Wednesday, June 10, 2020

Cross-country variation in Covid-19 deaths

With Surjit S Bhalla

Covid19 Deaths: Variation Across Countries

Existing literature on COVID deaths has focused mainly on identifying government measures which may have been successful in slowing the fatality rate or cumulative death rates. In this note we analyze exogenous factors which may explain divergences in the rate of mortality (deaths) from COVID.
Our analysis on deaths uses the same underlying model/framework as used in the estimation of COVID cases. Several contemporary and historical studies of virus diffusion have documented that the spread follows an S-shaped pattern. We are not aware of any studies showing that the number of deaths very likely also follows an S pattern – but we are not epidemiologists so we very likely have missed the studies – there must be some.

Framework and Approach

The prime exogenous determinants of COVID cases (Blog 4) were found to be urbanization, the share of elderly male population, and average temperature that each country had experienced during the last five months. Because of the imperatives of an S-shaped pattern, and in order to account for this influence, we had estimated, and reported, cross-country models as of day 40, 60, 80 and day 100 of virus diffusion in each country. We use the same model to explain cross country Corona deaths with the difference that the model is estimated for different death days, rather than different infection days.  The cross-sections are for day 30 through day 70 of deaths, in increments of 20 days. For completeness, we also add log population as a determinant of log deaths; as expected, its coefficient gets close to 1 for longer period samples (for 70 days, the coefficient is 0.94). The presence/absence of log population does not make any difference to the results, but it is econometrically correct to include it.
Death rates depend on the both the number of cases and on the variable that directly affect fatality of the disease, such as aged population and the quality of health infrastructure. As equations are estimated in reduced form, the difference in the number of deaths across countries will also be determined by the variables which affect the spread of the virus, even if they do so indirectly (via the influence on infections). The share of urban population was found to be an important determinant of case transmission (possibly operating as a social distancing proxy). It may however also be correlated with direct determination of death rates i.e. urbanization may be a proxy for better health care facilities. Analogously, higher temperature was found to slow the spread of virus infections, but higher temperatures may also make the patient less vulnerable to serious illness.


Table 1 (and Chart 1) reports our experiments with modelling (log) COVID death rates. Separate regressions are reported for day 30 thru day 80, in increments of 10 days; as for the case of infections, the separate day regressions are estimated to capture the likely different stages of the S-shaped (logistic) curve of diffusion. The first noteworthy result: all three determinants (age, temperature and urbanization) remain statistically significant – more importantly, their magnitude stays broadly the same. This is very encouraging because it suggests that our model specification is robust. 
The second important result: effect of aged population on deaths is much more pronounced than was the case with infections.  In the case of infections, we had found that the share of male population greater than 60 yielded worked the best in terms of statistical significance – in the case of deaths, the effect is the highest for males over the age of 80.  These results are consistent with limited data available for COVID death rates by age from hospital data in different countries, which has been partially attributed to co-morbidity and other pre-existing conditions (like diabetes and heart problems). Though conjectured, we are unaware of any empirical study documenting the vulnerability of deaths to differences in the share of aged population
The third result is that temperature variable is highly significant. This result parallels the effect of temperature on corona virus infections, but is even stronger, suggesting that higher temperature may have both direct and indirect effects on reducing death rates. 

Table 1:     Determinants of  of COVID19 deaths across countries

Dependent Variable: Log of Covid19 deaths

Days since the first day COVID19 death was observed


Log population


% population >= age80


% urban


Average temperature


(log) # Hospital Beds




# observations
adj. R-sq

t statistics in parentheses

="* p<0.05
 ** p<0.01

Sources: COVID19 data, JHU, World Bank, UN

The effect of share of urban population on deaths is also significant (Table 1). The magnitude of the effect is much less than was the case with the number of infections. This suggests that the correlation between urbanization and quality of health care is high, and that the quality of health care in urban areas may be directly helping reduce the incidence of death rates. We test for this possibility below.  If urban hospitals have better equipment and quality of health care than rural ones, as is true in most countries, then urbanization effect would be lower and could eventually turn negative (capturing transmission via lack of social distancing)   
The number of deaths (and infections) will obviously be higher the larger the population, ceteris paribus. Curiously, this result is the least noticed in the media (and even in some scholarly articles). The most common line in the media (and twitter and podcasts) is that the number of infections or deaths recorded a new high yesterday! But each new local high will mean (after the tide or curve has turned) less and less percent increase, and eventually a flat, or zero increase.
The effect of population on number of deaths has steadily increased over time from 0.57 on day 30 to reach 0.94 on day 70. In other words, if a country has 10 % higher population than another country, it will have ( 0.94*10) 9.4 % more deaths, again, after accounting for all other influences.
Quality of Health Services and Effect on deaths   We also analyzed the effect of quality of health services on the corona virus death rate across countries, by using the log magnitude of hospital beds as an indicator. This variable is found to be significant at the 5% level (last column of Table 1).  Its magnitude is -0.622 i.e. each 10 % increase in hospital beds would lead to, given average effects, a 6.2 % decline in COVID deaths.
And now for a picture of comparative performance in the war against COVID
Chart 1 documents the overall prediction vs reality on day 70 (again, day 70 is not the same date for different countries but different dates reflecting day 70 of the observed death rates within the country).  The most prominent outlier is USA with a higher than predicted number of Covid related deaths. Other countries with higher than predicted deaths are Great Britain, Italy, Spain and Brazil. On the other side China, Japan, Taiwan, and Hong Kong (among many others) stand out as countries with less than predicted number of deaths. 
Missing from the picture is Viet Nam – it is missing because it has zero recorded deaths to date (number of infections only 329). Cambodia also has zero deaths, and only 125 infections. Laos even fewer infections (19) and zero deaths. Myanmar 240 cases and 6 deaths; Costa Rica 12263 cases and 10 deaths.  Earlier, we had mentioned how population size can affect interpretations of performance.  COVID history of New Zealand and Australia is revealing. New Zealand with a 4.8 m population has only 22 deaths; Australia has a population of 25.3 million and only 103 deaths i.e. a population 5.3 times as large, with deaths only 4.7 times higher.
Chart 1: Actual and Model Predicted Corona Death Rates


Sadhana said...

The great thing about this post is quality information. I always like to read amazingly useful and quality content. Your article is amazing, thank you for sharing this article.
Bambino Agro Industries Limited
Bampsl Securities Limited
Banaras Beads Limited
Banas Finance Limited

Niveshi said...

The great thing about this post is quality information. I always like to read amazingly useful and quality content. Your article is amazing, thank you for sharing this article.
Bannari Amman Sugars Limited
Bansal Roofing Products Limited
Banswara Syntex Limited
Barak Valley Cements Limited

sudalai said...

synmac tax consultant in Delhi
doing company registration in Delhi,GST registration in Delhi,Tax consultant in Delhi...
if you have any requirement GST,ITR
click the below link...

tax consultant in Delhi
company register in Delhi
GST register in Delhi
tax consultant in Delhi

seeniwasan said...

Hi there, I read your blogs on a regular basis. Your humoristic style is witty, keep it up! Thank You for Providing Such a Unique and valuable information, If you are looking for the best covid ready hospitals in mogappair

covid ready hospitals in mogappair

Critical Care Hospital in Chennai