Bolivia is Incredibly Heterogeneous – Let’s Take Advantage of That When Fighting COVID-19

By: Lykke E. Andersen, Ph.D.*


“The only way to avoid ‘groupthink’ and blind spots is to ensure representatives with diverse backgrounds and expertise are at the table when major decisions are made.”
Devi Sridhar, Chair of Global Health at the University of Edinburgh Medical School


Bolivia is an amazingly diverse and heterogeneous country in every way. Within a million square kilometers we find both steamy Amazon jungle, large modern cities, mosquito infested swamps, melting glaciers, huge salt flats, and picturesque cloud forests. Some people live pretty much as their ancestors did hundreds of years ago, while others enjoy all the luxuries of the most advanced countries. According to our upcoming Municipal Atlas of the SDGs in Bolivia, the differences between municipalities within Bolivia are larger than the differences between all the countries in the world in terms of the Sustainable Development Index (SDI). And as within countries, there are also large inequalities within each municipality in Bolivia.

In this blog, I will argue that we should take advantage of this heterogeneity to reduce the mortality rate of COVID-19 in Bolivia.


Good News of the Week

About a month ago, the World Health Organization (WHO) came out with a worrying statement saying that “there is currently no evidence that people who have recovered from COVID-19 and have antibodies are protected from a second infection” [1]. If this were true, it would be very bad news for the many countries, including Bolivia and its neighbors, that have failed at containing the virus, have no test-trace-and-quarantine capacity, and whose only option to get through this pandemic therefore is to achieve herd immunity.

The good news out of the Korean Center for Disease Control (KCDC) last week was that the 263 people in Korea who tested positive for the disease a second time after previously being declared recovered and virus-free, was not due to re-infection nor re-activation of the SARS-CoV-2 virus. Rather, it was due to the PCR tests picking up old, inactive, and harmless virus RNA still present in the body a couple of months after the original infection. According to KCDC, the process in which COVID-19 produces a new virus takes place only in the cytoplasm of the host cells and does not infiltrate the nucleus. This means it does not cause chronic infection or recurrence, unlike viruses like HIV [2].

Thus, it seems that the herd immunity strategy might be viable after all. Once 60-70% of the population has become immune, the virus will die out, as it becomes difficult for it to find new hosts in which to multiply. The question now is how to apply that strategy with the least number of deaths and collateral damage possible.


Optimizing the Herd Immunity Strategy

I previously estimated that we will likely “end up with an Infection Fatality Rate (IFR) of around 1% for Bolivia (meaning anywhere between 0.3% and 2%, given the still high uncertainty). If 60% of 11.6 million people get infected, and 1% of those die, we would end up with about 70 thousand COVID-19 deaths in Bolivia” [3]. While this is an awful lot of deaths, the 1% IFR is a realistic estimate considering the age distribution of the population, underlying health conditions, the quality of the health care system, typical housing and work conditions, in addition to the level of education and trust in the population. It looks like the most likely outcome if we let the virus burn slowly through the population (through appropriate physical distancing measures) in a random way until reaching herd immunity after 12-18 months, and if we don’t actively make things worse than they have to be.

However, the IFR can be lowered considerably if we take advantage of the fact that the population is not homogeneous. Some people have far lower risk of dying from COVID-19 than others, so if we could secure that the first 60% to get infected are the ones least likely to suffer severe complications, then we could potentially reduce the total number of deaths considerably.

In the rest of this blog I will outline the main dimensions to consider when optimizing the herd immunity strategy.


1. The Demographic Dimension

The evidence from all over the world shows unequivocally that the risk of death from COVID-19 is higher in older people. In Italy, for example, until May 18th, only 4 people under 20 years of age had died, while the number of deaths of people over the age of 50 was almost 30,000 (see Figure 1).

Figure 1: COVID-19 deaths in Italy as of May 18, 2020, by age group

Istituto Superiore di Sanità (via Statista 2020).


There is also overwhelming evidence that men are almost twice as likely to die from COVID-19 compared to women. Figure 2 shows the case fatality rates observed in Italy to date, disaggregated by age and gender.


Figure 2: COVID-19 death rates in Italy as of May 2020, by gender and age group

The case fatality rates in this figure do not reflect the real Infection Fatality Rates, as there are errors
in both the nominator (un-counted COVID-19 deaths) and denominator (un-identified COVID-19 cases).
However, the overall pattern by age and gender is likely to reflect the actual differences in IFRs by age and gender.
Source: Istituto Superiore di Sanità (via Statista 2020).


Given this pattern, and similar patterns from all other countries with age and gender disaggregated COVID-19 death statistics, we can make the following rough demographic risk classification:

If we apply this classification to the population pyramid of Bolivia, it looks as shown in Figure 3.


 Figure 3: Population Pyramid for Bolivia, with risk categories

Author´s elaboration based on data from 


Based on just these basic demographic factors, 77% of the Bolivian population is at Low Risk of dying if contracting COVID-19; 16% is at Medium Risk; and 7% is of High Risk.

But there are more risk dimensions to take into consideration.

2. The Geographic Dimension

While age and gender are important determinants of risk, there are factors in the surrounding community that can either amplify or moderate the risk for each individual.

  • It is safer to live in a disperse rural area where you interact with few different people, than to live in a dense urban area touching public surfaces that thousands of other people touch every day.
  • It is safer to live alone, rather than in a three-generation extended household.
  • It is safer to live in a place that is not simultaneously plagued by other health threats, such as Dengue, Malaria, Tuberculosis and HIV.
  • It is safer to live in an area where there are basic water and sanitation services available.

In last week’s blog we developed a Municipal Vulnerability Index to COVID-19 [4], and while it is a continuous indicator, we can roughly divide it into Low, Medium and High Risk municipalities, with some admittedly arbitrary cut-offs. If we consider all municipalities with a Vulnerability Index higher than 36.7 High Risk, and lower than 31.7 Low Risk, then we get a municipal risk list as shown in Figure 4.


Figure 4: Municipal level COVID-19 Vulnerability Index, with risk categories.

(Click here to see details)
Source: Between a Wall and a Nasty Virus [4].


This Vulnerability Index only considers structural variables, and not actual infection rates, which would also be important to consider, but these suffer daily changes, and due to the limited testing capacity, many cases go undetected.


3. Occupational Dimension

Even for people of the same age and gender, living in the same municipality, risk will vary substantially depending on the kind of activities each of them engages in. The risk will be extremely high if you work as a dentist, but extremely low if you collect Brazil nuts alone in the forest.

In general, solitary outdoor activities are much safer than working indoors with lots of different people cramped together. The highest risk occupations would be those where you must be very close to many different, potentially infected people every day, such as dentists, doctors, and nurses. If you work at, or frequent, indoor locations where people are singing, screaming or breathing heavily, such as night clubs, karaoke places, churches, and gyms, you are also at high risk, as the virus spreads very effectively in this kind of places.


