The climate crisis has not gone away

The return of the summer wildfires in California reminds us that the climate crisis has not gone away just because there is a pandemic. The USA has recently recorded the hottest ever measured temperature of 54.1 C (130 F) in Death Valley and 2020 is on track to be the hottest year on record for global average temperatures. A recent paper (Slater et al 2020) estimated total ice loss over the last 25 years at 28 trillion tonnes, which matches the worst-case-scenario predictions of the Intergovernmental Panel on Climate Change (IPCC).

And yet a substantial proportion of people in many countries deny the science, not only for human-caused global warming, but for coronavirus and vaccines. A recently published reconstruction of the earth’s global climate over the last 12,000 years illustrates starkly how different the rising global temperatures of the last decades have been from natural cycles. The figure below was prepared by Alexander Radtke (@alxrdk) with captions added by using temperature data for the 12k->2k years ago from, for the 2k->170 years ago data: and for 1850-2020 it is an equal blend of HadCRUT 4, GISTEMP, Berkeley Earth, Cowtan and Way and NOAA datasets.

Climate change denialists like to argue that climate changes naturally. It does, but the graph makes clear how different the rate of change has been since the mid-20th century, and by hom much it exceeds the limits of natural climate change. The climate fluctuates naturally in a roughly 10,000 year cycle and its clear from the graph that we should still be in a cold period, not the warmest by far in the last 12,000 years.

Posted in Global health trends, Uncategorized | Tagged , , , , , , , , | Leave a comment

Trends in cultural values across 105 countries: 1980 to 2020

In a previous post, I examined variations in cultural values across 105 countries in year 2019 using data from all 7 waves of the World Values Survey (WVS)and all 5 waves of the European Values Survey (EVS) [1-4]. I developed an improved method for estimating latent variables f1 and f2 for two dimensions of human values: a traditional values (parents,state, religion) versus secular values dimension and a survival values versus emancipative values dimension. These two dimensions were defined for earlier waves of the WVS and EVS by Inglehart and Welzel using factor analysis.

The regression procedure described in the previous post (for projections to 2019 for all countries) was also used to interpolate a complete time series for each country from 1980 to 2020. For some countries where f1 and f2 varied substantially across waves, the time trends were not smoothly varying. I thus applied a Loess regression with bandwidth 0.5 to smooth trajectories to a moderate degree. Using 10 “culture zones” defined by Welzel to group countries, I computed population-weighted averages of f1 and f2 for the culture zones and these are shown in the Figure below.

Trends in cultural values from 1980 to 2020 for 10 culture zones (see below).

The dots on the trajectories denote values for years 1980, 1990, 2000, 2010, 2020. For all trajectories, except for the Islamic East, the leftmost dot denotes 1980. For the Islamic East, the lower of the two rightmost dots is for 1980 and the upper for 2020. The culture zones are defined as follows:

Reformed West — Western European societies strongly affected by the Reformation;
New West — English-speaking countries (UK, Ireland and former overseas colonies);
Old West — Mostly Catholic parts of Western Europe being core parts of the
Roman Empire;
Returned West — Catholic and Protestant parts of post-communist Europe returning
to the EU;
Orthodox East — Christian Orthodox or Islamic parts of the post-communist world,
mostly parts of former USSR;
Indic East — Parts of South and South East Asia under the historic influence
of Indian culture;
Islamic East — Regions of the Islamic world that have been parts of the Arab/Caliphate,
Persian and Ottoman empires;
Sinic East — Parts of East Asia under the historic influence of Chinese culture;
Latin America — Central and South America and the Caribbean;
Sub-Saharan Africa — African countries south of the Sahara.

Some of the kinks and twists in these trajectories relate to variation in f1 and f2 across waves for some individual countries.  The following graphs illustrate the cross-wave trends for selected countries.

Trends in values of f1 and f2 across waves (indicated by the dots) for selected countries, 1980-2020.

There are some countries where values are reasonably smoothly evolving apart from a single wave (eg. Hungary) and others where value vary substantially up and down across waves (eg. China, Turkey).  It is possible some of these represent real changes in cultural values over time, but also likely that in many cases it reflects variations in survey quality and representativeness. To get a better idea of the extent of this issue, I summarized the net deviations of intermediate waves values of f1, f2 from the values obtained by carrying out the covariate regression described in the previous post and adjusting its predictions so they ran through the first and last wave values for f1 and f2.

