In Crisis, We Pray: Religiosity and the COVID-19 Pandemic

Abstract

In times of crisis, humans have a tendency to turn to religion for comfort and explanation. The 2020 COVID-19 pandemic is no exception. Using daily data on Google searches for 95 countries, this research demonstrates that the COVID-19 crisis has increased Google searches for prayer (relative to all Google searches) to the highest level ever recorded. More than half of the world population had prayed to end the coronavirus. The rise amounts to 50% of the previous level of prayer searches or a quarter of the fall in Google searches for flights, which dropped dramatically due to the closure of most international air transport. Prayer searches rose at all levels of income, inequality, and insecurity, but not for the 10% least religious countries. The increase is not merely a substitute for services in the physical churches that closed down to limit the spread of the virus. Instead, the rise is due to an intensified demand for religion: We pray to cope with adversity.

Full text

In Crisis, We Pray:
Religiosity and the COVID-19 Pandemic
Jeanet Sinding Bentzen∗
University of Copenhagen, CEPR, CAGE
Aug 3 2020
Abstract
In times of crisis, humans have a tendency to turn to religion for comfort and explanation.
The 2020 COVID-19 pandemic is no exception. Using daily data on Google searches for
95 countries, this research demonstrates that the COVID-19 crisis has increased Google
searches for prayer (relative to all Google searches) to the highest level ever recorded.
More than half of the world population had prayed to end the coronavirus. The rise
amounts to 50% of the previous level of prayer searches or a quarter of the fall in Google
searches for flights, which dropped dramatically due to the closure of most international
air transport. Prayer searches rose at all levels of income, inequality, and insecurity,
but not for the 10% least religious countries. The increase is not merely a substitute
for services in the physical churches that closed down to limit the spread of the virus.
Instead, the rise is due to an intensified demand for religion: We pray to cope with
adversity.
Keywords: COVID-19 pandemic, disaster, psychological coping, religiosity, empirical
analysis, Google search data.
JEL codes : I15, D91, Q54, Z12, O57.
∗Associate professor, Department of Economics, University of Copenhagen, jeanet.bentzen@econ.ku.dk. I
thank Lourdes Cantarero Arevalo, Sascha Becker, Ryan Cragun, Carl-Johan Dalgaard, Pablo Duarte, Gunes
Gokmen, Jens Hjorth, Claus Thustrup Kreiner, Andrea Mattranga, Landon Schnabel, Lena Lindbjerg Sperling,
Tohit Ticku, and seminar participants at the Virtual COVID-19 Seminar Series at University of Copenhagen
for useful comments.

1 Introduction
The COVID-19 pandemic has brought sizeable costs for societies across the globe. Countries
follow different strategies for dealing with the pandemic and suffer varying degrees of losses.
Some of these losses potentially involve deteriorating mental health.1At the same time, re-
search has documented positive effects of religion on mental health.2The use of religion for
coping with the anxiety caused by the pandemic may therefore potentially reduce emotional
distress. This may even impact economic losses. Research has documented heightened eco-
nomic anxiety in the face of the pandemic, which arguably fuels the economic crisis.3If
religion provides stress relief, believers may experience less economic anxiety and be quicker
to recover from the crisis.4On the other hand, religion may also have exacerbated losses, as
some religious communities defied social distancing recommendations and continued to have
religious mass gatherings, potentially intensifying the spread of the virus.5Identifying the
impact of the coronavirus on the use of religion is therefore relevant.
This research identifies empirically the extent to which the COVID-19 pandemic has in-
duced people across the globe to pray, whether the phenomenon is global, and who prays
in times of crisis. Google searches for prayer, as a share of all Google searches, provides
a signal of peoples’ interest in prayer in real time. Previous research has documented that
Google searches reveal real offline behavior, such as consumption patterns (Goel et al., 2010),
unemployment rates (Askitas and Zimmermann, 2009; D’Amuri and Marcucci, 2017), predict-
ing the present in general (Choi and Varian, 2012), development of the flu (Ginsberg et al.,
2009), and also infections by COVID-19 prior to official statistics (Walker et al., 2020). Like-
wise, whether or not we google religious terms reflects our religious preferences (Yeung, 2019;
Stephens-Davidowitz, 2015). Events that instigate intensified actual prayer are clearly visible
in the data. Before the COVID-19 pandemic, the Ramadan contributed to the largest yearly
1Using surveys and activity on the Chinese microblogging site Weibo, research has identified mental health
effects of COVID-19 in China (Li et al., 2020; Wang et al., 2020; Xiao et al., 2020).
2Reviews in Pargament (2001); O. Harrison et al. (2001); Schnittker (2001). For instance, Miller et al.
(2014) found that religious individuals have thicker cortices, a measure of lower depression risk.
3Fetzer et al. (2020); Binder (2020). Andersen et al. (2020) show that the main part of the decline in
consumption in Scandinavia was due to fear rather than lockdowns. In general, economic anxiety has real
effects on the economy (Bailey et al., 2018, 2019; Coibion et al., 2019; Puri and Robinson, 2007).
4Research has documented that religiosity correlates with various socio-economic factors, such as labour
force participation, education, crime, health, and even aggregate GDP per capita growth (Guiso et al., 2003;
McCleary and Barro, 2006; Campante and Yanagizawa-Drott, 2015; Iannaccone, 1998; Lehrer, 2004; Iyer, 2016;
Kimball et al., 2009). Reducing economic fluctuation may be another unexplored correlate of religion.
5Examples include religious gatherings in a Dallas megachurch (https://www.wsj.com/
articles/despite-coronavirus-some-religious-services-continue-11584310003), secret
church services in Mexico and Brazil (https://www.theguardian.com/global-development/
2020/jun/17/mexico-churches-catholic-mass-covid-19-coronavirus), and encouragement
by Vatican Cardinal to continue having masses (https://www.ncronline.org/news/people/
catholic-cardinal-burke-says-faithful-should-attend-mass-despite-coronavirus).
1

rise in the global search intensity for prayer (Panel a of Fig. 1). Also, prayer search shares
spike up on Sundays everywhere (Stephens-Davidowitz, 2015). Searches for prayer surged in
Iran on January 7 2020, coinciding with the funeral of Qassem Soleimani, the Iranian ma-
jor general killed by American troops, in Australia on January 5 2020 when the movement
”Prayer for Australia” swept across Australia in the midst of the unprecedented bushfires,
and in Albania on November 26 2019 when a 6.4 magnitude earthquake stroke the country.
The countries that search more for prayer on the Internet are also ranked in surveys as being
more religious (Fig. A.5).
Figure 1: Worldwide Google searches for “prayer” during the past 4 years
(a) Jan 1 2016 - Apr 11 2020
(b) Feb 1 - Apr 1 2020
Google searches for prayer relative to the total number of Google searches. The maximum shares were set to 100 by Google
Trends. The searches encompass all topics related to prayer, including alternative spellings and languages. The vertical stippled
lines in Panel (a) represent the first week of the Ramadan. The period in Panel (a) is the longest period for which comparable
data was available at the time of writing. The period in Panel (b) is the period used in the main analysis (before COVID-19
became a pandemic and before the onset of Easter and the Ramadan). Data source: Google Trends. For the development since
2004, see Fig. A.15. Find more details in Appendix A.1 and C.
2

In March 2020, the share of Google searches for prayer surged to the highest level ever recorded,
surpassing all other major events that otherwise call for prayer, such as Christmas, Easter, and
Ramadan (Fig. 1, Fig. A.15, and Appendix C).6The World Health Organization declared
the COVID-19 a pandemic on March 11, 2020. The level of prayer search shares in March
2020 was more than 50% higher than the average during February 2020. For comparison, the
surge in Google searches for prayer was 1.3 times larger than the rise in searches for takeaway
and amounted to 12% of the rise in Netflix searches or 26% the fall in searches for flights,
which all saw massive changes globally, since most countries were in lockdown and air traffic
was shut down (cf. Appendix B.4).7
When googling prayer what you find is specific prayer texts to use when praying. Prayers
may be recited from memory, read from a book of prayers, or composed spontaneously as they
are prayed. In modern times, these books of prayer or verses of prayer can be found on the
Internet. The most common form of prayer in Christianity is to directly appeal to a deity
to grant one’s requests (Kurian and Smith, 2010). One of the most searched for prayers in
March 2020 was ”Coronavirus prayer”, which are prayers that ask God for protection against
the coronavirus, prayers to stay strong, and prayers to thank nurses for their efforts (Appendix
Figures B.2 and B.3). According to a Pew Research Center survey from March 2020, more
than half of Americans had prayed to end the coronavirus (Pew, 2020b).
Using daily data on Google searches for prayer for 95 countries across the globe, this
research documents that the rise visible in Fig. 1 is not driven by a few countries, but is a
global phenomenon. Google searches for prayer surged after March 11 for most countries, and
more so after their own populations had been infected. Prayer searches rose for all the major
religions, and so did earches for topics related to God, Allah, Muhammad, Quran, Bible, and
Jesus, and to a lesser extent Buddha, Vishnu, and Shiva. Prayer search shares rose more in
poorer, more insecure, and more unequal countries, but this heterogeneity is exclusively due
to these countries being more religious: Prayer searches rose more in more religious societies.
This means that what matters for whether people use religion for coping is not the state of the
economy, but the role played by religion. It also may indicate that the emotional distress was
not much larger in poorer countries, perhaps because the virus did not add much to existing
adversity. The growth rate of prayer searches also rose in the least religious countries, due to
low initial levels of prayer searches. A back-of-the-envelope calculation shows that more than
half of the global adult population had prayed to end the coronavirus by April 1 2020.
6Fig. 1 shows Google search data going back to 2016. The spike in prayer is even more dramatic when
including the entire series of available Google search data going back to 2004, but these series include a data
break (Fig. A.15).
7In an attempt to limit the spread of the virus, most countries implemented lockdowns and air traffic was
shut down. As a result, many people were at home ordering takeaway and watching Netflix more than usual.
3

The main reason for the rising interest in prayer on the Internet is religious coping: People
use their religion to cope with adversity.8They pray for relief, understanding, and comfort.
Research has documented that people struggling with cancer, death in close family, or severe
illness are more religious, and also that adversity in the form of actual natural disasters or
primed in experiments cause people to use their religion more intensely.9
People may google prayer for a reason unrelated to religious coping. They may be searching
for online forums to replace their physical churches that closed down in an attempt to enforce
social distancing. Theoretically, we would not expect this to be the main explanation for the
rising search shares for prayer. An extensive literature has documented that people tend to use
mainly their intrinsic religiosity (such as private prayer) rather than their extrinsic religiosity
(such as churchgoing) to cope with adversity.10 In addition, a recent survey shows that 24%
of Americans respond that their faith has strengthened since the coronavirus, which we would
not have predicted if people are simply replacing their physical churchgoing with online church
(Pew, 2020a). The empirical results reveal that replacement of physical churches is not the
main reason for the rise in Google searches for prayer. For instance, searches for ”internet
church” also rise, but follow a distinctly different pattern than the prayer searches and is of
a much smaller magnitude; the search shares on prayer continue to rise long after the church
closures; and the rise in prayer searches is not limited to Sundays, where most masses are
held, but occur on all days of the week, except Fridays.
One concern is that the lockdowns shifted the composition of internet users towards more
religious individuals. Perhaps more blue collar workers without access to the Internet at work
were sent home and blue collar workers may be more religious. This is unlikely to explain the
results. The rise in prayer searches persists if restricting the sample to Sundays or Saturdays,
where the composition of internet users is arguably nearly unchanged compared to prior to
lockdown. In general, taking the back-of-the-envelope calculation at face value, every second
adult on Earth had prayed to end the coronavirus. It seems unlikely that a rise this size is
driven by a specific population group.
The lockdowns also meant that people had more time on their hands. Perhaps people
searched Google for prayer out of boredom. This also is unlikely to explain the results. The
measure of prayer searches amounts to Google searches for the topic prayer as a share of all
other searches on Google. An alternative explanation to coping has to explain why the bored
8Pargament (2001); Norenzayan and Hansen (2006); Cohen and Wills (1985); Park et al. (1990); Williams
et al. (1991).
9Reviews by Ano and Vasconcelles (2005) and Pargament (2001) and more recent studies by Bentzen
(2019); Norenzayan and Hansen (2006).
10E.g., Johnson and Spilka (1991); Pargament (2001); Allport and Ross (1967); Koenig et al. (1988); Miller
et al. (2014); Schnittker (2001).
4

people chose to Google prayer rather than Jigsaw puzzles, for instance.11
There are reasons to believe that the rise in Google searches for prayer underestimates
the true increase in prayer intensity, which is potentially much larger than what is visible
from Fig. 1. First, most prayers are performed without the use of the Internet, instead
recited from memory or read from physical books. Second, among those who use the Internet
to find prayers, the data encompasses only those who google prayer, while those who enter
their preferred prayer websites directly are not included. Third, the elderly, who were most
severely affected by the pandemic, are not the most active internet users and thus their prayer
intensity will not show up in Google. Fourth, the month of March 2020 saw an even larger
rise in internet searches on topics related to COVID-19 and other topics since people across
the globe were at home due to lockdowns. These searches enter the denominator of all other
search shares, which mechanically reduces the search shares for these other searches, including
prayer. Fifth, the data includes only countries with enough internet users and thus the poorest
countries or countries with restricted internet access, such as China, are not included. Poorer
countries are on average more religious (Inglehart and Norris, 2003) and thus more prone to
engage in religious coping (Pargament, 2001).
The results inform the literature on the mental health effects of the COVID-19 pandemic.
Research has documented symptoms of stress and anxiety among health personnel and the
population in China (Li et al., 2020; Wang et al., 2020; Xiao et al., 2020). In the absence of
real-time measures of people’s mental health, Google searches for prayer reveal that people
from across the globe experience emotional distress in the face of the COVID-19 pandemic,
and they use religion to cope.
These emotional effects may have economic consequences. Studies have documented rising
economic distress in the face of the COVID-19 pandemic (Fetzer et al., 2020; Binder, 2020).
In Scandinavia, the main part of the economic downturn was due to the perceived risk of the
virus rather than government mandated lockdowns (Andersen et al., 2020). In more general,
part of the collapse in output during crisis is caused by anxiety and expectations (Bailey et
al., 2018, 2019; Coibion et al., 2019; Puri and Robinson, 2007). If religion provides stress
relief, believers may experience less economic anxiety, which may make them less likely to
change their economic behavior dramatically in response to the crisis. This may mean quicker
recovery in these societies.
This research also contributes to the literature on religious coping. While previous research
has documented a rise in religiosity in the aftermath of natural disasters (Belloc et al., 2016;
11Google searches for Jigsaw puzzles also rose during the lockdowns, but to a much smaller extent than
searches on prayer and for a quite different set of countries: https://trends.google.com/trends/explore?
date=2020-01-29\%202020-04-01&q=\%2Fm\%2F0_d6n,\%2Fm\%2F0687y.
5

