Unequal Adaptation to COVID Schooling

Posted:  https://www.nber.org/papers/w27555
By: National Bureau of Economic Research
Issued: July 2020, Revised: November 2020
Authors: Andrew Bacher-Hicks, Joshua Goodman, Christine Mulhern

ABSTRACT

This study uses high frequency internet search data to study in real time how US households sought out online learning resources as schools closed due to the Covid-19 pandemic. By April 2020, nationwide search intensity for both school- and parent-centered online learning resources had roughly doubled relative to pre-COVID levels. Areas of the country with higher income, better internet access and fewer rural schools saw substantially larger increases in search intensity. The pandemic will likely widen achievement gaps along these dimensions given schools’ and parents’ differing engagement with online resources to compensate for lost school-based learning time. Accounting for such differences and promoting more equitable access to online learning could improve the effectiveness of education policy responses to the pandemic. The public availability of internet search data allows this analysis to be updated when schools reopen and to be replicated in other countries.

INTRODUCTION

As schools across the United States closed in response to the Covid-19 pandemic, roughly 55 million K-12 students experienced a serious disruption to their school year. Though most schools quickly began offering some type of virtual education (Hamilton et al., 2020; Lake and Dusseault,
2020), there have been growing concerns about the effects of this unprecedented shift (Malkus, 2020; von Hippel, 2020). In particular, there are fears that low-income students will be unequally harmed by the shift to online learning, due to less access to online resources to compensate for lost
in-person instruction (Horowitz, 2020). As states and districts consider how to best educate students in the wake of the pandemic, it is critical to better understand the effect of pandemic-induced school closures on students’ access to online learning resources, particularly for low-income students.
This paper uses high frequency, nationally representative Google search data to document in real time how parents and students sought out online resources as schools closed in response to the Covid-19 pandemic.

Combining the online search measures with measures of demographic
characteristics by geography, we use a difference-in-difference strategy to estimate how Covidinduced demand for online resources varied by a range of geographic and socioeconomic factors, including income, internet access and school rurality. We document three findings new to the research literature. First, we show that pre-Covid search intensity for online learning resources can be usefully divided into two categories, which we call “school-centered resources” and “parent-centered resources”. School-centered resources are platforms typically used by schools to provide instruction, assign work, or communicate with students (such as Google Classroom or Schoology). Parent-centered resources are more generic
search terms likely indicating parents or students are seeking supplemental learning resources (such as home schooling or math worksheets). We show that search intensity for school-centered resources dwarfs that for parent-centered resources and that both follow the school calendar, peaking at the start of each school year and vanishing in the summer. Second, we show that the onset of Covid dramatically disrupted this usual school calendar
cycle of search intensity, as the pandemic triggered a very large increase in demand for online learning resources. By April 2020, nationwide search intensity for online learning resources had roughly doubled relative to pre-Covid levels. We find sharp increases in searches for both school and parent-centered resources, suggesting that increased demand for online support came not only from schools shifting their mode of instruction but also from parents and students seeking additional support to fill learning gaps as schools closed. Third, we show the pandemic substantially widened socioeconomic gaps in searches for online learning resources. Search intensity rose twice as much in areas with above median socioeconomic
status (measured by household income, parental education, and computer and internet access) as in areas with below median socioeconomic status. Search intensity for school-centered resources, for example, increased by 15 percent for each additional $10,000 in mean household income and by
roughly 5 percent for each percentage point increase in the fraction of households with broadband internet and a computer. Areas with more rural schools and Black students saw lower increases in search intensity. Socioeconomic gaps widened both between and within the country’s four Census regions (Northeast, Midwest, South, and West). We also show that changes in search behavior correlate with changes in students’ actual math progress, suggesting online search metrics may be a useful proxy for educational actions taken by parents and students.