Net deviations were calculated in f1-f2 space as the distance between the survey wave values and the regression-based values. The root mean square deviation δ for all intermediate waves was calculated by country. The following table shows the percent of countries within each culture zone with δ > 0.12 (a value one standard deviation above the mean δ for all countries).

It is clear that there are much larger deviations from smoothly evolving time trajectories in developing regions (except the Sinic East) than in high income regions with the exception of the Old West. It seems to me highly probable that most of the larger deviations across waves are likely due to variations in survey implementation, sampling and quality. Various factors such as lack of good population data to establish sampling frames, language translation issues, interviewer training, difficulties in finding funding, and security issues in some countries, can all contribute to systematic differences across surveys.

The following graph bypasses these issues of comparability across waves to show the net trend in culture values from 1980 to 2020 as straight lines joining those two points.

Net trends in vulture variables f1 and f2 from 1980 to 2020 for 10 culture zones.

This graphs shows a very clear contrast between the evolution of cultural values for the West plus Latin America and the other culture zones. The West regions and Latin America have all moved quite strongly towards more emancipative values and also away from traditional values to more secular-rational values. In contrast, while the other regions have also moved somewhat rightward in emancipative values, they have moved downwards away from secular-rational values towards to more traditional values. The Islamic East is the major exception with very little change in either dimension.

In his 1996 book, The Clash of Civilizations and the Remaking of the World Order, Samuel Huntington described a global revival of religion in the second half of the twentieth century (La Revenche de Dieu) claiming that the trend towards secularization went into reverse in the 1970s in every region of the world [5]. At least to the extent the second latent dimension (or indeed Inglehart and Welzel’s traditional-secular factor) measure the degree of secularization, the trends from 1980 to the present do not fit with his conclusion. He is correct in identifying a return to religion in the former Soviet countries with predominantly Orthodox Christian or Islamic religious tradition. He also points to an increasingly Hindu orientation of India, whereas the story from the WVS-EVS data is somewhat more complex with a fairly stable level of religiosity (f2) from 1990 to 2000 associated with an increasing level of emancipative values, but from 2000 to 2020 an increasing degree of religiosity associated with a declining level of emancipative values. This correlates broadly with the rise of Hindu extremism and the election of the BJP Party with its Hindu nationalist orientation in 2014.

As a result of this initial analysis, I plan to revisit the conceptualization of latent culture variables in order to develop clearer operationalizations of important culture dimensions such as religiosity/secularism that make better use of more of the data collected in the WVS. I will also look at developing better methods of extrapolating and interpolating time series for countries with considerable deviations across waves, either by leaving out a single deviant wave, or perhaps by more broadly averaging across the waves. And perhaps, to revisit the grouping of countries into culture zones.


  1. Inglehart, R., C. Haerpfer, A. Moreno, C. Welzel, K. Kizilova, J. Diez-Medrano, M. Lagos, P. Norris, E. Ponarin & B. Puranen et al. (eds.). 2014. World Values Survey: All Rounds – Country-Pooled Datafile Version: Madrid: JD Systems Institute.
  2. Haerpfer, C., Inglehart, R., Moreno,A., Welzel,C., Kizilova,K., Diez-MedranoJ., M. Lagos, P. Norris, E. Ponarin & B. Puranen et al. (eds.). 2020. World Values Survey: Round Seven–Country-Pooled Datafile. Madrid, Spain & Vienna, Austria: JD Systems Institute& WVSA Secretariat[Version:].
  3. Gedeshi, Ilir, Zulehner, Paul M., Rotman, David, Titarenko, Larissa, Billiet, Jaak, Dobbelaere, Karel, Kerkhofs, Jan. (2020). European Values Study Longitudinal Data File 1981-2008 (EVS 1981-2008). GESIS Datenarchiv, Köln. ZA4804 Datenfile Version 3.1.0,
  4. EVS (2020): European Values Study 2017: Integrated Dataset (EVS 2017). GESIS Data Archive, Cologne. ZA7500 Data file Version 3.0.0,doi:10.4232/1.13511
  5. Samuel Huntington. The Clash of Civilizations and the Remaking of the World Order. London: Simon and Schuster 1996.