4. Individual Risk Factors

Apart from all the above mentioned risk variations, there will be additional personal risks that can be either permanent or temporary. For example, anyone who suffers from high blood pressure, diabetes or asthma would automatically and permanently move to a higher risk category than the one suggested by their age, gender, location, and occupation.

Likewise, anybody who presents COVID-19-like symptoms, irrespective of their age, gender, location, and occupation, should immediately consider themselves at High Risk and take every precaution to protect themselves and others.

Individual risk factors should also take into account other people in the same household. One individual might be young and healthy, but if they live together with a High Risk person, their risk category increases, because their actions might carry risks not to themselves, but to their loved ones.


Central Planning is Unlikely to Work

From the analysis above it is obvious that risks vary by several orders of magnitude from place to place and from person to person. This makes centralized decisions extremely difficult, and uniform rules will likely be both inefficient and harmful.

The initial strict quarantine measures have served to educate people about the dangers of this virus and about the hand-washing and physical distancing measures that can help control contagion. But these strict measures are clearly not sustainable over the many months that this pandemic is likely to last [5], and it is time for a more nuanced approach.


Decentralization of Decisions and Responsibilities

If our goal is to reach herd immunity with the least number of deaths and collateral damage possible, then we need to decentralize decisions considerably. Each department, each municipality, each business, each school, and each family need to analyze their strengths and weaknesses in this new global context, and make a plan on how to get through the following 24 months with the slightest possible damage. Damage not only includes COVID-19 deaths, but also loss of education, income, freedom, agency, joy, and happiness; so everybody needs to engage in quite a holistic analysis, which is not easy. It will take patience, communication, collaboration, and many iterations.

The need for decentralization of decisions and responsibilities is even greater now that the central government is facing a precipitous decline in revenues from all sources (especially IDH, IVA, IT, IUE, ICE, and RC-IVA), and thus will have much fewer funds available for distribution to departments, municipalities, and individuals.

We need to recognize that there are no right answers. Nobody knows the best way to get through this, and there is no one-size-fits-all solution. Nobody knows what the world is going to look like on the other side. This is a good time to be flexible, think out-of-the-box, and try out new ways of learning, working, and living.

Learning quickly is more important than ever, and learning is only possible if we try different strategies and learn from their different results. The best way to do that is to let municipalities pursue different strategies and record results more or less in real time.


The Need for Timely, Geographically Disaggregated Data on Deaths from All Causes

Since we have extremely limited testing capacity all over the world, and especially in Bolivia, reported COVID-19 cases and deaths rarely reflect reality. It is more feasible and useful to simply register the total number of deaths (by age and gender) per week from all causes, and compare that to the expected number of deaths per week in each territory.

According to INE, we expected 66,760 deaths in Bolivia this year without the COVID-19 pandemic [6], corresponding to 1,284 deaths per week in the whole country. This data can be disaggregated to the department level by applying the departmental crude deaths rates calculated by INE to the population of each department (see Table 1).


Table 1: Expected weekly deaths, 2020, by department

Source: INE and


According to this data, Beni is the only department in Bolivia that has a serious COVID-19 outbreak at the moment, probably because it is particularly vulnerable to a COVID-19 outbreak (due to high levels of obesity, low coverage of water and sanitation, crowded housing, simultaneous Dengue, Malaria, Tuberculosis, and HIV outbreaks, and low local government capacity), as we showed in our blog a couple of weeks ago [4]. However, the real number of COVID-19 deaths is likely far higher, since only people with a positive COVID-19 tests are counted. The dedicated COVID-19 cemetery in Trinidad (Beni), for example, a few days ago held 148 deceased, of which only 57 were confirmed COVID-19 deaths, while 91 were suspected [7].

In the table above, there are question marks in the last column concerning the number of deaths from all causes. This information is currently not available from any government entity in Bolivia. My recommendation to the National Statistical Institute of Bolivia (INE) would be to quickly build up a system to record number of deaths in each municipality each week by age and gender.

Weekly, geographically disaggregated information on all deaths by age and gender is necessary to carefully monitor local outbreaks and take adequate precautions in the right places [8].. A fine-tuned, decentralized response to this pandemic, requires timely disaggregated data. It is well worth the effort to set up the information gathering and reporting system, as it can save tens of thousands of lives and avoid a lot of unnecessary economic costs and human suffering.




[2] See John Campbell’s video for an easy-to-understand interpretation of the findings: See MedCram for a much more detailed, intracellular, explanation of the same thing:



[5] The World is still in the very early phases of this pandemic, with even hard-hit countries still having a long way to go before reaching herd immunity.  By early May, it was estimated that Belgium was the country closest to herd immunity, with 6.4% of the population having been infected, while in other European countries the immune population is still less than 5%. In certain hot-spots, like Madrid, the rate is much higher, but still nowhere near immunity (


[7] This page by Our World In Data provides a collection of sites monitoring excess mortality ( The Economist, Financial Times, the New York Times and EUROMOMO all provide excellent examples of how this data can be presented in user-friendly ways.

* SDSN Bolivia.

The viewpoints expressed in the blog are the responsibility of the authors and do not reflect the position of their institutions. These posts are part of the project “Municipal Atlas of the SDGs in Bolivia” that is currently carried out by the Sustainable Development Solutions Network (SDSN) in Bolivia.

Nuestro aporte en tiempos de COVID-19

Durante el actual contexto de la pandemia del COVID-19, enfermedad provocada por el virus SARS-CoV-2, SDSN Bolivia ha realizado una serie de presentaciones haciendo uso de los indicadores del Atlas Municipal de los ODS en Bolivia, para ayudar a entender mejor la situación de Bolivia ante esta amenaza. A continuación, se encuentran los detalles para dos presentaciones que están disponibles en línea.


Economics Research Workshop No. 38

El día 6 de mayo, la Directora Ejecutiva de SDSN Bolivia, Lykke E. Andersen, Ph.D, realizó la presentación “Usando el Atlas Municipal de los ODS en Bolivia para analizar la vulnerabilidad ante el COVID-19”, durante la cual explicó cómo estos indicadores pueden ayudarnos a entender la variación espacial en los potenciales efectos sobre la salud que pueden derivar de la propagación del virus SARS-CoV-2.

Andersen presentó los avances del Atlas y mostró, en un ejercicio interactivo con los participantes, cómo los datos pueden ser usados para analizar la vulnerabilidad al COVID-19. La presentación se dio en el Economics Research Workshop No. 38, coorganizado por UPB, SEBOL, INESAD y ABCE, en colaboración con la Fundación Solydes.

Como resultado del aporte de los participantes, se eligieron los principales indicadores para analizar esta situación (vea el informe completo aquí).