Our work adds to three strands of the research literature. First, our paper shows that internet search behavior can provide useful, real-time information about education-related actions being taken by households. Prior work shows the utility of search data in predicting economic and social outcomes such as parents’ preferences for schools (Schneider and Buckley, 2002), disease spread (Polgreen et al., 2008), consumer behavior (Choi and Varian, 2012), voting (Stephens-Davidowitz, 2014), and fertility decisions (Kearney and Levine, 2015). Most recently, Goldsmith-Pinkham and Sojourner (2020) use the volume of online search for unemployment benefits to predict post-Covid unemployment claims. Our results suggest that search data contain valuable information about how households react to educational shocks, both in terms of overall use of educational resources
and in heterogeneity in such usage by socioeconomic characteristics.
Second, we measure a new aspect of the digital divide, namely the extent to which households seek out online learning resources either prompted by their schools or of their own accord. A large literature documents pre-Covid gaps in access to and proficiency in the use digital technologies
by income, education, and family background (Bucy, 2000; Rice and Haythornthwaite, 2006; Jones et al., 2009; Vigdor et al., 2014). Multiple post-Covid surveys show consistent socioeconomic gaps in self-reported engagement with remote learning at a single point in time (Barnum and Bryan, 2020). We complement this evidence with the first nationally representative revealed preference measure of such engagement, based on households’ actual behavior rather than self reports. Ours is also the first high frequency data brought to bear on this issue, allowing study of the evolution over time of engagement with online learning resources.
Third, we provide some of the clearest evidence on one channel through which the Covid-19 pandemic will likely widen socioeconomic educational gaps. Based on prior estimates of school closure effects from natural disasters and summer months, Kuhfeld et al. (2020) predict that Covidinduced closures will generate substantial learning losses, with the largest negative effects concentrated among low-achieving students. Aucejo et al. (2020) surveyed university students and find the pandemic lowered on-time graduation rates and job offers, with larger effects among low income students. Using data from one online learning platform, Chetty et al. (2020) provide perhaps the only direct measure of Covid-induced learning loss, showing that low-income students experienced substantially larger and more persistent reductions in learning progress relative to
high-income students. We show that socioeconomic gaps in engagement with online learning resources are not limited to a single platform or location but are a widespread and fundamental feature of the post-Covid landscape. Accounting for household responses to changing school inputs
will be critical for predicting educational effects of the pandemic and policy responses to it, given evidence that parental and school investments are often substitutes, both in the US (Houtenville and Conway, 2008) and developing countries (Das et al., 2013; Pop-Eleches and Urquiola, 2013).
Our findings provide insight into the mechanisms underlying learning losses that are beginning to emerge following pandemic-induced school closures and can help inform future policy responses to schooling disruptions, whether related to the pandemic or not. That search for school- centered resources increases more in high income areas suggests either that those areas’ schools are using online platforms more, that those areas’ parents are more likely to engage with such platforms, or both. That search for parent-centered resources increases more in high income areas suggests that, separate from schools’ actions, parents are differentially likely to seek out their own ways of compensating for their children’s lost learning time.
These results can help policymakers and school leaders formulate more effective responses to the educational disruptions caused by Covid-induced school closures. Students from lower income families and schools may require additional attention and resources given lower engagement with online learning resources during spring 2020. Moreover, because remote learning will likely remain a central piece of the public education system for the foreseeable future (Cleveland, 2020), preventing the widening of achievement gaps may require improving access to home computers and broadband internet for low income and rural students. Schools may also need to improve the deployment of remote learning platforms to more equitably engage students and parents in the use of those platforms.
Whether efforts to close gaps in online learning engagement succeed will only become clear as new data become available in subsequent school years. One advantage of using publicly available search data to measure household behavior is that our analyses can be easily updated in real time
when the school year begins in the fall. This will help reveal whether socioeconomic gaps in engagement with online learning have narrowed since the initial shock of schools closing or if different remote learning strategies across regions were particularly successful. In addition, the
set of search terms studied can be easily modified to accommodate new online learning resources as they emerge. Finally, our analyses can be replicated in other countries, particularly ones large enough to generate search data at sub-national levels such as provinces and cities. The flexibility of this approach shows promise for understanding the behavioral responses of households to school closures and developing policy responses in real time to address changing student needs.

DISCUSSION

We document a sharp increase in searches for learning resources as schools closed in response to the Covid-19 pandemic. By April 2020, nationwide search intensity for online learning resources had roughly doubled relative to baseline. The shock of school closures increased demand both for the specific online platforms schools shifted instruction to (such as Google Classroom) and for the supplemental resources that households sought out to fill gaps in their learning (such as math worksheets). The likelihood of future school closures or partial reopenings implies these supplemental online resources are likely to become important drivers of student learning.
Though demand for online resources increased in both high and low SES areas, the increase was substantially larger in high SES areas. Areas of the country with higher income, greater internet access, and fewer rural schools had substantially larger increases than less advantaged areas.
Along with results from several contemporaneous studies, these results suggest that academic gaps across students will be wider than normal in future school years, a result strengthened by our finding that changes in search behavior correlate with changes in students’ math progress.
Our results suggest the potential value of policy responses that directly address these documented inequalities in engagement with online learning resources. Students in low SES areas and rural communities are likely to need additional support to overcome the educational challenges created by Covid-19. Because online learning will likely remain a key component of school systems in the near future, school leaders and policymakers may want to prioritize access to home computers and broadband internet. Improving access to and engagement with online learning platforms will likely be an important step to equalizing learning opportunities and preventing a widening of achievement gaps. Publicly available, high frequency internet search data helps illuminate the evolution of educational choices made by households, as well as socioeconomic inequalities in those choices. Our analyses can be updated in real time to study future changes in engagement with online learning, can be modified to study different search terms, and can be replicated in other countries. Household adaptation to schooling shocks is an understudied phenomenon that can be readily observed in internet search data. Understanding and accounting for such behavioral responses by parents and students will be critical to predicting the long-term effects of the pandemic.

For access to the full article go to: https://www.nber.org/papers/w27555