key: cord-0701187-ou7pwtiu
authors: Ma, Mac Zewei
title: COVID-19 concern in cyberspace predicts human reduced dispersal in the real world: Meta-regression analysis of time series relationships across U.S. states and 115 countries/territories()
date: 2021-10-14
journal: Comput Human Behav
DOI: 10.1016/j.chb.2021.107059
sha: 0198b7969e9cdaca3739f088585986c4ffe13beb
doc_id: 701187
cord_uid: ou7pwtiu

Built upon the parasite-stress theory of sociality and the behavioral immune system theory, this research examined how concern about COVID-19 in cyberspace (i.e. online search volume for corona, coronavirus, covid-19, etc.) would predict human reduced dispersal in the real world (i.e. Google's COVID-19 Community Mobility Reports) between January 05, 2020 and May 22, 2021, accounting for COVID-19 cases per million, case fatality rate, death-thought accessibility, government stringency index, yearly trends, season, religious holiday, and serial autocorrelation. Meta-regressions analyzing the results of multiple regressions on weekly-level data showed that when people had higher levels of COVID-19 concern in cyberspace in a given week, the amount of time people spent at home increased from the previous week across U.S. states (Study 1) and 115 countries/territories (Study 2). Across studies, the association between COVID-19 concern and stay-at-home behavior was stronger in areas of higher levels of infectious-disease contagion risks. Compared with actual coronavirus threat, COVID-19 concern in cyberspace had a significantly larger effect on predicting human reduced dispersal in the real world, suggesting that online query data have an invaluable implication for predicting large-scale behavioral changes in response to life-threatening events in the real world and are indispensable for COVID-19 surveillance systems.

, the editor-in-chief of Computers in Human Behavior (CHB), proposes that virtual worlds would broaden people's understanding of large-scale human behavioral changes in response to critically threatening events. Accordingly, will cyberpsychology research deepen the public understanding of the ongoing Coronavirus Disease 2019 (COVID-19) and will online technologies help mitigate this pandemic?

According to a recent editorial of CHB (Guitton, 2020 ), the answer is yes. For example, online technologies, such as social media and Internet, help maintain social relationships during the pandemic (Guitton, 2020) and technological tracking techniques, such as Internet-based and mobile-based strategies, are critically useful in surveilling and managing the COVID-19 (Georgieva et al., 2021) . Indeed, the recently published studies in CHB reveal that analyzing big data related to online behavior could greatly broaden the public understanding of risks and crisis communications (Wang et al., 2021) , citizen engagement (Chen et al., 2020) , and death anxiety (Barnes, 2021) during the deadly pandemic. In addition to analyzing data sourced from social media websites, such as Twitter (Barnes, 2021; Chen et al., 2020; Wang et al., 2021) , another approach of studying the role of cyberpsychology in the ongoing pandemic is by focusing on how online query data would track the public attention to COVID-19 (Brodeur et al., 2021; Hu et al., 2020; Muselli et al., 2021; Springer et al., 2020) and how these data could be used for COVID-19 surveillance (Ayyoubzadeh et al., 2020; Ma, 2021; Mavragani & Gkillas, 2020; Nindrea et al., 2020) .

Because Internet search engines capture millions of people's information-seeking behaviors naturally and anonymously (Lai et al., 2017) , tracking the changes of public interest in specific search terms could be used as an acceptable means of technological tracking in the COVID-19 context (Georgieva et al., 2021) . As computer and other J o u r n a l P r e -p r o o f digital devices (e.g. smartphone) are mediums of getting information online and that conducting web searches is an everyday-life behavior (Cervellin et al., 2017; Lai et al., 2017) , online query data are indeed the outcomes of human interactions with computer and all other Internet-accessible devices, which could be used to track the anonymous online search behavior to reveal how computer and online technologies are important for professional practices, such as predicting human behavioral changes in response to large-scale catastrophic events (Guitton, 2013) . Thus, this line of research is indeed consistent with Guitton (2020) and Georgieva et al. (2021) that online technologies, such as the Internet-based strategies, will provide political decision makers with informative implications for mitigating the crisis.

As a response to Guitton (2020) and Georgieva et al. (2021) that online technologies are useful for mitigating the current crisis, this research attempted to investigate how online query data related to COVID-19 would predict societal-level mobility changes during the deadly pandemic, given that promoting stay-at-home behavior during the pandemic significantly contributed to the control of COVID-19 (Castillo et al., 2020; Gao et al., 2020; Medline et al., 2020; Padalabalanarayanan et al., 2020; Yilmazkuday, 2020) . As online query data are publicly available, easily accessible, and of high anonymity (Lai et al., 2017) , these features make tracking online interests in specific search terms to be a powerful and socially acceptable means of tracking the societal changes of human behaviors in response to large-scale life-threatening events (Guitton, 2013) , such as the COVID-19 (Georgieva et al., 2021) . In this regard, it is of high practicality if searching for COVID-19 online could predict human dispersal behavior during the pandemic, given that policy makers could include tracking search query data as one important part of preventive and control measures for the novel coronavirus.

In order to track millions of people's thoughts of COVID-19 in cyberspace, Google Trends, a big data tool tracking people's natural thoughts on specific topics over time (Arora et al., 2019) , would be employed. This tool has been widely used to investigate how human interactions with computers could contribute to the understanding of various important research topics, such pornography (e.g. Markey & Markey, 2011) , death-thought accessibility (e.g. Pelham et al., 2018) , mental health (e.g. Adam-Troian & Arciszewski, 2020), physical health (e.g. Flanagan et al., 2021) , religiosity (e.g. Ma & Ye, 2021) , human basic motivations (e.g. Ma, 2021) , and the ongoing COVID-19 (Brodeur et al., 2021; Husnayain et al., 2020; Mavragani & Gkillas, 2020) . For example, the method of tracking web search data on specific search terms has been successfully used to investigate how searching for major illnesses (e.g. cancer) online would induce people's death anxiety and activate terror management processes (Alper, 2019; Pelham et al., 2018) .

