How COVID-19 altered perceived household resource consumption in the United States: Results from a survey

The COVID-19 pandemic has led to unprecedented changes in the daily lives of people in the United States and across the world, particularly around how households consume critical resources. We fielded a survey to a national U.S. sample (n = 1214) to quantify the nature and extent of perceived change in household consumption of energy, water, information and communication technology (ICT) services, transportation, and grocery and non-grocery shopping during the COVID-19 pandemic. We find that most survey respondents report increased consumption of electricity and water, but they did not report altered heating and cooling energy consumption. Respondents reported sharp increases in work-related ICT usage for medium- and high-income respondents, and increased ICT usage for recreation and entertainment across all income categories during the pandemic. Nearly half of car-based commuters stopped commuting, with higher shares of medium- and high-income respondents shifting to working from home instead of commuting by car. Respondents shopped less frequently, spent more on both grocery and non-grocery items, and moved their shopping online—indicating that the pandemic hastened the ongoing shift to online modes of resource consumption. Low-income households and respondents of color reported different shifts in consumption in comparison to higher income households and white households. Finally, more than half of our respondents stated that their altered consumption patterns will persist post-pandemic, pointing to potential long-term shifts in consumption behaviour. These findings provide useful empirical evidence for perceived changes in household resource consumption during the pandemic, suggest that we need to better protect low-income and Black households from the effects of large-scale disruptions, and can inform more effective and equitable disaster response policies.


Introduction
The health, social, and economic disruptions associated with the COVID-19 pandemic have led to unprecedented changes in ways of living for much of humanity. Nearly two-thirds of the global population has faced different levels of lockdown measures during this crisis (Hale et al 2021), forcing many households to reimagine the home as a single space serving all their needs for work, school, entertainment, recreation, shopping, and other activities. In the United States, 94% of the population faced some form of shelter-in-place orders at the peak of the pandemic (Secon 2020), with a gradual patchwork of reopening policies in different states as vaccines are disseminated and new variants of the virus emerge. The resulting changes in household resource consumption, especially of common goods and services, have been sudden yet prolonged. There is much uncertainty around the nature and extent of these changes and in their persistence in the short and long terms, especially given continued ambiguity in future COVID-19 trends and lockdown policies.
Quantifying how the consumption of household resources, namely energy, water, information and communication technology (ICT) services, transportation, and shopping has changed during the pandemic can be critical to (1) informing public policies for effective disaster response, (2) prioritizing investments of limited public funding, and (3) creating long-term strategies to manage persistent shifts in household consumption and their environmental impacts. Understanding the nature and extent of changes in household energy consumption due to the pandemic is an evolving area of research. Recent work has shown that overall electricity demand decreased during the pandemic, especially in the commercial, retail, and industrial sectors (Hinson 2020, Buechler et al 2022. Studies for specific locations, like in Austin (TX), have shown increased residential energy use (John 2020), but this increase does not make up for the decrease in other sectors. It is not just energy consumption within the home that has changed, but also the consumption of water, ICT services, transportation, and shopping for grocery and non-grocery items. Nearly 42% of the U.S. labor force shifted to working from home full-time during the pandemic (Bloom 2020). Many organizations had to adapt to a digital workplace rapidly, making their workers dependent on ICT services that not only consume energy for the ICT devices themselves, but also throughout the entire communication network. Workers now rely heavily on videoconferencing applications and remote connections to servers that host work-related software (Koeze and Popper 2020). While the long-term effects of this shift are still being understood (Gorlick 2020), preliminary estimates show a 40%-70% increase in home broadband usage (PCMag 2020).
The shift to remote work has implications for transportation energy use and emissions as well-early research found a nearly 50% reduction in total commutes in major U.S. metropolitan areas (Klein et al 2020), and a 43% decrease in total vehicle miles traveled (VMT) as the pandemic began (Doucette et al 2020). VMT has since recovered in many cities (Olin 2020), but there might be permanent shifts towards remote or hybrid forms of work, especially for households whose occupations enable that transition.
While U.S. households across the socioeconomic spectrum have been impacted by the pandemic, there is now much concern about the potentially disproportionate negative impacts for low-income households and people of color (Brosemer et al 2020). Addressing issues of distributional justice and equity in resource consumption and providing adequate crisis response depends on a robust understanding of the differential implications of shocks on different types of households and the ways in which vulnerable sections of the population are likely to be impacted. This is true not just for governments putting together disaster response policies, but also for private sector operators managing supply chains. In the past, the measurement and quantification of gaps in access and affordability of key household resources has informed remedial policies at the federal, state, and local levels. Programs such as the Low Income Home Energy Assistance Program (LIHEAP), Supplemental Nutrition Assistance Program (SNAP), and state-level transit subsidies have attempted to address inequitable access to energy, food, and transportation respectively (Tiehen et al 2012, Bednar and Reames 2020, FTA 2021. During the COVID-19 pandemic, due to increased household energy consumption, energy security is a major concern for lowincome households who spend 5.4%-9% of their earnings on energy needs compared to 3.5% across the population (Graff and Carley 2020). Such effects can be exacerbated by greater job losses suffered by the same households during the pandemic (Parker et al 2020). At the same time, the shift to remote work and schooling necessitates access to affordable internet connections and internet-connected devices for all. The Emergency Broadband Benefit program, for instance, offers broadband subsidies to qualifying households during the pandemic (FCC 2021). There is thus a need to understand not just the overall impacts, but to explore the equity and distributional implications of these sudden shifts in resource consumption.
Given the continued disruptions from current and future waves of the pandemic, and to better prepare the United States and other nations for similar natural or human-induced catastrophes, there is a need for a comprehensive survey-based assessment of how households perceived changes in their consumption of critical resources during the pandemic. This study thus aims to answer two primary research questions: 1. What is the nature and extent of changes in perceived household-level consumption of critical resources-energy, water, internet, transportation, grocery, and non-grocery items-during the COVID-19 pandemic in the United States? 2. How do these changes in household resource consumption vary by key demographic characteristics such as income and race?
The novelty of this work is two-fold: first, while previous work has studied the impact of the pandemic on a single type of household consumption in isolation, we study changes across a broad range of resource categories for each household. We also study how persistent the respondents believe these changes will be post-pandemic. Second, our study aims to identify whether specific groups suffered disproportionate and inequitable changes in resource consumption during the pandemic. We developed and fielded a survey with 1214 U.S. respondents in late May 2020, when stay-at-home orders had been in effect long enough to enable an understanding of their impact, but there was continued uncertainty around its duration and long-term effects. Table 1 summarizes key demographic characteristics of the survey sample.