Posted in Uncategorized | Tagged , , , , , , , | 4 Comments

An update on coronavirus in Geneva and Victoria

Switzerland currently bans entry from countries with more than 60 new infections per 100,000 population in the last 14 days. Geneva passed that threshold on 27th July, and passed 100 new cases per 100,000 on 1 August. So if Geneva was a separate country, residents would be forced to quarantine upon entering the rest of Switzerland. Or if they were not considered to be Swiss citizens, they would not be able to enter at all. Belgium has banned travel to and from Geneva, along with Valais and Vaud, as a result of the rising infections.

The overall new infection rate for the last 14 days is 18 per 100,000 for the rest of Switzerland, excluding Geneva, and 23 per 100,000 for Switzerland as a whole (still well below the Swiss government’s high risk threshold). Why is it so much higher in Geneva?  As a Swiss epidemiologist explained: “population density, airport and border with France”. Geneva has the second highest population density after Basel, the second-busiest airport, and many cross-border workers (325,000 crossed the border into Geneva each day in 2019). However, the increase in cases in Geneva has as only reached 20% of the peak in the first wave.  With any luck, it will not reach the level of the first peak as much more has been learnt about how to use social distancing and masks to reduce the effective reproduction rate of the virus since then.

The state of Victoria in Australia is also experiencing a second wave of infections, much stronger than the first wave. As of Tuesday 4 August, Victoria experienced a total of 6101 new infections in the last 14 days, giving it a rate of 94 per 100,000 population, almost as high as Geneva. The new infection rate for the rest of Australia is very low at 2.4 per 100,000 in the last 14 days.

Posted in Global health trends | Tagged , , , , , , | Leave a comment

Variations in cultural values across 105 countries in 2019

Working for the World Health Organization over the last 20 years and dealing with national health authorities, I’ve been quite fascinated by the variations across cultures and countries in values and beliefs, and the huge influence they have on willingness to accept or implement evidence-based public health interventions. More broadly, it seems fairly clear that people’s beliefs play a key role in economic development, the emergence and flourishing of democratic institutions, the rise of gender equality, and the extent to which societies have effective government.

So I was particularly interested to learn recently that data from the most recent wave of the World Values Survey was being released in July. This wave has been interrupted by the coronavirus pandemic, but the data for 48 countries whose survey were completed by early 2020 has recently been released on the WVS website.

Data from previous waves of the World Values Survey were used by political scientists Ronald Inglehart and Christian Welzel to identify two major dimensions of cross cultural variation across the world. They refer to these as Traditional values versus Secular-rational values and Survival values versus Self-expression values. The map below uses data from the 6th wave of the WVS carried out during 2010-2014 to plot countries according to their factor values for these two dimension. Moving upwards on this map indicates increasing secular values and moving to the right increasing self-expression values.

Inglehart-Welzel Culture Map 2010-2014. Source:

Traditional values emphasize the importance of religion, parent-child ties, deference to authority and traditional family values. Secular-rational values place less emphasis on religion, traditional family values and authority.

Survival values place emphasis on economic and physical security. It is linked with a relatively ethnocentric outlook and low levels of trust and tolerance. Self-expression values are associated with gender equality, relative acceptance of divorce, abortion, and homosexuality and rising demands for participation in decision-making in economic and political life.

Inglehart and Welzel have used factor analysis to extract these two dimensions from WVS data for 12 questions. The final factors are rotated, and all 12 questions contribute to the factor scores, which are correlated. I was interested to examine their approach more closely and attempt to use it, or a related approach, for an analysis of all the WVS data, including the recent release.

There have been seven waves of the World Values Survey, the first in 1980-1982 and the seventh underway since 2017. Additionally, the recent 2017 wave of the European Values Survey includes many of the same items as the WVS and has data for 30 countries. Together the recent wave of the WVS+EVS include data for 77 countries.

I downloaded all unit record data for the WVS from their website [1-4]. In total, the World Values Surveys plus the European Values Surveys now includes data for 117 countries or territories and over 638,000 respondents.  The dates of the waves, and the number of respondents by wave and survey type are shown in the table below.