Bentzen, 2019; Bulbulia, 2004), these disasters do not hit all countries. For instance, Northern
European countries are rarely hit. One conclusion in the literature has been that mainly
the poor and vulnerable use religion in coping (Inglehart and Norris, 2003). The poor and
vulnerable are also hit more often by adversity, though. Instead, the COVID-19 pandemic hit
globally and thus provides a unique experiment to study which types of societies use religion
for coping. It turns out that the poor and vulnerable are not the only users of religion in
coping.
That people use religion in crisis further speaks to one of the puzzles of the social sciences:
Religion has not vanished as societies modernize as early scholars otherwise predicted.12 On
the contrary, the role of religion is on the rise in many places and today 83% of the global
population believe in God.13 The use of religion in crisis may be one explanation.
More generally, the results inform a literature on the socio-economic consequences of reli-
giosity. Scholars have found two opposing mechanisms: Prosocial behavior on the one hand
and more conservative values and behavior on the other.14 One potential outcome of the use
of religion in crisis is reduced economic fluctuation. Another may work through the potential
permanent effects of the crisis on religiosity. Whether the crisis has such permanent effects is
impossible to say at this time.
2 Religious coping
The tendency for people to use religion to deal with crisis can be understood within the
religious coping terminology.15 The theory states that people use religion as a means to cope
with adversity and uncertainty. They pray, seek a closer relation to God, or explain the tragedy
by reference to an Act of God. Research has documented that people who experienced adverse
life events, such as cancer, heart problems, death in close family, divorce, or injury are more
religious than others (Ano and Vasconcelles, 2005; Pargament, 2001). Studies have further
documented that the impact is causal: Priming subjects in an experiment with thoughts of
death makes them rank themselves as being more religious (Norenzayan and Hansen, 2006),
churchgoing rose in the aftermath of the 1927 Great Mississippi flood (Ager et al., 2016),
people hit by the Christchurch 2012 earthquake reported increased religiosity (Sibley and
Bulbulia, 2012), and natural disasters more broadly instigates people across the globe to use
their religion more intensively (Bentzen, 2019). They are more likely to rank themselves as
a religious person, find comfort in God, and to state that God is important in their lives
12Marx (1844); Weber (1905); Durkheim (1912); Freud (1927).
13Norris and Inglehart (2011); Stark and Finke (2000); Iannaccone (1998).
14Guiso et al. (2003); Barro and McCleary (2003); Squicciarini (2019); Belloc et al. (2016); B´enabou et al.
(2015); Inglehart and Norris (2003); Shariff and Norenzayan (2007); Norenzayan et al. (2016).
15Pargament (2001); Norenzayan and Hansen (2006); Cohen and Wills (1985); Park et al. (1990); Williams
et al. (1991).
6

when hit by earthquakes, tsunamis, and volcanic eruptions. This surge in average religiosity
occurs on all continents, for people belonging to all major religions, income groups, and from
all educational backgrounds. Recent research also found that people who experienced conflict
are more religious (Henrich et al., 2019) and that earthquakes increased the power of religious
authorities in Medieval Italy (Belloc et al., 2016).
Using religion for coping is part of what is termed emotion-focused coping, in which people
aim to reduce the emotional distress arising from a situation (Lazarus and Folkman, 1984).
While people use religion for coping with various types of situations, religion is used mainly
for coping with negative and unpredictable situations (Pargament, 2001; Bjorck and Cohen,
1993; Smith et al., 2000). On the other hand, when we face perceived negative, but predictable
events, such as an approaching job interview, we are more likely to engage in problem-focused
coping, where we aim to directly tackle the problem that is causing the stress. Folkman and
Lazarus (1980) found that work contexts favor problem-focused coping, and health contexts
favor emotion-focused coping. Also, religiosity increases more in response to unpredictable
natural disasters, such as earthquakes, tsunamis, and volcanic eruptions compared to more
predictable ones, such as storms and in response to earthquakes in areas that are otherwise
rarely hit compared to frequently hit areas (Bentzen, 2019). Being a negative and highly
unpredictable health-related event, the COVID-19 crisis certainly fits the criteria for being an
event that could instigate religious coping. As of April 20 2020, the COVID-19 had affected
210 countries and territories, infected more than 2.4 mio. individuals worldwide and taken
more than 165,000 lives.
People are more likely to use their intrinsic religiosity to cope with adversity rather than
their extrinsic religiosity (Johnson and Spilka, 1991; Pargament, 2001). Intrinsic religiosity
involves private prayer and one’s personal relation to God, while an example of extrinsic
religiosity is going to church for social needs or other more ultimate ends than beliefs per se
(Allport and Ross, 1967). When faced with adversity, people are thus more likely to use their
private beliefs to cope rather than to go to church. This is much like depressed individuals not
feeling like socializing (Jacobson et al., 2001). The most frequently mentioned coping strategies
among 100 older adults dealing with stressful events were faith in God, prayer, and gaining
strength from God. Social church-related activities were less commonly noted (Koenig et al.,
1988). Miller et al. (2014) found that individuals for whom religion is more important had lower
depression risk (measured by cortical thickness), while frequency of church attendance was not
associated with thickness of the cortices. Schnittker (2001) found that religious salience and
spiritual help-seeking exhibit significant stress-buffering effects, but find no such effects of
attendance at religious services. A recent survey reveals that 95% of Americans who pray,
pray alone, while only 2% pray collectively in a church (Barna, 2017). Also, natural disasters
7

increase private religious beliefs and affect churchgoing much less (Bentzen, 2019). We would
therefore expect the COVID-19 pandemic to impact private prayer more than churchgoing,
had the churches been open.
The use of religion for coping with adversity has improved the mental health of many
practitioners (reviews in Pargament (2001); O. Harrison et al. (2001); Schnittker (2001)).
Miller et al. (2014) found that religious individuals have thicker cortices, a measure of lower
depression risk. Lang et al. (2015) primed subjects in an experiment on Mauritius with
anxiety by having them plan for the next natural disaster. They measured heightened stress
levels for the subjects, but these stress levels were lower for subjects allowed to perform their
usual religious rituals afterwards, compared to subjects that were asked to sit and relax in a
non-religious space.
The intensified use of religion may translate into a permanently stronger role of religion,
even after disaster has passed. Through 129 retrospective interviews of elderly individuals,
Ingersoll-Dayton et al. (2002) found that events such as death of close family or friends intensi-
fied participants’ felt relationship with God permanently. Bentzen (2019) found that while the
main surge in religiosity occurred during the few years immediately following earthquakes, a
residual of elevated religiosity remained and was passed on to future generations. This resulted
in significant differences in religiosity depending on natural disaster risk in parents’ country
of origin, even for children of migrants who never lived in the disaster-prone countries. Thus,
natural disasters or death in close family can strengthen the role of religion permanently. Only
time will show whether the same is true for the COVID-19 crisis.
Examples abound of people using prayer as a way of dealing with the uncertainty and fear
surrounding the COVID-19 outbreak. While the title of a sermon by Pastor Robert Jeffress
at an Evangelical Christian megachurch in Dallas asks “Is the Coronavirus a Judgement from
God?”, political leaders from Mr. Akufo-Addo of Ghana to Mr. Morrison of Australia urge
their populations to pray as the coronavirus finds its’ way into their economies.16
16Newspaper articles: https://www.firstdallas.org/blog/is-the-coronavirus-a-judgment-from-god/,
https://www.iol.co.za/news/africa/ghana-president-nana-akufo-addo-calls-for-prayer-fasting-for-covid-19-
45421569, https://www.canberratimes.com.au/story/6692098/my-prayer-knees-are-getting-a-good-work-out-
pms-coronavirus-address/.
8

3 Data: The rise in prayer intensity
To identify which countries experienced an increased interest in prayer and whether some are
more likely than others to use religion for coping, four types of databases were constructed (see
also Appendix A). First, a database on Google searches for topics related to prayer as a share
of total Google searches for the 95 countries in the world with enough internet users (from
Google Trends).17 These searches include all topics related to prayer, including alternative
spellings and searches for prayer in other languages. Two series of data were constructed:
Daily data for all 95 countries for the period January 29 to April 1 2020 and global weekly
data from 2016 to 2020. The weekly data going back to 2016 is only used for Fig. 1. The
econometric analysis using weekly data will be based on the same period as the daily data.
The series stop on April 1, well before the onset of Easter 2020 (Palm Sunday was April 5)
and the Ramadan (first day of the Ramadan 2020 was April 24). The daily series start on
January 29, after the January holidays and after the fires in Australia. January 29 is chosen
to February 1 to get a sample consisting of full weeks, which does not matter for the analysis
using daily data, but could matter for the analysis aggregating the daily data to weeks.
Google Trends provides two types of data: Time-series data and cross-section data. The
time-series data is available for one a country at a time or as an average for the world.
The cross-section data is available for countries or subnational regions as an average over a
specified period of time.18 For the time-series data, Google Trends normalized the search
shares to equal 100 for the highest search share during the period for each country. For the
cross-country data, the search share was set to 100 for the country with the highest search
shares in the sample. This means that only the growth rates, and not the levels, of the time-
series data have a meaningful interpretation and can be compared across countries. For the
cross-country data, the levels can be compared across countries. The analysis includes country
fixed effects throughout and thus does not compare countries, but the analysis in Section C.2
identifies the characteristics of the countries who pray more, which means that comparison
across countries occurs. To construct a panel dataset, I combined the growth rates from the
time-series data with the levels from the cross-section data. I set the level of prayer search
shares on January 29 equal to the average prayer search shares for 2019, and calculate the
search shares for the rest of the period based on the growth rates from the time-series data.19
17Google Trends is used in previous research to predict real behavior, such as Goel et al. (2010); Askitas and
Zimmermann (2009); D’Amuri and Marcucci (2017); Choi and Varian (2012); Ginsberg et al. (2009); Walker
et al. (2020); Olivola et al. (2019); Goldstone and Lupyan (2016); Cavazos-Rehg et al. (2015).
18The current analysis uses countries instead of subnational regions, as no time-series data exist for the
regions.
19What I would really like to do is to combine the level of the search shares on January 29 with the growth
rates since January 29, but the daily data is much more inaccurate than the yearly average. Therefore, I use
the average for 2019 as a proxy for the level on January 29 2020. An example of the procedure is as follows.
9

Most tables and figures are based on these comparable data, except Fig. 1 (and other
figures in the Appendix that use world aggregates) which includes the raw data from Google
Trends, Fig. 2 where the search shares in all countries are instead normalized to 1 on February
15 2020, Panel (a) of Fig. 3 and Tables A.14-A.15, which are based exclusively on the growth
rates in prayer search shares. These results are not affected by the procedure described.
The Google searches for prayer will fall as people access their prayer websites directly with-
out googling them or memorize the prayers. Likewise, searches on prayer surge dramatically
on the first week of the Ramadan only to drop the week after, even though Muslims pray every
day during the Ramadan (cf. Fig. 1). An increasing prayer share reveals that new people are
searching for prayer or people who already searched for prayer are searching for prayer again
(one person googling prayer many times over a short period of time will not enter the search
data many times, though). Thus, falling search shares for prayer are difficult to interpret.
Therefore, observations are dropped after the prayer search shares reached their maximum
level in Figures 2 and 3. Most remaining figures and maps include the full dataseries, unless
stated otherwise. These are therefore conservative estimates.
A second database identifies what people are searching for when searching for the topic
prayer (see also B.2). Apart from searches for prayer in different languages, the four search
queries that contribute the most to the rise in search shares for prayer are ”prayer for coro-
navirus”, ”pray for the world”, ”spiritual communion prayer”, and ”pray for italy” (cf. Fig.
A.6). When googling ”prayer for coronavirus”, various websites offer prayers related to the
coronavirus. These include prayers to prevent the virus from spreading and prayers to thank
nurses and other care-takers for their work in relation to the pandemic (see Appendix B.2).
The third database consists of daily data on registered cases and deaths by COVID-19 for
each country of the world (see also Appendix A.2). These numbers depend on the amount of
testing in each country and general policies regarding registration of cases and deaths, and are
therefore neither comparable across countries nor across time (where policies may change).
Inclusion of country fixed effects throughout takes care of the difficulty of comparison across
countries, but does not account for the difficulty of comparison over time. As an attempt to
account of the latter, measures of the timing of the first case or death will be used, but the
main results will depend on a measure independent of the registered cases and deaths: The
point in time when COVID-19 was declared a pandemic; March 11 2020.
Fourth, to identify the characteristics of those who search more for prayer, the database
with Google searches for prayer was combined with data on various characteristics of the
The average prayer search share in 2019 was 3 for Denmark, while that in Morocco was 87. I therefore set the
prayer search share on January 29 to 3 in Denmark and 87 in Morocco. From January 29 to 30, prayer search
shares rose by 68% in Denmark and by 6% in Morocco. The prayer search share on January 30 2020 therefore
amounts to 5.1 in Denmark and 92.4 in Morocco, and so forth.
10

countries, such as religiosity levels before COVID-19, the share of Christians, Muslims, Hindus,
and Buddhists, and various socio-economic characteristics.
4 Results
To parsimoniously illustrate the findings, Fig. 2 shows daily search-shares for prayer during
the period February 15 to April 1 2020 for all 95 countries, split into fourteen regions. Each
panel shows two groups of countries within the particular region. The darker curves represent
the average for the particular group, while the lighter curves represent the raw data for each
country. The search-shares are set to 1 on February 15, which means that the figure shows the
change in search-shares, relative the initial level of searches for prayer in the particular country.
The vertical line represents March 11, where WHO declared the COVID-19 a pandemic.
Search-shares for prayer rose around mid March for most regions, even for the most secular
regions of Northern Europe.
The map in Panel (a) of Fig. 3 also shows the relative changes in prayer search shares.
The map illustrates the growth rate in prayer search shares from February to the highest
level reached in March: prayermarch −prayerf eb
prayerf eb .20 The growth rates are large for the more secular
Northern Europe, where few people searched for prayer before COVID-19. Likewise, the
somewhat smaller increases in growth rates in Northern Africa are due to the high initial
levels of prayer searches. Panel (b) of Fig. 3 documents the absolute increases in prayer
search shares, which is the relevant metric to identify the global spread of intensified prayer.
The largest absolute increases occur in South America and Africa, some of the most religious
regions of the world. The econometric analysis will rely on the absolute changes, while Tables
A.14 and A.15 document the statistically significant rises in growth rates.
Fig. 2 also showed that prayer search intensity rose just after March 11 for most of the
fourteen regions, the date when WHO declared the COVID-19 a pandemic. More formally,
Appendix B.5 documents that 94 of the 95 countries in the sample experienced significant
increases in prayer search shares between January 29 and April 1. For the majority of these
countries, the rise occurred on March 11 or later.
20To prevent a general rise in prayer search shares from being associated with COVID-19, potential in-
creases in February were subtracted from the numerator in both panels. The map is very similar without this
correction.
11

Figure 2: Daily Google searches for the topic ”prayer” by region
(a) Northern Europe (b) Southern Europe
(c) North and Middle America (d) South America
(e) SE Asia (f) Rest of Asia
(g) Africa
12

Google searches for prayer as a share of the total number of Google searches on the particular day, set to 1 on February 15 2020.
A country drops out of the sample after it reaches its’ peak during the period Feb 15 to Apr 1. The searches encompass topics
related to prayer, including alternative spellings and languages. The light-coloured lines represent a country. The darker-coloured
lines represent the average prayer intensity for the particular group. The countries behind the blue curves are italicized in the
following. Northern Europe:Belgium, Denmark, Finland, Netherlands, Norway, Sweden, Austria, France, Germany, Ireland,
Switzerland, United Kingdom. Southern Europe:Italy, Spain, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Czech
Republic, Greece, Moldova, Poland, Portugal, Romania, Slovak Republic, Ukraine, Yugoslavia. North and Middle America:
Canada, USA, Costa Rica, Dominican Republic, El Salvador, Guatemala, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama,
Trinidad and Tobago, South America,Brazil, Argentina, Bolivia, Chile, Colombia, Ecuador, Paraguay, Peru, Puerto Rico,
Uruguay, Venezuela, SE Asia:Australia, New Zealand,Rest of Asia:Cyprus, Iran, Israel, Jordan, Kuwait, Lebanon, Qatar,
Saudi Arabia, Turkey, United Arab Emirates, Azerbaijan, Bangladesh, Georgia, India, Kazakhstan, Pakistan, Russia, Sri Lanka,
Africa:Egypt, Morocco, Tunesia, Algeria, Angola, Cameroon, Ghana, Ivory Coast, Kenya, Nigeria, South Africa, Tanzania,
Uganda, Zambia. See more details in Appendix A.1 and C.
Figure 3: The rise in prayer search shares across the globe in March 2020
(a) Growth rate (%)
(b) Absolute increase
The map shows the rise in Google search shares from February 29 2020 to the highest level reached in March 2020 subtracted
the rise in February 2020 in Panel (b). Panel (a) shows the rise as a percentage of the average level during February 2020 (the
average rise was 91%, the minimum was 19%, and the maximum was 282%). Panel (b) shows the absolute rise (the average rise
was 24, the minimum was 0.5, and the maximum rise was 81 units). Darker green indicates larger rises in prayer search shares.
Missing data is indicated with grey. Find more details in Appendix A.1 and C.
13