Because Google search terms served as a reliable proxy for researchers to gain insight into the thoughts of millions of people (Alper, 2019; Du et al., 2020; Husnayain et al., 2020; Lai et al., 2017; Mavragani & Gkillas, 2020; Pelham et al., 2018; Senecal et al., 2020) , Google search data can provide much natural and ecological evidence for the relationships investigated. Recently, Ma and Ye (2021) and Adam-Troian and Bagci (2021) have used Google Trends to track search volume for coronavirus-related keywords (e.g. coronavirus, etc.) to obtain a measure of perceived (= subjective) coronavirus threat, given that online searches could assess the group-level spontaneous exposure to coronavirus-related information online (Adam-Troian & Bagci, 2021) and that searching for COVID-19 related information and reading these information Kim et al., 2021 ) induced people's perceived threat of COVID-19. Although people might have different J o u r n a l P r e -p r o o f purposes for conducting online searches, searching for coronavirus online is at least capturing people's psychological concern about COVID-19 in cyberspace (Du et al., 2020; Muselli et al., 2021) , given that people conduct online searches for finding answers, reducing uncertainty, and sensemaking (Lai et al., 2017) . Indeed, there was a strong and positive correlation between online query data on COVID-19 and epidemiological data on the novel coronavirus across different countries (Du et al., 2020; Mavragani & Gkillas, 2020; Muselli et al., 2021) . Moreover, as searching for religious terms online (e.g., Jesus, God, and prayer) is a terror management process (Alper, 2019; Pelham et al., 2018) , the evidence that searching for COVID-19 in cyberspace was accompanied by searching for religious terms online (Ma & Ye, 2021) suggests that online query data on COVID-19 optimally capture COVID-19 concern in cyberspace.

Although there is an increasing trend of examining how online data, such as Twitter and Google Trends data, would predict behavioral changes in response to COVID-19, little research has provided an ecologically-sound theory or a well-framed social psychological model to deepen the public understanding of the relationships investigated. Indeed, as most of the studies using online query data are conducted at a population level to investigate how infectious-disease threat would influence human psychology and behaviour (e.g. Adam-Troian & Bagci, 2021; Brodeur et al., 2021; Du et al., 2020; Mavragani & Gkillas, 2020) , theories accounting for group-level phenomenon or models explaining human psychological and behavioral responses to infectious-disease threat are needed. Similar to Ma and Ye (2021) , this research would employ the parasite-stress theory of sociality, which is a theory about how pathogen prevalence shapes human sociality and cultural diversity , 2012b Thornhill & Fincher, 2014b) , and the behavioral immune J o u r n a l P r e -p r o o f system theory, which is a theory about the psychological and behavioral changes in response to pathogen threat, to examine the relationship between COVID-19 concern in cyberspace and stay-at-home behaviour in the real world.

The parasite-stress theory of sociality proposes that strong ingroup assortative sociality is favored by natural selection for avoiding and managing novel infectious diseases in areas of high pathogen-stress (Thornhill & Fincher, 2014a) , as ingroup assortative sociality creates strong intergroup boundaries to block outgroup communication (e.g., xenophobia, prejudice, collectivism, philopatry, etc.) to prevent the transmission of novel infectious diseases from outgroup people. The propositions that features of ingroup assortative sociality are adaptive (ancestrally) preferences/values and behaviors for infectious-disease avoidance and management (Thornhill & Fincher, 2014b) are consistent with the proactive responses postulated by the behavioral immune system theory (Ackerman et al., 2018) , as proactive responses aim at managing infectious diseases in the long run. The proactive role of ingroup assortative sociality in avoiding infectious diseases is supported by recent empirical studies showing that U.S. states (Ma & Ye, 2021) and countries (Gelfand et al., 2021; Gokmen et al., 2020; Maaravi et al., 2021; Rajkumar, 2021) of stronger ingroup assortative had better control of the COVID-19.

However, it remains unknown how ingroup assortative sociality is a reactive response to heightened parasite-stress at a group level. According to the behavioral immune system theory, presence of pathogen threat triggers reactive responses (Ackerman et al., 2018) . For example, experimentally manipulating individuals to perceive an increased risk of parasitic infection resulted in their greater ingroup assortative sociality at an individual level (Faulkner et al., 2004; Navarrete & Fessler, 2006; Wu & Chang, 2012) .

As recent studies investigated the relationship between ingroup assortative sociality and J o u r n a l P r e -p r o o f COVID-19 from a proactive response perspective (i.e. the effect of ingroup assortative sociality on COVID-19 pandemic) (Gelfand et al., 2021; Gokmen et al., 2020; Maaravi et al., 2021; Rajkumar, 2021) , it provides little insight for understanding this relationship from a reactive response perspective at a group level (i.e. the effect of COVID-19 pandemic on ingroup assortative sociality) (Ma & Ye, 2021) .

One central feature of ingroup assortative sociality is reduced dispersal, which is the "behaviours that reduce movements away from a central location" and "keeps people near to their natal locale and social community, and hence contributes to collectivism, ethnocentrism, and in-group assortative in general" (Thornhill & Fincher, 2014a) . According to , reduced dispersal serves as an infectious-disease avoidance strategy by increasing people's association with immunologically similar individuals to prevent getting parasitized from more immunologically distant others. found that people embedded more in their extended family in parasite-stressed societies and dispersed over shorter distances annually. Thornhill and Fincher (2014a) further showed that U.S. states of higher pathogen prevalence showed fewer residential emigration events. These findings suggest that infectious-disease contagion risks could promote reduced dispersal by keeping people near to their natal locale as a means of avoiding novel infectious diseases.