I first set out to replicate the Inglehart-Welzel analysis of the two cultural dimensions based on data for 12 questions. I immediately ran into problems as the various documentation on the website has several differing lists of variables included in the factor analysis, and trying all three, none replicated the values for the factors in the data files. And indeed, there were at least three different sets of factor values in various places. I think what has happened is that the Welzel analysis and summary indicators have evolved over time.

I did my best to match his description of the analysis of waves 1-6 using factor analysis, and my resulting factor values for the wave 6 countries correlate highly with Welzel’s secular values index (svi) and emancipative values index (evi) with correlation coefficients of 0.92 and 0.97 respectively. See scatterplots below. So my factor analysis gives somewhat different results but very highly correlated with Welzel values. Its quite possible that the factor procedure I am using in stata does not exactly replicate the factor analysis or scoring produced by Welzel using SPSS.

In his recent book, Freedom Rising [6], Welzel mostly abandons factor analysis in favour of explicitly constructed indicators based on conceptual alignment of variables with the target latent construct. However, as with the factor analysis, he is assigning equally spaced scores between 0 (lowest) and 1 (highest) for categorical response categories to questions in the WVS/EVS. So for example,

How important is religion in your life?       Questionnaire Responses     Welzel rescale
Very important                     0
Rather important                  0.333
Not very important               0.667
Not at all important              1

When jobs scarce: men should have more right to job

Agree                                      0
Neither                                   0
Disagree                                 1

This a priori forces the factor analysis to treat all the categories for each question as equally spaced on the underlying factor even though responses for different questions may have very different means and be skewed across the response categories quite differently. In other words, the end categories for questions may map to very different values on an underlying latent construct, but are all forced to values 0 and 1.

A recent paper has highlighted the dangers of using factor analysis on ordered-categorical survey data (e.g., Likert items). The authors conclude that the common practice of factor analysing ordered-categorical data such as Likert items leads to very high risks of incorrect and misleading diagnoses of the latent structure of the items, with dire consequences for conceptualisation, replication and comparability, and evidence-based interventions, behaviour and policy [5]. They advise analysts to investigate latent structure with IRT models instead.  When the latent trait is continuous, but the data is a set of ordered categorical responses, the appropriate approach is to use Item Response Theory (IRT) models; this is the most common term in applications in educational or psychological testing, where these models are very widely used.

I have implemented a form of IRT analysis using the stata procedure gsem, for structural equation modelling, with ordered probit regression. For f1, the first latent variable (survival-emancipative dimension), I used data for three questions of gender equality (jobs, politics, education) and three questions on acceptance of homosexuality, abortion, and divorce. I estimated the second latent variable (f2 — traditional-secular dimension) using data for three questions on sources of authority (nation pride, government, parents) and three questions on religion (importance, belief, practice

Country-wave average values for f1 and f2 were computed taking sampling weights into account. Because the year of survey varied within waves, and also not all countries were in each wave, I used a projection procedure to estimate culture values for year 2019 by projecting forward from the most recent country-year available since 2005. Values were regressed on time and the following covariates: GDP per capita (PPP dollars), average years of education (adults 25+), general fertility index, UNDP gender inequity index and regional indicator variables, with random effects at country level. The following map shows the resulting estimated cultural latent variables f1 and f2 for 2019.

Because there are 105 countries displayed on this map, it becomes harder to draw nice coherent cultural groupings as in the Inglehart-Welzel map shown before. I’ve also adopted and adapted the 10 cultural groupings used by Welzel in his more recent book [6], into 8 groupings as follows:

New West and West —  Western Europe and overseas offshoots of Western Europe

Returned West —  Catholic and Protestant parts of post-communist Europe  returning to the EU

Orthodox —  Christian Orthodox or Islamic parts of the post-communist world, mostly parts of former USSR

South Asia —  Parts of South Asia under the historic influence of Indian culture

South East Asia —  Parts of South East Asia excluding those under historic influence of Chinese culture

Sinic East —  Parts of East Asia under the historic influence of Chinese culture

Latin America —  Central and South America and the Caribbean

African-Islamic —  African countries south of the Sahara, together with regions of the Islamic world that have been parts of the Arab/Caliphate, Persian and Ottoman empires

The general topology of this map is similar to the Inglehart-Welzel map, with Scandinavian countries to the top right, Sinic countries to the top left, and African-Islamic countries to the bottom left. However, there are some considerable differences in the locations of countries relative to each other, and the positions of some individual countries – no doubt reflecting the difference between arbitrary metrization of Likert scales versus estimation of latent values using IRT methods.

