key: cord-0810741-er6codzm authors: Aparicio, Ainoa; Grossbard, Shoshana title: Intergenerational residence patterns and Covid-19 fatalities in the EU and the US*() date: 2020-10-29 journal: Econ Hum Biol DOI: 10.1016/j.ehb.2020.100934 sha: 25591e0eaa84e06feacdf89cd87ee9770b872006 doc_id: 810741 cord_uid: er6codzm We study how patterns of intergenerational residence possibly influence fatalities from Covid-19. We use aggregate data on Covid-19 deaths, the share of young adults living with their parents, and a number of other statistics, for 29 European countries associated with the European Union and all US states. Controlling for population size, we find that more people died from Covid in countries or states with higher rates of intergenerational co-residence. This positive correlation persists even when controlling for date of first death, presence of lockdown, Covid tests per capita, hospital beds per capita, proportion of elderly, GDP per capita, government’s political orientation, percentage urban, and rental prices. The positive association between co-residence and fatalities is led by the US. Covid is a relatively deadly epidemic (Yang et al., 2020) that is particularly likely to kill older people (Abdulamir and Hafidh, 2020) . It is also very contagious (Wu et al., 2020) . One strategy that individuals and policy-makers have used to protect the more vulnerable elderly from the virus is to minimize contacts between older adults and younger people (Koh, 2020) . Intergenerational co-residence makes it harder to avoid such intergenerational contacts. This paper's main goal is to test whether intergenerational coresidence is positively associated with Covid fatalities. To our knowledge there is limited research assessing the association between mortality from Covid and co-residence of older and younger adults based on international comparisons and across US states. Some recent studies examining the question by comparing various regions of European countries have not established a positive association between intergenerational coresidence and fatalities from Covid (e.g. Belloc et al. 2020 , Arpino et al. 2020 . Tests are performed using data from 29 countries associated with the European Union (EU) as well as the fifty states of the USA, which adds up to 79 areas (countries or US states). 1 By studying 79 geographic areas associated with only two federations (the EU and the USA) we add to previous cross-country or cross-region studies of the association between intergenerational co-residence and Covid fatalities in multiple ways. First, we expand on Bayer and Kuhn's (2020) study based on data from 24 countries from four continents; our sample is a more homogeneous set of 79 countries associated with the EU or part of the USA. 2 Second, our period of observation is longer than most other studies, including Bayer and Kuhn (2020) : we collected number of deaths from February 15 (first death in France) to August 3 2020, one hundred and twelve days after the first Covid death was reported in Wyoming. Third, as pointed out by Belloc et al (2020) , Bayer and Kunz pull together countries at different stages of their epidemic curve. In contrast, we analyze cumulative fatalities attributed to Covid measured at fixed intervals after an area's onset of the epidemic: 20, 40, 60, 80 and 100 days after the area's first death. We investigate not only whether intergenerational co-residence contributes to deaths from Covid-19 but also whether such contribution (if positive) varies with the timing of the epidemic's onset. Many cross-country studies of fatalities from Covid such as Bayer and Kuhn (2020) and Sorci et al (2020) have used variation in CFR (case fatality rate) as their measure of mortality. This requires data on both Covid cases and fatalities. Our fourth contribution is that we use cumulative fatalities as our principal COVID outcome. We complement it using cumulative confirmed cases but we place more emphasis on explaining cumulative deaths, as it minimizes measurement error coming from differences in country data publishing methods and testing behavior. In our sample number of tests per capita after 86 days ranged from 0.01 in Bulgaria to 0.18 in Rhode Island and New Mexico. Fifth, we make a methodological contribution to the literature on fatalities and intergenerational co-residence by controlling for the following variables: test rates per capita measured 14 days prior to the dates at which fatalities are measured, number of days separating the first death in France 3 from the onset date in a state or country, number of hospital beds, and a number of other economic, demographic and political factors likely to be associated with Covid fatalities. We find that intergenerational co-residence, defined as the share of 18-34 years old living with their parents, is associated with more cumulative deaths after 20, 40, 60, 80 and 100 days. For example, an extra one percentage point in such co-residence is associated with 4% more cumulative deaths from Covid 40 days after the area's onset of the epidemic, 3.5% more cumulative deaths 60 days after onset and 3% more Covid deaths at either 80 or 100 days after onset. We find that the association between coresidence and cumulative deaths is larger for US states than for EU countries. Lovett et al. (2020) have suggested that in the USA the association between co-residence and deaths is related to