As practicing stay-at-home behavior suggests embedding more in one's own family and dispersing for shorter distances from the natal locale (Alesina & Giuliano, 2010; Thornhill & Fincher, 2014a) , staying at home captures strong philopatric values. In this regard, people's increased average amount of time spent at places of residence in the context of COVID-19 (Bavadekar et al., 2020; Saha et al., 2020) is indeed a proxy for reduced dispersal during the pandemic. As it was J o u r n a l P r e -p r o o f straightforward that increasing the relative frequency, time and duration of visits related to places of residence (Bavadekar et al., 2020; Saha et al., 2020) would fundamentally reduce the contact with immunologically distant individuals outside the residence, this provides an ecologically-sound perspective to track how psychological concern about COVID-19 in cyberspace would predict mobility changes at a societal level during the pandemic.

Because reduced mobility could be a consequence of the lethargy and incapacitation associated with high levels of pathogen prevalence , reduced dispersal during the pandemic might not always reflect increased philopatric values serving to avoid the infectious diseases (Thornhill & Fincher, 2014a) . Thus, the present research of investigating how online query data on COVID-19 would predict stay-at-home behavior could greatly enhance one's understanding of how changes in the collective concern about the novel coronavirus would promote philopatry at a group level (Thornhill & Fincher, 2014a) . More importantly, findings in the current research could show how Google Trends would be used as an easily-implemented tool of predicting human behavioral changes in response to large-scale catastrophic events (Guitton, 2013) and how online query data are informative to policy makers in the context of COVID-19 (Georgieva et al., 2021) . As time series data were analyzed, the effects of seasonality and autocorrelation were addressed by accounting for season, religious holidays, yearly trends, and reduced dispersal in the prior week (Alper, 2019; Pelham et al., 2018) . This research also accounted for the effects of stringency of COVID-19 policy (Saha et al., 2020) and terror management during the pandemic (Pyszczynski et al., 2020) to rule out alternative explanations.

Given that perceiving an increased risk of parasitic infection had a unique effect on ingroup assortative sociality (Faulkner et al., 2004; Karwowski et al., 2020; Navarrete J o u r n a l P r e -p r o o f & Fessler, 2006; Sorokowski et al., 2020; Wu & Chang, 2012) and that online query data captured millions of people's thoughts across a wide variety of topics to effectively predict important psychological and behavioural changes in the real world (e.g. Adam-

Troian & Arciszewski, 2020; Brodeur et al., 2021; Flanagan et al., 2021; Husnayain et al., 2020; Ma, 2021; Ma & Ye, 2021; Markey & Markey, 2011; Mavragani & Gkillas, 2020; Pelham et al., 2018) , this study predicted that:

H1: COVID-19 concern in cyberspace (i.e. search volume for coronavirus-related keywords) would uniquely predict stay-at-home behavior in the real world during the pandemic.

As ingroup assortative sociality is predominately valued in parasite-stressed areas due to the benefits associated with strong ingroup assertive sociality when it comes to defending novel infectious diseases Thornhill & Fincher, 2014b; Thornhill et al., 2009) , reduced dispersal is proposed to be favored by natural selection in areas of high pathogen-stress Thornhill & Fincher, 2014a ). Thus, it was further hypothesized that:

H2: The association between COVID-19 concern in cyberspace and stay-at-home behavior in the real world would be stronger in areas of higher infectious-disease contagion risks historically.

This study examined how COVID-19 concern in cyberspace would predict stay-athome behavior in Washington D.C. and each U.S. state between January 05, 2020 (the first week of 2020) and May 22, 2021 (i.e. 72 consecutive weeks). Times dataset (https://github.com/nytimes/COVID-19-data), respectively. New cases per million and case fatality rate (i.e. a ratio of COVID-19 deaths to COVID-19 cases)

were calculated for each week to indicate the prevalence and lethality of the novel coronavirus to capture coronavirus threat (Mazumder et al., 2020; Verity et al., 2020) .

For weeks (January and early February) whose epidemiological data were not available, the values were fixed to 0.

Recent studies used online query data on COVID-19 to obtain a measure of perceived coronavirus threat (Adam-Troian & Bagci, 2021; Ma & Ye, 2021) . Because people could have different purposes to search for COVID-19 online and that searching for specific information online is an information-seeking behavior, which is for finding answers, reducing uncertainty, and sensemaking (Lai et al., 2017) , online query data on COVID-19 could also capture people's concern about the novel coronavirus in cyberspace. Indeed, as there was a strong and positive correlation between online query data on COVID-19 and epidemiological data on the novel coronavirus across different countries (Du et al., 2020; Mavragani & Gkillas, 2020; Muselli et al., 2021) , recent studies have suggested COVID-19 search volume captured people's psychological concern about COVID-19 (Du et al., 2020; Mavragani & Gkillas, 2020; Muselli et al., 2021) and related death anxiety (Ma & Ye, 2021) . Therefore, operationalizing the current index as an index about COVID-19 concern in cyberspace would make the research findings to be more valid and comprehensive.