I have also made a more recent post which examines regional and country-specific trajectories in cultural values over the forty-year period 1980 to 2020.  A more interesting issue, and one that I may spend some time on, is to develop a better set of parsimonious values latent variables from the rich data of the World and European Values Survey.  At a time when democratic values, free speech and human rights seem to be under increasing attack in all regions, it would be nice to develop better tools to describe how these are valued across the world.


  1. Inglehart, R., C. Haerpfer, A. Moreno, C. Welzel, K. Kizilova, J. Diez-Medrano, M. Lagos, P. Norris, E. Ponarin & B. Puranen et al. (eds.). 2014. World Values Survey: All Rounds – Country-Pooled Datafile Version: Madrid: JD Systems Institute.
  2. Haerpfer, C., Inglehart, R., Moreno,A., Welzel,C., Kizilova,K., Diez-MedranoJ., M. Lagos, P. Norris, E. Ponarin & B. Puranen et al. (eds.). 2020. World Values Survey: Round Seven–Country-Pooled Datafile. Madrid, Spain & Vienna, Austria: JD Systems Institute& WVSA Secretariat[Version:].
  3. Gedeshi, Ilir, Zulehner, Paul M., Rotman, David, Titarenko, Larissa, Billiet, Jaak, Dobbelaere, Karel, Kerkhofs, Jan. (2020). European Values Study Longitudinal Data File 1981-2008 (EVS 1981-2008). GESIS Datenarchiv, Köln. ZA4804 Datenfile Version 3.1.0,
  4. EVS (2020): European Values Study 2017: Integrated Dataset (EVS 2017). GESIS Data Archive, Cologne. ZA7500 Data file Version 3.0.0,doi:10.4232/1.13511
  5. van der Eijk C, Rose J (2015) Risky Business: Factor Analysis of Survey Data – Assessing the Probability of Incorrect Dimensionalisation. PLOS ONE 10(3): e0118900.
  6. Welzel C. Freedom Rising. Human Empowerment and the. Quest for Emancipation. 2013. Cambridge: Cambridge University Press.
Posted in Global health trends, Projections | Tagged , , , , , , , | 4 Comments

Drug overdose deaths – another exceptional US epidemic

Earlier this week, CDC released provisional figures for drug overdose deaths in the USA in 2019. They estimate almost 72,000 deaths, the highest annual number yet seen. Initially driven by prescription opioid painkillers, users migrated first to heroin and then to fentanyl, which is cheaper (much of it illicitly made). Synthetic opioids accounted for an estimated 36,500 deaths.

The WHO Global Health Estimates  included estimates of direct deaths due to drug use disorders, including accidental poisoning by drugs and overdose deaths, as well as deaths coded to drug dependence (see Technical Note for methods and data).   Last year I updated these to year 2017 using the latest death registration data from the WHO Mortality Database, as well as recent reports for selected countries, and estimates of drug deaths from the IHME GBD 2017. It was apparent at the time that the USA was a massive outlier globally, responsible for a disproportionate proportion of global overdose deaths. I’ve increasingly noticed that the USA is an outlier from other developed countries for a number of health and social indicators, most recently the coronavirus pandemic, police killings, gun homicides and incarceration rates. I was interested to see how the USA compared with other countries for drug-related deaths, given its enormous investment in and the human cost of the “war on drugs”. I’ve done a quick update as described below, but I am fairly sure it gives a reasonably good general picture, not too different from the earlier 2017 update for WHO or the IHME GBD 2017 estimates.

I estimate that globally, drug use disorders caused close to 181,000 deaths in 2019, of which 40% were in the USA. The global average death rate for drug use disorders was 15 per million population, compared with 219 per million for the USA. In other words, the USA rate is 15 times higher than the global average.