4.1 Econometric analysis
To identify formally what Figures 1, 2, and 3 showed visually, the following equation was
estimated:
prayerct =β+γcovid19ct−1+δtc+κc+εct (1)
where prayerct measures the number of google searches on prayer in country con day tas a
share of total google searches on the same day for the same country. covid19ct−1captures the
exposure to COVID-19 using different measures: A dummy variable, pandemic, equal to one
on March 11 where WHO declared COVID-19 a pandemic, measures of the total number of
registered people infected by COVID-19 and the total number of deaths, a dummy equal to
one after the country registered its’ first case or death, days since the first case or death, and
days since March 11 (cf. Table 1 and Tables A.4-A.7). These variables are lagged a day in
the main analysis. Alternative specifications are investigated, such as adding squared terms,
aggregating to weekly data (cf. Tables A.4-A.7), and examining growth rates (Section C.4).
tcis a country-specific time-trend. This variable captures the general upward or downward
trend in prayer search shares for each country.21 κcis a list of country fixed effects, ensuring
that results are only compared within one country at a time. When the covid19ct−1variable is
the pandemic dummy or the first case or death dummy, γcan be interpreted as the average rise
in prayer search shares after March 11 or after the first registered case or death, respectively.
While it is theoretically probable that causality in equation (1) instead runs from religiosity
to COVID-19 exposure, this seems a highly unlikely explanation for the results. The increases
in prayer search shares documented here are the largest ever recorded. For reverse causality
to explain the results, one would have to come up with another explanation for this sudden
rise in prayer intensity. Also, the main results are based on the pandemic dummy, which does
not suffer from reverse causality or other endogeneity issues, as the WHO announcement was
done centrally and independent of country-specific conditions.22
Table 1 documents the estimates of equation (1), including country-fixed effects and
country-specific time trends throughout. The model in column (1) of Panel A documents
that prayer search shares rose with 5.1 units since March 11. This amounts to 16.9% of the
average prayer search shares over the period (30.2, calculated at the bottom of each Panel
in Table 1). The model in column (2) adds a measure of the number of days passed since
COVID-19 was declared a pandemic and documents that prayer search shares continued to
21This is a generalization of the subtraction of ∆prayerf eb done for Fig. 3 (cf. Appendix C).
22One may imagine that the general level of religiosity impacts the spread of COVID-19, which may poten-
tially influence results based on the infection and death rates. However, these results are based on the changes
in religiosity.Thus, the levels do not impact these results.
14

rise daily after March 11. After 10 days, prayer search shares had risen by 19.7% of the
mean ((2.45+3.5)/30.2), after 20 days 31.3% of the mean. The increase will probably not
continue linearly, especially since those who start to access their prayer websites directly with-
out googling them are not captured by the google search shares (analyzed more formally in
Tables A.4 and A.5). Only time will show how much further the search shares for prayer will
continue to rise.
Columns (3) and (6) document that prayer search shares rose after a country registered
its’ first case or death, but nearly half of this is due to WHO’s announcement on March 11
(columns 4 and 7). Fig. A.17 shows that these results are not caused by a distinct cluster
of observations. Instead, the likelihood of rising prayer shares varies very homogenously with
having passed March 11 or having registered the first case or death.
Columns (5) and (8) show that prayer search shares rose more after March 11 in countries
where the COVID-19 had already arrived. This result is even stronger when restricting the
sample to the sample where observations after prayer search shares reached their maximum
level are dropped (Table A.5). 23
In an attempt to circumvent endogeneity issues and the arguable lack of comparability of
registered deaths and cases over time and space, the remainder of the analysis will use the
pandemic dummy to measure the impact of COVID-19.
Panel B of Table 1 splits the sample into the different regions of the world and documents
that prayer search shares rose significantly in all regions after March 11. Again, the absolute
rise is larger in the Americas and Africa, where the overall search shares for prayer are also
higher (MeanDepVar at the bottom of the table shows the mean prayer search shares within
each region).24 The fact that the rise in prayer search shares is larger in Africa than in Europe,
even though very few cases had been detected in Africa at the time (cf. Fig. A.14), further
illustrates that the rise in prayer is caused by fear more than coping with the actual infection
or death. This is consistent with the finding by Bentzen (2019) that people used religion to
cope with natural disasters in neighboring districts, even when not hit directly by the disaster
themselves. It is also consistent with the spatial spread of the ”pray for italy” movement
which surged in several African countries located far from Italy (Fig. A.7).
23Regressions in columns (2), (4), and (7) all have Variance Inflation Factors around 3-4, which is above
the critical value of 1. This means that the precision of the estimates may be biased by multicollinearity, and
interpretation of the coefficients should be done with care. The VIF rises further in columns (5) and (8), but
this is exclusively due to the inclusion of interaction terms and thus causes no further cause for concern.
24The standard errors in columns (2)-(8) of Panel B are potentially biased, as they are clustered at only
11-18 clusters. Table A.6 keeps the full sample of 95 countries throughout, including instead interaction terms
between the pandemic dummy and each of the regions. The same conclusion is reached: Prayer search shares
rise within all regions. This regression further documents that the rise is significantly smaller in Europe than
the rest of the world.
15

Table 1: The impact of COVID-19 on prayer search shares
Dependent variable: Prayer search shares
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A
Pandemic dummy 5.11*** 2.45*** 4.77*** 2.07 4.49*** 3.87***
(0.752) (0.762) (0.752) (1.835) (0.795) (0.839)
Days since Pandemic 0.35***
(0.059)
First case dummy 2.92*** 1.62** 1.05
(0.848) (0.764) (0.750)
Pandemic x first case dummy 3.16*
(1.849)
First death dummy 3.89*** 2.47** 0.34
(0.960) (0.968) (1.532)
Pandemic x first death dummy 2.84*
(1.677)
R-squared 0.84 0.84 0.84 0.84 0.84 0.84 0.84 0.84
Observations 6080 6080 6066 6066 6066 6080 6080 6080
Countries 95 95 95 95 95 95 95 95
MeanDepVar 30.2 30.2 30.2 30.2 30.2 30.2 30.2 30.2
Panel B All N Europe S Europe N America S America SE Asia Rest Asia Africa
Pandemic dummy 5.11*** 2.50** 2.87** 7.11*** 8.72** 5.14* 4.34*** 5.98***
(0.752) (0.912) (1.281) (2.054) (3.906) (2.499) (1.458) (1.503)
R-squared 0.84 0.91 0.77 0.73 0.78 0.92 0.63 0.70
Observations 6080 768 960 832 704 768 1152 896
Countries 95 12 15 13 11 12 18 14
MeanDepVar 30.2 10.0 24.5 50.4 40.7 15.5 19.3 53.0
OLS estimates. Units: Days ×countries. Period: January 29 to April 1 2020. All regressions include a constant, country-specific time
trends, and country fixed effects. The sample includes the full sample in Panel A and in column (1) of Panel B, but varies across the
remaining columns of Panel B: Northern Europe in column (2), Southern Europe (3), North America (4), South America (5), South
East Asia and Oceania (6), the rest of Asia (7), and Africa (8). Robust standard errors clustered at the country level in parentheses.
*, **, and *** indicate significance at the 10%, 5%, and 1% level. Find more details in Appendix A.1.
Result: Prayer intensity increases day-by-day after the WHO announces COVID-19 a pandemic for all regions. The rise is marginally
larger in countries where the COVID-19 had physically arrived.
Google searches for prayer may rise for a reason unrelated to religious coping. Since the
churches closed down to prevent the disease from spreading, part of the intensified prayer
searches may be replacing physical church attendance. Theoretically, we would not expect
this to be the main explanation for the rising search shares for prayer, as churchgoing is and
extrinsic type of religiosity which is not the main type of religiosity used for coping with
adversity. Instead, the majority of people use their intrinsic religiosity, which includes private
prayer, when coping with adversity. Surely, the pandemic would most likely have resulted
in more churchgoers had the churches been open, as experienced in the USA after the 9-11
attacks. The theory on religious coping suggests, though, that the rise in private prayer would
be larger.
The econometrics also indicate that the rise in searches for prayer is due to religious coping
and not merely a shift from physical church to online church (Appendix C.3). First, data on
the specific contents of the Internet searches reveal that searches for topics related to ”internet
church” also rise, but exclusively in Sundays and compared to the rise in prayer search shares,
the increase is indistinguishable from zero (Fig. A.18). Second, the search shares on prayer
continue to rise long after the church closures (Fig. 2). Third, the rise in prayer searches does
16

not only occur on Sundays, where most masses are held. While the rise on Sundays is higher
than other days, the search shares rise on all days of the week, except Fridays.
4.2 Characteristics of those who pray more
This section examines who uses religion for coping. Reviewing 26 studies on religious cop-
ing, Pargament (2001) finds that religious coping is unsurprisingly used mainly by the more
religious. Seven of the studies examine heterogeneity with respect to income, education or
social status. Five of these find that there is no difference in the use of religious coping based
on income or education, while the remaining two studies find that poorer and less educated
individuals use religion more in coping. Based on this review, we would expect that prayer is
used more intensively in more religious societies, and not necessarily used more in poorer or
less educated societies. These studies, though, are based on rather small samples, the majority
below 300 individuals.
In a sample of appr. 100 countries, Norris and Inglehart (2011) documents a negative cor-
relation between more religious societies and various measures of development and insecurity.
Norris and Inglehart conclude that religion provides a sense of existential security, which is
most needed among vulnerable populations, especially those living in poorer nations, facing
personal survival-threatening risks. Based on this, one would expect religion to be used more
to cope emotionally with COVID-19 in poorer, more unequal, and more insecure states.
To examine the heterogeneity in the use of prayer to cope with COVID-19, I estimate the
following equation (see also Tables A.9 - A.12):
prayerct =β+γpandemict−1+λpandemict−1×characteristicc+δtc+κc+εct (2)
where characteristiccincludes different measures of country characteristics: the religiosity
level in country cbefore the onset of the COVID-19 pandemic, a dummy equal to one for each of
the major religions, and various socio-economic characteristics meant to measure development,
inequality, and insecurity. These measures of socio-economic characteristics are chosen from
the Quality of Government Dataset constructed by Teorell et al. (2020a). Measures available
for at least 70 countries in the baseline estimation were chosen. The rise in prayer search
shares after March 11 now equals γ+λcharacteristicc. If the rise in prayer search shares
after March 11 is larger for the more religious, certain religious denominations, or certain
socio-economic characteristics, this is captured by λ > 0. Apart from the interaction term,
the regression is the same as before (with country-specific time-trends, δtcand country-fixed
effects, κc).
17

Panel (a) of Fig. 4 shows the estimates of equation (2) for different religiosity levels in
2019, measured by the average search shares for prayer in 2019.25 Prayer search shares rose
significantly for all quartiles of previous prayer intensity, but rose more for the countries that
prayed more in 2019. For instance, prayer search shares rose more than five times more in the
most religious quarter of countries, compared to the least religious. This pattern is confirmed
when using instead global surveys conducted well before the COVID-19 pandemic (the World
Values Survey and European Values Study). Prayer searches rose more in countries where
a larger share of the population answered that they prayed more, went more to church, or
felt that God was important in their lives (Tables A.9 - A.11). Also, prayer searches rose
significantly even in the 15% least religious countries for most measures of previous religiosity
(Table A.9). Among the 10% least religious countries, prayer search shares rose significantly
for only 4 out of 9 religiosity measures. The 10% least religious countries are the Czech
Republic, Denmark, Finland, Germany, Japan, the Netherlands, Norway, Sweden, Taiwan,
Thailand, and Vietnam. Thus, Northern European countries, formerly communist countries
(that prohibited religion), and Buddhist majority countries that were hit early by COVID-19
did not experience significant increases in search shares.
Previous religiosity levels may be endogenous and correlate with various other country-
characteristics. To exploit instead exogenous variation in religiosity, I rely on previous research
that documented that earthquakes increase religiosity due to religious coping (Belloc et al.,
2016; Bentzen, 2019; Bulbulia, 2004). Confirming the results in Panel (a), Panel (b) of Fig.
4 shows that prayer search shares rose more in countries with more earthquake risk. The
measure of earthquake risk is based on the distance to high-risk earthquake zones constructed
by Bentzen (2019). Earthquake risk is calculated as 2,512−D istance
2,512 , where 2,512 is the maximum
distance to high-risk earthquake zones in the sample. The sample is further restricted to
countries located within 1,500 km of an earthquake zone, as varying degrees of earthquake
risk does not matter for religiosity at these low levels of earthquake risk.26
Panel (c) of Fig. 4 shows that prayer search intensity rose for Christians (particularly
Catholics), Muslims, Hindus, and Buddhists, but insignificantly so for the latter two. The
countries are categorized into the major denominations based on there being at least 25%
adherents to the particular denomination. The insignificance for Hindus and Buddhists is
mainly due to the larger standard errors, which are caused by rather few countries in these
two groups. The two only countries defined as Hindu in the sample are India and Trinidad
and Tobago, while countries defined as Buddhist are Japan, Sri Lanka, Taiwan, Thailand,
25Four dummy variables were constructed based on the quartiles of the prayer search share in 2019 and
equation (2) was run for each of them. Each dot in Fig. 4 represents γ+λfor each of the dummies.
26More details on the earthquake risk measure are available in Section A.3.
18

and Vietnam. The parameter estimate of prayer search shares for Hindus and Buddhists is
of similar size as that for Protestants. Fig. A.16 documents that Google searches for god,
allah, jesus, mohammad, bible, quran, buddha, vishnu, and shiva also rose in March 2020.
Combined with the results in Fig. 4, this indicates that Hindus and Buddhists most likely
also use religion for coping. For the search terms on buddha, vishnu, and shiva, the rise in
March is not larger than other holy events during the year, such as Buddhas birthday or
Hindu holidays for Lord Shiva or Lord Vishnu. Thus, while Hindu and Buddhist traditions
may also use religion for coping, these traditions seem more focused on celebration. The
finding that all major religions use religion in coping is consistent with findings by Abu-Raiya
and Pargament (2015) and observations by Pargament (2001) (p.3): ”while different religions
envision different solutions to problems, every religion offers a way to come to terms with
tragedy, suffering, and the most significant issues in life.”
Another interpretation of the lower impact among the Buddhist countries is that these
countries were infected by COVID-19 before the virus was declared a pandemic and they have
experienced more epidemics than the rest of the world during the past 50 years. Thus, the
rise in fear and emotional distress around March 11 may have been lower in these countries,
instigating a lower need for coping.
19

Figure 4: The rise in prayer search shares for different religiosity and denominations
(a) Prayer search shares in 2019
(b) Earthquake risk
20

Figure 4: cont. The rise in prayer search shares for different religiosity and denominations
(c) Religious denominations
The rise in prayer intensity for different prayer search shares in 2019 in Panel (a), different earthquake risk intensities in Panel
(b), divided into quartiles, and different major religious denominations in Panel (c). Each dot represents the estimate of the
rise in prayer search shares after March 11 in an OLS regression, where the rise in prayer search shares is allowed to vary with
initial religiosity levels in Panels (a) and (b) and with religious denominations in Panel (c), controlling for country-specific time
trends and country fixed effects. The denominations are defined based on there being at least 25% adherents of the particular
denomination in a country. The horizontal lines represent the 95% confidence bounds. See more details in Appendix A.1 and C.2.
Result: Prayer intensity rose at all levels of previous religiosity and major religious denominations, but more in more religious
countries.
To test whether prayer has been used more to cope emotionally with COVID-19 in poorer,
more unequal, and more insecure states (call this the vulnerability hypothesis) or whether
poor and rich use prayer equally much for coping, Table 2 shows estimates of Equation (2)
adding interaction terms with different measures of economic development, inequality, and
mortality.
The model in Panel A of Table 2 seems to confirm the expectation. Prayer searches rose
more in poorer countries, where development is measured by GDP per capita (col 1) and
the share of people living below 1.9US$ a day (col 2), less educated countries (col 3); more
unequal countries, where inequality is measured by the degree of economic inequality (the Gini
coefficient, col 4) and a measure of the degree to which economic development is unevenly
distributed (col 5); more fragile states (col 6); states with higher adult mortality rates (col 7);
and in states with lower quality property rights institutions (col 8). That is, prayer searches
rose more in poorer, more unequal, and more insecure states. Table A.12 finds similarly that
the rise in prayer search shares is larger in countries with lower security and lower quality
public services, larger demographic pressures, lower development measured by the human
21

development index, and higher infant mortality rates.
A competing hypothesis to the vulnerability hypothesis is that religion is used more for
coping in societies where religion plays a more prominent role. According to this hypothesis,
Canadians would use religion for coping as much as Kazakhstanis, since their religiosity levels
are very similar.27 According to the vulnerability hypothesis, however, Kazakhstanis would
use religion much more in coping, as they rank within the bottom third in the global income
distribution, while Canada belongs to the top 10%. The remaining panels of Table 2 attempt
to disentangle the two hypotheses.
Panel B documents that the differential effects in Panel A are due to poorer and more
insecure countries being more religious: When adding an interaction term with religiosity
(measured by average prayer search shares in 2019), the mentioned effects fall by nearly
a factor 10 in most columns and the estimates turn insignificant (without much different
standard errors). The only significant variable is the interaction between the pandemic dummy
and the prayer search shares in 2019: Religious countries are more likely to search for prayer
on the Internet in the face of COVID-19.
However, some societies may be more religious because they are poor, unequal, and un-
certain (in keeping with the secularization hypothesis)28 and thus some of the impact of the
socio-economic confounders may work through religiosity. To obtain variation in religiosity
that is independent of other confounders, I again exploit variation caused by earthquake risk.
Panel C restricts the sample to countries within 1,500 km of high-risk earthquake zones
and replicates the estimation in Panel A on this sample. Panel D instruments the interaction
between the pandemic dummy and prayer search shares in 2019 with an interaction between
the pandemic dummy and earthquake risk. The First stage F statistic is above 10 in most
specifications, which means that the instrument is valid.29 The standard errors on the interac-
tion term between the pandemic dummy and the socio-economic confounders are again nearly
unchanged, compared to those in Panel C. Instead, the size of the parameter estimate drops
by nearly a factor 5 across all columns and becomes again insignificant. The instrumented
interaction with prayer search shares in 2019 remains highly significant.30 The results confirm
that the heterogeneity with respect to the socio-economic characteristics found in Panel A is
due to the fact that poorer, more unequal and insecure countries are also more religious. The
2789% and 90% believe in God, respectively, according to the World Values Survey.
28The secularization hypothesis states that the role of religion diminishes as countries modernize, original
put forth by Marx (1844); Weber (1905); Durkheim (1912); Freud (1927).
29The exclusion restrictions are rather unlikely to be violated: Earthquake risk most likely does not influence
the rise in prayer search shares after March 11 2020 through other channels than previous religiosity levels.
30The OLS estimate on the interaction with prayer search shares in 2019 in the reduced sample is 0.16
throughout. The IV estimate is therefore double the size of the OLS estimate, but this difference is statistically
insignificant.
22

results in Table A.12 confirm this, as all the additional characteristics turn insignificant in
Panels B and D.
This means that prayer shares rose in all countries, independent of their economic status,
whether or not they are unequal, fragile or more mortal. The only thing that matters for
whether people use religion for coping or not is how religious they are to start, consistent
with the competing hypothesis. Kazakhstanis use religion in coping as much as Canadians
do. Indeed, the changes in prayer search shares in the two countries are similar: 7 and 11,
respectively in a distribution ranking from 0.5 to 81 with a median of 18. These results
are consistent with the majority of the studies surveyed by Pargament (2001) and with the
results by Bentzen (2019). One way to reconcile the results is that COVID-19 did not increase
emotional distress more in poorer countries than in their richer counterparts, as mortality rates
are already high in the former. If emotional distress did rise more in poorer societies, another
interpretation is that the availability of religion as a coping tool seems to be more important
than the need for such a tool. Either way, COVID-19 generates a need for emotional coping,
and societies use religion to cope, independent on whether they are rich or poor, uncertain
or secure. These results also mean that studies documenting differential effects of religious
coping for poor and vulnerable societies should be aware that these differential effects could
be simply a result of higher religiosity levels in these societies.
23