La presentación se encuentra disponible en el siguiente enlace: o en el reproductor:



WEBINAR |Bolivia: Pobreza y Desigualdad en el entorno del #COVID19

El 14 de mayo, Lykke E. Andersen realizó la presentación “Variación espacial de los impactos del COVID-19 en Bolivia” a través de la cual se intentó dar una respuesta a las siguientes interrogantes:

  • ¿Quién es más vulnerable a los impactos directos del virus SARS-CoV-2?
  • ¿Quién es más vulnerable a las medidas de mitigación del virus?
  • ¿Qué estrategias podemos implementar para evitar grandes aumentos en pobreza y desigualdad?

En análisis se realiza tomando en cuenta tres dimensiones: geográfica, demográfica y sectorial. Le invitamos a conocer las respuestas a estas interrogantes siguiendo el enlace: o en el reproductor:



El webinar fue moderado por Oscar Molina Tejerina, Ph.D, Vicerrector Nacional de la Universidad Privada Boliviana, y Presidente de la Sociedad de Economistas de Bolivia, y contó con tres conferencistas internacionales: María Eugenia Dávalos, Ph.D (Economista Senior – Banco Mundial), Lykke E. Andersen, Ph.D (Directora Ejecutva – SDSN Bolivia) y Maria Alejandra Gonzalez-Perez, Ph.D (Profesora Titular – Universidad EAFIT).


Entradas de Blog:


Entre la espada y la pared: el dilema del COVID-19

Por: Lykke E. Andersen, José Acuña y Luis Gonzales | Publicado el 12 de mayo de 2020

Cuarenta días de cuarentena: ¿qué hemos aprendido? 

Por: Lykke E. Andersen | Publicado el 01 de mayo de 2020

Between a Wall and a Nasty Virus

Lykke E. Andersen*,
José Acuña**,
and Luis Gonzales***



During the present COVID-19 pandemic, most countries in the world have failed at implementing precise measures of testing, contact tracing and quarantining of sick and infectious people. Instead they have implemented the rather crude and desperate strategy of locking everybody up for many weeks.

The latter is obviously not a sustainable strategy, and many countries are beginning to let people out of their homes, knowing full well that the virus is still out there, so opening up will inevitably lead to higher infection rates and more deaths. They just hope to keep severe cases at manageable levels. It is a big balancing act that requires good information in real time, and that is incredibly scarce.

Ideally, we should first let out the people who are least likely to get infected, least likely to infect others, and least likely to die from COVID-19. The idea is to gradually build up herd immunity over the next 12 months with the least number of COVID-19 deaths possible and the lowest collateral damage. That is, we are aiming to minimize total damage from this pandemic.

Who can most safely get out and resume work, education, and leisure activities? The relevant probabilities have three main dimensions: geographic, demographic, and sectoral, and they interact in complicated ways.

In this blog we will provide some empirical evidence to inform strategies to gradually get out of the extremely strict lockdowns in Bolivia.


Geographic variations in risk

Once the virus arrives to a new location, it can have very different impacts depending on a variety of factors. If it arrives to a sparsely populated region with young, healthy, well-informed people practicing good hygiene, the virus will spread slowly, and the vast majority of infected people will likely have either mild symptoms or none at all. But if it hits a densely populated area with malnourished and frail people already suffering from other diseases and without adequate access to water, soap, and sanitation (such as a refugee camp), the impacts could be devastating.

In this section we will present a municipal level analysis of differences in the likely impact of the arrival of SARS-CoV-2 to different municipalities in Bolivia. We considered several dozen potentially relevant indicators from the upcoming Municipal Atlas of the SDGs in Bolivia, and grouped them into the following three broad categories:


  1. Risk of rapid spread
    • Number of main roads entering the municipality
    • Centrality of migration (an index reflecting how many other municipalities each municipality is connected with through recent migration)
    • Percentage of population living in urban areas
    • Percentage of population living in crowded homes (more than 2 persons per room)
    • Public transportation intensity
  2. Underlying health situation
    • Obesity level
    • Chronic malnutrition level
    • Incidence of Dengue, Chagas, Malaria, Tuberculosis and HIV
    • Percentage of population older than 60 years
  3. Response capacity
    • Water and sanitation coverage
    • Electricity, Phone and Internet coverage
    • Share of population who do not speak Spanish
    • Education inequality
    • Local government budget execution capacity
    • Public investment per capita
    • Number of doctors per 10,000 inhabitants.


Each variable was converted into an index from 0 to 100, and these indices were aggregated together using weights derived from an online consultation process [1]. The Vulnerability Index reported in this blog uses the 15 indicators that more than 50% of participants agreed had a strong effect on the probability of dying from COVID-19, and equal weights were given to each of them. The 15 indicators chosen were equally distributed between the three groups and they are highlighted in italics in the list above [2].

Table 1 shows the Vulnerability Index for the 339 municipalities of Bolivia ranked from the most vulnerable to the least vulnerable.

Table 1: Vulnerability Index to COVID-19 (based on 15 indicators)
Click on image to expand.

Source: Authors’ elaboration.
Note: The 9 state capitals + El Alto are highlighted in bold.


Map 1 shows the spatial distribution of the same Vulnerability Index.


Map 1: Vulnerability Index to COVID-19 (based on 15 indicators)

Source: Authors’ elaboration.


Demographic variations in risk

Within each municipality there is considerable variation in risk between different population groups.

It is clear from studies carried out in areas that were infected by SARS-CoV-2 early on that Infection Fatality Rates (IFR) increase exponentially with age and are considerably higher for men than for women [3]. In addition, people with one or more underlying health problems, especially hypertension, obesity, and diabetes, are much more likely to die [4]. Both in the UK and the US, racial differences in IFRs have been observed. Even after controlling for socio-economic and occupational factors, people with darker skin are more likely to die than people with lighter skin [5], possibly due to vitamin D deficiencies that weaken the immune system [6]. This is unlikely to be a significant factor in Bolivia, as long as everybody can get out in the sun every day.

Fortunately, children rarely get seriously ill from COVID-19. Of more than 27 thousand COVID-19 deaths in Italy to date, only 3 were under the age of 18 [7]. There is also some emerging evidence that infected children may not spread the virus as much as adults [8]. Thus, many countries are starting to reopen schools, with increased hand-washing and social distancing efforts, daily disinfection and cleaning of school surfaces, and careful monitoring of the results [9].


Sectoral variations in risk

For people of similar demographic characteristics within the same municipality, risk will vary depending on what kind of activities they carry out. In general, solitary work outdoors is much less risky than working indoors with exposure to lots of different people. Thus, agriculture and construction are relatively safe, while working in a supermarket, a hospital or a dentist office is of high risk [10]. Similarly, solitary outdoor leisure activities, like hiking, golf or tennis, are much safer than indoor team sports like volleyball, basketball or handball. Potentially super-spreading places include night clubs and karaoke bars, among others.