Given that Google Trends uses a natural language classification engine to categorize search terms that share similar concepts into specific topics (Choi & Varian, 2012) , irrespective of which languages are used to search these terms (Dilmaghani, 2020; Yeung, 2019) , recent studies have utilized categorized search terms to improve the reliability and validity of their research findings (Brodeur et al., 2021; Flanagan et al., 2021; Gianfredi et al., 2018; Kamiński et al., 2020; Ma & Ye, 2021; Strzelecki, 2020; Yeung, 2019) . In the context of COVID-19, using only exact search terms, such as "covid-19", had at least two disadvantages: 1) it was impossible to obtain search volume yielded from people performing online searches with other languages within the investigated geographic region (e.g. people in the United States may use multiple languages to search information related to COVID-19 such as "coronavirus (English)"

or "coronavírus (Portuguese)", and 2) search volume of other search terms which shared similar concepts to the exact search terms used (e.g. "SARS-COV-2" was a related search term to "coronavirus") would be excluded. (Ma & Ye, 2021) to improve the reliability and validity of the COVID-19 concern index. Since search volume for some of the keywords were often less than 1 in January weeks, these less-than-one values were fixed to 0. At a national level, the search terms had a Cronbach α of .97, suggesting that it was plausible to average the relative-search-volume (RSV) scores across the search terms to capture COVID-19 concern at a weekly level. As the lethality of the novel coronavirus was associated with people's concerns about COVID-19 (Muselli et al., 2021) , this study found that the current index had a significant and positive correlation with COVID-19 case fatality rate analyzed with time series data, r = .37, ptwo-tailed = .001.

Furthermore, if searching for the novel coronavirus in Google did indeed reflect people's concern about COVID-19 in cyberspace, the current index should be significantly and positively associated with search volume for fear-related emotions and help-seeking behaviors (i.e. helpline fear, worry, panic, and death) and infectiousdisease avoidance behaviors (i.e. hand washing, social distancing, and quarantine) (Du et al., 2020) . Table S1 (below the diagonal; Supplementary File) shows that the correlations between the COVID-19 concern index and the external variables were significant (all ps < .001) and positive (.65 ≤ rs ≤ .96) at a national level. Table S1 (above the diagonal; Supplementary File) shows that accounting for the actual COVID-19 threat in a partial correlation analysis did not alter the significant and positive relationships between the COVID-19 concern index and the external variables (.64 ≤ rs ≤ .93, all ps < .001), suggesting that the psychological process in cyberspace was unique and reflected people's concern about COVID-19 in the virtual world. Thus, the COVID-19 concern index was computed for each U.S. state.

To track mobility changes during the pandemic, this study sourced the daily COVID- (Bavadekar et al., 2020) and the "same world-class anonymization technology" (https://www.google.com/covid19/mobility/?hl=en) to aggregate and anonymize the individual data. Therefore, the mobility data released by Google are aggregated and anonymized at a regional level, as it is important to protect the privacy of the individual users when it comes to releasing the data to the public (Bavadekar et al., 2020) . Bavadekar et al. (2020) describes the detailed processes of aggregation and anonymization.

Huynh (2020) showed that residential movement was strongly and negatively related to all other types of movement. Thus, this study averaged the daily mobility scores of retail & recreation, grocery & pharmacy, parks, transit stations and workplace to compute a mean score of non-residential movement, which was subtracted from the mobility score of residential movement to create a daily reduced dispersal index, which was further used to calculate a weekly-level index so that a higher score would indicate a greater level of reduced dispersal in a given week (i.e. spending more time at home; Alesina & Giuliano, 2010; Thornhill & Fincher, 2014a) . Given that religiosity is an important element of ingroup assortative sociality (Fincher & Thornhill, 2012a Thus, the weekly-level reduced dispersal index of each U.S. state was calculated by using the U.S. regional reports of the Google CMR. Because Google CMR provided no data for weeks between 01 January and 14 February, 2020, the reduced dispersal scores of these weeks were fixed to 0.

According to Pelham et al. (2018) , search volume for hypertension, cancer, and diabetes induced death-thought accessibility at a group level. As reduced dispersal could also serve as a proximal defensive strategy during the deadly pandemic to manage the terror of death (Pyszczynski et al., 2020) , this study obtained the weekly-level RSV scores of the categorized search terms of Hypertension (Medical condition), Cancer (Disease) and Diabetes (Disorder) to compute a major-illness index (Alper, 2019; Pelham et al., 2018) to account for the effect of death-thought accessibility on reduced dispersal.

Because mobility changes were also the consequences of the societal responses to COVID-19 lockdown policy (Saha et al., 2020) , the government stringency index from

The University of Oxford (https://www.bsg.ox.ac.uk/research/research-projects/covid-19-government-responsetracker) was controlled, as this index has been widely used to account for the stringency of COVID-19 policy (Kapoor et al., 2021; Sorci et al., 2020; Wang, 2021) . For weeks whose stringency indices were not available, the values were fixed to 0. 

The statistical approach in Pelham et al. (2018) and Alper (2019) were followed. First, as week was the analyzed unit, all weekly-level cases were the individual data points.

Second, this study specified a regression model to predict level of reduced dispersal in the present week (week x), from several predictors: 1) year, 2) season and religious holidays, 3) reduced dispersal in the prior week (i.e. week x -1), which allowed the present study to account for autocorrelation and assess changes in reduced dispersal rather than simple variation in it, 4) major-illness index and stringency index in the J o u r n a l P r e -p r o o f present week, 5) COVID-19 threat in the present week, and 6) COVID-19 concern in the present week. This regression model would be first analyzed with national-level data. Then, the analytical procedures would be separately replicated in Washington D.C. and each U.S. state. Specifically, variables were Z-transformed to obtain standardized betas and standard errors estimated from these scores to improve the legibility and allow comparison. Based on the results of the multiple regressions, meta-regression was conducted to examine the moderating effect of state-level infectious-disease contagion risk on the combined association between COVID-19 concern in cyberspace and stayat-home behavior across U.S. states (Alper, 2019) . Two-tailed tests were conducted to avoid inaccurate and biased statistical results.

The national level analysis showed that while actual coronavirus threat had no significant effect on reduced dispersal in the real world, the effect of COVID-19 concern in cyberspace was significant, β = .34, ptwo-tailed < .001, 95% CI = [.28, .40], accounting for a series of covariates ( The first meta-regression examined the combined effect of COVID-19 concern in cyberspace on reduced dispersal in the real world, which showed a non-significant heterogeneity in the associations across states, after accounting for covariates, Q(50) = 50.19, ptwo-tailed = .466. Thus, a fixed-effect method was used (Alper, 2019) , showing a significant combined effect of COVID-19 concern in cyberspace on reduced dispersal concern in cyberspace and reduced dispersal in the real world was stronger in states of higher infectious-disease contagion risks (Figure 1 ).