Table 2: The rise in prayer search shares across country characteristics
Dependent variable: Prayer
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A GDP Poverty Educ Gini Uneven Fragile AdultMort PropRights
Pandemic dummy 18.4*** 5.34*** 8.29*** -4.45 0.34 1.76 2.91** 8.03***
(5.096) (0.948) (1.238) (3.525) (1.516) (1.159) (1.139) (1.900)
Pandemic x Variable -1.48*** 0.32*** -0.17*** 0.25** 0.95*** 0.057** 0.018** -0.060**
(0.519) (0.083) (0.046) (0.100) (0.355) (0.024) (0.008) (0.027)
R-squared 0.84 0.85 0.85 0.84 0.84 0.84 0.84 0.84
Observations 6016 4544 5504 5248 5824 5824 5888 5952
Countries 94 71 86 82 91 91 92 93
Panel B
Pandemic dummy 4.21 1.68* 2.69* -2.16 1.08 1.73 1.67 1.57
(5.310) (0.951) (1.355) (3.588) (1.460) (1.085) (1.046) (2.295)
Pandemic x Variable -0.29 0.040 -0.051 0.11 0.080 -0.0081 -0.0046 -0.0047
(0.518) (0.158) (0.045) (0.115) (0.460) (0.027) (0.012) (0.028)
Pandemic x Prayer 2019 0.14*** 0.16*** 0.14*** 0.12*** 0.14*** 0.15*** 0.16*** 0.15***
(0.037) (0.052) (0.043) (0.042) (0.049) (0.041) (0.046) (0.040)
R-squared 0.84 0.85 0.85 0.84 0.84 0.84 0.84 0.84
Observations 6016 4544 5504 5248 5824 5824 5888 5952
Countries 94 71 86 82 91 91 92 93
Panel C
Pandemic dummy 15.9** 4.37*** 6.96*** -5.55 -0.81 1.24 1.38 7.85***
(6.857) (0.874) (1.214) (3.775) (1.527) (1.401) (1.313) (2.084)
Pandemic x Variable -1.21* 0.88*** -0.11** 0.28** 1.19*** 0.061* 0.029** -0.062**
(0.701) (0.218) (0.046) (0.109) (0.388) (0.031) (0.012) (0.029)
R-squared 0.87 0.88 0.88 0.89 0.87 0.87 0.87 0.87
Observations 4352 3456 4096 3904 4352 4352 4416 4416
Countries 68 54 64 61 68 68 69 69
Panel D
Pandemic dummy -7.39 -1.75 -4.12 -3.21 -0.0036 -0.050 -2.04 -3.15
(8.748) (1.760) (3.688) (3.659) (1.693) (1.443) (1.623) (3.592)
Pandemic x Variable 0.56 0.20 0.073 0.075 -0.66 -0.060 0.0052 0.019
(0.796) (0.281) (0.085) (0.114) (0.868) (0.049) (0.013) (0.031)
Pandemic x Prayer 2019 0.32*** 0.31*** 0.36*** 0.24** 0.36** 0.38*** 0.28*** 0.32***
(0.102) (0.118) (0.128) (0.093) (0.154) (0.131) (0.094) (0.113)
R-squared 0.87 0.88 0.88 0.89 0.87 0.87 0.87 0.87
Observations 4352 3456 4096 3904 4352 4352 4416 4416
Countries 68 54 64 61 68 68 69 69
FirstStageF 12.2 14.9 11.4 12.0 7.17 8.62 21.3 8.87
OLS estimates. Units: Days ×countries. Period: January 29 to April 1 2020. All regressions include a constant, country-
specific time trends, and country fixed effects. Panel A includes an interaction between the Pandemic dummy and various
socio-economic variables described in the text and in Appendix A.4. Panel B includes also an interaction between the
Pandemic dummy and prayer search shares in 2019. Panel C restricts the sample to countries within 1500 km of high-risk
earthquake zones and estimates the regressions in Panel A again. Panel D instruments the interaction between prayer
search shares in 2019 and the pandemic dummy with an interaction between earthquake risk and the pandemic dummy. The
scalar FirstStageF is the Kleibergen-Paap first stage F statistic. Robust standard errors clustered at the country level in
parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% level. See more details in Appendix A.1 and C.2.
Result: Prayer search shares rose more in poor, unequal, and insecure countries. But this is exclusively because these
societies are more religious.
24

5 The relative size of the rise in prayer
To get a sense of the relative size of the rise in actual prayer, the following back-of-the-envelope
calculation was made. A conclusion from Section C.2 is that the factor that matters most
for the difference in the size of the rise in prayer search shares is existing religiosity levels.
Combining this insight with results from a Pew Research Center survey from March 2020
showing that 55% of American adults had prayed to end the coronavirus (Pew, 2020b), the
global average rise in prayer related to COVID-19 can be backed out. The average religiosity
across all religiosity measures used in the analysis (cf Table A.9), weighted with the population
size in each country yields the number 0.654 (with a standard deviation of 0.19).31 The
religiosity level in the USA is 0.671. Since previous religiosity levels is the only factor that
matters for the intensity of the use of religious coping, a back-of-the-envelope estimate of the
share of the people in the sample that prayed for the coronavirus is therefore very close to
55%.32 The sample of 95 countries represents 68% of the world population33 and the average
religiosity level in the sample is no different from the average in the countries outside the
sample with information on the survey-based religiosity measures. Thus, the back-of-the-
envelope calculation shows that more than half of the world adult population have prayed to
end the coronavirus. This large number is reconcilable with the finding that the rise in Google
searches for prayer is larger than searches for the topic takeaway and amounts to 12% the rise in
searches for Netflix, and 26% the fall in searches for flights, which all changed tremendously in
the month of March 2020, where most of the world’s countries were in lockdown (cf Appendix
B.4).34
6 Conclusion
Google searches for prayer provides a measure of the interest in prayer in real time. In March
2020, Google searches for prayer rose to the highest level ever recorded, which indicates an
increased interest in prayer in the midst of the COVID-19 pandemic. A back-of-the-envelope
calculation shows that more than half of the world’s adult population have prayed to end the
coronavirus. This increased interest in prayer occurs on all continents, for all major religious
denominations, and for all levels of development and insecurity.
The rising prayer intensity is a result of religious coping: When faced with uncertainty
31All religiosity measures were scaled between 0 and 1.
32The share of Catholics in the USA is 23%, close to the global average of 17%, but the share of Protestants
is 48.9%, much higher than the global average of 12%. Since the rise in prayer shares for Protestants is lower
than both Catholics and Muslims, the estimate is conservative.
335.15 bio. / 7.55 bio. people.
34If the socio-economic variables analyzed in Table 2 do matter for coping after all, the estimated share of
people praying would be even larger, as the USA is at the top of the world’s income distribution.
25

and adversity, humans have a tendency to use religion for comfort and explanation. The
results reveal that the majority of the world experienced emotional distress in the face of the
COVID-19 pandemic, and they used religion to cope. The use of religion for coping is - not
surprisingly - more pervasive for more religious societies. Religious coping, though, occurs
at all levels of religiosity, except for the 10% least religious. The use of religion in coping
may explain the puzzle that the role of religion has not declined as early scholars otherwise
predicted: All types of societies use religion for coping with adversity, even modern secular
societies, and adversity has not vanished.
If religion dampens the emotional distress caused by COVID-19, this influences the general
well-being of societies. All other things equal, we would expect COVID-19 to cause less
emotional distress in more religious societies. Religious coping may even reduce economic
anxiety and through that reduce economic fluctuation caused by economic anxiety. Whether
religion has had such an effect on economic fluctuation is up for future research to determine.
Previous research found that disasters leave a long-lasting impact on religiosity, which can
influence the economy in various ways. Whether the COVID-19 pandemic will have similar
long-term effects remains an open question. More speculatively, if the COVID-19 pandemic
can have such a dramatic impact on one of the deepest rooted of human behaviors, what else
can it influence?
26

References
Abu-Raiya, Hisham and Kenneth I Pargament, “Religious coping among diverse re-
ligions: Commonalities and divergences,” Psychology of Religion and Spirituality, 2015, 7
(1), 24.
Ager, P., C. W. Hansen, and L. Loenstrup, “Church Membership and Social Insurance:
Evidence from the Great Mississippi Flood of 1927,” University of Southern Denmark Dis-
cussion Papers on Business and Economics, 2016, 7.
Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and
Romain Wacziarg, “Fractionalization,” Journal of Economic growth, 2003, 8(2), 155–194.
Allport, G.W. and J.M. Ross, “Personal religious orientation and prejudice.,” Journal of
personality and social psychology, 1967, 5(4), 432.
Andersen, Asger Lau, Emil Toft Hansen, Niels Johannesen, and Adam Sheri-
dan, “Pandemic, Shutdown and Consumer Spending: Lessons from Scandinavian Policy
Responses to COVID-19,” Unpublished manuscript, 2020.
Ano, Gene G and Erin B Vasconcelles, “Religious coping and psychological adjustment
to stress: A meta-analysis,” Journal of clinical psychology, 2005, 61 (4), 461–480.
Askitas, Nikolaos and Klaus F Zimmermann, “Google Econometrics and Unemployment
Forecasting,” Applied Economics Quarterly, 2009, 55 (2), 107.
Bailey, Michael, Eduardo D´avila, Theresa Kuchler, and Johannes Stroebel, “House
price beliefs and mortgage leverage choice,” The Review of Economic Studies, 2019, 86 (6),
2403–2452.
, Ruiqing Cao, Theresa Kuchler, and Johannes Stroebel, “The economic effects of
social networks: Evidence from the housing market,” Journal of Political Economy, 2018,
126 (6), 2224–2276.
Barna, Research, “Silent and Solo: How Americans Pray,” Barna Group, 2017, June 5.
Barro, Robert J and Rachel M McCleary, “Religion and economic growth across coun-
tries,” American Sociological Review, 2003, 68 (5), 760.
Belloc, Marianna, Francesco Drago, and Roberto Galbiati, “Earthquakes, Religion,
and Transition to Self-Government in Italian Cities,” The Quarterly Journal of Economics,
2016, 131 (4), 1875–1926.
27

B´enabou, Roland, Davide Ticchi, and Andrea Vindigni, “Forbidden fruits: the po-
litical economy of science, religion, and growth,” Technical Report, National Bureau of
Economic Research 2015.
Bentzen, Jeanet Sinding, “Acts of God? Religiosity and Natural Disasters Across Subna-
tional World Districts,” The Economic Journal, 2019, 129 (622), 2295–2321.
Binder, Carola, “Coronavirus fears and macroeconomic expectations,” Available at SSRN
3550858, 2020.
Bjorck, Jeffrey P and Lawrence H Cohen, “Coping with threats, losses, and challenges,”
Journal of Social and Clinical Psychology, 1993, 12 (1), 56–72.
Bulbulia, J., “The cognitive and evolutionary psychology of religion,” Biology and Philoso-
phy, 2004, 19 (5), 655–686.
Campante, Filipe and David Yanagizawa-Drott, “Does religion affect economic growth
and happiness? Evidence from Ramadan,” The Quarterly Journal of Economics, 2015, 130
(2), 615–658.
Cavazos-Rehg, Patricia A, Melissa J Krauss, Edward L Spitznagel, Ashley Lowery,
Richard A Grucza, Frank J Chaloupka, and Laura Jean Bierut, “Monitoring of
non-cigarette tobacco use using Google Trends,” Tobacco control, 2015, 24 (3), 249–255.
Choi, Hyunyoung and Hal Varian, “Predicting the present with Google Trends,” Eco-
nomic record, 2012, 88, 2–9.
Cohen, S. and T.A. Wills, “Stress, social support, and the buffering hypothesis,” Psycho-
logical bulletin, 1985, 98 (2), 310–357.
Coibion, Olivier, Dimitris Georgarakos, Yuriy Gorodnichenko, and Maarten Van
Rooij, “How does consumption respond to news about inflation? Field evidence from a
randomized control trial,” Technical Report, National Bureau of Economic Research 2019.
D’Amuri, Francesco and Juri Marcucci, “The predictive power of Google searches in
forecasting US unemployment,” International Journal of Forecasting, 2017, 33 (4), 801–
816.
Durkheim, ´
Emile,Les formes ´el´ementaires de la vie religieuse: le syst`eme tot´emique en
Australie, Vol. 4, Alcan, 1912.
28

Fetzer, Thiemo, Lukas Hensel, Johannes Hermle, and Chris Roth, “Coronavirus
perceptions and economic anxiety,” arXiv preprint arXiv:2003.03848, 2020.
Folkman, Susan and Richard S Lazarus, “An analysis of coping in a middle-aged com-
munity sample,” Journal of health and social behavior, 1980, pp. 219–239.
Freud, Sigmund,The future of an illusion, Broadview Press, 1927.
Ginsberg, Jeremy, Matthew H Mohebbi, Rajan S Patel, Lynnette Brammer,
Mark S Smolinski, and Larry Brilliant, “Detecting influenza epidemics using search
engine query data,” Nature, 2009, 457 (7232), 1012–1014.
Goel, Sharad, Jake M Hofman, S´ebastien Lahaie, David M Pennock, and Dun-
can J Watts, “Predicting consumer behavior with Web search,” Proceedings of the National
academy of sciences, 2010, 107 (41), 17486–17490.
Goldstone, Robert L and Gary Lupyan, “Discovering psychological principles by mining
naturally occurring data sets,” Topics in cognitive science, 2016, 8(3), 548–568.
Guiso, L., P. Sapienza, and L. Zingales, “People’s opium? Religion and economic atti-
tudes,” Journal of Monetary Economics, 2003, 50 (1), 225–282.
Haken, N, J Messner, K Hendry, P Taft, K Lawrence, T Anderson, and A White-
head, “Failed states index 2006-2019. The Fund for Peace (FFP),” 2019.
Harrison, Myleme O., Harold G Koenig, Judith C Hays, Anedi G Eme-Akwari,
and Kenneth I Pargament, “The epidemiology of religious coping: A review of recent
literature,” International review of psychiatry, 2001, 13 (2), 86–93.
Henrich, Joseph, Michal Bauer, Alessandra Cassar, Julie Chytilov´a, and Ben-
jamin Grant Purzycki, “War increases religiosity,” Nature human behaviour, 2019, 3(2),
129–135.
Iannaccone, Laurence R, “Introduction to the Economics of Religion,” Journal of economic
literature, 1998, 36 (3), 1465–1495.
Ingersoll-Dayton, Berit, Neal Krause, and David Morgan, “Religious trajectories and
transitions over the life course,” The International Journal of Aging and Human Develop-
ment, 2002, 55 (1), 51–70.
Inglehart, R. and P. Norris,Rising tide: Gender equality and cultural change around the
world, Cambridge University Press, 2003.
29