Some high-risk activities are essential and must remain functional even during the strictest lockdowns. In these cases, implementing measures to reduce risk as much as possible is important. Mask wearing in supermarkets, banks, and public transportation is a highly effective way of preventing the virus spread from asymptomatic people to others. Extended opening hours and staggered work schedules reduce crowding and facilitate physical distancing. Switching to online services and home delivery is also possible in many cases. Even many medical services have been successfully transformed and turned into much safer and more convenient telemedicine systems.


Recommendations for loosening the lockdown in Bolivia

A total and uniform lockdown turns out to be inefficient given the large geographic, demographic, and sectoral variations in COVID-19 risk. It is also highly damaging to people’s mental, physical and economic health, and clearly unsustainable.

Given that a vaccine is very unlikely to become available at a global scale until late next year at the earliest [11], and given that we have failed at eradicating the virus even after 50 days of the strictest lockdown that could possibly be enforced in Bolivia, the only remaining option to get through this pandemic is to let the virus burn through most of the population in a controlled fashion over the next 12 to 24 months [12].

If we abandon all precautionary measures, infection rates will skyrocket and we will end up with many more daily deaths than we can physically and psychologically handle. Instead we should continue rational precautionary measures and carry out a gradual reopening, starting with the municipalities, the demographic groups, and the sectors with the lowest risk.


Our recommendations for the immediate future are the following:

  1. Everywhere, we should maintain the following simple measures to limit the infection rate:
    • No kissing, hugging or handshaking, except with your closest circle of family members and loved ones;
    • No unnecessary gatherings of a lot of people, meaning no sports events, no concerts, no carnivals, no festivals, no graduation events, and no religious gatherings;
    • Maintain a 2-metre distance from strangers, interact with as few different people as possible, and wear a mask if you have to be close to them (in supermarkets, public transportation, banks, etc.);
    • Avoid touching surfaces that a lot of other people touches, and wash your hands thoroughly after touching a potentially infected surface;
    • Work and study from home as much as possible, and limit interactions to as few different people as possible.
    • When working from home is not possible, implement flexible working hours and staggered work schedules to reduce peak occupancy in public transportation systems and workplaces.
  1. Everywhere, outdoor work and leisure activities should be permitted, as long as physical distancing is possible. Indeed, people should be encouraged to get fresh air, sunlight and moderate exercise in order to optimize their immune systems. Masks should not be compulsory when exercising outdoors, because they limit the optimal oxygen intake.
  2. In the vast majority of municipalities, children can get back to school if adequate hygiene facilities are available. It would be safest if teachers are female, under the age of 60, and generally healthy. Teachers with high risk of a severe COVID-19 reaction (older males with high blood pressure, diabetes, or other risk factors) should not get into contact with children. WHO has guidelines on how to reduce risk in schools [13].
  3. In the vast majority of municipalities, most shops can reopen, as long as clients can maintain adequate physical distancing. Risks would be lowest if shops are attended by young women. Opening hours should be expanded rather than reduced, in order to reduce crowding.


On the other hand, the most vulnerable municipalities need to prepare for a major impact of COVID-19. In all the variations of our Vulnerability Index, Riberalta jumps out as the most vulnerable municipality in Bolivia. It is a big urban municipality (about 100,000 inhabitants) with lots of connections to other municipalities through recent migration, and it provides health services for many surrounding municipalities in the departments of Beni, La Paz and Pando. However, it has a very low coverage of water and sanitation, and high levels of obesity, malnutrition, Dengue, Malaria, Tuberculosis, and HIV. It is a ticking bomb. But for some reason, the government’s index does not flag it as high risk.

Finally, since this is going to take many months to get through, we should use this opportunity to get universal coverage of electricity, telecommunications, and online services. Now is the time for AGETIC to really push forward with electronic government services.


[1] See recording of webinar held on the 6th of May 2020, and this PDF of the polling results.

[2] We are working on a working paper with much more details and which includes a sensitivity analysis, since there are many possible ways of aggregating the indicators. But the results presented here are highly correlated with the other aggregate indices analyzed.

[3] In Germany, for example, the death rate for men between 50 and 80 years is at least double the death rate for women in the same age group ( In Italy, Spain, China, Peru and Greece, the gender difference is even more pronounced (

[4] See, for example, Richardson, S., Hirsch, J. S., Narasimhan, M., et al. (2020). “Presenting Characteristics, Comorbidities, and Outcomes Among 5700 Patients Hospitalized with COVID-19 in the New York City Area.” JAMA. Published online April 22, 2020. doi:10.1001/jama.2020.6775 (







[11]See this article for a discussion of what it takes to develop, test, produce and distribute a new vaccine:[0]=18743&tl_period_type=3

[12] See

[13] See

* Executive Director, SDSN Bolivia

** Economic Research at Centro Latinoamericano de Políticas Económicas y Sociales – Pontificia

Universidad Católica de Chile – CLAPES-UC.

*** Head of Energy, Environment and Economics at Centro Latinoamericano de Políticas Económicas y

Sociales –Pontificia Universidad Católica de Chile – CLAPES-UC


* SDSN Bolivia.

The viewpoints expressed in the blog are the responsibility of the authors and do not reflect the position of their institutions. These posts are part of the project “Municipal Atlas of the SDGs in Bolivia” that is currently carried out by the Sustainable Development Solutions Network (SDSN) in Bolivia.

Forty Days of Quarantine – What Have We Learned?

By: Lykke E. Andersen, Ph.D.*

 “There are lies, damn lies, and COVID-19 statistics”
Johan Norberg



The term quarantine comes from a venetian word meaning “forty days” which was the number of days in which ships and people had to be isolated before being admitted to the Republic of Venice in Medieval times in order to be sure that they were not infected by deadly infectious diseases, such as the Plague, Cholera, Syphilis, or Yellow Fever.

In Bolivia we are nearing our forty days of “reverse quarantine” (keeping the healthy people locked up, in order to prevent them from getting infected by an external threat).

This reverse quarantine has bought us some time to get to understand the novel SARS-CoV-2 virus (Severe Acute Respiratory Syndrome Corona Virus No. 2), and the COVID-19 disease it is causing. Initial data out of China and Italy was so scary that it was worth the high costs to buy us some time to formulate a strategy on how to deal with the threat, and prepare the population and the health system to manage it as well as possible. We have very little data from Bolivia, but since this is a global pandemic, we can learn a lot from other countries, despite extremely flawed data.