The second meta-regression analysis was conducted to investigate the combined effect Because the combined effect of COVID-19 concern (standardized estimate = .29) was larger than that of actual coronavirus threat (standardized estimate = .07), H1 was supported. Moreover, since the association between COVID-19 concern and reduced dispersal was stronger in states of higher infectious-disease contagion risks, H2 was supported. As the parasite-stress theory of sociality is supported cross-nationally , 2012a Thornhill et al., 2010) , it was expected that findings in Study 1 could be replicated across different countries/territories.

Study 2 examined whether the effect of COVID-19 concern on reduced dispersal could be found in different countries/territories and investigated how this effect would depend on the country-level infectious-disease contagion risk by replicating the research design and analytical procedure of Study 1. Thus, Study 2 was conducted to examine the generalizability of the findings in Study 1.

Country selection was based on 1) Google CMR data availability, 2) Google Trends data availability, 3) COVID-19 epidemiological data availability, 4) the availability of data on covariates. Accordingly, a total of 115 countries/territories were investigated.

The period examined was identical to Study 1 (i.e. between January 05, 2020 and May 22, 2021; 72 consecutive weeks).

All measures were identical to those employed in Study 1 except for several minor modifications for 1) religious holidays, 2) autumn and winter weeks, 3) languages used for searching the word "coronavirus", and 4) historical infectious-disease contagion risk index. First, the CIA WORLD FACTBOOK (https://www.cia.gov/the-worldfactbook/) was used to identify the predominant religion of each country. Then, 1)

Easter Day and Christmas Day were controlled for Christian countries/territories, 2)

Ramadan, Eid al-Fitr, and Eid al-Adha were controlled for Islamic countries/territories, and 3) Diwali was controlled for countries/territories predominated by Hinduism (Alper, 2019; Pelham et al., 2018) . For Buddhist countries, Buddha's Birthday was controlled.

Passover, Rosh Hashanah, Yom Kippur, and Hanukkah were controlled for Judaism.

The Bon festival was controlled for Japan (Pelham et al., 2018) . For Nepal, the Dashain festival was controlled. For Vietnam, the Hung Kings Temple Festival was controlled.

The exact dates of the religious holidays in each year were tracked by an online calendar tool (https://www.officeholidays.com/). As for season, autumn and winter weeks (from September to February for countries/territories in the northern hemisphere; from March to August for countries/territories in the southern hemisphere) were coded as "1" while other weeks were coded as "0".

Because people in different countries might have different expressions for the word "coronavirus", this study used two strategies to minimized the linguistic influences. word of "coronavirus" in other languages and reveals that the internal reliability of the COVID-19 concern index was satisfactory across different countries/territories (mean Cronbach α = .92, SD = 0.04).

The historical parasite-stress index in Murray and Schaller (2010) and the infectious disease richness index in were used to create a countrylevel infectious-disease contagion risk index (Cronbach α = .74) for each country/territory (Table  SSS1 https://osf.io/w25kb/?view_only=cf42e658bceb4791b4c20d4a4d535d01). A higher score indicated a higher level of infectious-disease contagion risk historically.

The analytical procedures were identical to Study 1.

Results of the 115 separately conducted multiple regression analyses are presented from Tables  SSS2  to  SSS116  (Study  2  Supplementary  Tables https://osf.io/w25kb/?view_only=cf42e658bceb4791b4c20d4a4d535d01). Across the 115 multiple regressions, the mean Durbin-Watson value was 2.01 (SD = 0.27) and the median value was 2.02, suggesting that autocorrelation was not an issue (Brocklebank & Dickey, 2003; Pelham et al., 2018; Yaffee & McGee, 2000) .

The first meta-regression examined the combined effect of COVID-19 concern in cyberspace on reduced dispersal in the real world. Because the heterogeneity in the associations across countries/territories was significant, after accounting for covariates, Study 2 showed that a higher level of COVID-19 concern in cyberspace was a major cause of reduced dispersal in the real world across different countries/territories, which supported H1. Compared with the effect of COVID-19 concern in cyberspace (standardized estimate = .28), COVID-19 threat in the real world showed an extremely small effect (standardized estimate = .06), revealing that tracking people's COVID-19 concern in cyberspace could be a more effective approach of predicting people's behavioral response to COVID-19. Figure 3 presents the combined time series association between COVID-19 concern in cyberspace and reduced dispersal in the real world for a 72-consecutive week period across countries/territories. As the association between COVID-19 concern and reduced dispersal was stronger in countries/territories of higher infectious-disease contagion risks historically, reduced dispersal could be a mechanism of the behavioral immune system at a group level and is favored by natural selection in areas of high pathogen-stress, which supported H2.

Insert Figure 3 about Here.