Iyer, Sriya, “The new economics of religion,” Journal of Economic Literature, 2016, 54 (2),
395–441.
Jacobson, Neil S, Christopher R Martell, and Sona Dimidjian, “Behavioral activation
treatment for depression: Returning to contextual roots,” Clinical Psychology: science and
practice, 2001, 8(3), 255–270.
Johnson, Sarah C and Bernard Spilka, “Coping with breast cancer: The roles of clergy
and faith,” Journal of Religion and Health, 1991, 30 (1), 21–33.
Kimball, Miles S, Colter M Mitchell, Arland D Thornton, and Linda C Young-
Demarco, “Empirics on the origins of preferences: The case of college major and religios-
ity,” National Bureau of Economic Research w15182, 2009.
Koenig, H.G., L.K. George, and I.C. Siegler, “The use of religion and other emotion-
regulating coping strategies among older adults,” The Gerontologist, 1988, 28 (3), 303–310.
Kurian, George Thomas and James D Smith,The Encyclopedia of Christian Literature,
Vol. 2, Scarecrow Press, 2010.
Lang, Martin, Jan Kr´atk`y, John H Shaver, Danijela Jerotijevi´c, and Dimitris
Xygalatas, “Effects of anxiety on spontaneous ritualized behavior,” Current Biology, 2015,
25 (14), 1892–1897.
Lazarus, R.S. and S. Folkman,Stress, appraisal, and coping, Springer Publishing Com-
pany, 1984.
Lehrer, Evelyn L, “Religion as a determinant of economic and demographic behavior in the
United States,” Population and development review, 2004, 30 (4), 707–726.
Li, Sijia, Yilin Wang, Jia Xue, Nan Zhao, and Tingshao Zhu, “The impact of COVID-
19 epidemic declaration on psychological consequences: a study on active Weibo users,”
International journal of environmental research and public health, 2020, 17 (6), 2032.
Marx, Karl, “Contribution to the Critique of Hegel’s Philosophy of Right,” Deutsch-
Franz¨osische Jahrb¨ucher, 1844, 7(10), 261–271.
McCleary, R.M. and R.J. Barro, “Religion and economy,” The Journal of Economic
Perspectives, 2006, 20 (2), 49–72.
30

Miller, Lisa, Ravi Bansal, Priya Wickramaratne, Xuejun Hao, Craig E Tenke,
Myrna M Weissman, and Bradley S Peterson, “Neuroanatomical correlates of re-
ligiosity and spirituality: A study in adults at high and low familial risk for depression,”
JAMA psychiatry, 2014, 71 (2), 128–135.
Norenzayan, A. and I.G. Hansen, “Belief in supernatural agents in the face of death,”
Personality and Social Psychology Bulletin, 2006, 32 (2), 174–187.
Norenzayan, Ara, Azim F Shariff, Will M Gervais, Aiyana K Willard, Rita A
McNamara, Edward Slingerland, and Joseph Henrich, “The cultural evolution of
prosocial religions,” Behavioral and brain sciences, 2016, 39.
Norris, Pippa and Ronald Inglehart,Sacred and secular: Religion and politics worldwide,
Cambridge University Press, 2011.
Olivola, Christopher Y, Helen Susannah Moat, and Tobias Preis, “Using big data
to map the relationship between time perspectives and economic outputs,” Behavioral and
Brain Sciences, 2019, 42.
Pargament, K.I.,The psychology of religion and coping: Theory, research, practice, Guilford
Press, 2001.
Park, C., L.H. Cohen, and L. Herb, “Intrinsic religiousness and religious coping as
life stress moderators for Catholics versus Protestants.,” Journal of personality and social
psychology, 1990, 59 (3), 562.
Pew, Research Center, “Few Americans say their house of worship is open, but a quarter
say their faith has grown amid pandemic,” Fact Tank, 2020, April 30.
, “Most Americans Say Coronavirus Outbreak Has Impacted Their Lives,” Social and De-
mographic Trends, 2020, March 30.
Puri, Manju and David T Robinson, “Optimism and economic choice,” Journal of finan-
cial economics, 2007, 86 (1), 71–99.
Schnittker, Jason, “When is faith enough? The effects of religious involvement on depres-
sion,” Journal for the scientific study of religion, 2001, 40 (3), 393–411.
Shariff, A.F. and A. Norenzayan, “God Is Watching You Priming God Concepts Increases
Prosocial Behavior in an Anonymous Economic Game,” Psychological Science, 2007, 18 (9),
803–809.
31

Sibley, Chris G and Joseph Bulbulia, “Faith after an earthquake: A longitudinal study
of religion and perceived health before and after the 2011 Christchurch New Zealand earth-
quake,” PloS one, 2012, 7(12), e49648.
Smith, Bruce W, Kenneth I Pargament, Curtis Brant, and Joan M Oliver, “Noah
revisited: Religious coping by church members and the impact of the 1993 Midwest flood,”
Journal of Community Psychology, 2000, 28 (2), 169–186.
Squicciarini, Mara, “Devotion and development: Religiosity, education, and economic
progress in 19th-century france,” CEPR Discussion Paper No. DP13877, 2019.
Stark, Rodney and Roger Finke,Acts of faith: Explaining the human side of religion,
Univ of California Press, 2000.
Stephens-Davidowitz, Seth, “Googling for God,” Opinion, New York Times, 2015.
Teorell, Jan, Stefan Dahlberg, S¨oren Holmberg, Bo Rothstein, Felix Hartmann,
and Richard Svensson, “The quality of government standard dataset, version jan20,”
University of Gothenburg: The Quality of Government Institute, http://www. qog. pol. gu.
se, 2020.
, , Soren Holmberg, Bo Rothstein, Natalia Alvarado Pachon, and Sofia Axels-
son,The Quality of Government Standard Dataset, version Jan20, University of Gothen-
burg. The Quality of Government Institute, 2020.
Walker, Abigail, Claire Hopkins, and Pavol Surda, “The use of google trends to in-
vestigate the loss of smell related searches during COVID-19 outbreak,” in “International
Forum of Allergy & Rhinology” Wiley Online Library 2020.
Wang, Cuiyan, Riyu Pan, Xiaoyang Wan, Yilin Tan, Linkang Xu, Cyrus S Ho, and
Roger C Ho, “Immediate psychological responses and associated factors during the initial
stage of the 2019 coronavirus disease (COVID-19) epidemic among the general population
in China,” International journal of environmental research and public health, 2020, 17 (5),
1729.
Weber, M., “The Protestant Ethic and the Spirit of Capitalism,” trans. Talcott Parsons
1930, 1905.
Williams, D.R., D.B. Larson, R.E. Buckler, R.C. Heckmann, and C.M. Pyle,
“Religion and psychological distress in a community sample,” Social Science & Medicine,
1991, 32 (11), 1257–1262.
32

Xiao, Han, Yan Zhang, Desheng Kong, Shiyue Li, and Ningxi Yang, “The effects
of social support on sleep quality of medical staff treating patients with coronavirus dis-
ease 2019 (COVID-19) in January and February 2020 in China,” Medical science monitor:
international medical journal of experimental and clinical research, 2020, 26, e923549–1.
Yeung, Timothy Yu-Cheong, “Measuring Christian Religiosity by Google Trends,” Review
of Religious Research, 2019, 61 (3), 235–257.
33

Appendix
A Data
A.1 Google searches for prayer and other religious terms
Google Trends provides access to a sample of actual search requests made on Google. It is
anonymized (no one is personally identified), categorized (determining the topic for a search
query) and aggregated (grouped together). The Google Trends data thus displays interest in
a particular topic from around the globe. The data is available back to 2004, but there was
a trend break on Jan 1 2016, where the data was improved. The data is downloadable from
google.trends.
Google Trends normalizes the search data in the following way: 1) Each data point is
divided by the total searches of the geography and time range it represents to compare relative
popularity, 2) The resulting numbers are then scaled on a range of 0 to 100 based on a topic’s
proportion of all Google searches.
Google Trends filters out some types of searches: 1) Searches made by very few people:
Google Trends only shows data for popular terms, so search terms with low volume appear
as ”0”. 2) Duplicate searches: Google Trends eliminates repeated searches from the same
person over a short period of time. 3) Special characters: Google Trends filters out queries
with apostrophes and other special characters.
Google Trends provides two methods of accessing what people search for. Search terms
show matches for all terms in a query, in the language given. If you search the term ”prayer,”
results include terms like ”prayer” or ”coronavirus prayer”. If you specify ”coronavirus
prayer,” results include searches for ”coronavirus prayer,” as well as ”prayer for coronavirus”.
Topics are a group of terms that share the same concept in any language. If you search the
topic ”London,” results include topics such as ”Capital of the UK” or ”Londres,” which is
”London” in Spanish.
Google Trends provides two types of data: Time-series data and cross-section data. The
time-series data is available for one a country at a time or as an average for the world.
The cross-section data is available for countries or subnational regions as an average over a
specified period of time.35 For the time-series data, Google Trends normalized the search
shares to equal 100 for the highest search share during the period for each country. For the
cross-country data, the search share was set to 100 for the country with the highest search
shares in the sample. This means that only the growth rates, and not the levels, of the time-
35The current analysis uses countries instead of subnational regions, as no time-series data exist for the
regions.
34

series data have a meaningful interpretation and can be compared across countries. For the
cross-country data, the levels can be compared across countries. The analysis includes country
fixed effects throughout and thus does not compare countries, but in Section C.2 I identify the
characteristics of the countries who pray more, which means that comparison across countries
occurs. To construct a panel dataset, I combined the growth rates from the time-series data
with the levels from the cross-section data. For each country, I downloaded the average prayer
search shares for 2019 based on the cross-section data, set this to the search share on January
29 2020, and calculate the search shares for the rest of the period based on the growth rates
from the time-series data.36
The data used throughout this paper are based on time-series data with searches for
the topic ”prayer”. This means that the data is independent of languages and includes all
searches related to the topic ”prayer”. The main data includes search shares for prayer during
the period January 29 to April 1 2020. The period starts well in advance of the onset of
COVID-19 as a pandemic and before the onset of Easter and the Ramadan, where prayer
search shares may rise for reasons other than the COVID-19.
Some fluctuations in the data are too extreme to represent real fluctuations in the interest
on prayer. Fluctuations are defined as ”too extreme” when prayer search shares spike up
(or down) on one day with more than 25 percentage points, only to fall down (or rise) again
with 25 percentage points or more on the following day. For the data behind all Figures and
Tables, except Figure 1, these extreme fluctuations were cut in half. Figure 1 shows the raw
data from Google Trends. Single-day spikes that last more than one day are not affected by
this correction, but were kept unchanged throughout. The correction affects 7.6% of the data,
which includes mainly a few countries that each have many such extreme fluctuations. The
countries with most of these extreme fluctuations are Tanzania, Qatar, and Finland with 22,
19, and 16, days, respectively with these extreme fluctuations out of a total of 61 days in the
sample. These corrections matter mainly for the visual presentation of the results in Fig. 2.
The econometric analyses would treat these fluctuations in the data as noise, which would
enter the error term and produce slightly larger standard errors on the parameters estimated.
Since standard errors throughout are quite small, this does not change the econometric results
in any important way.
There are 99 countries with both time-series data for Google searches on prayer, prayerct
and globally comparable prayer search shares for 2019, averageprayer2019c. Four of these are
small islands or countries with many large fluctuations in the search share data: Martinique,
36For instance, the average prayer search share in 2019 was 3 for Denmark, while that in Morocco was 87. I
therefore set the prayer search share on January 29 to 3 in Denmark and 87 in Morocco. From January 29 to
30, prayer search shares rose by 68% in Denmark and by 6% in Morocco. The prayer search share on January
30 2020 therefore amounts to 5.1 in Denmark and 92.4 in Morocco, and so forth.
35

Mauritius, Reunion, and Senegal. These four countries were excluded from the dataset, mean-
ing that the final dataset on prayer search shares includes 95 countries, listed in the notes for
Figure 2.
The main period of analysis is January 29 to April 1 2020. The data thus ends one week
before the onset of Easter and three weeks before the onset of the Ramadan, where search
shares for prayer rise for other reasons that the COVID-19. January 29 was chosen to get as
large a pre-period as possible, but still be able to zoom in on the COVID-19 pandemic. Some
of the figures show longer periods.
A.2 Measures of the impact of COVID-19
Data on affected cases and deaths by the COVID-19 for the globe are provided by the European
Centre for Disease Prevention and Control (ECDC). The data is available on a daily basis
since December 31 2019 for all countries that were affected by the COVID-19. The main
measure of cases measures the total number of registered people infected by the COVID-19.
The variable does not account for who had recovered again, which means that the variable
can only increase with time. Likewise, deaths by COVID-19 measures the total number of
registered deaths by COVID-19. These two measures are both dependent on the extent of
testing being done in the particular countries. Testing strategies vary across countries in terms
of how much they test, both before and after death.
Pandemic dummy is a dummy equal to one after March 11 when the WHO declared
COVID-19 a pandemic, and zero otherwise.
Days since Pandemic measures the number of days passed since March 11. The variable
is equal to zero on March 11 and before.
First case dummy is a dummy equal to one after the country had its’ first registered
case of COVID-19, zero otherwise.
First death dummy is a dummy equal to one after the country had its’ first registered
death by COVID-19, zero otherwise.
Days since first case measures the days passed since the country had its’ first registered
case of COVID-19. The variable is equal to zero before that.
Days since first death measures the days passed since the country had its’ first registered
death by COVID-19. The variable is equal to zero before that.
A.3 Previous levels of religiosity
The analysis includes the following measures of religiosity before COVID-19. These are used
mainly in Fig. 4 and Tables A.9, A.10 and A.11:
Prayer 2019: Average Google searches for prayer as a share of total Google searches from
36

January 1 2019 to December 31 2019.
The remaining measures of religiosity in Table A.9 are based on answers to questions asked
by the World Values Survey and European Values Study. These are surveys distributed to a
total of 505,000 individuals across the globe over the period 1981-2014. The two surveys ask
the same questions and the responses are therefore comparable.
Moments of prayer: The share of respondents in a country who answered yes to the
question ”Do you take some moments of prayer, meditation or contemplation or something
like that?”.
Ever prayed: This variable is based on the question ”Apart from weddings and funerals,
about how often do you pray these days?” Respondents can answer ”More than once a week”,
”Once a week”, ”Once a month”, ”Only on special holy days”, ”Once a year”, ”Less often”,
or ”Never, practically never”. The variable ”Ever prayed” measures the share of respondents
in a country who answered anything but ”Never, practically never”. This variable was only
asked in Muslim countries.
Weekly pray: The share of respondents in a country who answered ”More than once a
week” or ”Once a week” to the above question.
God: This variable is based on the question ”How important is God in your life? Please
use this scale to indicate. 10 means “very important” and 1 means “not at all important”.
The variable ”God” measures the share of respondents in a country who answered anything
but ”not at all important”.
Very God: The share of respondents in a country who answered “very important” to the
above question.
Ever church: This variable is based on the question ”Apart from weddings and funerals,
about how often do you attend religious services these days?” Respondents can answer ”More
than once a week”, ”Once a week”, ”Once a month”, ”Only on special holy days”, ”Once a
year”, ”Less often”, or ”Never, practically never”. The variable ”Ever church” measures the
share of respondents in a country who answered anything but ”Never, practically never”.
Weekly church: The share of respondents in a country who answered ”More than once
a week” or ”Once a week” to the above question.
Earthquake risk: This variable is the inverse of the distance to the highest earthquake
risk zones. Data on earthquake risk zones are provided by the United Nations Environmental
Programme as part of the Global Resource Information Database (UNEP/GRID), who divided
earthquake risk into five categories based on various parameters such as ground acceleration,
duration of earthquakes, subsoil effects and historical earthquake reports. High risk earthquake
zones are defined by Bentzen (2019) as the two zones with highest risk, zones 3 and 4. The
reasoning for using distances instead of the average of earthquake risk zones is that the measure
37

is meant to provide exogenous variation in religiosity. The impact of earthquake risk on
religiosity is psychological and the use of religion for coping can be strong in areas close
to high-risk zones, even though these areas face low risk of earthquakes (Bentzen, 2019).
Therefore, distances are more relevant than averages across the earthquake risk zones. When
using this measure, the sample is restricted to countries within at least 1500 km of a high-risk
earthquake zone.
A.4 Data on economic and political uncertainty
The variables in Table 2 are chosen from a comprehensive dataset provided by the Quality of
Government Institute (Teorell et al., 2020b), which gathers data from various studies on the
quality of government and related matters. The search was limited to variables available for
at least 70 of the countries in the sample in Panel A of Tables 2 and A.12.
Education measures the human capital index by Penn World Tables.
Fraction measures the degree of ethnic fractionalization in 2000, measured by Alesina et
al. (2003). The definition of ethnicity involves a combination of racial and linguistic charac-
teristics. Fractionalization reflects the probability that two randomly selected people from a
given country will not share a certain characteristic, the higher the number the less probability
of the two sharing that characteristic.
Fragile States Index produced by Haken et al. (2019), 2016 at The Fund for Peace
(http://ffp.statesindex.org/) measures the pressures on states, their vulnerability to internal
conflict, and societal deterioration. The index is based on twelve primary social, economic
and political indicators (each split into an average of 14 sub-indicators). For each indicator,
the ratings are placed on a scale of 0 to 10, with 0 being the lowest intensity (most stable)
and 10 being the highest intensity (least stable). Table 2 shows results using the index, but
also some of the subcomponents of the index: 1) Economic Decline Indicator considers
factors related to economic decline within a country. For example, the indicator includes
patterns of progressive economic decline of the society as a whole as measured by per capita
income, Gross National Product, unemployment rates, inflation, productivity, debt, poverty
levels, or business failures. 2) Security includes measures related to internal conflict, small
arms proliferation, riots and protests, fatalities from conflict, military coups, rebel activity,
bombings, and political prisoners. The measure increases as security deteriorates. 3) Pub-
lic Service includes measures related to policing, criminality, education provision, literacy,
water and sanitation, infrastructure, quality healthcare, telephony, internet access, energy re-
liability, roads. The measure increases as public service deteriorates. 4) Uneven Economic
Development measures the extent to which economic development is unevenly distributed.
Includes measures related to the GINI coeffcient, income share of highest 10%, income share
38

of lowest 10%, urban-rural service distribution, access to improved services, and slum popu-
lation. 5) Demographic Pressure includes measures related to natural disasters, disease,
environment, pollution, food scarcity, malnutrition, water scarcity, population growth, youth
bulge, and mortality.
GDP per capita measures the real PPP adjusted GDP per capita in 2000, provided by
the Penn World Tables. The logarithm is taken in Table 2, while the level is shown without
logs in Table A.12.
Gini is a dummy equal to one if the average Gini coefficient over the period 1991 to 2010
exceeded the median level. The Gini coefficient measures the degree of economic inequality.
Human Development Index measures the Human Development Index in 2010 from the
U.N Human Development Report.
Mortality measures the adult mortality rate per 1000 population, provided by the World
Health Organization.
Poverty measures the poverty gap at Purchasing Parity Adjusted 1.9US$ a day, 2011,
measured by the World Development Indicators.
Property rights institutions measures property rights institutions by Heritage Foun-
dation.
Rule of law measures rule of law by Freedom House.
TFP measures total factor productivity by the Penn World Tables.
The table below documents the summary statistics for the included variables.
39