Data and methods
We collected survey data through an online survey form created on Qualtrics and deployed through two online platforms-Amazon Mechanical Turk (Amazon 2020) (henceforth, MTurk) and Prolific (Prolific 2020). Online surveys offer researchers the advantage of reaching many respondents within a shorter timeframe and at lower cost compared to inperson surveys. This was especially important during the COVID-19 pandemic, where shelter-in-place orders prevented physical interaction between surveyors and respondents due to public health concerns. We conducted the survey in late May 2020 through an online questionnaire accessible through laptop or desktop computers, smartphones, and tablets. The survey contained six main sections including energy use, water use, ICT use, transportation, shopping, well-being, and other issues in addition to questions on demographics (the full survey questionnaire is provided in the SI). The respondents were paid $5 if they started the survey and completed it and $2 if they started the survey and did not complete it. We acknowledge that the household-level consumption patterns reported by survey respondents are based on their perceived consumption, which is subjective by nature and can differ from actual consumption (Attari et al 2010, Lesic et al 2019. A total of 1250 responses were collected. Data were collected for a U.S. national sample of 650 respondents through MTurk and for a U.S. national sample of 600 respondents through Prolific matching U.S. Census benchmarks for age, sex, and ethnicity. Respondents were free to choose either 'do not know' or 'do not wish to answer' in response to any question. After data cleaning, a total of 1214 responses were analyzed using R Statistical Software (v3.6.1) (R Core Team 2019). Regression analysis for the anticipated changes in resource consumption was conducted using logistic regression models, which are best suited to the multi-level ordinal variables obtained from the survey. Regression models were run using the 'clm' function from the 'ordinal' package (Christensen 2019), and using the stepAIC function from the 'MASS' package in R (Venables and Ripley 2002).
Table S1 (SI) summarizes key demographic characteristics of survey participants, disaggregated by the data source. The comparable national statistic from the U.S. Census Bureau is provided where available. The state-level share of survey respondents closely matches the national population distribution, with the maximum difference between the two shares being 2% (see SI table S5) (U.S. Census Bureau 2019). We also present some key results separately for each sample in the SI. The findings are consistent across samples.