I have projected recent trends forward to 2019 and updated drug death estimates for the USA, Canada, Australia and China based on recent reports. The analysis of death registration data is complicated by the need to distribute the accidental poisoning category for “other and unspecified chemicals and noxious substances” (X49) to the specific categories for alcohol and drug use disorders (opioids, cocaine, amphetamines, cannabis and “other drugs”) and to accidental poisoning (non-drug and non-alcohol). Additionally, there is a category F19 in the mental health chapter for “multiple drug use and unspecified drug use disorders” which is used to code deaths in some countries and also must be redistributed appropriately. So there is some uncertainty associated with the estimates for countries with death registration data, and even more for countries without good data. The GHE includes cannabis use disorders as a category, but I have not shown it in the graphs here as there are no overdose deaths.

In terms of the overall drug disorder death rate in 2019, the four leading countries were the USA (219 per million), Canada (121), Australia (71) and Russia (63). Opioids are responsible for many more deaths than other types of drugs.  The following graph shows the drug use death rates for the 15 leading countries in 2019, with the contributions of opioids, cocaine, amphetamines and other drugs.

The following graph shows time trends for opioid death rates in selected countries from 2000 to 2019.

The USA and Canada stand out as having substantially increasing opioid overdose death rates over the last decade. In fact, the opioid death rate for the USA has been growing exponentially since the 1980s until 2017, when there was a 5% drop for 2018, followed by a rise to slightly higher than the 2017 peak in 2019.  Its still too early to say whether the US drug death rates are stabilizing or will start to fall. A recent paper by Jalal et al (Science 2018) shows that the US drug overdose death rate fits an exponential curve from 1980 to 2016 resulting from multiple subepidemics with changing patterns of overdose deaths by age distribution (see graph below).

Source: Jalal et al. Changing dynamics of the drug overdose epidemic in the United States from 1979 through 2016. Science. 2018 

The following regional graphs highlight how different the situation in North America is to the rest of the world.

In this post, I have focused on direct deaths associated with drug use. However, drug use also raises risks for road injury deaths and suicide, and injecting drug users also face increased risks of death from HIV, hepatitis B and hepatitis C. Together, direct and indirect deaths associated with drug use account for close to half a million deaths annually in the world, and the great majority of these (74%) are associated with opioid use [conference-presentations #131].

Posted in Global health trends, World Health Organization | Tagged , , , , , | 1 Comment

Covid-19: which countries are succeeding and not succeeding?

There is now sufficient data on social distancing, lockdown, and other measures that researchers are starting to assess the effectiveness of these measures. The July 4th edition of The Economist has an article summarizing what is becoming clear.  Extensive testing and tracing of contacts was carried out by the Asian countries earliest affected, China and South Korea. China also locked down Wuhan and largely succeeded to confining the epidemic there. Italy was the earliest European country affected, and it locked down Lombardy on February 22, and succeeded in largely confining the epidemic to the north of the country. Some European countries, Australia and New Zealand locked down hard and very early while there were still very few deaths.  These were the most successful. As were Switzerland, Germany and Austria which locked down while they still had fewer than 60 deaths. Other European countries that waited till they had hundreds of deaths before locking down, such as Britain, were much less successful and the virus spread throughout the country.

The Imperial College modelling group have estimated that Europe’s lockdown policies prevented 3 million deaths by May 4th. The first wave in the USA, which primarily hit New York and New Jersey, was fairly successfully contained as in the European countries, but other states which had not closed their borders or successfully reduced the spread of the virus, opened prematurely and the US is now facing a runaway growth in new cases and deaths (see my previous post).

The country plots below show that it only takes about 5-7 weeks of strong interventions to get rid of the majority of cases. Taking half-measures as the US has done does not work. Particularly when the President repeatedly contradicts the advice of the public health experts and refuses to provide national guidance.

A couple of months ago, I was producing country plots of time trends in cases and deaths to see how the epidemic was evolving across the world. Since then, a very nice site has been doing these plots and updating them almost daily. The site groups countries into three groups: (1) those that are beating Covid-19, (2) those that are nearly there, and  (3) those that need to take action. The plots show new cases per day averaged over the last week and each graph has been normalized so all the peaks are the same height. This better shows the trend in the data and the degree to which the epidemic is being controlled, though not the relative magnitudes of the epidemics in different countries.

I’ve selected the plots for some countries of interest to me, to show differences in the impact of control measures in the three groups of countries, using data up to 4 July. The trends in some countries in the second group look quite similar to those in the first. The reason they are still in the second group is that the total number of new cases per day is higher that in the first group, and that can allow the epidemic to still escape control fairly easily.