As a response to Guitton (2020) and Georgieva et al. (2021) that online technologies are useful for mitigating the ongoing COVID-19 pandemic, this research is the first to document that when people were concerned about COVID-19 in cyberspace in a given

week, the amount of time people spent at home in the real world increased from the previous week across U.S. states (Study 1) and 115 countries/territories (Study 2), controlling for a series of covariates. Because both online query data and mobility data are outcomes of human interactions with computers (i.e. people use computers or other digital devices to get access to online information and the virtual world), the present research findings are consistent with the recent editorial of CHB (Guitton, 2020 ) that cyberpsychology research and online technologies will deepen the public understanding of the ongoing COVID-19 and help mitigate this pandemic. Indeed, as the effect of COVID-19 concern in cyberspace on predicting reduced dispersal in the real world was significantly larger than that of actual coronavirus threat (i.e. COVID-19 cases per million and case fatality rate) across studies, online query data have an invaluable implication for predicting large-scale behavioral changes in response to catastrophic events (Guitton, 2013) and are indispensable for COVID-19 surveillance system (Georgieva et al., 2021) . From a theoretical perspective, the present research extends the parasite-stress theory of sociality (Thornhill & Fincher, 2014b ) by showing that a high level of concern about parasitic infection promoted a high level of reduced dispersal, which is a reactive response of the behavioral immune system (Ackerman et al., 2018) . Moreover, as the combined association between COVID-19 concern and reduced dispersal was stronger in U.S states (Study 1) and countries/territories (Study 2) of higher infectious-disease contagion risks historically, the present research J o u r n a l P r e -p r o o f supports the proposal that reduced dispersal is favored by natural selection in areas of high pathogen-stress Thornhill & Fincher, 2014a) .

As reduce dispersal in the form of practicing stay-at-home behavior is shown to decrease the transmission of the novel coronavirus (Castillo et al., 2020; Gao et al., 2020; Medline et al., 2020; Padalabalanarayanan et al., 2020; Yilmazkuday, 2020) and

that searching for coronavirus-related keywords in Google uniquely predicted dispersal behavior in the current research, the present findings are consistent with early (Carneiro & Mylonakis, 2009; Zhou et al., 2011) and recent studies (Fantazzini, 2020; Peng et al., 2021; Venkatesh & Gandhi, 2020) showing that web search data have important implications for the control of novel infectious diseases. As online query data are publicly available, easily accessible, and of high anonymity (Lai et al., 2017) , the present significant effect of COVID-19 concern in cyberspace on dispersal behavior in the real world is indeed suggesting that tracking online interests in specific search terms could be a powerful and socially acceptable means of tracking societal changes of human behaviors in response to large-scale life-threatening events (Guitton, 2013) , such as the COVID-19 (Georgieva et al., 2021) . In this regard, policy makers could use COVID-19 search query data to predict the likelihood of future mobility changes at a societal level to reach better control of the COVID-19. Moreover, as the mobility data analyzed in the current study were obtained from millions of people's individual devices, such as smartphones which allowed recording of location history (Sulyok & Walker, 2020) , the current research suggests that the Internet-based strategies (e.g.

tracking search volume for coronavirus-related keywords) and mobile-based strategies (i.e. tracking citizens' moves) could serve as effective tracking and technological surveillance strategies of the population in the context of COVID-19 (Georgieva et al., 2021) .

The present research design is consistent with the growing literature on utilizing big data to capture the thoughts of (Alper, 2019; Du et al., 2020; Husnayain et al., 2020; Lai et al., 2017; Mavragani & Gkillas, 2020; Pelham et al., 2018; Senecal et al., 2020) and behaviors of (Huynh, 2020; Saha et al., 2020; Wang, 2021; Yilmazkuday, 2020) millions of people to test a wide range of important research topics. Thus, utilizing big data makes the current research findings to have relatively higher objectivity and ecological validity in comparison to studies relying on traditional self-report measures and experimental designs. In addition, the validity of the current research findings was strengthened by ruling out alternative explanations via controlling constructs that were relevant to reduced dispersal in the context of COVID-19 (Pyszczynski et al., 2020; Saha et al., 2020) , thereby decreasing the chance of drawing fallible conclusion.

Moreover, the present findings could not be interpreted as artifacts of common method variance and serial autocorrelation. First, reduced dispersal was estimated with Google mobility data, which were anonymized data aggregated from the number of requests made to Google maps for directions (Saha et al., 2020) , and the COVID-19 concern index was estimated with online query data on COVID-19. Thus, the present findings are not the artifact of common method variance (Lindell & Whitney, 2001) . Second, as the effects of season, religious holidays, yearly trends, and reduced dispersal index in the prior week were well controlled, it was found that the Durbin-Watson values across studies were close to the ideal value of 2.00, suggesting absence of autocorrelation (Brocklebank & Dickey, 2003; Pelham et al., 2018; Yaffee & McGee, 2000) . Thus, COVID-19 concern in cyberspace did indeed uniquely predict reduced dispersal in the real world, which supports that big data, such as online query data, can be applied to account for many group-level processes which are hardly examined using traditional research methods (Alper, 2019; Du et al., 2020; Husnayain et al., 2020; Huynh, 2020;  J o u r n a l P r e -p r o o f Lai et al., 2017; MacInnis & Hodson, 2015; Mavragani & Gkillas, 2020; Pelham et al., 2018; Saha et al., 2020; Senecal et al., 2020; Wang, 2021; Yilmazkuday, 2020) .

In addition to their practical implications, the present findings also provide implications for social psychological theories about the role of pathogen threat in shaping human psychology and behaviors. The present research found that reduced dispersal was a reactive response of the behavioral immune system (Ackerman et al., 2018) at a group level, which is consistent with the individual-level finding that inducing a higher level of risk perception on infectious diseases resulted in an increased level of ingroup assortative sociality (Faulkner et al., 2004; Karwowski et al., 2020; Navarrete & Fessler, 2006; Sorokowski et al., 2020; Wu & Chang, 2012) . As reduced dispersal is proposed to contribute to ingroup assortative sociality (Thornhill & Fincher, 2014a) , the present research findings and the previously found proactive role of ingroup assortative sociality in the COVID-19 pandemic (Gelfand et al., 2021; Gokmen et al., 2020; Maaravi et al., 2021; Rajkumar, 2021) could jointly suggest that the reactive and proactive responses of the behavioral immune system at a group level may sustain the predictive power of the parasite-stress theory of sociality when it comes to explaining ingroup assortative sociality phenomenon and infectious-disease avoidance at a group level , 2012a Thornhill et al., 2010) .