Energy services
More than half (58%) of the respondents reported an increase in electricity consumption during the pandemic (figure 1), with a larger share of medium-(63%) and high-income (73%) households reporting an increase compared to low-income (51%) households. A third each of the respondents with no income (note the small n of 12) reported an increase or decrease in electricity consumption, in alignment with recent research (Memmott et al 2021). By race, Black and Asian households are more likely to report an increase compared to White households.
Home heating and cooling accounts for about half of U.S. residential energy consumption (EIA 2018), making it critical to understand how their usage changed. Of the respondents with heating at home, about 75% of them did not change their heating and cooling behavior (see SI, figures S1 and S2). A slightly larger share (80%) of respondents with no household income (note the small n of 12) reported no change in typical hours of heating compared to low (77%), medium (71%), and high (78%) income respondents. We note that the shelter-in-place orders at the time of the survey coincided with the seasonal change from winter to spring/ summer, which is typically marked by mild and ple n asant weather. We also asked respondents about household appliance ownership and frequency of use before and during the pandemic. We find that respondents increased their usage of most appliances, as shown in SI figure S3.

Water services
More than half of the respondents (51%) reported increased water usage during the pandemic (see SI, figure S4). Medium and high-income respondents are more likely to perceive an increase in water consumption compared to low-income respondents. Among households with no income (note the small n of 12), more than 20% of respondents report decreased water consumption, suggesting a potential equity concern. Asian respondents are more likely to report an increase, while Black respondents are more likely to report a change (either increase or decrease) in their water consumption.

Information and communication technologies (ICT)
ICT infuses every sector of the U.S. economy, from work-related software and online learning to digital civic services and telemedicine. The pandemic induced a sharp increase in ICT usage as well as the accelerated adoption of services such as videoconferencing (Koeze and Popper 2020). Our survey asked about hours spent on various activities on ICT devices at home before and during the pandemic by respondents and the children in their household.
Since it is difficult for a respondent to precisely specify the hours spent on ICT activities by other adults, the survey asked about the broad similarity in ICT usage between the respondents and other adults in the household. Figure 2(1) shows the typical daily hours (x-axis) spent on work-related software, videoconferencing, and video streaming activities (y-axis, right) on ICT devices at home by respondents (see SI figure S5 for all ICT activities). Not surprisingly, hours spent on ICT devices increased across all activities. For instance, more than half of the respondents (54%) spent no time on videoconferencing before, whereas the vast majority (72%) used videoconferencing daily during the pandemic.
Work-related ICT use varied by income-lowincome respondents saw very similar software use before and during the pandemic while we observe higher daily hours of software use for medium and high-income respondents. Nearly 40% of lowincome respondents continued to spend zero hours on software use during the pandemic, compared to middle (21%) and high income (17%) respondents. Similarly, while 18% of low-income respondents spent 1-3 h on videoconferencing during the pandemic, the share was much higher for medium (31%) and high (44%) income respondents. This likely reflects the nature of jobs, where medium and high income households with more digitalized jobs could work from home but low-income households were more likely to be physically present at their workplace or potentially stop working during the pandemic (Kantamneni 2020).
Increased hours are reported not just for work but also for recreation and entertainment activities such as social media, online shopping, and video streaming. For example, while the share of respondents that video-streamed more than 3 h daily before the pandemic was 30%, it grew to 46% during the pandemic. In contrast to work-related ICT use, there are no major differences in recreational and entertainment ICT use by income category. Figure 2(2) shows the typical daily hours (x-axis) that children spent on online learning, videoconferencing, and video streaming activities (y-axis, right) on ICT devices (see SI figure S6 for all ICT activities by level of schooling). During the pandemic, a greater share of children from medium-and highincome households spent more time on online learning compared to low-income households. For video conferencing and streaming, there are no major differences across groups. Finally, a larger share of children from Asian households spent more hours on online learning and videoconferencing compared to children from White and Black households. Figure 3(A) shows that the number of people using cars to commute to work changed from 47.8%   (n = 581) of respondents before to 27.3% (n = 331) of respondents during the pandemic, while the share of respondents working from home increased from 14.3% before to 39.1%. Further, we find that 13% of people who were previously commuting did not commute during the pandemic because they were no longer working (see SI figure S7 for travel time by primary travel mode before and during the pandemic). Figure 3(B) shows the travel mode during the pandemic for respondents who commuted by car before the pandemic, by income category. While 40% of medium-income and 50% of high-income respondents who commuted by car before switched to working from home during the pandemic, only 25% of low-income workers were able to do so. A larger share (11%) of low-income respondents who commuted by car before the pandemic reported 'Not working' during the pandemic, compared to 7% and 1% of medium-and high-income respondents.