Some countries that are beating Covid-19

Some countries that are nearly there

Countries that need to take more action

Posted in Global health trends | Tagged , , | 1 Comment

An out-of-control pandemic in most world regions

I downloaded the Covid-19 data up to 5 July from the JHU Github today and plotted smoothed new cases per day (average over last 7 days) as rates per million population by world region. The dramatic difference in trajectories for Western Europe and the Americas is obvious. Rates in all other regions are rising except for East Asia and Pacific, which is dominated by China. Australia, New Zealand, Japan and South Korea are also included in that region.

I thought it would be interesting to take a closer look at the situation in the USA, as North America appears to have the most out-of-control epidemic.  I classified the US states into blue and red, according to the popular vote in presidential elections. A state was blue if a majority voted for the Democrat candidate in more than half of the last four elections and red if it voted for the Republican candidate in more than half.  For three states which voted two times for each party’s candidate, I used the five last elections to break the tie.

The first peak in April for the blue states is dominated by New York. However from early June with premature re-opening of blue states there has been dramatic growth in new cases at a similar rate to the early stage in late March, and all benefits of the lockdowns have been wasted. There is now an upwards trend in blue states as well starting later in June.

I’ve done some back of the envelope calculations of the worst case scenario where growth in cases is not brought under control and the infected population reaches the estimated herd immunity level of around 60%.  Recent data from Geneva confirms other studies that have estimated the overall infection mortality rate at 0.6%.  Assuming infection fatality rates do not decrease due to improving treatment regimes, that would mean there would be approximately 1 mllion further deaths in the USA. After I had done this calculation, I came across a post by an Australian infectious disease expert which also predicts a worst case death toll of 1 million deaths for the USA.

Posted in Global health trends | Tagged , , , | 2 Comments

Two different pandemics: the USA and Europe

The trajectories of COVID-19 have been dramatically different in the USA and Europe. Currently there are close to 30,000 new cases per day in the USA, compared with around 3,000 per day in the European Union (and around 20 per day in Switzerland which is not an EU member). Testing rates per million are similar in USA and EU, and the EU population is 446 million compared to 332 million in USA. Almost half of US states are reporting significant upward trend in new daily cases. In most cases, this is not a “second wave” but a restart of growth after social distancing was relaxed while there had not been any substantial downturn in new cases.

Posted in Global health trends | Tagged , , , , , , | Leave a comment

Imperial study estimates 3.1 million deaths averted by the lockdown in Europe

I’ve stopped doing my own graphs to check trends for coronavirus. The website has added daily cases and daily deaths time series as well as moving 7-day average time trends to its visualizations, and has plots for countries for absolute numbers and rates per million population, and also for US States.  Looking at these yesterday, it is clear that apart from New York the US epidemic is not declining and for at least a dozen States, including Texas, North Carolina and California, the daily cases are trending upwards.  Likely as a result of the relaxation of social distancing with media reports of clusters of cases associated with various gatherings.

The Imperial College coronavirus modelling team published a new study on Monday which estimated that the lockdowns and other restrictions on daily life prevented around 3.1 million deaths in 11 European countries (Flaxman et al, The study worked back from observed deaths (with all their limitations) up to 4 May to estimate transmission that occurred several weeks prior and hence the reproduction number R. They found that the various lockdowns and other restrictions on public life had reduced the reproduction number by an average 82%, bringing it below 1. They then ran the model to predict the number of deaths if no restrictions had been implemented to estimate the restrictions prevented 3.1 million deaths.

Overall, they estimated that between 12 and 15 million people had been infected in the period, or between 3.2 and 4 percent of the population of the 11 countries. The table below shows the estimated infection rate for each of the 11 countries.

I’d recently seen a plot of US states coronavirus rates against population density showing a strong correlation. I did a similar plot below for the 11 European countries, with linear trend line. Belgium and Sweden are outliers on the high side for infection rates, and Germany, Denmark, Austria and Switzerland on the low side. Of course, not too much should be read into this, as this analysis should really be done at subnational level, rural/urban at minimum or preferably, city, town, rural, which would require more detailed geographic breakdown within countries of the deaths.

Of course, the situation in Switzerland has been of particular interest to me as I live in Geneva, one of the hardest hit cantons. Regularly updated time series data are available for Swiss cantons at Covid-19 Information for Switzerland.