a urban/rural divide, with multi-generational co-residence being more common in rural areas. We don't find that. We also explore whether one of the channels that links intergenerational coresidence and cumulative death rates is an association between intergenerational coresidence and number of cases. We find that co-residence rates are also positively associated with cumulative cases, a proxy for the number of infected individuals. Even though it was not our primary goal, our study also offers a more sophisticated comparison of fatalities due to Covid in the US and the EU than a simple comparison between European countries and the USA as one country. From a statistical point of view it may be incorrect to pool all the states of the US. 4 We check the robustness of our results by controlling for different social distance measures, not only lockdowns, and using a 16 days difference between tests and deaths instead of a 14 days difference. The next section discusses the models and data used in this study. Results are presented in Section 3 and Section 4 concludes. 2a. The model. Our main variables of interest are cumulative Covid-caused deaths and intergenerational co-residence rates. We estimate the conditional correlation of Covid deaths and intergenerational co-residence rates using log-linear regressions, with log of number of deaths being the left-hand side variable. We use a logarithmic transformation, for logarithms allow us to interpret coefficients in percentage terms which favors comparability across areas (countries and states) that are highly heterogeneous. 5 This leads to the following estimating equation: where Deaths are the number of cumulative Covid-caused deaths 20, 40, 60, 80, or 100 days after the first death in area r, and Co-resid is the proportion of 18 to 34 years old individuals living with their parents in that area. X is a vector of variables related to Covid and Z is a vector of demographic and other controls. Vector X includes a dummy equal to one if the government imposed a lockdown and days from first death to lockdown: imposition of a lockdown and the speed at which a lockdown was imposed have been associated with lower death rates from Covid (e.g. Friedson et al 2020, Huber and Langen 2020) . We also control for number of tests per capita 14 days prior to the day deaths were measured and for number of days that elapsed between onset of the epidemic in France and in each country/state. 6 For example, for deaths at 20 days we include tests per capita at 6 days past onset; for deaths at 100 days past onset, we include tests 86 days past onset. Vector Z includes demographic variables, starting with total population in the country/state. Population's effect may be mechanical: more people implies a potential for more deaths. In addition, it is possible that population affects number of deaths because it affects population density. In turn, such density may facilitate the spread of infections such as Covid. It is not just the total population that matters, but also the share of individuals over age 65 who are more likely to die from Covid (Abdulamir and Hafidh, 2020) . We thus add proportion of individuals over 65 as a control. Furthermore, our regressions control for proportion urban: density in large cities may be particularly conducive to Covid infections and fatalities (Florida 2020 Croatia has the highest co-residence rate in Europe (75%); Denmark the lowest (18.8%). In the US co-residence rates range between 14.5% (North Dakota) and 46.25% (New Jersey). From Table 2B ). It can also be seen that 30% of all areas did not have a lockdown, implying that a lockdown was imposed in 70% of the areas. EU countries were less likely to have a lockdown than US states: 62% of EU countries did not have a lockdown, but this was only the case with 12% of US states. On average it took an area 6.4 days from onset to lockdown. We now proceed to the regression analysis allowing us to examine the partial association between fatalities and co-residence, after controlling for many other relevant factors. Main findings. parents is associated with 4 % more deaths after 40 days, 3.5% more deaths after 60 days, 3.2% after 80 days, and 3.1% after 100 days. Starting at 40 days, all associations between deaths and co-residence are highly significant statistically. The magnitude of the estimated associations between the logarithms of deaths at different time intervals and co-residence rates is similar. This may be attributed to the similar distributions of the five logarithmic dependent variables. Averages range between 4.2 to 6.6 and standard deviations grow from 1.5 to 1.9. Also the supports overlap significantly (values go from 0 to 10.6). Moreover, the five dependent variables are very highly correlated: correlations range from 0.802 for log of deaths at 20 and 100 days to almost one for log of deaths at 60 and 80 days, log of deaths at 60 and 100 days, and log of deaths at 80 and 100 days. Next, we assess whether the association between mortality and co-residence is more applicable to the US or to Europe. In particular, we test the hypothesis that the relationship between co-residence and Covid fatalities for US states differs from that for countries associated with the EU. This involves testing whether the coefficient of the interaction of co-residence and the US dummy ( 2 ) equals zero in the following equation: where the