A few basic things have become clear:

  • The SARS-CoV-2 virus has infected people in pretty much every country and territory on the planet, so suppression and eradication of the virus is unfortunately no longer a realistic option (1);
  • Many people who catch the virus have no symptoms, which is why this virus has managed to spread so easily across the globe (2).
  • There is no previous immunity, nor any treatment available (3), so the virus will not go away until we have achieved herd immunity, either through vaccination, or by 60-70 percent of the population having been infected (4);
  • Although many potential vaccines have been developed in record time, they have to be tested for safety and efficacy, which means that vaccines at a global scale will not be available for at least another 12 months. By the time a safe and effective vaccine becomes available for billions of people, it may not be needed any more (5);
  • The Infection Fatality Rate (IFR) is likely to be somewhere between 0.1% and 10%, depending on the health of the population, the age composition of the population, the quality of the health care system, the policies enacted to confront the problem, and possibly the type of the virus that dominates, because there seem to be different strains circulating already (6). It is clear that men are more likely to die than women, older people are far more likely to die than younger people, and people with underlying health problems, especially hypertension, obesity and diabetes, are more likely to die (7).

Given these facts, it is clear that we are facing some difficult decisions. Short of simultaneously locking up everybody on the planet for many months, there is no way we can prevent that millions of people will die from COVID-19. In the absolute best case scenario (IFR like the common flu at 0.1% and 60% of the world population getting infected), we will see 4.2 million people die from this disease, and we should consider ourselves very lucky if that is the number we converge towards over the next 24 months. More likely, there will be at least 5 times more deaths than that, meaning at least 20 million people will die. So far, only about a quarter of a million have died, so the world is still in the very early stages of the pandemic (98% still to come).


How are we doing in Bolivia?

In Bolivia, we have barely started the process, because we locked down early and thoroughly. To date we have only 62 confirmed COVID-19 deaths, out of a minimum of 7,000 and a maximum of 800,000 to be expected. That is a frustratingly large range, and it is difficult to make wise decisions until we narrow down the likely path of this epidemic. The whole point of locking down is to obtain more information and figure out which end of the range is most likely, and thus which kind of policies are appropriate to get us through this pandemic.

In a normal year, about 24,100 people die from all causes in Bolivia. In the absolute best case scenario, this virus would kill off 6,600 old and frail people who would have died from other causes this year anyway, thus implying no excess mortality. Unfortunately, we already know that this best case scenario will not play out, because among the first to die of COVID-19 in Bolivia were a pregnant nurse, and several otherwise healthy people below 70 years of age.


How bad is it going to be?                                                                                       

We have waited for 40 days in order to rule out the worst case scenario, which would be a 10% IFR. The original data coming out of Wuhan suggested that 20% of infected people would need hospital care in order to survive, and that almost 4.9% of infected people died anyway. Also, of more than 1 million closed cases to date worldwide, 18% have died (8). However, recent antibody tests carried out in California (9), Germany (10), Denmark (11), and the Netherlands (12) suggest that many people have been infected without any symptoms, which means that the true number of infections is several times higher than the confirmed cases, implying that the IFRs are much lower than the official Case Fatality Rates (CFR = Deaths/Confirmed Cases) suggest.

In the USA, the New York State recently carried out random anti-body testing on 3,000 individuals to figure out how many people had really been infected, and they found that 13.9% of the population, or about 2.7 million people in the state, had already been infected at a time when “only” 19,453 COVID-19 deaths had been registered. This suggests an IFR in New York state of around 0.72%, or a little bit higher, since some of these infected people are still critically ill and more will unfortunately die (13).

Of course, New York is one of the richest places on the planet, so their IFR may not be relevant for Bolivia. Data from Peru is probably more relevant here, and fortunately Peru has somehow managed to carry out more than 300,000 tests, while Bolivia has only done around 6,000. In Peru, more than 37,000 people have been confirmed to have the virus, but only 2.8% of confirmed cases have died so far (8). However, Peru, like all other countries, has limited testing capacity, so in reality there will be many more infected, and thus the IFR will be a lot lower.

With the still very incomplete information available at the moment, I estimate that we will end up with an IFR of around 1% for Bolivia (meaning anywhere between 0.3% and 2%, given the still high uncertainty). If 60% of 11.6 million people get infected, and 1% of those die, we would end up with about 70 thousand COVID-19 deaths in Bolivia. The number could be lower if a vaccine becomes available before we reach herd immunity through infection, but I consider that unlikely (5). The good news is that more than 11 million Bolivians will not die from COVID-19.


Proportionate interventions

We are facing an undeniably difficult situation, like all other countries. What we definitely have to make sure is not to make things even worse than they have to be. 70,000 dead people is bad. But these people dying alone, shunned and isolated in designated COVID hospitals, without family, friends and funerals, seems much worse. If at the same time even more people are losing their livelihoods, their investments and their dreams due to quarantine, that would be awful. If children start dying from hunger because their parents are not allowed to work (14), it would be a full-blown disaster. If we lose our basic human rights and freedoms, and we cannot see and hug our loved ones for years (15), that is simply an unbearable thought.

Thus, we have to make sure our interventions are well thought through and based on the best evidence possible. We are lucky that our country got infected relatively late (first confirmed case on the 10th of March, 2020), and we managed to keep numbers low for the first couple of months through strict quarantine measures, which means we got the gift of time to enable us to learn from good and bad experiences in other countries, and from all the new scientific research that is coming out to help us understand our options better.


Flattening the curve is clearly necessary

I am not suggesting that we should flatten the curve so that our health care system does not get overwhelmed, because it got overwhelmed by the very first patient (16). But I do suggest we flatten the curve enough to make sure that we can physically, mentally and socially handle every diseased person in a dignified manner. If we do not spread our expected 70,000 deaths out as evenly as possible over at least a year, we will experience the horrors of dead bodies piling up in the streets, like we are seeing in Guayaquil in Ecuador (17). If we could spread our expected 70,000 deaths perfectly evenly over the next 12 months, we would have around 1,350 COVID-19 deaths per week. Hopefully some of these would have died from other causes anyway, but it is clear that we have to be prepared to increase our funeral capacity, because Bolivia is used to handle only about 1,300 deaths per week from all causes.


What does successful management of a global pandemic look like?

Ideally, we should have nipped this epidemic in the bud, like we managed to do with the first SARS outbreak in 2003, the MERS outbreak in 2012, the Ebola outbreak in 2014, and hopefully most future similar outbreaks. However, this time the world screwed up big time, and with millions of people being infected all over the globe, eradication is just not realistic anymore. A few rich island nations may be able to test, trace and isolate cases and keep it under control until a vaccine is available, but for most of the world’s countries, including Bolivia, that is just not a realistic aspiration.

My criteria for success are much less ambitious: If less than 0.6% of the population die from COVID-19 within the next 12 months, and if the unfortunate 0.6% die with loved-ones holding their hands, and family members and close friends get the opportunity to pay their respects and process their losses, and if the economy contracts less than 5% (a setback of less than 2 years), then I would consider that a successful management of an unescapable pandemic with no known cures available.