As the relationship between COVID-19 concern in cyberspace and reduced dispersal in the real world was stronger in areas of higher infectious-disease contagion risks historically, the mitigating effect of ingroup assortative sociality on COVID-19 (Gokmen et al., 2020; Maaravi et al., 2021; Rajkumar, 2021) may root in the reactive responses activated in these high parasite-stressed regions. For example, as collectivism reflects ingroup assortative sociality and is predominately valued in areas of high J o u r n a l P r e -p r o o f pathogen-stress (Fincher & Thornhill, 2012a; Thornhill et al., 2010) , people in more collectivistic countries/territories could react more strongly to avoid the novel coronavirus, such as wearing medical masks (Lu et al., 2021) , albeit Lu et al. (2021) interpreted their results from the perspective of cultural differences. Thus, the present research findings reveal that studies examining the relationship between ingroup assortative sociality and COVID-19 need to consider alternative explanation(s) for their research findings.

Because Twitter data also captured psychological and behavioral changes to COVID-19 (Barnes, 2021; Chen et al., 2020; Wang et al., 2021) , the validity and reliability of the present findings could be examined by analyzing Twitter data on COVID-19

concern. Moreover, although Google is a representative search engine, it is not widely used in some countries (Jun et al., 2018) , suggesting that using only Google search volume data may not fully explain the variation in COVID-19 concern during the deadly pandemic. Future research is encouraged to collect web search data from other search engines to cross-validate the present findings. For example, as the Baidu Index tracks Chinese people's online search behaviors (Liu et al., 2019) , Baidu search volume data could be utilized in future studies. Because online query data were widely used across different disciplines to predict important research topics (Adam-Troian & Arciszewski, 2020; Brodeur et al., 2021; Flanagan et al., 2021; Husnayain et al., 2020; Ma, 2021; Ma & Ye, 2021; Markey & Markey, 2011; Mavragani & Gkillas, 2020; Pelham et al., 2018) , it is expected that online query data on COVID-19 could be used to explain other COVID-19 related issues. For example, future studies may investigate whether a higher level of COVID-19 concern in cyberspace would predict a higher likelihood of wearing medical masks in the real world. Indeed, as search engines are

used for finding answer, reducing uncertainty, and sensemaking and that searching for J o u r n a l P r e -p r o o f information online is an everyday-life behaviour (Lai et al., 2017) , online query data could inform important human psychological and behavioural changes. Furthermore,

given the significant cultural differences in COVID-19 severity (e.g. Gelfand et al., 2021; Gokmen et al., 2020; Maaravi et al., 2021; Rajkumar, 2021) and COVID-19 preventive behaviors (e.g. Lu et al., 2021) , future studies may investigate how cultural factors would moderate the association between COVID-19 concern and dispersal behavior. Another suggestion for future studies relates to the parasite-stress theory per se. As individual-level findings have shown that inducing a higher level of risk perception on coronavirus resulted in greater levels of ingroup favoritism and outgroup avoidance Sorokowski et al., 2020) , future studies may test how COVID-19 concern in cyberspace would predict other ingroup assortative sociality features.

In sum, this research shows that online query data on COVID-19 are effective in predicting stay-at-home behavior in the real world at a population level across U.S.

states (Study 1) and 115 countries/territories (Study 2). Across studies, the association between COVID-19 concern in cyberspace and reduced dispersal in the real world was stronger in areas of higher infectious-disease contagion risks historically, suggesting that reduced dispersal is favored by natural selection in areas of high pathogen-stress.

Thus, the present research responses to Guitton (2013) , Guitton (2020) (Ma & Ye, 2021) to suggest that strong ingroup assortative sociality is a reactive response to a high level of concern about parasitic infection at a group level, which serves to defend people against the transmission of novel infectious diseases from outgroup people, particularly in more parasite-stressed areas.

J o u r n a l P r e -p r o o f

The behavioral immune system: Current concerns and future directions

Absolutist words from search volume data predict state-level suicide rates in the United States

The pathogen paradox: Evidence that perceived COVID-19 threat is associated with both pro-and anti-immigrant attitudes

The power of the family

Does the Association Between Illness-Related and Religious Searches on the Internet Depend on the Level of Religiosity?

Google Trends: Opportunities and limitations in health and health policy research

Predicting COVID-19 incidence through analysis of google trends data in iran: data mining and deep learning pilot study. JMIR public health and surveillance

Understanding terror states of online users in the context of J o u r n a l P r e -p r o o f COVID-19: An application of Terror Management Theory

Search Trends Symptoms Dataset: Anonymization Process Description (version 1.0)

SAS for forecasting time series

COVID-19, lockdowns and well-being: Evidence from Google Trends

Google trends: A web-based tool for realtime surveillance of disease outbreaks

The effect of state-level stay-at-home orders on COVID-19 infection rates

Is Google Trends a reliable tool for digital epidemiology? Insights from different clinical settings

Unpacking the black box: How to promote citizen engagement through government social media during the COVID-19 crisis

Predicting the present with Google Trends

Who is Not Afraid of Richard Dawkins? Using Google Trends to Assess the Reach of Influential Atheists across Canadian Secular Groups

COVID-19 Increases Online Searches for Emotional and Health-Related Terms

Short-term forecasting of the COVID-19 pandemic using Google Trends data: Evidence from 158 countries

Evolved diseaseavoidance mechanisms and contemporary xenophobic attitudes

Assortative sociality, limited dispersal, infectious disease and the genesis of the global pattern of religion diversity

Parasite-stress promotes in-group assortative sociality: The cases of strong family ties and heightened religiosity

The parasite-stress theory may be a general theory of culture and sociality

Pathogen prevalence predicts human cross-cultural variability in individualism/collectivism

Utilizing Google Trends to assess worldwide interest in irritable bowel syndrome and commonly associated treatments

Association of mobile phone location data indications of travel and stay-at-home mandates with covid-19 infection rates in the us

The relationship between cultural tightness-looseness and COVID-19 cases and deaths: A global analysis

Ensuring social acceptability of technological tracking in the COVID-19 context

Monitoring public interest toward pertussis outbreaks: An extensive Google Trends-based analysis

The Impact of National Culture on the Increase of COVID-19: A Cross-Country Analysis of European Countries

Developing tools to predict human behavior in response to large-scale catastrophic events

Cyberpsychology research and COVID-19

More effective strategies are required to strengthen public awareness of COVID-19: Evidence from Google Trends

Applications of google search trends for risk communication in infectious disease management: A case study of COVID-19 outbreak in Taiwan

Does culture matter social distancing under the COVID-19 pandemic?