Table A.3: Summary statistics
Variable Mean Std. Dev. N
Google search share for prayer 30.17 26.383 6080
Growth rate in prayer search shares 0.172 1.691 6080
Pandemic dummy 0.328 0.47 6080
Case dummy 0.545 0.498 6080
Death dummy 0.219 0.413 6080
Average prayer search share 2019 26.158 20.03 6080
Earthquake risk 0.812 0.231 4864
Moments of prayer 0.732 0.188 4096
Ever prayed 0.828 0.174 2880
Pray weekly 0.589 0.261 2880
God important 0.917 0.096 5056
God very important 0.559 0.301 5056
Ever went to church 0.783 0.162 4992
Go to church weekly 0.339 0.23 4992
Fraction Christians 0.604 0.363 6080
Fraction protestants 2000 0.113 0.19 6080
Fraction catholics 2000 0.349 0.357 6080
Fraction Muslims 0.207 0.33 6080
Fraction Hindu 0.017 0.08 6080
Fraction Buddhist 0.038 0.137 6080
Poverty gap at 1.95USD a day 1.734 4.214 4544
GDP per capita 2000 (PPP) 8.987 1.161 6016
Human Development Index 0.712 0.145 5888
Avg gini 1991-2010 39.69 9.325 5248
Uneven Economic Development 5.188 2.08 5824
Fragile States Index 61.489 23.619 5824
Adult mortality rate 130.359 73.039 5888
Property Rights 47.957 26.001 5952
40

B Google searches
B.1 Correlation between Google searches and surveys
Fig. A.5 shows the correlation between average Google search shares for prayer in 2019 and
the share of survey respondents who replied that they pray weekly in Panels (a) and (c)
and the share of respondents who replied that they take moments of prayer, meditation, and
contemplation. Panels (a) and (b) show the raw correlation, while Panels (c) and (d) removes
variation across continents (i.e. a regression of Google searches for prayer on the particular
survey measure and a list of continent dummies). The correlation is high and significant and
prayer search shares alone explain around 30% of the variation in religiosity based on surveys.
This substantiates that Google searches capture real prayer intensity stated in surveys.
41

Figure A.5: Relation between survey answers on prayer and Google search shares for prayer
(a) Share praying weekly (b) Share taking moments of prayer
(c) Including continent fixed effects (d) Including continent fixed effects
Correlation between the share of Google searches for prayer in 2019 and the share of survey respondents answering that they pray
weekly in Panels (a) and (c) and the share of survey respondents answering that they take moments of prayer, meditation, and
contemplation in Panels (b) and (d). Panels (a) and (b) depict the raw correlation, while Panels (c) and (d) depict the correlation
after controlling for continent fixed effects. The grey line represents the fitted line, while the stippled lines represent the 95%
confidence intervals. The measures are described in Section A
B.2 Contents of Google searches for prayer
Fig. A.6 documents the development in the specific Google searches that contributed to the
most to the rise in searches for the prayer topic. For each topic, Google Trends provides
information on the top-25 search terms and the top-25 rising search terms. The combination
of the two lists provides a list of search terms that are both large in levels and rising over the
period. Four main search terms dominate the global pattern in searches for the topic ”prayer”.
Fig. A.6 shows the development over time in these search terms. The ”Pray for Italy” trend
swept across the globe in March 2020 as Italy was the first country outside Asia affected by the
COVID-19 virus. Spiritual Communion is a Christian practice of desiring union with Jesus
Christ. Searches for spiritual communion spike every Sunday, particularly after March 11 and
42

are examples that some Google searches for prayer are replacing physical church attendance.
The map in Fig. A.7 shows the global spread in Google searches for ”pray for italy”. The
map illustrates that searches that are specific to the situation in one country can surge in
other countries, even far from the country in question.
Figure A.6: Top search terms within the topic ”Prayer”
The three spikes in the search terms for ”Spiritual communion prayer” are Sundays. Searches for ”prayer for coronavirus” includes
searches for ”prayer for COVID-19”.
Figure A.7: Geographic spread of searches for ”pray for italy” March 5-30 2020
.
43

B.3 Examples of prayer websites
Figures A.8 , A.9, A.10, and A.11 show screenshots of websites that one encounters when
googling ”coronavirus prayer”. The websites contain instructions on how to pray as well as
specific prayer texts.
Figure A.8: Example of a guide to a coronavirus prayer
The website of 24/7 Prayer: https://www.24-7prayer.com/60-minute-coronavirus-prayer
44

Figure A.9: Example of a coronavirus prayer
The website of World Vision: https://www.worldvision.org/disaster-relief-news-stories/prayers-people-affected-new-coronavirus
Figure A.10: Example of website with COVID-19 prayers
The website of the Church of England: https://www.churchofengland.org/more/media-centre/coronavirus-covid-19-liturgy-and-
prayer-resources
45

Figure A.11: Example of website with list of COVID-19 prayers
The website of website of My Catholic Life: https://mycatholic.life/catholic-prayers/a-prayer-for-healing-and-hope/
B.4 The relative size of the increase
Fig. A.12 shows the increase in Google searches for prayer relative to searches for other topics
that rose during the COVID-19 pandemic. The purpose is to illustrate the relative size of
the rise in prayer searches. The COVID-19 pandemic resulted in massive lock downs and
quarantines across the globe, meaning that people were at home and not allowed to go out.
In addition, most international air traffic was shut down.
Fig. A.12 shows that searches for topics related to take-out and Netflix rose during the
month of March 2020, while searches for flights fell. The volume of searches for prayer was
higher than searches for takeaway (by a factor 4.8), but lower than searches for Netflix (25%)
and flights (28%). Like prayer, the Google searches for take-out, Netflix, and flights encompass
all searches for topics related to these in all languages.
The relative sizes of the increases in the searches are calculated using the following formula
for Netflix and take-out:
∆prayer
∆other =maxprayermar −avgprayerf eb
maxothermar −avgotherfeb
(3)
where maxprayermar is the maximum level of search shares for prayer reached during the
month of March 2020 and maxothermar is the maximum level of search shares for either
Netflix or Take-Out reached during the month of March 2020. avgprayerf eb is the average
46

level of search shares for prayer during February 2020 and avgotherf eb is the average level of
search shares for Netflix or take-out during February 2020.
Instead of maxothermar, the calculation for flights included the minothermar, which is the
minimum level of search shares for flights reached during the month of March 2020. This way,
the spike in searches for flight in early March does not influence the calculation. This surge
may be due to people anticipating a change in rules for flight traffic.
Searches for prayer rose by 134% the rise in Google searches for take-out, by 12% the rise
in searches for Netflix, and by 26% the fall in searches for flights.
Figure A.12: Google searches for other terms affected by COVID-19
(a) Take-out (b) Prayer and take-out
(c) Prayer and Netflix (d) Prayer and flight
Global average of Google searches on different topics over the period Feb 1 to April 1 2020. The searches are set to 100 for the
largest search within each panel. The size of the increases are therefore not comparable across panels, but they are comparable
within one panel. Result: Google searches for prayer compares in size to movements in other tendencies that were impacted by
COVID-19.
B.5 The timing of the rise in prayer searches
Fig. A.13 documents the distribution of the countries based on when the prayer search shares
rose for the first time in each country. The figure illustrates the timing of the surge in Google
searches for prayer. The following calculations define which increases in the prayer search
47

shares are significant based on whether the increase exceeds one standard deviation.
prayerct > prayerct0+sd(prayerc) (4)
where prayerct is the daily share of Google searches for the topic prayer for country c, as
described in Section A.1. prayerct0measures the average prayer search shares in the first week
of February (period t0) and sd(prayerc) measures the standard deviation of the prayer search
shares over the entire period from January 29 to April 1. Equation (4) defines an increase
in prayer search shares as significant when the search share rose more than one standard
deviation above the initial level in the beginning of February. For each country and for each
day, one can calculate whether or not search shares rose above this level. Fig. A.13 depicts
the first day that this level was reached for at least two consecutive days or with maximum
one day in between. This occurred within the window of analysis (Feb 1 - Apr 1 2020) for 94
out of the 95 countries in the sample. The 94 countries are represented by the density mass in
Fig. A.13. For 51 of these 94 countries, the significant rise occurred on March 11 or thereafter
(the density mass at or to the right of March 11 in the figure).
Of the 43 countries, where the first day with significant increases in prayer search shares
occurred before March 11 (the density mass to the left of March 11 in the figure), 19 were
located in Asia, where the COVID-19 virus first hit, cf. Fig. A.14 which shows the development
in registered reported cases worldwide.
Morocco is the only country in the sample not included in Fig. A.13. The search shares
for prayer did rise above the mean of the first week of February for several days during the
window of analysis, but with two or more days in between the spikes.
All in all, Fig. A.13 shows that most of the countries in the sample experienced their first
large increases in search shares for prayer during the period March 14 to March 25.
48

Figure A.13: Distribution of the countries based on first day with two-days rise in prayer
search shares
The histogram shows the distribution of 94 countries in the sample, based on the day when their prayer search shares first rose
more than one standard deviation above the level in the first week of February for two consecutive days or with maximum one
day in between. All countries, except Morocco fulfil this criteria and are included in the figure.
Figure A.14: Distribution of COVID-19 cases worldwide as of 16 April 2020
Distribution of cases of COVID-19 by continent (according to the applied testing strategies in the affected countries). Source:
ECDC, https://www.ecdc.europa.eu/.
Fig. A.15 shows that the spike in Google searches for prayer is even visible in the data back
to beginning of the Google Trends series, starting in Jan 1 2004. Note, however, that there is
a trend break in the data on Jan 1 2016, where Google Trends’ data collection method was
improved.
49

Figure A.15: Global Google searches for prayer Jan 2004 to Apr 2020
The vertical line represents an improvement of Google Trends’ data collection method.
B.6 Alternative Google searches for religious topics
Figure A.16 documents rising search shares for other religious search topics. The period
includes the full year from Apr 14 2019 to Apr 14 2020, the latest date at the time of writing.
The end date coincides partly with Easter 2020, which may influence the rise for the Christian
search terms, but should not matter for the remaining religious terms.
50

Figure A.16: Google searches for religious topics Apr 14 2019 - Apr 14 2020
(a) Topics quran, muhammad, allah (b) Topics jesus and bible
(c) Topics buddha and vishnu (d) Topic shiva
51

Figure A.16: Cont. Google searches for religious topics Apr 14 2019 - Apr 14 2020
(e) Topic god (f ) Topics god, quran, bible, buddha, shiva
(g) Topics god and prayer
Global average of Google searches on religious topics over the period April 14 2019 to April 14 2020. Google Trends sets the
searches to 100 for the largest search within each time series. The search shares are therefore not comparable across panels, but
they are comparable within one panel.
Result: Search shares rise in March 2020 for all religious terms. In March 2020, searches for muhammad, allah, bible, jesus, and
god surpass the search shares across all other religious events during the year. Searches for buddha peak on May 12, Buddhas
birthday, quran peaks on the first day of the Ramadan, vishnu peaks on Nov 10, Vaikuntha Chaturdashi, the Hindu holiday for
Lord Vishnu and Lord Shiva, shiva peaks on Feb 21, Maha Shivratri, the worshipping of Lord Shiva.
C The motivating figures
This section contains supplementary information for Figures 1, 2, and 3. Fig. 1 documents
the development in the global average of Google searches for the topic prayer. The data shown
is the direct download from Google Trends, before the data corrections described in Section
A.1. The time-line in Panel A is chosen as the longest possible window without data breaks.
This means that the series starts on Jan 1 2016. Fig A.15 documents that the same picture
emerges when extending the timeline back to 2004, which is the earliest available data from
Google Trends. The series ends at the latest date available at the time of writing, April 11.
Panel B of Fig. 1 restricts the period to the period used in the main analysis, starting on
52

January 29 2020 and ending on Apr 1 2020, before the onset of Easter 2020.
Fig. 2 is constructed based on the data on google searches for prayer, described in Section
A.1. To construct the figure, data points were dropped after each country reached its’ maxi-
mum search share for prayer during the period January 29 to April 1 2020. Interpretation of
a fall in search shares is not straight forward, since the search shares mechanically fall when
people start entering their preferred prayer websites directly instead of googling them first,
even if the interest in prayer stays constant. The means within each group were calculated
only when at least 2-5 countries had information on prayer search shares on the given day.
Third, to increase comparability across the panels, the y-axis was cropped at prayer share val-
ues of 4, even in cases where some data points exceeded this value. These large fluctuations
in the data occur mainly in Asia, particularly outside of South East Asia.
Fig. 3, Panel (b) documents the absolute change in prayer search shares in March 2020,
∆prayermarch, which is constructed using the following formula:
∆prayermar =maxprayermar −avgprayerf eb −∆prayerf eb (5)
maxprayermar measures the highest prayer search share reached after March 1. avgprayerf eb
measures the average prayer share during February 2020. ∆prayerfeb measures the change
in prayer search shares from the first to the last week of February 2020. The rationale is to
remove the general trend in prayer search shares, ensuring that we do not attribute a potential
general rise in prayer search shares to the COVID-19 pandemic. If prayer search shares rose
in February, this may be due to other things than the pandemic or the fact that COVID-19
started in Asia well before it was declared a pandemic. ∆prayerf eb is set to zero if prayer
search shares fell during February 2020 to get a conservative measure of the rise in prayer
search shares after March 1. The measure ∆prayermar now measures the absolute change in
prayer search shares from March 1 to the day with the highest prayer search shares in March
2020, where the rise in prayer search shares in February are subtracted.
The relative increase in prayer search shares in Panel (a) of Fig. 3 is calculated by dividing
the absolute change in prayer search shares, ∆prayermarch subtracted the rise in February, with
the average level in February, avgprayerfeb.
C.1 The impact of COVID-19 on prayer search shares
This section explores further the link between COVID-19 exposure and the rise in prayer
search shares. First, Fig. A.17 shows that the results in Table 1 are not driven by specific
observations. The figure shows the added variables plot of columns (1), (3), and (6) of Table
1, where observations are binned into 100 equally sized bins.
53

Table A.4 tests for non-linearities in the measures of days since COVID-19 was declared
a pandemic, registered cases, and deaths. Panel A includes the full sample, while Panels B
and C restricts the sample to only include observations until the prayer search shares reaches
its maximum over the period. This is to account for the fact that falling prayer search shares
are difficult to interpret as some of the fall may be caused by people accessing their prayer
websites directly instead of googling them. The rise in prayer search shares slows down as
time passes or more cases and deaths are registered, except in the restricted sample where the
rise in prayer search shares does not slow down as time passes. Panel C documents that the
number of cases and deaths do not impact prayer search shares once days since COVID-19
was declared a pandemic is accounted for.
Table A.5 documents that the rise in prayer search shares after March 11 is larger when
the country had also one or more registered cases or deaths in the restricted sample where
observations are dropped after prayer search shares reached their maximum. This interaction,
though, becomes insignificant once time fixed effects are accounted for.
Table A.6 documents that the rise in prayer search shares is similar across continents. The
table mimics the estimations in Panel B of Table 1, but instead of throwing away countries that
are not located on the particular continent, the regressions include an interaction between the
pandemic dummy and a dummy for each of the continents. If the estimate on this interaction
is statistically different from zero, the rise in prayer search shares is different on this continent,
compared to the rest of the world.
So far, the analysis has included daily data. This means that the estimate on registered
cases and deaths (γin equation (1)) measures the daily change in prayer shares as cases or
deaths go up on the day before. Fluctuations in the Google data may make these day-to-day
comparisons rather imprecise. Table A.7 documents similar results when the data is aggregated
up to weekly averages. Panel A documents that cases and deaths increase prayer shares, but
not when accounting for the date when the WHO declared COVID-19 a pandemic (March
11). Panel B includes instead days since the first case or death with similar conclusions. The
inclusion of country-specific trends may be too restrictive in this much smaller sample. Table
A.8 shows similar results when including a global time trend, instead of the country-specific
trends.
54