coefficient of co-residence ( 1 ) now refers to EU countries only. As shown in Table 4 , 2 takes values between 6 and 11% and is statistically significant in all regressions. In contrast, 1 ranges between 1 and 2% and is not statistically different from zero. Hence, we conclude that US states lead our estimated relationship between co-J o u r n a l P r e -p r o o f residence and COVID fatalities. The total association between co-residence and fatalities in the US is calculated by adding up the two coefficients of co-residence and coresidence*US. Consider deaths at 100 days after onset (col. 5 in Table 4 ). We add up 0.109 and 0.00852 which equals 0.11752, implying that a one extra percentage point in the share of co-residence in the US (let us say from the mean of 37.5% co-residing in the state to 38.5%) is associated with an increase in deaths at 100 days of 11.8 percent. This sum is significantly different from zero and is economically significant. Our conclusions are robust to an alternative estimation strategy in which we run separate regressions for subsamples of US states and EU countries. 12 All results remain invariant if we use population over 65 instead of overall population. In our main specification we accounted for days since onset of the epidemic to lockdown and absence of lockdown. Lockdown implies that the government requires citizens to shelter in place all day long and that they are allowed to come out only to buy essential items. We used these extreme limitations on social activities because with 79 observations we need to limit the number of controls. However, governments changed social distancing rules as the epidemic progressed and measures softer than lockdowns were often introduced. In additional regressions, we included dummies for each separate limitation (no social events, no schools, no shops, partial lockdown and lockdown) being in place at the time that deaths were measured. 13 We also include number of days from the first death to the implementation of each type of limitation. Results in Table 5 show that coefficients of co-residence remain mostly unaltered. Our estimates pass the Oster (2019) test for selection on unobservables. In particular, we obtain a measure of the extent of selection on unobservables relative to 12 We do not include them in the paper since the regressions for EU countries are based on few observations which harms inference. They are available upon request. 13 We classify various degrees of lockdown and limitations following Olivier Lejeune (https://github.com/OlivierLej/Coronavirus_CounterMeasures). selection on observables (delta) equal to 1.55. This value implies that selection on unobservables would have to be 1.55 times larger than selection on observables to explain the observed correlation between intergenerational co-residence and Covid-related deaths. This threshold exceeds the rule-of thumb cutoff of 1 in observational studies. Throughout our analysis we treat states/countries as independent statistical units. However, our variables of interest are characterized by significant spatial autocorrelations (Miron, 1984) . We present some evidence related to these autocorrelations. We first calculated average COVID fatalities and average co-residence rates in neighboring states/countries (those countries/states that share a border) for each country/state. For instance, to the state of Kansas we assigned the average values for the states of Colorado, Nebraska, Missouri, and Oklahoma. We then computed correlations between COVID fatalities in a particular area and average COVID fatalities in neighboring states/countries as well as between between co-residence rate in a particular area and average co-residence rate in neighboring countries/states. Our sample size is reduced to 77 because Alaska and Hawaii do not have neighboring states. All the correlations are high: those between fatality rates are between 0.38 and 0.54, depending on the days since onset, and that between co-residence rates is 0.88. The spatial correlations may be due to a significant amount of individuals commuting across borders or to the fact that areas that are geographically close tend to share cultural and institutional traits. These high correlations imply that the joint analysis of deaths and co-residence in a state/country and in its neighboring states/countries is problematic. 14 To circumvent this problem we performed regressions substituting an area's co-residence rate by its neighbors' average co-residence rates and found that the association between deaths and neighboring co-residence rates ranges from 4% to 5.6%, depending on the time interval at which deaths were measured. 