COVID-19 lockdowns & demographically-relevant Google Trends: Changes in searches for terms such as wedding and those related to condom use, emergency contraception, pregnancy tests, and abortion

Berger LM, Ferrari G, Leturcq M, Panico L, Solaz A (2021) COVID-19 lockdowns and demographically-relevant Google Trends: A cross-national analysis. PLoS ONE 16(3): e0248072. https://doi.org/10.1371/journal.pone.0248072

Abstract: The spread of COVID-19 and resulting local and national lockdowns have a host of potential consequences for demographic trends. While impacts on mortality and, to some extent, short-term migration flows are beginning to be documented, it is too early to measure actual consequences for family demography. To gain insight into potential future consequences of the lockdown for family demography, we use cross-national Google Trends search data to explore whether trends in searches for words related to fertility, relationship formation, and relationship dissolution changed following lockdowns compared to average, pre-lockdown levels in Europe and the United States. Because lockdowns were not widely anticipated or simultaneous in timing or intensity, we exploit variability over time and between countries (and U.S. states). We use a panel event-study design and difference-in-differences methods, and account for seasonal trends and average country-level (or state-level) differences in searches. We find statistically significant impacts of lockdown timing on changes in searches for terms such as wedding and those related to condom use, emergency contraception, pregnancy tests, and abortion, but little evidence of changes in searches related to fertility. Impacts for union formation and dissolution tended to only be statistically significant at the start of a lockdown with a return to average-levels about 2 to 3 months after lockdown initiation, particularly in Europe. Compared to Europe, returns to average search levels were less evident for the U.S., even 2 to 3 months after lockdowns were introduced. This may be due to the fact, in the U.S., health and social policy responses were less demarcated than in Europe, such that economic uncertainty was likely of larger magnitude. Such pandemic-related economic uncertainty may therefore have the potential to slightly increase already existing polarization in family formation behaviours in the U.S. Alongside contributing to the wider literature on economic uncertainty and family behaviors, this paper also proposes strategies for efficient use of Google Trends data, such as making relative comparisons and testing sensitivity to outliers, and provides a template and cautions for their use in demographic research when actual demographic trends data are not yet available.

5. Conclusion

The COVID-19 pandemic is likely to have a variety of short-, medium-, and long-term consequences for society that may vary across nations, at least in part, in accordance with governmental responses vis-à-vis lockdowns and other health and social policies. Notably, we hypothesized that consequences for demographic behaviors might be more marked in countries with less well managed outbreaks and with fewer safety nets, exacerbating feelings of uncertainty and economic insecurity. Pandemic-related research to date has primarily focused on mortality [2], migratory patterns [3, 4], economic wellbeing and government responses thereto [1, 15, 38, 45–48, 66], and physical and mental health and wellbeing [11–13, 36–38, 46, 48, 63]. [1] provide a comprehensive review of the pandemic-related research to date. Understanding more fully how population dynamics and, in particular, those related to family demography, may be affected is crucial to predicting subsequent demographic trends. To this end, we use Google Trends data and event study and difference-in-differences techniques to examine the influence of national and, in the case of the U.S., state-level lockdowns during the pandemic on demographically-relevant Google search trends. We focus specifically on sexual behavior-, contraceptive use-, pregnancy termination- and fertility-related searches; coupling, romantic relationship, and union formation-related searches; and union dissolution-related searches. On the whole, we find modest evidence of changes in search patterns related to family planning, but little evidence of changes in search patterns related to fertility, with the exception that current parents may have increased searches for information about having subsequent children. Using Google Trends data, Wilde et al. [61] predict a 15% decrease in fertility; these results are based on keywords linked to unemployment rather than fertility, with the assumption of a causal link between unemployment and fertility, as reported in previous years. When based solely on fertility key-related words, their results suggest no large negative effect on fertility and are in line with our results, in spite of using different fertility key-related. We do find some changes in search patterns related to relationship and union formation, as well as union dissolution. However, these tended to attain statistical significance only in the period immediately surrounding lockdown initiation and to return to average pre-pandemic levels within 2 to 3 months, particularly in the European context; divergence in U.S. trends tended to last longer. This fits with our initial hypothesis that lockdowns will lead to more marked demographic consequences in countries that managed the economic fallout from the pandemic less well and provided weaker financial safety nets.