How do we successfully manage this epidemic?

The key is avoiding huge and unmanageable spikes in deaths. That will require carefully calibrated social distancing measures.

Some “easy” social distancing measures should be implemented by everybody at all times until this pandemic is over:

  • No kissing, hugging and handshaking, but try to be kind to everybody anyway;
  • No unnecessary gatherings of a lot of people, meaning no sports events, no concerts, no carnivals, no festivals, no graduation events, and no religious gatherings; but try to have fun in new creative ways;
  • Maintain a 2-metre distance from strangers, interact with as few different people as possible, and wear a mask if you have to be close to them;
  • Avoid touching surfaces that a lot of other people touches, and wash your hands thoroughly after touching a potentially infected surface;
  • Work and study from home as much as possible, and limit interactions to as few different people as possible.
  • When work from home is not possible, implement flexible working hours and staggered work schedules to reduce peak occupancy in public transportation systems and work places.

These simple measures substantially reduce infection rates, but they may not be enough. Even tougher measures may be necessary in certain locations if infections spike for some reason.


Monitoring of outbreaks

In order to know when tougher measures are necessary, we need extremely good monitoring of the epidemic. Ideally, we would have massive testing capacity like Iceland or South Korea, but Bolivia has the lowest testing capacity in South America at less than 1 in a thousand people (18), and we have to be realistic about what is actually feasible.

There are two alternative options that could provide us with valuable information in real time about how the pandemic is evolving:

  • A daily symptom tracker app on our phones, which could alert authorities to a local outbreak, and help individuals get the help they need. A simple such app exists in the UK, and much more elaborate apps are used in many places in Asia (19). It is trickier in Bolivia, as it requires high levels of trust in the government, and the population would need to perceive concrete benefits of using the app. For example, it could be linked to a generous donation of free phone minutes and Internet access, telemedicine consultations, free medicine delivery, and more. Such an app would be technically relatively easy to develop, but it would require serious thoughts on how to get a large share of the population to trust and use the app daily.
  • A less demanding option is to monitor weekly deaths from all causes at a sub-national (ideally municipal) level, so as to alert us if any region is beginning to spike, and would therefore need to implement stricter social distancing measures and receive more support from the central government. EuroMOMO would be a good model for this (20).

Both of these options would be much less expensive and damaging than shutting down the entire country for many months. A traffic light system could be designed to clearly communicate the current levels of restrictions in different parts of the country. Indeed, this should be part of the app mentioned above.


Sustainable transition towards a new Bolivia

Whatever systems we do implement, we have to make sure they can be sustained over time, because this is going to take at least a year to get through, and the world will look different on the other side. Families and firms will have to adapt to these new circumstances, and the government needs to support them through this transition. At the very least, the government has to make sure nobody starves to death (people should be able to request help through the app if they have urgent needs, and the government needs to develop the infrastructure needed to respond). The government also has to accelerate investments in absolutely crucial infrastructure, such as water, sanitation, electricity and Internet. In order to facilitate a more agile transition and response to the rapidly changing market conditions, now is a good time to eliminate the very strong rigidities in the Bolivian labor market, because many of the changes we will see are not going to be transitory. It would also be a good idea to make it much easier to close companies that have become unsustainable, so that people can spend their time and money on starting up new businesses, instead of spending months or years going through all the ridiculously difficult procedures to close a company.


(1) Bolivia, as well as several other female-led countries, such as Taiwan, Hong Kong, New Zealand, Iceland, Norway, Finland and Germany, potentially could suppress and eliminate the virus, but that doesn’t help us much in this globalized world, if there are major male-led countries around us that fail to do that (e.g. United States, United Kingdom and Brazil).

(2) For example, 408 residents at a homeless shelter in New York was tested for the virus, and 36% of them was found to have the virus, but 87.7% of the people who had the virus, did not have any symptoms (

(3) Countries around the globe have been scrambling to buy mechanical ventilators, but it is not a treatment, it only provides life-support while the body’s own immune system battles the virus. A recent study of outcomes in 12 New York hospitals show that the vast majority of COVID-19 patients on ventilators die. Indeed, of those aged 65+, 97.2% of COVID-19 patients with an outcome by the end of the study, had died. For patients aged 18-65, 76.4% had died. See: Richardson S, Hirsch JS, Narasimhan M, et al. Presenting Characteristics, Comorbidities, and Outcomes Among 5700 Patients Hospitalized With COVID-19 in the New York City Area. JAMA. Published online April 22, 2020. doi:10.1001/jama.2020.6775 (

(4) Some people are questioning whether we will even be able to achieve herd immunity, as some people who had previously been diagnosed and then cleared, have caught the virus a second time within a few months ( See also this note about Corona virus immunity research at Columbia University:

(5) See this article for a discussion of what it takes to develop, test, produce and distribute a new vaccine:[0]=18743&tl_period_type=3

(6) The Chinese scientist who originally proposed the lock down of Wuhan, Dr. Li Lanjuan, has carried out ultra-deep sequencing of the RNA in different samples, and says that the SARS-CoV-2 virus mutates faster than previously thought, and that some strains are more infectious and more lethal than others (

(7) Richardson S, Hirsch JS, Narasimhan M, et al. Presenting Characteristics, Comorbidities, and Outcomes Among 5700 Patients Hospitalized With COVID-19 in the New York City Area. JAMA. Published online April 22, 2020. doi:10.1001/jama.2020.6775 (

(8) See

(9) Researchers at Stanford conducted antibody tests on 3,300 volunteers (a non-random sample obtained through facebook ads) in Santa Clara, California, and found that  1.5% of the sample tested positive for the antibodies, suggesting that the real number of COVID-19 infection was 50-85 times higher than the official numbers by April 1st, 2020 ( However, it should be noted that this was not a random sample. Volunteers who responded to the ad, might be people who had experienced COVID-19 symptoms, and were eager to find out if they have already had the virus.

(10) Researchers conducted anti-body tests on inhabitants of Gangelt, a German municipality near the border with the Netherlands, which was hard hit by covid-19 after a February carnival celebration. They found that 14% of the population had already been infected by late March (

(11) On 6-8 April 2020, Denmark tested 3,898 blood donations from asymptomatic people and found that 1.9% had COVID-19 antibodies (

(12) Between the 6th and the 12th of April 2020, the Netherlands tested 4,194 blood donations and found that 3.4% had COVID-19 antibodies (

(13) See Dr. John Campbell’s discussion of these results here:

(14) Unfortunately this has already started happening in Bolivia. A 12-year-old girl in the municipality of Montero killed herself after being quarantined without food for several days with her mother and her 7 siblings (

(15) While other countries help, or at least allow, their nationals to return home, Bolivia has closed its borders so tightly that many Bolivians have been stranded either at border crossings, or wherever they happened to be at the time of the lock down. Especially inhuman was the treatment of a group of Bolivians, including pregnant women and women with babies, who tried to get home from Chile late March ( I was also horrified to read than an opposition mayor in Cochabamba was arrested in her home last week for playing loud music and drinking “chicha” (a fermented beverage made from corn) together with her closest family. Although she tested negative in the alcohol test made on the “scene of crime”, all eight people present were arrested, and the youngest child was sent to a center for homeless children (



(18) See

(19) Here is the symptom app used in the UK: Google and Apple are also working to develop a contact tracing app that can alert you if you have been near a confirmed COVID-19 infected person (, but that would have to be used in conjunction with extensive testing and it requires a very disciplined population for people to self-isolate for two weeks, just because their phone indicates they have passed by an infected person, so it does not seem ideal for Bolivia.