14 We thank an anonymous referee for pointing this out. These numbers are not substantially different from the estimates reported in Table 3 . We interpret this as additional evidence that co-residence is strongly related to COVID deaths, for average co-residence rates of neighboring states/countries constitute a good proxy for actual co-residence rates. Results are available from the authors upon request. Discussion of the main findings: Our cross-country and cross-state analysis confirms a finding first reported by Bayer and Kuhn (2020) based on data for 24, mostly European, countries: more people die from Covid where intergenerational co-residence is more common. Our finding is based on regressions controlling for multiple factors that may influence fatalities (their only control variable is an East Asia dummy) and on a larger sample of 79 areas, of which 29 are European countries associated with the UE and 50 are US states. While we find that the association between Covid deaths and intergenerational co-residence across European countries is positive, we find that it is weak and not significant statistically, which is consistent with analyses by Belloc et al (2020) and Arpino et al (2020) at either the cross-country level, the European regional level or based on an analysis of Italian regions. A surprising result is that the positive association between intergenerational coresidence and Covid fatalities is considerably larger in the US than in Europe, and in contrast to our results for European countries, it is statistically significant. This was already apparent from simple correlations presented graphically in Figures 1B and 1C . After controlling for many factors likely to influence fatalities from Covid our regression results indicate that across US states cumulative deaths vary positively with share of individuals age 18-34 living with their parents, while this not the case across EU countries. For example, in the US a one extra percentage point in the share of co-residence is J o u r n a l P r e -p r o o f associated with an 11.8 percent increase in cumulative deaths from Covid 100 days after the onset of the pandemic in that state As for the meaning of this positive association between fatalities and intergenerational co-residence, we are reluctant to jump to conclusions about causality based on this limited evidence. A possible interpretation is that when adult children live with their parents this exposes more vulnerable populations to a dangerous virus, but it is certainly premature to conclude that some deaths could be avoided if older adults and their adult children live in separate households rather than share the same residence. It is possible that more people die from Covid when more young adults live with their parents because this raises the likelihood that a household is exposed to Covid (see Harris 2020b for some possible evidence about this). If so, we should find that this type of co-residence is also associated with more Covid cases. In Table 6 we present regressions of cumulative cases at the same intervals used in Tables 3-5: 20, 40, 60, 80 and 100 days after onset in a particular state or country. The magnitude of the coefficients shows that one additional percentage point in the share of co-residents is associated with a 4.6% increase in the number of cumulative cases 40 days after onset, a 3.5% increase at 60 days past onset, and a 3% increase at 80 and 100 days after onset. Next, we list a number of alternative interpretations of the positive deaths/coresidence association. First, the association could indicate an income effect not captured by the inclusion of area-level Gross Product and average rent in the regressions. Other income, wealth, or cost-of-living components associated with intergenerational co-residence could simultaneously affect fatalities and co-residence. Lovett et al (2020) noted the association between multigenerational co-residence and high fatality rates in poverty-stricken parts of the US such as Indian reservations. That many of the US states with above-average co-J o u r n a l P r e -p r o o f residence also have above-average state income (as in the case of New Jersey and Connecticut) suggests that poverty is not a major factor here, but the possible association between co-residence and poverty is worthy of further investigation based on more detailed individual or regional data. Second, intergenerational co-residence may reflect a more widespread presence of active religious communities encouraging physical contacts conducive to the spread of Other components of such intergenerational relationships include geographic proximity, contact frequency and provision of grandchild care. Future research may want to estimate how geographic variations in Covid fatalities also vary with these other dimensions of intergenerational relationships. Fifth, intergenerational co-residence may indicate psychological dispositions such as perceived sociability, as suggested by Oksanen et al. (2020) . They showed that Covid mortality was significantly positively associated with such sociability. It could be that countries with cultures that encourage sociability also have higher rates of intergenerational co-residence. This interpretation is reinforced by Albertini et al.'s (2020) simulation experiments suggesting that high intergenerational connectedness alone is not sufficient to rapidly contaminate a large fraction of the elderly, and thus cause high fatalities. Only when they introduce social connectedness among the elderly do they find that a virus infects large fractions of the elderly. Other findings: The regression results reported in Table 3 confirm that extent of testing helps explain variation in deaths from Covid, in line with previous research e.g. by Sung and Kaplan (2020) and Terrieau et al (2020). We find that a 1 percentage point increase in tests per capita 6 days after the epidemic's onset is associated with a 253% reduction in fatalities after 20 days. Associations between tests per capita and fatalities become statistically insignificant starting at 40 days past onset. Associations between our measures of