Figure A.17: Binned added variables plots of the rise in prayer search shares after different
dates
(a) The rise in prayer after March 11 (b) after the first registered case
(c) after the first registered death
The binned added variables plot of regressions of the prayer search share on the pandemic dummy in Panel (a), the dummy equal
to one after the first case is registered in Panel (b), and after the first death is registered in Panel (c). The regressions mirror
those in columns (1), (3), and (6) of Table 1 and include country fixed effects and country-specific trends. The observations are
binned into 100 equally sized bins.
Result: The results are not driven by specific observations.
55

Table A.4: The impact of COVID-19 on prayer search shares I
Dependent variable: Prayer searches
(1) (2) (3) (4) (5) (6)
Panel A: Full sample
Days since Pandemic 0.44*** 1.25***
(0.057) (0.180)
Days since Pandemic squared -0.043***
(0.008)
COVID-19 infected cases 0.090** 0.30***
(0.044) (0.101)
Infected cases squared -0.0019***
(0.001)
COVID-19 deaths 1.35*** 3.47**
(0.427) (1.364)
Deaths squared -0.25**
(0.117)
R-squared 0.84 0.84 0.83 0.83 0.83 0.83
Observations 6080 6080 6066 6066 6080 6080
Panel B: Restricted sample
Days since Pandemic 1.02*** 0.87***
(0.129) (0.239)
Days since Pandemic squared 0.011
(0.016)
COVID-19 infected cases 0.36** 0.90***
(0.157) (0.190)
Infected cases squared -0.0094***
(0.002)
COVID-19 deaths 2.71** 9.62**
(1.259) (4.382)
Deaths squared -1.07*
(0.563)
R-squared 0.84 0.84 0.83 0.83 0.83 0.83
Observations 4345 4345 4331 4331 4345 4345
Panel C
Days since Pandemic 1.02*** 1.02*** 1.05*** 1.07*** 1.02*** 1.02***
(0.129) (0.129) (0.138) (0.143) (0.133) (0.135)
COVID-19 infected cases -0.13 -0.37
(0.095) (0.302)
Infected cases squared 0.0039
(0.004)
COVID-19 deaths -0.57 0.36
(0.560) (2.153)
Deaths squared -0.14
(0.260)
R-squared 0.84 0.84 0.84 0.84 0.84 0.84
Observations 4345 4345 4331 4331 4345 4345
Countries 95 95 95 95 95 95
OLS estimates. Units: Days ×countries. Period: January 29 to April 1 2020. All regressions include a
constant, country-specific time trends, and country fixed effects. The sample is restricted to the sample
where observations are dropped after the maximum prayer search share over the period is reached.
Panel B replicates the regressions in Panel A, but includes the variable ”Days since Pandemic”. Robust
standard errors clustered at the country level in parentheses. *, **, and *** indicate significance at the
10%, 5%, and 1% level.
Result: The number of cases and deaths do not matter for prayer search shares when controlling for
the number of days passed since WHO declared the COVID-19 a pandemic.
56

Table A.5: The impact of COVID-19 on prayer search shares II
(1) (2) (3) (4) (5) (6) (7) (8)
Dep var: Prayer search share
Pandemic dummy 5.49*** 0.069 5.22*** 3.24***
(0.923) (2.665) (1.038) (0.951)
First case dummy 2.53** 1.82* 2.74*** 2.62** 3.39***
(1.019) (0.945) (1.003) (1.043) (0.975)
Pandemic x first case dummy 6.19** 0.93
(2.626) (2.497)
First death dummy 3.10* -0.47 -1.33
(1.572) (1.363) (2.056)
Pandemic x first death dummy 5.78*** 1.33
(1.553) (2.289)
R-squared 0.83 0.83 0.84 0.84 0.83 0.83 0.84 0.84
Observations 4331 4331 4331 4331 4345 4331 4345 4345
Countries 95 95 95 95 95 95 95 95
TimeFE No No Yes Yes No No Yes Yes
OLS estimates. Units: Days ×countries. Period: January 29 to April 1 2020. All regressions include a constant,
country-specific time trends, and country fixed effects. In addition, time fixed effects are added in columns 3, 4, 7,
and 8. The sample is restricted to the sample where observations after prayer searched shares reached their max are
dropped. Robust standard errors clustered at the country level in parentheses. *, **, and *** indicate significance at
the 10%, 5%, and 1% level.
Result: Prayer search shares rose more after March 11 for countries that had already had their first case or death.
Table A.6: The impact of COVID-19 on prayer search shares III
(1) (2) (3) (4) (5) (6) (7) (8)
Dep var: prayer search shares All N Europe S Europe N America S America SE Asia Rest Asia Africa
Pandemic dummy 5.11*** 5.48*** 5.52*** 4.79*** 4.63*** 5.10*** 5.28*** 4.96***
(0.752) (0.844) (0.854) (0.807) (0.678) (0.788) (0.865) (0.845)
Pandemic x continent -2.98** -2.65* 2.33 4.09 0.037 -0.94 1.02
(1.218) (1.509) (2.142) (3.805) (2.531) (1.667) (1.684)
R-squared 0.84 0.84 0.84 0.84 0.84 0.84 0.84 0.84
Observations 6080 6080 6080 6080 6080 6080 6080 6080
Countries 95 95 95 95 95 95 95 95
MeanDepVar 30.2 30.2 30.2 30.2 30.2 30.2 30.2 30.2
OLS estimates. Units: Days ×countries. Period: January 29 to April 1 2020. All regressions include a constant, country-specific
time trends, and country fixed effects. Compared to Panel B of Table 1, this table includes the full sample throughout and includes
instead an interaction term between the pandemic dummy and a continent dummy equal to one for the particular continent. Robust
standard errors clustered at the country level in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% level.
Result: Prayer search shares rose on all continents, but significantly less than the world average in Europe.
57

Table A.7: The impact of COVID-19 on prayer search shares IV
Dependent variable: Prayer searches
(1) (2) (3) (4) (5) (6)
Panel A
Pandemic dummy 6.49*** 9.15*** 9.06*** 9.08***
(0.854) (1.176) (1.218) (1.192)
COVID-19 infected cases 0.34** 0.056
(0.143) (0.069)
COVID-19 deaths 2.50*** 0.79+
(0.864) (0.488)
R-squared 0.95 0.97 0.96 0.97 0.96 0.97
Panel B
Pandemic dummy 6.49*** 9.15*** 7.82*** 8.89***
(0.854) (1.176) (1.316) (1.242)
Days since first case 0.64*** 0.18
(0.122) (0.136)
Days since first death 0.75*** 0.082
(0.190) (0.194)
R-squared 0.95 0.97 0.96 0.97 0.96 0.97
Observations 950 661 661 661 661 661
MeanDepVar 30.6 31.6 31.6 31.6 31.6 31.6
RestrictedSample No Yes Yes Yes Yes Yes
OLS estimates. Units: Weeks ×countries. Period: January 29 to April 1 2020. All regres-
sions include a constant, country-specific time trends, and country fixed effects. The sample
consists of the full sample in column (1), but is restricted to the sample where observations
are dropped after the maximum prayer search share over the period is reached in remaining
columns. Panel A includes the number of registered cases and deaths. Panel B includes a
measure of the days since a country had its’ first case or death by COVID-19. Robust standard
errors clustered at the country level in parentheses. +, *, **, and *** indicate significance at
the 15%, 10%, 5%, and 1% level.
Result: The number of cases and deaths do not matter for prayer search shares when con-
trolling for the dummy indicating when the WHO declared the COVID-19 a pandemic.
58

Table A.8: The impact of COVID-19 on prayer search shares V
Dependent variable: Prayer searches
(1) (2) (3) (4) (5) (6)
Panel A
Pandemic dummy 6.49*** 9.01*** 8.99*** 8.93***
(0.811) (1.179) (1.233) (1.203)
COVID-19 infected cases 0.16*** 0.0089
(0.052) (0.077)
COVID-19 deaths 1.75*** 0.77**
(0.434) (0.351)
R-squared 0.93 0.95 0.94 0.95 0.94 0.95
Panel B
Pandemic dummy 6.49*** 9.01*** 9.77*** 8.39***
(0.811) (1.179) (1.118) (1.116)
Days since first case 0.035 -0.099*
(0.064) (0.058)
Days since first death 0.37*** 0.17
(0.139) (0.184)
R-squared 0.93 0.95 0.94 0.95 0.94 0.95
Observations 950 661 661 661 661 661
MeanDepVar 30.6 31.6 31.6 31.6 31.6 31.6
RestrictedSample No Yes Yes Yes Yes Yes
The estimations mimic those of Table A.7, except that a global time trend is included instead
of country-specific time-trends.
Result: The number of cases and deaths do not matter for prayer search shares when con-
trolling for the dummy indicating when the WHO declared the COVID-19 a pandemic.
C.2 Who is praying more?
Table A.9 documents that the rise in prayer intensity is generally higher for populations that
are already more religious, using different measures of religiosity: Average Google searches for
prayer in 2019 in column (1) and measures based on various questions asked in global surveys
(conducted before 2015) in columns (2)-(8): Whether or not respondents take moments of
prayer, meditation or contemplation (col 2),37 ever prayed (col 3), pray weekly (col 4), rank
God as anything but unimportant in their lives (col 5), rank God as very important (col 6),
ever went to church (col 7), or go to church on a weekly basis (col 8). Last, to obtain exogenous
variation in religiosity, column (9) interacts instead with earthquake risk.
The rise in prayer searches is larger in countries where a larger share of the population
initially prayed more, went more to church, or answered that God is important in their lives,
and faced higher earthquake risk. Prayer search intensity rose by more than 50% of the mean
37While this measure captures more than religiosity, it correlates with more than .8 with the remaining
measures of religiosity listed. This indicates that the majority of the affirmative answers cover some sort of
religious prayer, meditation or contemplation.
59

for the countries with the highest initial religiosity.38
Prayer search shares rose much less in the less religious countries, but even in the least
religious countries, prayer searches rose for five out of the nine measures of religiosity (Mini-
mumImpact in the bottom of the table).
The p-values, PvalueAtXPct, in the bottom of the table indicate the p-value of the following
test, where the parameter values are indicated in equation (2): γ+λreligiosityc= 0. Thus,
the test indicates at what level of religiosity, the prayer search shares rose significantly. The
value of religiositycis the value at the 1st, 10th, 15th, and 20th percentiles, respectively. The
calculations show that prayer search shares even rose significantly for the 1% least religious
countries for 2 out of 9 measures of religiosity. These two are the measures available for most
countries, indicating that the lack of a significant rise for the remaining measures is most likely
due to lower precision in the smaller sample. For the 10% least religious, prayer searches rose
significantly for 4 out of 9 religiosity measures, while they rose for 8 out of 9 measures within
the 15% least religious countries. The measure that does not show a significant rise for this
group is the religiosity measure available for the least countries. At religiosity levels as low as
the 20th percentile, prayer search shares rose significantly for all measures of religiosity.
Tables A.10 and A.11 document the differential effects across different religiosity measures
for more detailed levels of religiosity, based on the prayer and churchgoing measures in Table
A.10 and based on the importance of God measure in Table A.11.
Table A.12 replicates Table 2, but with interactions with different characteristics. The
regressions in Panel A and C seem to confirm the vulnerability hypothesis: The rise in prayer
search shares is larger in countries with lower security and lower quality public services, larger
demographic pressures, higher infant mortality, and lower scores on the Human Develop-
ment Index. These differential effects, though, disappear once previous religiosity levels are
accounted for (Panels B and D).
38This calculation is based on the MaximumImpact scalar provided in the bottom of the table. This scalar
measures the impact of the pandemic dummy in countries with the maximum level of the particular religiosity
measure. In col (1), the maximum level of prayer search intensity in 2019 was 87, reached by Morocco. Thus,
MaximumImpact was -0.086+0.08*87=6.85.
60

Table A.9: The rise in prayer search shares for different religiosity levels I
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Dep var: Prayer searches Pray2019 MomentPray EverPray WeekPray God VeryGo d EverChurch WeekChurch EarthqRisk
Pandemic dummy 1.26* -4.84** -4.82 -0.84 -16.9*** -0.13 -6.37** 0.67 -7.76**
(0.747) (2.106) (3.623) (2.056) (4.844) (0.919) (2.734) (0.875) (3.777)
Pandemic x Religiosity 0.15*** 12.5*** 12.0** 10.1** 23.9*** 9.15*** 14.4*** 12.6*** 14.4***
(0.035) (3.254) (4.972) (4.065) (5.709) (2.237) (3.938) (2.728) (4.890)
R-squared 0.84 0.88 0.89 0.90 0.87 0.87 0.87 0.87 0.87
Observations 6080 4096 2880 2880 5056 5056 4992 4992 4416
Countries 95 64 45 45 79 79 78 78 69
MeanDepVar 30.2 23.8 27.6 27.6 25.9 25.9 26.0 26.0 25.1
MinimumImpact 1.41 -1.28 0.33 0.93 -3.08 0.72 -0.93 1.00 -1.14
MaximumImpact 14.0 7.45 7.19 8.94 6.98 8.77 7.95 11.8 6.61
PvalueAt1Pct 0.054 0.30 0.84 0.52 0.058 0.34 0.48 0.23 0.47
PvalueAt10Pct 0.0030 0.66 0.24 0.20 0.16 0.041 0.13 0.062 0.0020
PvalueAt15Pct 0.0010 0.016 0.039 0.13 0.0020 0.0080 0.0070 0.032 0
PvalueAt20Pct 0 0 0.0030 0.059 0 0.0020 0 0.012 0
OLS estimates. Units: Days ×countries. Period: January 29 to April 1 2020. All regressions include a constant, country-specific time trends, and country
fixed effects. The pandemic dummy is interacted with average google searches for prayer in 2019 (col 1), the share of the populations taking moments
for prayer, meditation, or contemplation (col 2), ever prayed (col 3), pray weekly (col 4), answered that God is anything but important in their lives (col
5), answered that God is very important in their lives (col 6), ever went to church (col 7), or went to church weekly (col 8), and average earthquake risk
(col 9). MinimumImpact (MaximumImpact) indicates the impact of the pandemic at the minimum (maximum) level of the particular religiosity measure.
PvalueAtXPct indicates the p-value of the test that the impact of the pandemic dummy is insignificant, evaluated at the X percentile of the religiosity
measure. Robust standard errors clustered at the country level in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% level.
Result: Prayer search shares rose at most levels of religiosity and rose more for more religious countries.
Table A.10: The rise in prayer search shares for different religiosity levels II
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Dep var: Prayer searches Moments Never Ever Yearly Weekly Daily Never Ever Yearly Weekly Daily
Pandemic dummy -4.84** 7.19*** -4.82 -3.80 -0.84 0.75 8.06*** -6.37** -4.80** 0.67 3.59***
(2.106) (1.642) (3.623) (2.721) (2.056) (1.413) (1.383) (2.734) (1.955) (0.875) (0.875)
Pandemic x Pray 12.5*** -12.0** 12.0** 11.8*** 10.1** 9.26**
(3.254) (4.972) (4.972) (4.288) (4.065) (3.731)
Pandemic x Church -14.4*** 14.4*** 14.3*** 12.6*** 9.08**
(3.938) (3.938) (3.424) (2.728) (3.465)
R-squared 0.88 0.89 0.89 0.90 0.90 0.90 0.87 0.87 0.87 0.87 0.87
Observations 4096 2880 2880 2880 2880 2880 4992 4992 4992 4992 4992
Countries 64 45 45 45 45 45 78 78 78 78 78
OLS estimates. Units: Days ×countries. Period: January 29 to April 1 2020. All regressions include a constant, country-specific time trends, and
country fixed effects. Robust standard errors clustered at the country level in parentheses. *, **, and *** indicate significance at the 10%, 5%, and
1% level.
Result: Prayer search shares rose more for more religious countries.
Table A.11: The rise in prayer search shares for different religiosity levels III
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Dep Var: Prayer search share Very:10 9 8 7 6 5 4 3 2 Not:1
Pandemic dummy 0.57 -0.13 -1.35 -2.51* -4.06** -6.85*** -8.46*** -11.7*** -16.9*** 6.98***
(0.832) (0.919) (1.122) (1.325) (1.663) (2.248) (2.599) (3.406) (4.844) (1.050)
Pandemic x Importance of God 8.98*** 9.15*** 9.90*** 10.7*** 12.0*** 14.3*** 15.8*** 18.8*** 23.9*** -23.9***
(2.307) (2.237) (2.327) (2.434) (2.723) (3.221) (3.538) (4.307) (5.709) (5.709)
R-squared 0.87 0.87 0.87 0.87 0.87 0.87 0.87 0.87 0.87 0.87
Observations 5056 5056 5056 5056 5056 5056 5056 5056 5056 5056
Countries 79 79 79 79 79 79 79 79 79 79
CountryFE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
MeanDepVar 25.9 25.9 25.9 25.9 25.9 25.9 25.9 25.9 25.9 25.9
OLS estimates. Units: Days ×countries. Period: January 29 to April 1 2020. All regressions include a constant, country-specific time trends, and
country fixed effects. Robust standard errors clustered at the country level in parentheses. *, **, and *** indicate significance at the 10%, 5%, and
1% level.
Result: Prayer intensity rose more in countries where larger shares of the population rank God as important.
61

