How a school holiday led to persistent COVID-19 outbreaks in Europe

This paper investigates the role of large outbreaks on the persistence of Covid-19 over time. Using data from 650 European regions in 14 countries, I first show that winter school holidays in late February/early March 2020 (weeks 8, 9 and 10) led to large regional outbreaks of Covid-19 in the spring with the spread being 60% and up-to over 90% higher compared to regions with earlier school holidays. While the impact of these initial large outbreaks fades away over the summer months, it systematically reappears from the fall as regions with school holidays in weeks 8, 9 and 10 had 30–70% higher spread. This suggests that following a large outbreak, there is a strong element of underlying (latent) regional persistence of Covid-19. The strong degree of persistence highlights the long-term benefits of effective (initial) containment policies, as once a large outbreak has occurred, Covid-19 persists. This result emphasizes the need for vaccinations against Covid-19 in regions that have recently experienced large outbreaks but are well below herd-immunity, to avoid a new surge of cases.

since the holidays both ends first and the state is the most remote of the week 8 areas in Germany, this would likely lead to travelers starting their return earlier than those in other week 8 states in Germany (roughly 400 km south).
As discussed above in section 2 the spread was taking off during these pivotal days and finishing a school holiday earlier within week 8 may therefore lead to variation in likelihood of exposure. Secondly, the 2020 winter school holiday in Mecklenburg-Vorpommern started a week later than the three preceding years (begun on 4-6th of February). Saxony has for comparison begun breaks starting from 12-18th of February in previous years. If the timing of travel is somewhat sticky between years, it may lead to travel being relatively skewed to the early part of Austria is an example of a known hot-spot in late February and early March [29]. It is therefore natural to investigate the travel patterns to Austria during this period. From official Austrian tourism data [30] we can see that just below 2,7 million tourists visited the Austrian alps in February (three largest areas only: Tyrol, Salzburg, Vorarlberg). Investigating the origin country/region breakdown and duration of stay, we can see that the largest groups of visitors come from Germany, Netherlands, Belgium, Denmark and Sweden. Together they account for over 2 million visitors. From table C1 we can see considerable regional variation in the intensity of travel to Austria across German regions. Eastern-Germany and Berlin have the highest intensity of travel, as measured by average time spent in the Austrian Alps (120 and 116 number of nights per thousand inhabitants). A combination of many guests, relative to population, and long average duration is consistent with the fact that all the eastern regions and Berlin had extended holidays sometime during February. With the high intensity of travel but relatively early school holidays during (3-8 February, week 6), Berlin likely escaped a large initial outbreak.
We can also compare these regions to the similarly distant North Rhine-Westphalia and northern-Germany, which do not have a break in February. From table C1 we can see that the eastern regions and Berlin have a three-times higher level of travel (nights per capita), consistent with clustered school holiday travel. North Rhine-Westphalia and northern-Germany also share Week 7-10. Stockholm week 9.
Note: The data from Statistics Austria groups German states in the following way: Central Germany: Hesse, Rhineland-Palatinate, Saarland. Northern Germany: Lower Saxony, Hamburg, Bremen, Schleswig-Holstein. Eastern Germany: Saxony, Saxony-Anhalt, Thuringia, Brandenburg, Mecklenburg-Vorpommern. The numbers in columns 5-6 and 8 are per inhabitant of the origin region/country (in thousands). Columns 7 and 8 (for March) can be compared to columns 4 and 6 (for February). *Hamburg in Northern Germany, is the only German NUTS 3 region that has a break in the beginning of March. See discussion in appendix C.
a border with the Netherlands and Denmark, which have much higher level of travel. From table C1 we can see that people from the Netherlands spend on average the most nights in the Austrian alps during February (157 nights per thousand inhabitants). A notable difference is that the Netherlands has a school holiday in week 8 or 9. We can even see high level of travel from Denmark during the winter school holiday season in February consistent, with Bluhm et al.
[6] who trace the bulk of the genome sequences in Denmark to travel from Austria. Even in the south of Germany, relatively close to Austria, a difference can be seen in the travel between Bavaria (week 9) and Baden-Württemberg (none  figure F2 and tables F14 and F15. Second, as noted above, by including distance to Ischgl we can also investigate the role of relative within country distance to Ischgl. After adding distance to Ischgl and including the country specific fixed effect, the results are broadly similar. Note that initial impact in March/April is still large and highly significant for week 9 (around 50%) but appears somewhat dampened. The large week 10 effect is unchanged. This likely reflects the widespread testing in Germany from the first initial stages, compared to other countries, combined with the fact that week 9 regions in Germany are in the most southern part of the country (relatively closest to Ischgl). Importantly, we see that the results on post-summer persistence are unchanged, consistent with the results in figure F2 which shows minimal role of absolute distance. See figure   F3 and tables F14 and F15 for the full results. The inclusion of distance creates some odd patterns when used with multiple countries, as regions that have the same absolute distance to Ischgl will be considered relatively far and relatively close, depending on the domestic internal distance. This is since the country fixed effect will capture the part of the distance that is common to all regions in a country. An example is Netherlands and Denmark, who border Germany. Many of these areas will have the lowest internal distance, while the bordering areas in Germany will have relatively high measure of internal distance, since other German areas are closer.
Third, we may be worried that the results are driven by German regions that are within a few hundred kilometers from the Austrian ski-resorts. Hence, regions where it may be possible to take day-trips to the Alps rather than longer extended periods during the school holidays. To investigate this potential issue, we drop the two regions in Germany that are closest to Austria (Bavaria and Baden-Württemberg). The results are very similar, see figure F4. Fourth, as Covid-19 spread over time to more areas, travel to non-ski related areas may lead to exposure in the later holiday break weeks (9-10).
Distance is therefore not included in the main specification since: 1) the absolute distance has a no impact on the results suggestive of a dominant role of the timing of travel, 2) the relative internal distance to Ischgl has in general a small and unclear role in our setting given country fixed effects and other controls (e.g. typology), 3) the spread becomes more broad over time resulting in travel in general being likely to lead to exposure and not only travel to Austria.    Standard errors, clustered at NUTS 2 level, in parenthesis. c p < .1, b p < .05, a p < .01 Same results as in figure 4, 4a. Data from two areas is missing in March and from Portugal in August-October.      Standard errors, clustered at NUTS 2 level, in parenthesis. c p < .1, b p < .05, a p < .01 Same results as in figure 5.   Standard errors, clustered at NUTS 2 level, in parenthesis. c p < .1, b p < .05, a p < .01