lockdown and fatalities are mostly statistically insignificant. As for hospital beds, it also shows a large negative association with deaths at 20 days past onset: the magnitude of the coefficient implies that one more bed per 1,000 inhabitants is associated with 36% fewer Covid deaths 20 days after onset. However, number of beds per capita does not seem to matter much at 40 days or later. We also include days it took for the epidemic to reach a country or state after the epidemic's onset in France on February 15 (the variable is called Days Post France in Table 3 ) and find that once an area has passed 60 days since onset, the more time elapsed between a state or country's epidemic and the onset in France, the smaller the number of fatalities. At 80 days or 100 days past onset cumulative fatalities are 7.9% lower for every extra day between the onset in a particular area and the onset in France. This possibly indicates that countries or states that started their COVID pandemic later may have learned considerably from the experience of countries who got hit by the pandemic at an earlier date. For example, they could have adopted better techniques and strategies in taking care of patients, as suggested by Landoni et al (2020) . Deaths are higher in more urban states or countries 40 days or longer after the local onset of the epidemic. For example, at 40 days after the first death one extra percentage point in percent urban is associated with 2.8% more cumulative deaths from Covid. As the pandemic progresses urban areas suffer slightly more: 100 days after onset cumulative deaths are 3% higher for every point increase in percent urban. Not surprisingly, the larger its population the more deaths are recorded in a country or state. As for Gross Domestic Product or Gross State Product, it is associated positively with cumulative deaths for deaths at 20 days (perhaps because people travel more and have more visitors). We find that intergenerational co-residence is associated with a higher number of deaths from Covid using a sample of 29 European countries and 50 US states. This finding is applicable to cross-state comparisons in the US more than to cross-country comparisons in the EU. If the association is causal, it implies that reductions in such co-residential arrangements may protect an area's inhabitants from dying of Covid-19. However, we are not able to establish such causality. Further research is needed that will include better statistics on deaths from Covid, a longer period for the measurement of cumulative deaths, and more countries. It would also be useful to further explore the question we address using more detailed data, such as US counties, European provinces, or other sub-national data. Days post France** Same source as above Same source as above 2020 Same as above DEMOGRAPHICS J o u r n a l P r e -p r o o f Notes: Data sources: see Table 1 . Deaths20, Deaths30 etc refer to cumulative deaths 20, 30, etc days after the first death in the country/state. Co-residence using the share of 18-34 year-olds living with their parents. "Days post France" is the number of days between first death in France and first death in country/state. Days to lockdown reflects the number of days from the first COVID-related death to lockdown. It is missing for those countries/states without lockdown. Population is expressed in millions of citizens. J o u r n a l P r e -p r o o f J o u r n a l P r e -p r o o f The Possible Immunological Pathways for the Variable Immunopathogenesis of COVID--19 Infections among Healthy Adults, Elderly and Children Intergenerational contacts and Covid-19 spread: Omnipresent grannies or bowling together? Are intergenerational relationships responsible for more COVID-19 cases? A cautionary tale of available empirical evidence Federalism makes the US unique in its pandemic response Intergenerational Ties and Case Fatality Rates: A Cross-Country Analysis Cross-country correlation analysis for research on COVID-19 Reliable, real-world data on excess mortality are required to assess the impact of Covid-19 The Geography of Coronavirus. Bloomberg City Lab Did California's Shelter-in-Place Order Work? Early Coronavirus-Related Public Health Benefits. NBER Working Paper w26992 Reopening Under COVID-19: What to Watch For Data From the COVID-19 Epidemic in Florida Suggest That Younger Cohorts Have Been Transmitting Their Infections to Less Socially Mobile Older Adults The Impact of Response Measures on COVID-19-Related Hospitalization and Death Rates in Germany and Switzerland Occupational risks for COVID-19 infection Spreading the Disease: The Role of Culture Is time our ultimate ally in defying the pandemic? Pathogens and Global Health Covid-19 Stalks Large Families in Rural America Spatial autocorrelation in regression analysis: A beginner's guide Regulation and Trust: 3-Month Follow-up Study on COVID-19 Mortality in 25 European Countries Unobservable Selection and Coefficient Stability: Theory and Evidence Why does COVID-19 case fatality rate vary among countries? medRxiv preprint doi Why Do Countries' COVID-19 Death Rates Vary So Much? Medpage Today Impact of virus testing on COVID-19 case fatality rate: estimate using a fixed-effects model Estimating clinical severity of COVID-19 from the transmission dynamics in Wuhan, China The deadly coronaviruses: The 2003 SARS pandemic and the 2020 novel coronavirus epidemic in China