Starting with our results for sexual behavior-, contraceptive use-, pregnancy termination- and fertility-related searches, we conclude that the introduction of lockdown measures resulted in a short-lived decline in relative searches for condoms in the U.S., but no change in Europe, and that searches for the morning after/emergency pill declined substantially in both Europe and the U.S. at the initiation of lockdowns, returning to average levels in Europe within a few months, but remaining below average after 3 months in the U.S. In both Europe and the U.S., relative searches related to “pregnancy test” declined slightly a few weeks prior to lockdowns but returned to average within 6 to 8 weeks of lockdown initiations. Searches for abortion declined in the U.S. and, to a much lesser extent in Europe, with substantial rebound within 3 months of lockdown initiations. These findings may indicate some changes in sexual behaviors during lockdowns, such that unprotected sex and/or contraceptive failure may have temporarily declined, likely predominantly among non-coresident couples, during the early months of the pandemic and associated lockdowns, but tended to rebound relatively quickly.

We find no change in relative searches regarding overall planning for childbirth associated with lockdowns and their timing, but some suggestive evidence that couples who are already parents may take the opportunity of lockdown to consider having an additional child. We can place our results within the wider literature on drivers of fertility trends. First, these results speak to the literature on the impact unemployment or job insecurity [19, 22, 23]. Second, they fit with the literature of the response of fertility levels to “shocks”, such as the Great Depression (which resulted in a “baby bust”), or World War II (and its resulting “baby boom”). Seminal work by Rindfuss, Morgan, and Swicegood [25] found, for instance, that economic recessions often lead to a postponement of childbearing, especially for first births. The literature emphasizes the importance of actual versus anticipated financial and employment situations. While actual financial losses and unemployment do correlate with fertility trends, an important pathway to understand the impact of macro-level business cycles on fertility is consumer confidence [20, 27]. Financial incertitude might be crucial to understanding fertility responses to the lockdowns: while unemployment levels have increased as a result of lockdowns, our results are often stronger in the very first weeks before or after lockdowns, that is, before any substantial impacts on jobs and income losses. Yet, we find no evidence that initial lockdowns lead to significant changes in overall fertility intentions, and only a slight suggestion that higher order births may be being considered by current parents. Our conclusions, however, only relate to fertility intentions (rather than realization) and initial lockdown implementation (rather than potential subsequent economic recessions or the length of the lockdown). Interestingly, the (small) potential impact of lockdowns on higher-order fertility suggested by our results show a relative decline in searches related to higher-order childbearing in the month preceding and the month following lockdowns, with the suggestive evidence of an increase in searches only appearing thereafter. It is possible that this pattern is, at least partially, driven by expectations of future financial and employment instability, particularly if such uncertainty was highest in the earliest months of the pandemic.

Turning to our results on relationships, union formation, and union dissolution, relative searches related to dating (and for dating/relationship applications) and coupling/relationship formation appear to be negatively affected by lockdowns. However, these impacts were often short-lived and, again, concentrated in the weeks immediately surrounding the introduction of shelter-in-place measures. However, fitting in with our initial hypothesis, we find some differences between the U.S. and Europe in these impacts and, in particular, their timing. In Europe, relative searches returned to pre-lockdown levels after only a modest change, whereas changes in searches in the U.S. appeared to be more long lasting. As we had hypothesized, this may reflect that, around the time of national lockdowns, the pandemic was considered to be better controlled in many European settings that in the U.S., and also that economic consequences may have been expected be better attenuated by stronger welfare state policies and faster and more comprehensive economic responses in Europe.

European and U.S. trends were similar for searches related to union formation, with relative declines in searches related to marriage and weddings that persisted in both contexts even three months after lockdowns were initiated, though the magnitude of decline was larger in Europe. To the extent that trends in union formation parlay into subsequent trends in fertility, this may suggest slight declines in near-term first-birth fertility. However, the direction of causality between marriage and fertility is ambiguous. That is, lower marriage rates may reflect individuals or couples not (yet) wanting a child and therefore choosing to not (yet) marry or enter into cohabitation, rather than fertility being driven by changes in marriage or cohabitation formation behaviors [28, 29]. Moreover, delaying or opting out of marriage will only have an effect on fertility in countries where the tie between marriage and childbearing is strongest [28]. Furthermore, longer-term fertility and family formation plans will also be affected by post-lockdown economic and policy landscapes, which we do not observe.

It is less clear how lockdowns and resulting economic downturns might affect separation and divorce. Two contrasting hypotheses can be made [31]: economic instability increases financial and psychological stress for couples, which increase risk of separation and divorce, and, given considerable financial costs of separation and divorce (including legal costs and the loss of economies of scale), couples might be “priced out” of being able to separate during financially unstable or uncertain times. While evidence suggests that well-being might have decreased and stress increased during the pandemic [36], our results point—if anything—toward near-term declines in divorce, which is consistent with recent evidence from the 2008 recession, which resulted in couples putting off divorcing [30, 32]. Indeed, we find a modest short-lived decline in divorce-related relative search interest in Europe and a small short-lived decline in the U.S., particularly in the first weeks of lockdowns. Notably, however, we find little evidence of a change in searches for break-up in Europe and a slight, short-lived increase in break-up-related search interest in the U.S. in the wake of the lockdowns.