Ten years of research change using Google Trends: From the perspective of big data utilizations and applications

Global and local diet popularity rankings, their secular trends, and seasonal variation in Google Trends data

Innovation in isolation? COVID-19 lockdown stringency and culture-innovation relationships

When in danger, turn right: Does covid-19 threat promote social conservatism and right-wing presidential candidates

The impact of the COVID-19 threat on the preference for high versus low quality/price options

Research on web search behavior: How online query data inform social psychology

Accounting for common method variance in cross-sectional research designs

Collectivism predicts mask use during COVID-19

Group-level human values estimated with web search data and archival data explain the geographic variation in COVID-19 severity in the United States

The role of ingroup assortative sociality in the COVID-19 pandemic: A multilevel analysis of google trends data in the United States

The tragedy of the commons": How individualism and collectivism affected the spread of the COVID-19 pandemic

Do American states with more religious or conservative populations search more for sexual content on Google?

Pornography-seeking behaviors following midterm political elections in the United States: A replication of the challenge hypothesis

COVID-19 predictability in the United States using Google Trends time series

Geographical variation in case fatality rate and doubling time during the COVID-19 pandemic

Evaluating the impact of stay-at-home orders on the time to reach the peak burden of Covid-19 cases and deaths: Does timing matter?

Historical prevalence of infectious diseases within 230 geopolitical regions: A tool for investigating origins of culture

Coronavirus (Covid-19) pandemic: How may communication strategies influence our behaviours?

Disease avoidance and ethnocentrism: The effects of disease vulnerability and disgust sensitivity on intergroup attitudes

Validation: the use of google trends as an alternative data source for COVID-19 surveillance in Indonesia

Association of State Stay-at-Home Orders and State-Level African American Population With COVID-19 Case Rates

Searching for God: Illness-related mortality threats and religious search volume in Google in 16 nations

Real-time Prediction of the Daily Incidence of COVID-19 in 215 Countries and Territories Using Machine Learning: Model Development and Validation

A GPS Data-Based Index to Determine the Level of Adherence to COVID-19 Lockdown Policies in India

Terror Management Theory and the COVID-19 Pandemic

the impact of COVID-19 across nations

Lockdown for COVID-19 and its impact on community mobility in India: An analysis of the COVID-19 Community Mobility Reports

Google Trends Insights Into Reduced Acute Coronary Syndrome Admissions During the COVID-19 Pandemic: Infodemiology Study

Explaining among-country variation in COVID-19 case fatality rate

Can information about pandemics increase negative attitudes toward foreign groups? A case of COVID-19 outbreak

Google Trends provides a tool to monitor population concerns and information needs during COVID-19 pandemic

The second worldwide wave of interest in coronavirus since the COVID-19 outbreaks in South Korea

Community movement and COVID-19: A global study using Google's Community Mobility Reports

Collectivism-individualism, family ties, and philopatry

The Parasite-Stress Theory of Values

Parasites, democratization, and the liberalization of values across contemporary countries

Zoonotic and nonzoonotic diseases in relation to human personality and societal values: Support for the parasite-stress model

Prediction of COVID-19 outbreaks using google trends in India: A retrospective analysis

Estimates of the severity of coronavirus disease 2019: A model-based analysis. The Lancet Infectious Diseases

Government policies, national culture and social distancing during the first wave of the COVID-19 pandemic: International evidence

Examining risk and crisis communications of government agencies and stakeholders during early-stages of COVID-19 on

The social impact of pathogen threat: How disease salience influences conformity

An introduction to time series analysis and forecasting: With applications of SAS® and SPSS®

Measuring Christian Religiosity by Google Trends

Stay-at-home works to fight against COVID-19: International evidence from Google mobility data

Tuberculosis surveillance by analyzing Google trends

J o u r n a l P r e -p r o o f Figure 1 . The distribution of standardized coefficients predicting reduced dispersal in the real world from COVID-19 concern in cyberspace in Study 1 (N = 51). Whiskers represent 95% CIs for the standardized coefficients. Grey diamonds indicate the prediction from historical infectious-disease contagion risk of the state. States are ranked from the most parasite-stressed (Mississippi) to the least parasite-stressed (Maine). State-level infectious-disease contagion risk of Washington D.C. was estimated using an EM method.J o u r n a l P r e -p r o o f Figure 2 . The distribution of standardized coefficients predicting reduced dispersal in the real world from COVID-19 concern in cyberspace in Study 2 (N = 115). Whiskers represent 95% CIs for the standardized coefficients. Grey diamonds indicate the prediction from historical infectious-disease contagion risk of the country. Countries are ranked from the most parasite-stressed (Brazil) to the least parasite-stressed (Luxembourg). Figure 3 . Graph representing the combined time series association between COVID-19 concern in cyberspace and human reduced dispersal behavior in the real world across 115 countries/territories between January 05, 2020 and May 22, 2021 (Study 2). An upward trend of COVID-19 concern in cyberspace indicates an increasing online search volume for coronavirus-related keywords. An upward trend of the reduced dispersal in the real world indicates an increased amount of time people spend at home.