Grocery and non-grocery shopping
Respondents shopped for groceries less frequently during the pandemic but spent more on groceries monthly. Given the necessity of avoiding social contact during the pandemic, this decrease in shopping frequency is in keeping with recent research (Shamim et al 2021). Importantly, the shelter-in-place orders led to a shift to online shopping (figure 4) for both grocery and non-grocery items, with the share of respondents who shopped entirely in-store for groceries decreasing from 64% before to 38% during the pandemic. SI figure S8 shows the change in frequency of grocery shopping and figure S9 shows typical spending as share of monthly income by resource category before and during the pandemic.

Perceived persistence of altered consumption
51% of survey respondents either agreed or strongly agreed that their altered consumption patterns for various goods and services will persist after the COVID-19 pandemic is under control, while 19% disagree, and 24% are neutral (neither agree nor disagree). Importantly, the perceived persistence is similar across income and race categories (see SI figure S10).

Regression analysis
In this work, we rely primarily on the descriptive statistics shown in previous sections to understand changes in perceived consumption. In addition, we test a series of logistic regression models to understand associations between perceived changes in resource consumption as well as perceived persistence of behavior and demographic characteristics of survey respondents. The general model equation is as follows: logit (P (Y ⩽ j)) = β j0 + β j1 x 1 + · · · + β jp x p where the dependent variable Y is an ordinal variable (j = 1, . . ., j−1) and β are the coefficients for the p demographic predictors (x p ). The dependent variable Y for an anticipated change in a resource consumption category (electricity use, water use, ICT use, grocery shopping frequency, and monthly grocery spending) is a 3-level ordinal variable ('Decrease' , 'No change' , 'Increase') and for perceived persistence of change is a 3-level ordinal variable ('Disagree' , 'Neutral' , 'Agree'). Default categories for the demographic indicators are as follows: Stay-at-home: 'No' , Race: 'White'; Income level: 'Low'; Education: 'High school or less'; Area type: 'Urban' . After running the full regression models with all demographic indicators, we conduct backward stepwise selection to identify the models that minimize Akaike Information Criterion (AIC), explaining the most variation in the data while penalizing the use of excess parameters.
The results of our models are succinctly summarized through odds ratios in figure 5, which presents the models with the optimal set of features identified through stepwise selection. The dots in figure 5 represent the odds ratios and the whiskers represent the 95% confidence intervals. Full regression estimates for the optimized models are presented in SI table S2 and for the models with all demographic indicators are presented in SI table S3. For all models except grocery spending, the AIC for the selected model is lower than that of the null model, suggesting that demographic variables can help explain the variation in the dependent variable. For grocery spending, we find that the null model is best, meaning that demographic  SI table S2. variables do not help explain the variation in grocery spending, so this model is excluded from figure 5. We provide a brief interpretation of significant model coefficients below, but we wish to emphasize that while some of the demographic characteristics are statistically significantly associated with the dependent variables, the models tested do not offer great performance as indicated by low Pseudo-R 2 values in SI  table S2. A general guide to the interpretation of figure 5 is as follows, similar to recent work (Memmott et al 2021): the regression coefficient in a logistic regression is the estimated increase in the logarithmic odds of the dependent variable per unit increase in the independent variable. Therefore, the exponential function of the regression coefficient is the odds ratio associated with a unit increase in the independent variable (if it is a continuous variable) or with the specific category of the independent variable compared to its default category (if it is a categorical variable).
As illustrated in figure 5(A), respondents are more likely to anticipate an increase in electricity use if staying at home, with higher number of household members, and with higher income and education levels, controlling for all other independent variables respectively. They are less likely to anticipate increased electricity use if they live in a rural area. Respondents are more likely to anticipate increased water use if staying at home, with higher number of household members, and with higher education levels, controlling for all other independent variables respectively.
Respondents are more likely to anticipate increased ICT use if staying at home, if they selfidentify as White or Asian, and at any income level other than low-income, controlling for all other independent variables respectively. Respondents are less likely to anticipate an increase in grocery shopping frequency with higher education levels, controlling for all other independent variables respectively. Finally, as illustrated in figure 5(B), respondents are more likely to agree that their altered usage patterns will persist with increasing household size, if they self-identify as Black or Asian, and if they have higher levels of education, controlling for all other independent variables respectively.

