School holidays accounted for up to half of the increase in Covid-19 infections in Germany over the summer
In April 2020, Germany, like many European countries, had managed to repress the spread of Covid-19, allowing its citizens a summer of seeming normality. Yet, the party was short-lived. The second wave of Covid-19 started much earlier and more ferociously than many epidemiologists had expected (Iacobucci 2020, Ohrling 2020). Seasonal effects may have played a role of course, but the rise in the number of new infections began too early to be entirely caused by lower temperatures and an increase in indoors activities. In Germany, the public debate on the root causes of the second wave focuses on two factors: large private social events and international holiday travel to high-risk areas (RKI 2020a, RKI 2020b). Our recent research shows that, on average, the effect of school summer holidays accounts for about half of the increase in growth rates in infections in German local districts towards the end of the holiday period (Plümper and Neumayer 2020a).
The substantial effect of summer holidays on infections was widely predicted. Travelling via bus, train, or plane adds to the risk of becoming infected, though probably only to a small extent (Gonne and Hubert 2020, Hu et al. 2020, Schwartz et al. 2020). More importantly, holidaymakers and migrant workers, regardless of whether they travel internationally or internally (Valsecchi and Durante 2020), change their social behaviour and engage in more intense social interactions, often with people that they do not know. Thus, holidaymaking also reduces the ability of health authorities to successfully trace the close contacts of people that are infected with Covid-19.
A paucity of evidence on the effects of holidaymaking on the pandemic
Since the accelerating effect of holidaymaking on the Covid-19 pandemic was to be expected, it is surprising that we know of no academic study besides ours that actually assesses the impact that the school summer holiday season on the pandemic.1 To be sure, we are not the first to report relevant numbers for Germany. In reports of positive cases that local health authorities send to Germany’s Robert Koch Institute (RKI), the most likely place of infection is typically (though not always) noted, with the notification of multiple places possible. In its daily update from 13 October (RKI 2020a), the Institute reports:
“In the initial phase (…) the share of all cases with an exposure location abroad was 46% in reporting week 11. Due to travel restrictions, this share has steadily fallen, to 0.4% in reporting week 19. Since the 25th reporting week, there have been the first border openings, initially in Europe. From then onwards, the proportion of cases with a likely exposure location abroad increased again and reached its peak in reporting week 34 at 49%.”
After the peak in week 34 (mid-August), the share of infected with a likely infection location abroad declined to about 4% in mid-October. Based on numbers reported in RKI (2020), we calculate that, on average, in about 27% of weekly cases over the entire summer school holiday period in Germany, a foreign country was mentioned as the most likely place of infection.
This does not necessarily mean the effect of holidaymaking is on average 27% and at most 49% of all infections. On the one hand, infections may not have occurred abroad or may not have been linked to holidaymaking. On the other hand, the RKI excludes the impact of holidaying within Germany by design. The RKI numbers also understate the effect of holidaymaking via second-order infections. Consider the case of a person who became infected while being abroad and then infects one or more contacts after returning to Germany. Implicitly, the second-order infections not only do not count as foreign travel-related, they even bring down the share of infections occurring abroad.
Exploiting an idiosyncratic feature of school holidays in Germany
Our research design estimates the effect of the holiday season on the growth rate in all new infections. We exploit an idiosyncratic feature of school holidays in Germany, namely, the ex-ante coordination of holidays across the 16 German states. States stagger their school holidays in order to reduce the probability and scale of traffic jams on Germany’s motorways. As a consequence, school holidays are spread over a period of approximately 12 weeks, though the maximum length of holidays in each state is six weeks or a few days more. This feature allows us to estimate the effect of summer school holidays on the growth rate in districts that are on holiday with the growth rate in districts that are not on holiday as the presumed counterfactual. To mitigate the impact of potential confounders, we control via district fixed effects for time-invariant heterogeneity and we account for the common trend by including a lagged dependent variable. Our sample covers 401 German districts (combining districts in Berlin into a single unit) over ten weeks (the six holiday weeks plus two weeks before and after the holidays). We include the two weeks after holidays officially end to capture infections from those returning toward the end of the holiday period, the incubation period, and delays in getting tested.
The effects by week, by state, and by district
Disaggregating the effect week by week of the holiday season, we find that the estimated effect equates to 48.7% (95% confidence interval: 36.6–60.6) of the average growth rate across German districts during their respective final week of holidays and to 49.5% (95% confidence interval: 40.0–61.0) and 46.1% (95% confidence interval: 34.8–57.4) of the average growth rate during their respective first two weeks after holidays end. However, this effect is not identical for all German states. In contrast, if we allow for causal heterogeneity across states, we find virtually no holiday effect for Thüringen, Brandenburg, and Saxony-Anhalt. On the other end of the effect size continuum, we find the strongest holiday effects in Lower Saxony, Bavaria, Baden-Württemberg, Schleswig-Holstein, Bremen, and Hesse.
A plausible explanation for this heterogeneity comes from another of our findings, namely, that the holiday effects are stronger in districts that are richer and in which foreigners make up a larger share of the resident population. Richer people are more likely to go on holiday for longer (Gokovali et al. 2017) and foreign citizens are likely to use the holiday season for returning to their home country for family visits (possibly in addition to taking other holidays). The states that see the highest effects also tend to be ones with richer districts and districts with an above average share of foreign residents.