Overall, differences in demographically-relevant searches in the U.S. and Europe may reflect both the differential handling of the pandemic in the spring of 2020 and differences in underlying social welfare safety nets. Indeed, international heterogeneity in individual perceptions of the effects of the pandemic is remarkable—and tends to vary with both of these factors. For example, while threat perception of Covid-19 to oneself, one’s family, one’s local community, one’s country, and the world were high across most countries at the start of the pandemic, they decreased over time in countries such as Germany and Italy, while increasing in the U.K. and U.S. [39]. Similarly, confidence in institutions such as the health system was lowest in the U.K. and U.S., and highest in European countries, particularly Spain [39]. Also, the economic consequences of the pandemic and the resulting lockdowns have not been shared equally within societies [66]. Of particular concern, inequality in job losses during the pandemic have been found to be particularly stark in the U.S. and U.K. [66]. If a main pathway to understanding possible impacts of lockdowns on future family demographic trends is economic hardship and uncertainty, then we might expect increasingly polarized behaviors, particularly in the U.S. where our results, though still small-to-modest in magnitude, tend to be stronger and longer-lasting than in Europe. Trends in increasing polarization of family formation behaviors have been well underway in the U.S. for several decades [67]. Of course, only time will tell what, if any, influence the pandemic and resulting economic downturn with have on family demographic trends, and where.

Our results must be interpreted in the light of a number of limitations. First, although Google search trends cannot be assumed to fully represent the interests, concerns, or preoccupations of a nation’s population as a whole [65, 68], the popularity of Google as a search engine in Europe and the U.S., combined with quick and free data access, makes it a useful tool for observing immediate changes in search behaviors in response to the pandemic. Zagheni and Weber [69] recommend using difference-in-differences techniques, such as those used in this study, to assess relative trends in internet search data over time as an efficient means of reducing bias when the population-representativeness of internet searches by a subset of the population is unknown.

Second, as discussed in detail above, Google Trends only reports the relative (not absolute) search trends, specifically the number of searches for a particular term at a given time in a particular geographic unit relative to the total search volume in the geographic unit at a given time. This search index is then normalized to the maximum search volume observed over the period in the geographic unit. In a usual period, as the number of total Google searches is very large in comparison to the number of searches for a given term, a small change in the denominator (all searches) is negligible for the interpretation of the index. However, some periods, particularly those characterized by large shocks—like the introduction of national lockdowns—may generate a large increase in overall Google searches. In this case, the variation in the total number of Google searches might influence the relation of particular searches to the relative search index. This may be particularly true during lockdowns when individuals are required to stay at home and may therefore be more likely to use the internet. This implies that, while the raw number of searches for a particular term may be constant over a period, the number of searches relative to the overall number searches will change if the total number of Google searches increases (decreases) over the period. While we suspect that the total volume of Google searches increased during lockdown, at the time of writing Google could not confirm this hypothesis. Google Trends data must, therefore, be interpreted as reflecting searches for a specific term relative to all Google searches, and not as reflecting absolute search intensity, per se. Nonetheless, a change in relative search intensity during lockdowns can be identified and evaluated.

Third, Google searches do not directly measure behaviors. Future research based on actual levels of demographic behaviors is needed to corroborate whether and how such searches may predict trends in family demography.

Fourth, our results represent average national/state level searches and may mask considerable heterogeneity in the way countries (and states) imposed lockdown restrictions, which in turn may influence perceptions and reactions to these measures. For country comparison purposes, we use a binary treatment: whether the lockdown was implemented or not. This choice does not allow differentiating the extent of different lockdowns in terms of length or stringency, nor considering other forms of restrictions than stay-at-home orders or their changing nature over time. Our analytical framework also does not allow considering intra-country variation within the European countries considered. Further decompositions by smaller geographical areas are warranted.

In sum, using Google Trends data, we find evidence that pandemic-induced lockdowns appear to have small and likely short-lived impacts on family demography-related searches, particularly in the U.S. We interpret these findings within the context of important economic and social policy variation across countries, such that social and economic uncertainty was likely of larger magnitude in the U.S. than Europe. Whereas it is possible that such uncertainty may result in potentially (small-to-modest in magnitude) impacts of the pandemic on future family demography in the U.S., whether any such impacts occur is a question for future research.