(20) EuroMOMO is a European mortality monitoring activity, aiming to detect and measure excess deaths related to seasonal influenza, pandemics and other public health threats (


* SDSN Bolivia.

The viewpoints expressed in the blog are the responsibility of the authors and do not necessarily reflect the position of their institutions. These posts are part of the project “Municipal Atlas of the SDGs in Bolivia” that is currently carried out by the Sustainable Development Solutions Network (SDSN) in Bolivia.

Education Inequality in Bolivia

While inequality indicators are usually calculated using monetary income or consumption data, they can be calculated on any variable of interest, thus informing us about other types of inequality in society. In this blog we present a new education inequality indicator to be included in our Municipal Atlas of the SDGs in Bolivia.

Using data from the 2012 Population Census we calculated years of education for every person. We then chose the age group 25-65 years, since that is the age group in which most people will have finished their education and are highly likely to be working and using their education. Having learned from our experience analyzing electricity consumption inequality, we calculated a series of different inequality measures in order to figure out which measure would be the best choice. Unlike what we found for electricity consumption inequality, all education inequality measures were highly correlated, so it didn’t matter much which one we use, so we decided to use the Gini coefficient of Years of Education.

The Gini coefficient of Years of Education ranges from 0.205 in Coipasa (Oruro) to 0.642 in Ocurí (Potosí). This is a range in which the Gini coefficient works well and correlates closely with other measures of inequality. La Paz has very low education inequality (Gini = 0.207), while Sucre has the highest level of education inequality (Gini = 0.321) of the 10 main cities.

Many smaller municipalities have much higher education inequality, and as can be seen in Figure 1, this has a lot to do with women having been denied the same education opportunities as men in the past. Very few municipalities have a Gender Parity Index close to 1 (meaning that men and women have the same level of education). In many small, rural municipalities women have only between half and two-thirds of the education of men, and this situation correlates with high education inequality.

Figure 1: Education inequality versus gender inequality in education, by municipality, 2012

Source: Authors’ elaboration based on data from the 2012 Population Census.

Although returns to education are low in Bolivia, (in terms of future income) and the quality is suspected to be low as well, Bolivians are studying more than ever, so education is clearly considered an important investment by the population. We consider this education inequality variable an important indicator to be included in the Municipal Atlas of the SDGs in Bolivia.


* SDSN Bolivia.

The viewpoints expressed in the blog are the responsibility of the authors and do not necessarily reflect the position of their institutions. These posts are part of the project “Municipal Atlas of the SDGs in Bolivia” that is currently carried out by the Sustainable Development Solutions Network (SDSN) in Bolivia.

Measuring Inequality at the Sub-National Level in Bolivia

The scale of inequality in the world is staggering. In 2018, the average inhabitant of Denmark, Norway, Sweden, Iceland, and Ireland earned more in two days than what the average inhabitant of Malawi and Burundi earned during an entire year [1].

However, the inequality within some countries is even bigger than the inequality between countries. As we show in the upcoming Municipal Atlas of the SDGs in Bolivia, there are larger differences in the Sustainable Development Index between municipalities within Bolivia than there are between all the countries in the world. This means that within Bolivia we find municipalities with levels of development similar to the most advanced countries in the World, but also municipalities similar to some of the least developed countries in the World.

In addition to this astounding inequality between municipalities within Bolivia, there are also huge inequalities within each municipality, and that is what we focus on in this blog.

Measuring inequality at the sub-national level is difficult, because it requires income/consumption data that is representative at the municipal level, and this is neither available from household surveys nor population censuses. We get around this problem by using data on electricity consumption by each household in Bolivia, under the assumption that electricity consumption in each household is a reliable predictor of general consumption/income/welfare in the household. The analysis was made possible due to a research project sponsored by the Centre for Social Research (CIS) of the Bolivian Vice-presidency [2].

The CIS project calculated Gini coefficients of electricity consumption to estimate the level of inequality within each municipality. Figure 1 shows two examples. To the left is the Lorenz curve and the Gini coefficient for a big urban municipality, Santa Cruz de la Sierra. To the right is the Lorenz curve and Gini coefficient for a poor rural municipality, Tinguipaya, with low electricity coverage and low electricity consumption for the ones that actually do have access to the electricity grid.


Figure 1: Examples of electricity consumption Lorenz curves, 2016

Source: Andersen, Branisa y Guzman (2019) [2].


According to these Lorenz curves, the big, urban municipality of Santa Cruz de la Sierra is clearly more equal in electricity consumption (and thus likely consumption in general) than the small, rural municipality of Tinguipaya, as the Lorenz curve of the former is much closer to the diagonal line of perfect equality.

Indeed, as can be seen in Figure 2, this is the general tendency across all municipalities in Bolivia. Urban municipalities tend to have lower levels of poverty (poverty being measured as extremely low electricity consumption) and lower levels of inequality (as measured by the Gini coefficient of electricity consumption).


Figure 2: Energy inequalty versus energy poverty in Bolivian municipalities, 2016

Source: Authors’ elaboration based on data from Andersen, Branisa y Guzman (2019).


This strong positive correlation between energy poverty and energy inequality seems suspicious, and perhaps even counter-intuitive. Consider this extreme hypothetical example: If all households in a municipality had zero electricity consumption, except the mayor, who had a moderate electricity consumption covering a few light bulbs and a refrigerator, then the Gini coefficient would be 0.99, suggesting extremely high inequality in the municipality, whereas common sense would suggest that the population is very equal in their extremely high level of poverty.

The Gini coefficient is by far the most widely used measure of inequality, and easy to interpret, so it was the logical metric to use. But these results made us wonder if the Gini coefficient tends to confound extreme poverty with extreme inequality.

Thus, in this blog we set out to explore whether there are alternative measures of inequality that might correspond better to intuition. There are surprisingly many different measures of inequality, so it became another long post.