We illustrate the two conditioning effects in Figures 1a and 1b, in which we use a heatmap with dark and red colours indicating low levels of growth and yellow and white indicating high levels of growth in district infection numbers. The x-axis displays the period from seven days before the summer school holidays begin in a district to eight weeks after the beginning of holidays at a daily level. On the vertical axis of Figure 1a, the 401 German districts are sorted in descending order by taxable income so that districts with higher taxable income appear at the top and districts with a lower taxable income appear at the bottom. Similarly, on the vertical axis of Figure 1b districts are sorted by the share of foreigners amongst residents so that the districts with the highest share appear at the top and districts with the lowest share appear at the bottom of Figure 1b.
Figure 1 The conditioning effects of taxable income and share of foreigners on the holiday effect
a) By income
b) By share of foreigners
Figures 1a and 1b look fairly similar in that both become lighter as we move from left to right; Figure 1a in particular becomes lighter as we move from bottom to top of the figure. The conditioning effect of taxable income thus becomes more clearly visible in Figure 1a than the conditioning effect of the share of foreigners amongst residents in Figure 1b.
School summer holidays contributed substantially to the rise in infections in Germany, which is now facing a second wave of the pandemic. In this respect, the holiday effect reminds us of the very early stage of the pandemic in Germany, when ski holidays spread the virus to Germany and other countries in Europe (Plümper and Neumayer 2020b). The school summer holiday effects were entirely predictable and yet public health authorities largely failed to mitigate the impact. While the federal or state governments could not have prohibited holiday-related travel given the low infection levels in early summer and the relatively minor role strict border controls played during the first wave (Eckhart et al. 2020), the summer holiday effect caught authorities strangely unprepared. It took authorities till the end of the summer holiday to drive up testing capacities for returning travellers and it took even longer to implement compulsory free tests for travellers returning from high-risk areas. Holiday-related travels are of course not the only reason why infections started to rise again over the summer. People grew increasingly tired of social distancing rules and started to revert back to their pre-pandemic social behaviour. Nevertheless, historians will look back at the summer of 2020 as the period when things started to go pear-shaped again in Germany and other European countries.
Britton, T and F Ball (2020), “Summer vacation and COVID-19: effects of metropolitan people going to summer provinces”. arXiv preprint arXiv:2006.00579.
Matthias E, K Kappner and N Wolf (2020), “Covid-19 across European regions: The role of border controls”, VoxEU.org, 4 November.
Ewing A, EC Lee, C Viboud , S Bansal (2017), “Contact, travel, and transmission: The impact of winter holidays on influenza dynamics in the United States”, The Journal of Infectious Diseases 215(5):732-9.
Gokovali U, O Bahar, M Kozak (2007), “Determinants of length of stay: A practical use of survival analysis”, Tourism Management 28(3):736-46.
Gonne N, O Hubert (2020), “Spatial distancing: air traffic, COVID-19 propagation, and the cost efficiency of air travel restrictions”, Covid Economics 24: 111-125.
Hu M, H Lin, J Wang, C Xu, AJ Tatem, B Meng, X Zhang, Y Liu, P Wang, G Wu, H Xie (2020), “The risk of COVID-19 transmission in train passengers: an epidemiological and modelling study”, Clinical Infectious Diseases, 29 July.
Iacobucci, G (2020), “Covid-19: UK must prepare now for winter peak or risk many more deaths, scientists warn”, British Medical Journal (Online) 370.
Klausner, Z, E Fattal, E Hirsch and S C Shapira (2020), “A single holiday was the turning point of the COVID-19 policy of Israel”, Medrxiv.
Ohrling, M, J Øvretveit, U Lockowandt, M Brommels and V Sparring (2020), Management of the emergency response to the SARS-CoV-2 (COVID-19) outbreak in Stockholm, Sweden, and winter preparations”, Journal of Primary Health Care 12(3): 207-214.
Plümper, T, E Neumayer (2020a), “Summer School Holidays and the Growth Rate in Sars-CoV-2 Infections Across German Districts”, Medrxiv.
Plümper, T, E Neumayer (2020b), “Wealthier districts were hit by Covid-19 first in Germany, but their lockdowns were more effective”, VoxEU.org, 11 June.
Robert-Koch-Institute (2020a), Täglicher Lagebericht, 13 October.
Robert-Koch-Institute. (2020b), Täglicher Lagebericht, 27 October.
Schwartz KL, M Murti, M Finkelstein, JA Leis, A Fitzgerald-Husek, L Bourns, H Meghani, A Saunders, V Allen and B Yaffe (2020), “Lack of COVID-19 transmission on an international flight”, Canadian Medical Association Journal 192(15).
Valsecchi, M and R Durante (2020), “Internal migration and the spread of COVID-19”,VoxEU.org, 02 September.
1 See Ewing et al. (2017), Klausner et al. (2020) and Britton and Ball (2020) for relevant studies, none of which is directly comparable however.