I plotted the total reported coronavirus deaths as at 8 June 2020 as rates per million population by canton in the graph below. I thought about doing a density plot, but as some cantons have populations confined to valleys between Alpine peaks, the density issue may not be straightforward. Instead, I highlighted the language differences. The yellow bar denotes the canton in which Italian is the dominant language, Ticino, and it has the highest death rates because it borders northern Italy where the epidemic hit early and hard, and the border was not closed until well into the epidemic. The blue denotes cantons where French is the dominant language, and red where German is the dominant language.  Clearly Romandy (the Swiss French population in the west of Switzerland) was hit much harder than Alemannic Switzerland (the part that speaks the Alemannic dialect of German known as Schweitzer Deutsche.

The Swiss Government has issued a statement explaining that the Swiss lock-down rules were applied equally in all cantons. The main reason for the high rates in French and Italian speaking cantons were because of the high rates in the neighbouring countries and high proportions of cross-border workers. Geneva has a very high population of frontaliers (people who live in neighbouring France and commute to Geneva to work or go to school) as well as a high level of international visitors associated with tourism, skiing, and business and NGO headquarters. Additionally, when the epidemic hit Europe there were large numbers of skiers heading through Geneva to the Alps, from countries including France and Britain. In contrast, infection rates in Germany and Austria, neighbouring the German-speaking parts of Switzerland, have been much lower than in France and Italy.

Social distancing has been progressively relaxing since April 27th and most children are back at school, and businesses open but with relatively unenforced limits on interpersonal distances. If rates start to go back up, as is happening in many US States now, I have no doubt the Swiss will efficiently and without polarizing debate reintroduce relevant restrictions.

Posted in Global health trends | Tagged , , , , , | Leave a comment

Machine learning projection model for Covid-19

Youyang Gu, an independent data scientist, has created a statistical projection model for Covid-19 ( that uses machine learning techniques to fit a classical SEIR infectious disease model to the data for daily confirmed cases and deaths, taking into account the effects of social distancing and other factors. From the results I’ve looked at, it appears to be one of the better performing models around. The plots below show results for Switzerland, USA and United Kingdom based on data up to 31 May.

The second plot for each country shows R-t, the effective reproduction number at time t. When R is greater than 1, the epidemic is growing exponentially, and when R is less than 1 it is declining. The basic reproduction number in the absence of interventions to reduce transmission, R0, is typically around 2 for most countries, depending on factors such as population density and crowding. R0 was close to 4 for New York for example.

Looking across the country projections, it is interesting that R-t is currently slightly below 1 for countries such as Switzerland and UK, and marginally above 1 for the USA. It is more substantially below 1 for a few countries such as Norway and Australia, and above 1 for some countries, eg. Brazil, Russia and Nigeria.

A lot of people have now published strong criticisms of the IHME modelling, many identifying the major problem of fitting a mathematically symmetric curve to the epidemic which I noticed early on. Youyang Gu also compares IHME projections with his and shows severe under- and over-estimation issues with the IHME projections, which change wildly with model updates and iterations. See the plot below for a comparison.

Gu concludes:

“Models are going to make wrong predictions, but it’s important that we correct them as soon as new data shows otherwise. The problem with IHME is that they refused to recognize and update their wrong assumptions for many weeks. Throughout April, millions of Americans were falsely led to believe that the epidemic would be over by June because of IHME’s projections.

“On April 30, the director of the IHME, Dr. Chris Murray, appeared on CNN and continued to advocate their model’s 72,000 deaths projection by August. On that day, the US reported 63,000 deaths, with 13,000 deaths coming from the previous week alone. Four days later, IHME nearly doubled their projections to 135,000 deaths by August. One week after Dr. Murray’s CNN appearance, the US surpassed his 72,000 deaths by August estimate. It seems like an ill-advised decision to go on national television and proclaim 72,000 deaths by August only to double the projections a mere four days later.

“Unfortunately, by the time IHME revised their projections in May, millions of Americans have heard their 60,000-70,000 estimate. It may take a while to undo that misconception and undo the policies that were put in place as a result of this misleading estimate.”

Posted in Global health trends, Projections | Tagged , , , , , | 2 Comments