Table A.12: The rise in prayer search shares across country characteristics
Dependent variable: Prayer
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A Security Public DemoPres RuleLaw InfMort Fraction HDI GDP
Pandemic dummy 7.76*** 9.56*** 1.08 7.08*** 4.09*** 4.12*** 13.0*** 6.38***
(1.938) (2.010) (1.187) (1.810) (0.852) (1.331) (2.981) (1.350)
Pandemic x Variable -0.51* -0.78*** 0.86*** -0.22 0.076** 2.78 -11.0*** -0.050
(0.287) (0.268) (0.280) (0.147) (0.038) (2.933) (3.602) (0.036)
R-squared 0.84 0.84 0.84 0.84 0.84 0.84 0.84 0.84
Observations 5824 5824 5824 5952 5888 5888 5888 5952
Countries 91 91 91 93 92 92 92 93
Panel B
Pandemic dummy 1.14 1.93 0.40 0.29 1.44* 1.99* 0.48 1.03
(2.389) (2.793) (1.125) (1.975) (0.790) (1.032) (3.630) (1.341)
Pandemic x Variable 0.036 -0.077 0.28 0.088 -0.042 -2.41 0.94 0.0066
(0.309) (0.322) (0.318) (0.139) (0.050) (3.391) (4.113) (0.025)
Pandemic x Prayer 2019 0.15*** 0.14*** 0.13*** 0.16*** 0.17*** 0.16*** 0.15*** 0.15***
(0.039) (0.043) (0.040) (0.038) (0.045) (0.042) (0.041) (0.036)
R-squared 0.84 0.84 0.84 0.84 0.84 0.84 0.84 0.84
Observations 5824 5824 5824 5952 5888 5888 5888 5952
Countries 91 91 91 93 92 92 92 93
Panel C
Pandemic dummy 7.57*** 11.0*** -0.0071 7.31*** 2.73*** 2.61** 13.7*** 5.74***
(2.303) (2.541) (1.493) (2.111) (0.878) (1.198) (3.841) (1.522)
Pandemic x Variable -0.55 -1.04*** 1.07*** -0.29* 0.18*** 5.92* -12.2*** -0.037
(0.344) (0.345) (0.398) (0.170) (0.065) (3.058) (4.582) (0.043)
R-squared 0.87 0.87 0.87 0.87 0.87 0.87 0.87 0.87
Observations 4352 4352 4352 4416 4416 4352 4352 4416
Countries 68 68 68 69 69 68 68 69
Panel D
Pandemic dummy -4.69 -0.96 -2.47 -3.70 -1.56 -1.49 -4.84 -3.14
(3.590) (5.416) (1.659) (3.438) (1.520) (1.466) (7.607) (2.140)
Pandemic x Variable 0.40 -0.064 0.46 0.17 0.039 -0.73 3.83 0.033
(0.379) (0.499) (0.373) (0.180) (0.067) (3.742) (7.728) (0.025)
Pandemic x Prayer 2019 0.34*** 0.28** 0.24** 0.32*** 0.27** 0.30*** 0.31*** 0.32***
(0.101) (0.129) (0.095) (0.106) (0.107) (0.103) (0.113) (0.093)
R-squared 0.87 0.87 0.87 0.87 0.87 0.87 0.87 0.87
Observations 4352 4352 4352 4416 4416 4352 4352 4416
Countries 68 68 68 69 69 68 68 69
FirstStageF 13.2 8.54 14.2 13.1 14.8 15.1 11.1 13.8
The table replicates Table 2 with different characteristics interacted with the pandemic dummy.
Result: Prayer search shares rose more in poor, unequal, and insecure countries. But this is exclusively because these
societies are more religious.
C.3 Replacement of physical church or rise in prayer intensity
Around mid-March 2020, most churches across the globe closed in af effort to enforce social
distancing. A concern for the analysis is whether the rise in prayer search shares is simply a
replacement for the physical churches. According to the theory on religious coping we would
not expect this to be the case, since people tend to use their intrinsic religiosity rather than
their extrinsic religiosity to cope with adversity. Thus, even if the churches had been open,
62

we would not expect churchgoing to rise as much as private prayer. This section tests this
prediction empirically.
Fig. A.18 documents that Google searches for the topic ”internet church” also rises during
the month of March 2020, but in a very different pattern. The three large spikes in Panel (a)
of Fig. A.18 are the three last Sundays in March. These rises coincide with the closure of
the physical churches and they follow a very different pattern than the general rise in prayer
shares documented throughout this research. Panel (b) shows that the rise in searches on
internet church is insignificant compared to the total rise in prayer searches, indicating that
the rise in demand for internet churches does not explain the rise in prayer shares.
Table A.13 shows that prayer search shares rose every day of the week, except Fridays. The
rise on Sundays is larger than the other days, which could be due to most masses being held
on Sundays or simply that Sundays are the most holy day of the week for Christians, and thus
the day of the week, where most choose to pray. Panel A includes the full set of controls, while
Panel B excludes the country-specific time-trends and includes a global time-trend instead.
The results are nearly identical to those in Panel A.
63

Figure A.18: Global Google searches for prayer and internet church
(a) Searches for the topic ”internet church”
(b) Searches for the topics ”internet church” and ”prayer”
Global average of Google searches for prayer Jan 1 - Apr 1 2020. Searches for internet church also rise during the month of March
2020, but the share is minuscule compared to the size of the search shares for prayer. Furthermore, the searches for internet
church rise mainly every Sunday and thus have a distinctly different pattern than the search shares for prayer.
64

Table A.13: The rise in prayer search shares, by weekdays
Dependent variable: Prayer
(1) (2) (3) (4) (5) (6) (7)
Panel A Sun Mon Tue Wed Thu Fri Sat
Pandemic dummy 11.0*** 3.22** 4.38*** 4.58*** 3.71** 1.13 6.77***
(1.740) (1.542) (1.418) (1.190) (1.436) (1.590) (1.481)
R-squared 0.88 0.87 0.87 0.87 0.88 0.88 0.85
Panel B
Pandemic dummy 11.0*** 3.22** 4.38*** 4.58*** 3.71*** 1.13 6.76***
(1.642) (1.455) (1.338) (1.130) (1.355) (1.501) (1.398)
R-squared 0.83 0.84 0.84 0.84 0.85 0.83 0.80
Observations 855 855 855 950 855 855 855
Countries 95 95 95 95 95 95 95
MeanDepVar 30.7 29.6 29.6 30.1 30.0 31.3 29.9
OLS estimates. Units: Days ×countries. Period: January 29 to April 1 2020. All regressions
include a constant and country fixed effects. Panel A also includes country-specific time trends as
in the daily analysis, while Panel B includes only a global time trend. The sample includes only
Sundays in column (1), Mondays (2), Tuesdays (3), Wednesdays (4), Thursdays (5), and Fridays
(6), and Saturdays (8). Robust standard errors clustered at the country level in parentheses. *,
**, and *** indicate significance at the 10%, 5%, and 1% level.
Result: Prayer search shares increased on all weekdays, except Fridays.
C.4 Growth rates
Instead of identifying the impact on the levels of prayer search shares, this section documents
the impact on the growth rates, estimating the following equation:
gprayerct =β+αprayerct−1+γpandemict−1+λpandemict−1×characteristicc+δt+κc+εct (6)
where gprayerct is the growth rate in prayer search shares from time t-1 to time t in country
c. prayerct−1is the prayer search share at time t−1. To prevent day-to-day fluctuations
in the search data to impact results, the data was aggregated to weekly averages, so that t
represents a week. characteristiccis either religiosity or socio-economic characteristics. To
mimic the analysis across days, country-specific time-trends are included, but as this is a
rather restrictive test in this smaller sample, results are also shown with a global time-trend
instead. The main results from Table 1 are reproduced: The growth rate in prayer search
shares rise after March 11 and after the first case was registered. The rise in growth rates,
though is independent of there being any deaths registered and the rise after March 11 is not
larger for countries where COVID-19 had infected the populations (Table A.14). The results
are nearly identical including country-specific time-trends or not (Panel A vs Panel B).
Tables A.15 (with country-specific time-trends) and A.16 (with only a global time-trend)
65

document that prayer search shares grew at similar rates, independent of the previous level of
religiosity. If anything, the growth rates may be somewhat lower in more religious societies.
This is unsurprising, as small increases in prayer searches can increase the growth rate easier in
societies that previously did not search much for prayer on the Internet, compared to societies
where inhabitants already searched for prayer before the pandemic. This is also consistent
with what we saw in Panel (a) of Fig. 3.
When accounting for the initial level of religiosity in Panel B, the growth rate in prayer
search shares is higher in the more religious countries, consistent with the results on the
absolute rise in prayer search shares.
Table A.14: The rise in weekly prayer growth rates I
Dependent variable: Prayer
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A
Pandemic dummy 0.051* 0.16*** 0.16*** 0.077 0.17*** 0.17***
(0.026) (0.024) (0.024) (0.086) (0.024) (0.031)
Prayer t-1 -0.026*** -0.024*** -0.026*** -0.026*** -0.024*** -0.025*** -0.025***
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
Case dummy 0.089** 0.066* 0.058
(0.037) (0.035) (0.035)
Pandemic x First case dummy 0.085
(0.089)
Death dummy 0.040 -0.0096 0.0057
(0.038) (0.038) (0.040)
Pandemic x First death dummy -0.023
(0.047)
R-squared 0.12 0.37 0.35 0.38 0.38 0.34 0.37 0.37
Panel B
Pandemic dummy 0.051** 0.13*** 0.12*** 0.12** 0.13*** 0.13***
(0.025) (0.023) (0.022) (0.048) (0.024) (0.030)
Prayer t-1 -0.018*** -0.017*** -0.018*** -0.018*** -0.017*** -0.018*** -0.018***
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
Case dummy 0.060** 0.050* 0.049*
(0.026) (0.025) (0.026)
Pandemic x First case dummy 0.0081
(0.052)
Death dummy 0.012 -0.021 -0.037
(0.031) (0.031) (0.050)
Pandemic x First death dummy 0.021
(0.056)
R-squared 0.039 0.24 0.22 0.24 0.24 0.22 0.24 0.24
Observations 855 855 855 855 855 855 855 855
Countries 95 95 95 95 95 95 95 95
MeanDepVar 0.045 0.045 0.045 0.045 0.045 0.045 0.045 0.045
RestrictedSample No No Yes Yes Yes Yes Yes Yes
OLS estimates. Units: Weeks ×countries. Period: January 29 to April 1 2020. All regressions include a constant and country fixed
effects. Panel A also includes country-specific time trends as in the daily analysis, while Panel B includes only a global time-trend. The
sample is the full sample in columns (1)-(2), but is restricted to the sample where observations are dropped after the maximum prayer
search share over the period is reached in columns (3)-(8). Robust standard errors clustered at the country level in parentheses. *, **,
and *** indicate significance at the 10%, 5%, and 1% level.
Result: The growth rate in prayer search rises after March 11 and after the first case is registered.
66

Table A.15: The rise in weekly prayer growth rates II
Dependent variable: Prayer growth
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A Pray2019 MomentPray EverPray WeekPray God VeryGo d EverChurch WeekChurch EarthqRisk
Pandemic dummy 0.080* 0.073 0.026 0.037 0.25 0.064 0.20* 0.061 0.20
(0.041) (0.122) (0.142) (0.070) (0.256) (0.062) (0.115) (0.051) (0.172)
Pandemic x Religiosity -0.0011 -0.0096 0.068 0.075 -0.22 -0.034 -0.19 -0.024 -0.15
(0.001) (0.149) (0.164) (0.109) (0.273) (0.094) (0.138) (0.113) (0.192)
R-squared 0.12 0.12 0.16 0.16 0.13 0.13 0.13 0.13 0.13
MeanDepVar 0.045 0.044 0.045 0.045 0.043 0.043 0.044 0.044 0.047
MinimumImpact 0.079 0.070 0.055 0.050 0.12 0.061 0.13 0.061 0.13
MaximumImpact -0.017 0.063 0.093 0.11 0.027 0.031 0.014 0.040 0.051
Panel B
Pandemic dummy 0.10*** -0.043 -0.072 0.019 -0.12 0.068 0.039 0.080 0.021
(0.039) (0.119) (0.137) (0.066) (0.262) (0.062) (0.125) (0.053) (0.178)
Pandemic x Religiosity 0.0023** 0.29* 0.31* 0.29** 0.30 0.17* 0.16 0.26** 0.18
(0.001) (0.147) (0.166) (0.112) (0.280) (0.095) (0.154) (0.120) (0.201)
Prayer t-1 -0.026*** -0.029*** -0.026*** -0.026*** -0.027*** -0.027*** -0.027*** -0.027*** -0.027***
(0.002) (0.004) (0.004) (0.004) (0.003) (0.003) (0.003) (0.003) (0.004)
R-squared 0.37 0.33 0.36 0.37 0.34 0.35 0.35 0.35 0.34
Observations 855 576 405 405 711 711 702 702 621
Countries 95 64 45 45 79 79 78 78 69
MeanDepVar 0.045 0.044 0.045 0.045 0.043 0.043 0.044 0.044 0.047
MinimumImpact 0.11 0.041 0.062 0.069 0.058 0.083 0.10 0.087 0.11
MaximumImpact 0.31 0.25 0.24 0.29 0.19 0.23 0.20 0.31 0.20
OLS estimates. Units: Weeks ×countries. Period: January 29 to April 1 2020. All regressions include a constant, country-specific time trends, and
country fixed effects. Panel B also includes a control for the average prayer search share during the previous week. Robust standard errors clustered at
the country level in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% level.
Result: The growth rate in prayer search shares is not larger for the more religious. Only when accounting for prayer search shares during the previous
week.
Table A.16: The rise in weekly prayer growth rates III
Dependent variable: Prayer growth
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A Pray2019 MomentPray EverPray WeekPray God VeryGod EverChurch WeekChurch EarthqRisk
Pandemic dummy 0.037 0.072 0.23*** 0.16*** 0.21 0.089** 0.15** 0.087** 0.074
(0.034) (0.089) (0.083) (0.045) (0.153) (0.041) (0.067) (0.035) (0.118)
Pandemic x Religiosity 0.00054 -0.0085 -0.18* -0.14* -0.18 -0.078 -0.12 -0.099 -0.0046
(0.001) (0.109) (0.104) (0.075) (0.162) (0.058) (0.080) (0.071) (0.130)
R-squared 0.040 0.045 0.065 0.066 0.043 0.044 0.045 0.045 0.041
MeanDepVar 0.045 0.044 0.045 0.045 0.043 0.043 0.044 0.044 0.047
MinimumImpact 0.037 0.069 0.15 0.14 0.11 0.081 0.10 0.084 0.071
MaximumImpact 0.084 0.063 0.050 0.029 0.030 0.012 0.028 -0.0012 0.069
Panel B
Pandemic dummy 0.034 -0.021 0.063 0.067 -0.088 0.081* 0.018 0.095** -0.097
(0.031) (0.092) (0.082) (0.051) (0.175) (0.044) (0.089) (0.038) (0.128)
Pandemic x Religiosity 0.0039*** 0.22** 0.11 0.15* 0.23 0.080 0.15 0.11 0.28*
(0.001) (0.111) (0.101) (0.086) (0.188) (0.069) (0.109) (0.080) (0.142)
Prayer t-1 -0.019*** -0.021*** -0.017*** -0.018*** -0.019*** -0.019*** -0.019*** -0.019*** -0.019***
(0.002) (0.003) (0.004) (0.004) (0.003) (0.003) (0.003) (0.003) (0.003)
R-squared 0.26 0.22 0.23 0.23 0.21 0.21 0.22 0.22 0.21
Observations 855 576 405 405 711 711 702 702 621
Countries 95 64 45 45 79 79 78 78 69
MeanDepVar 0.045 0.044 0.045 0.045 0.043 0.043 0.044 0.044 0.047
MinimumImpact 0.038 0.043 0.11 0.092 0.047 0.089 0.073 0.098 0.032
MaximumImpact 0.37 0.20 0.17 0.21 0.14 0.16 0.16 0.19 0.18
The estimations mimic those in Table A.15, except that a global time-trend is included instead of country-specific time-trends.
Result: The growth rate in prayer search shares is not larger for the more religious. Only when accounting for prayer search shares during the previous
week.
67