Discussion and conclusions
This study aimed to (1) provide comprehensive empirical evidence for the nature and extent of perceived changes in household consumption of critical resources during the COVID-19 pandemic, and (2) highlight how the changes differed for different sections of the U.S. population. The insights from this study have direct policy implications that can help the United States and other regions create more equitable adaptation strategies for the ongoing pandemic as well as future large-scale disruptions.
First, in terms of energy consumption, mediumand high-income households were more likely to anticipate increased consumption, consistent with shifts to hybrid work and higher daytime occupancy of residences during the pandemic. From the electric grid perspective, there is a need to understand spatiotemporal shifts in load from commercial to residential sectors, the requisite upgrades in substations and distribution infrastructure to support the continued shift to hybrid work, and the ability of distributed energy resources at the household level to address energy inequities during a crisis. Further, while low-income households were less likely to report increased energy consumption, a majority of low-income households still reported increased consumption, pointing to the need for policy support to address exacerbated energy insecurity issues for these households (Graff and Carley 2020).
We find that ICT usage for middle-and highincome households for work-related activities was higher than that of low-income households not just during the pandemic but also before it, and the pandemic further exacerbated this inequity. So, while remote or hybrid work patterns might become more permanent for some, this shift is only possible for specialized and digitalized professions that employ relatively higher-income individuals (Muro et al 2017). The increase in school-related ICT usage for children suggests the need to provide children from disadvantaged households with access to both internet services and ICT devices. We can further assume that this study underestimates the ICT problem because the most affected households were likely unable to participate in the online survey due to lack of internet access. Given the persistent digital divide in the United States (Tomer et al 2018), continued support for affordable broadband access will be needed to bridge this divide.
In terms of transportation, low-income households continued to commute by car in higher proportions than medium and high-income households, many of whom shifted to working from home during the pandemic. In terms of shopping, the significant shift to online patterns of shopping suggest that the pandemic hastened a change that was already underway in the U.S. retail sector (McKinsey 2020). The inability to shop in person at brickand-mortar store locations has possibly forced consumers to go online, a shift that might become more permanent in the future, especially with climaterelated disruptions such as floods or power outages.
Coupled with prior research that shows the existence of food deserts in low-income communities (Ploeg et al 2011), the selective spread of online shopping to wealthier neighborhoods (Jackson 2020), and the lack of adoption of online shopping from households with food stamp benefits (Rogus et al 2020), our findings point to the enhanced need to ensure inclusive access to grocery and non-grocery items and prioritized access to high-priority essential goods during a crisis.
This study has several limitations and potential biases. First, our respondents reported perceived patterns of consumption, which can vary from actual consumption (Attari et al 2010, Lesic et al 2019. Future research can retrospectively explore the accuracy of perceived changes and the degree to which perceived persistence is indeed observed through crossvalidation with actual consumption data. Second, our survey was conducted in the earlier stages of the pandemic. Since then, lockdown policies have changed, and households have likely adapted to pandemic-related constraints differently in different places. Follow-up surveys can explore how household resource consumption has changed in the months since and in the future. Third, this survey was conducted online. Online surveys offer the advantages of rapid deployment and completion, lower cost, and wider audiences (Callegaro et al 2021), can increase response rates due to convenience (Callegaro et al 2021) and reduce social desirability bias (Phillips and Clancy 1972). However, there are other important sources of bias in online surveys due to differences in internet access and usage for lower income and digitally disconnected households (Huff and Tingley 2015), the lack of transparency in online panel recruitment (Miller et al 2020), the voluntary nature of online survey participation (Huff and Tingley 2015), and greater online participation from younger and more educated respondents (Smith et al 2007). While we chose to conduct this survey online to rapidly track the effects of a global pandemic where social distancing is a public health necessity, further internet panel-based and population research will be needed to explore how household consumption continues to evolve.
There are several opportunities to extend the insights from this study, particularly around the environmental impacts of changing household consumption. Future work can explore the emissions impacts of altered consumption of goods and services at the household level, model how hybrid work can impact residential and commercial building energy use, and explore the transportation emissions implications of changes in commuting and online shopping (Belavina et al 2014). There is growing recognition that extensive ICT usage can not only increase household energy consumption, but also create large emissions impacts for the electric grid (Su et al 2020), pointing to the need for a deeper understanding of this issue. And finally, while this study has described the nature and extent of changes in household consumption, further work is needed to understand the mechanisms that drive these changes in response to disruptive or extreme events.

Data availability statement
The data cannot be made publicly available upon publication because they contain sensitive personal information. The data that support the findings of this study are available upon reasonable request from the authors.