethnic and social class patterns in shadow education use during the pandemic. Specifically, we assessed SAT and ACT preparation activities, which has been the focus of much research on shadow education in the United States, as well as online education and tutoring. We also make the case that pandemic learning pods should be considered a form of shadow education and examined whether race and social class patterns in pod use were similar to that of more traditional forms of shadow education.

Our results also indicated that existing patterns of racial/ethnic differences in shadow education persisted during the pandemic. Consistent with prior research, African American students were more likely than Whites and other racial/ethnic groups to use higher levels of SAT/ACT preparation services. They also were more likely to use free online supplementary education and private tutoring for general academic enrichment. South Asian youth were more likely than Whites to utilize free online supplementary education. Whether this is because marginalized groups feel that they need to do even more than Whites to secure a competitive edge, or whether they feel that the “mainstream” education system in the U.S. is failing them, our findings suggest that some families perceive the need for additional or alternative learning support outside the school walls.

Perhaps surprisingly given prior research, we did not observe social class differences in the use of test preparation or online education, but we did find that those with higher household incomes were more likely to use paid private tutoring. These findings are consistent with the idea that SAT/ACT preparation and other supplemental education opportunities have become more accessible in recent years, largely due to the internet and various efforts to make these materials free or low-cost for students. Additionally, with the onset of the COVID-19 pandemic, more families and schools became dependent upon online supplemental learning activities. Online learning websites like Khan Academy and commercial learning centers like Sylvan and Kumon quickly expanded to meet (and maybe even increase) the demand for online test preparation and learning services. Buchmann et al. (2010a), for example, used data from NELS-88, capturing shadow education outcomes among a cohort of students who were in eighth grade in 1988—long before the advent of the internet. Our results indicate that, today, students are accessing SAT/ACT preparation and online education regardless of their family’s economic standing. Paid tutoring (which, by definition, requires financial resources) remains out-of-reach for those in lower socioeconomic groups.

We also examined patterns of inequality in learning pods, which we argue was a form of shadow education that was utilized during the pandemic. Like tutoring and test preparation, we found that African Americans were more likely to use learning pods, and East Asians were less likely to use learning pods than White families, even after taking into account learning modality. Notably, these pods were primarily supplemental, in the sense that they did not replace schooling but instead were used for additional support and assistance during remote schooling. Thus, we argue that early concerns about pods exacerbating social class and race inequalities largely did not come to fruition. Instead, pods primarily were in large part used by parents who needed to pool and draw upon community resources for childcare, with only a small fraction of pods being used to bypass traditional schools.

While not the focus of this article, our findings additionally shed some light on the relationship between student disability status and shadow education. Across all forms of shadow education, students who have been diagnosed with a disability were more likely to participate in the more costly forms of shadow education, as well as learning pods. While this is certainly reflective of the additional supports needed to raise children with disabilities, particularly during the pandemic when school resources were limited, it also underscores the lack of resources in many schools and highlights the need to provide these supports for lower-income families who may not be able to afford to use shadow education to the same extent.

One limitation to keep in mind is the cross-sectional nature of this study and, relatedly, the long gaps in time between earlier research on shadow education and today. As we mentioned earlier, Buchmann et al. (2010a) influential research on shadow education used data from thirty years ago, before the internet delivered more cost-effective shadow education options to students and families. Although we suspect that the relationship between social class and shadow education genuinely has dissipated in recent years with the advent of the internet and increased efforts to provide more accessible and affordable options during the pandemic, more research is needed to know whether these findings represent periodic shifts or more sustained patterns.

Despite these limitations, our study contributes to a strong tradition of research on shadow education in the U.S. and worldwide. While focused on the 2020–21 school year, we suspect the implications of our study extend beyond the pandemic. The pandemic highlighted the limitations of schools, exacerbated by lack of funding and resources, and how swiftly private forms of education were able to step in and capitalize on the situation. Although some of these new forms of supplementary education were developed in response to the Covid-19 pandemic, these changes may well be long-lasting and contribute to the increasing privatization of education. In addition, these findings demonstrate the importance of examining shadow education beyond test preparation—especially as more colleges adopt test-optional policies or disregard SAT scores altogether. It remains to be seen whether SAT test prep remains a prominent form of shadow education, or if it will be replaced by other supplemental learning activities, to the extent that other components of the college portfolio become more important.

Appendix A. Supplementary material

Supplementary material

.