There are four basic principles that one would expect from an inequality measure [3]:

  • Symmetry (or anonymity): If two people switch incomes, the index level should not change.
  • Population invariance (or replication invariance): If the population is replicated or “cloned” one or more times, the index level should not change.
  • Scale invariance (or mean independence): If all incomes are scaled up or down by a common factor (for example, doubled), the index level should not change.
  • The Pigou-Dalton Transfer Principle: If income is transferred from one person to another who is richer, the index level should increase. In other words, in the face of a regressive transfer, the index level must rise.

It can be shown that the Gini coefficient satisfies these four basic principles, as does several other inequality measures, such as the Atkinson Index, the Theil entropy measure, and the Theil mean log deviation measure. These inequality measures are all called strongly Lorenz-consistent [4].

Some frequently used inequality measures are only weakly Lorenz-consistent, however, as they do not fully comply with the fourth principle. When a poorer person makes a transfer to a richer person, the inequality indicator need not rise, but at least it should not fall [4]. The Palma ratio (the income of the 10% richest divided by the income of the 40% poorest) and other Kuznets ratios (X% richest/Y% poorest) are only weakly Lorenz-consistent.

Other inequality measures are plainly Lorenz-inconsistent. This is the case for quantile ratios (p90/p10, p75/p25, etc.) and the Variance of Logarithms. For both of these measures it is possible that a regressive transfer from a poorer person to a richer person would cause a fall in the inequality measure, which is clearly counter intuitive [4]. The Absolute Gini proposed by Jason Hickel [5] is even worse, as it also violates the third principle of scale invariance.

From the theoretical arguments mentioned above, we would expect all the Lorenz-consistent measures to yield pretty much the same results as the Gini coefficient, so we decided to also include weakly Lorenz-consistent and Lorenz-inconsistent measures in our comparisons.

However, we quickly ran into an important problem: Many inequality measures cannot handle zero values (e.g. Theil, Atkinson, percentile ratios). Some algorithms circumvent this problem simply by ignoring zero values. But we do not consider this a reasonable strategy, since the lack of access to electricity is one of the fundamental problems we want to highlight rather than ignore. An alternative way around this problem is simply to add a small value, for example 1 kWh per year, to all observations, which does not change overall patterns, but will make all computational algorithms work [6].

During the following sections, we will compare many different inequality measures with the Gini coefficient. All inequality measures are calculated on household level data on annual electricity consumption + 1 kWh for a selection of 25 municipalities that span the whole range shown in Figure 2. For each of the alternative inequality measures we discuss the intuition behind the measure, its advantages and disadvantages, and show its relationship to the Gini coefficient.


Atkinson Indices

The Atkinson Index, A(e), is actually a whole class of inequality measures, differentiated by a parameter, e, that measures the degree of inequality aversion. When e = 0, there is no aversion to inequality, and A(0) = 0. When e = ∞, there is infinite aversion to inequality and A(∞) = 1. The Atkinson Index thus varies between 0 and 1, like the Gini coefficient, which facilitates interpretation. The Atkinson Index has the advantage of being sub-group decomposable, which the Gini coefficient is not.

The Stata package ineqdeco (created by Stephen P. Jenkins at London School of Economics) calculates the Atkinson Index for three different parameters: 0.5, 1 and 2. Figure 3 shows how the Gini coefficient (of annual electricity consumption + 1 kWh) compares to these three Atkinson Indices.


Figure 3: Comparing the Gini coefficient with three Atkinson Indices for a sample of 25 municipalities

Source: Authors’ elaboration.


We see that for moderate inequality aversion (e = 0.5), the Atkinson Index behaves very similarly to the Gini coefficient. For stronger inequality aversion (e = 1), the Atkinson index increases at all levels, but especially for the ones that had high Gini coefficients, thus exaggerating the counter-intuitive results of the Gini coefficient. For very high inequality aversion (e = 2), the Atkinson Index is very close to maximum for all our municipalities, and thus do not provide any useful information about differences in inequality.

In conclusion, the A(0.5) index seems the most useful of the three, but it behaves very much like the Gini coefficient and thus does not solve our initial problem of counter-intuitive results.


Generalized Entropy Indices (including mean log deviation and Theil index)

The Generalized Entropy Index, GE(α), is another class of inequality measures. As with the Atkinson Indices, the GE Indices involve a parameter, α, that can shift the sensitivity to different parts of the distribution. For lower values of α, GE is more sensitive to differences in the lower tail of the distribution, and for higher values GE is more sensitive to differences that affect the upper tail. The most common values of α used are 0, 1 and 2 [7].

The GE Index has several other inequality metrics as special cases. For example, GE(0) is the mean log deviation (also sometimes called Theil’s L), GE(1) is the Theil index, or Theil’s T, and GE(2) is half the squared Coefficient of Variation.

The values of GE indices can vary between 0 and ∞, with zero representing an equal distribution and higher value representing a higher level of inequality.

Like the Atkinson Index, the GE Index is decomposable, which is an advantage over the Gini coefficient. However, it is not bounded above, and the interpretation is not at all intuitive.

The Stata package ineqdeco calculates several different GE Indices. Figure 4 shows how the Gini coefficient (of annual electricity consumption + 1 kWh) compares to the three most common ones.


Figure 4: Comparing the Gini coefficient with the three most common
Generalized Entropy Indices for a sample of 25 municipalities

Source: Authors’ elaboration.


In all three cases the GE indices agree with the Gini coefficient that the poorest municipalities with the highest Gini coefficients have the highest levels of inequality. The GE(0) index moderates the relationship slightly, whereas the GE(1) exaggerates it, and the GE(2) exaggerates it even more. Thus, none of the three most commonly used GE indices suggests that poorer municipalities would be more equal.

However, α can also take on negative values, making it more sensitive to differences in the lower end of the distribution. Figure 5 shows that the GE(-1) index is completely different from the Gini coefficient. One municipality that stands out with extremely high inequality (GE(-1) = 250) is Santa Cruz de la Sierra (Bolivia’s most populous municipality, home to both extremely rich and extremely poor households), whereas the poorer municipalities in our sample have the lowest levels of inequality according to this measure (still high, though, as the scale is completely different for this GE measure).


Figure 5: Comparing the Gini coefficient with the GE(-1) index for a sample of 25 municipalities

Source: Authors’ elaboration.


This rarely used inequality measure, GE(-1), potentially fits better with common intuition about which municipalities are more unequal. We will explore it in more detail further below.


Income shares, Kuznets ratios and Palma ratio

Thomas Piketty’s best-selling book “Capital in the Twenty-First Century” made the use of percentile shares popular for analyzing inequality. Piketty and collaborators focused on top-percentage shares, using varying percentages as thresholds (top 10%, top 1%, top 0.1%, etc.).

A related concept is the Kuznets ratio, which compares the income of the top X% with the bottom Y% of the population. A special case of this is the Palma ratio (the income of the 10% richest divided by the income of the 40% poorest) (8). The top-percentage shares are a special case of Kuznets ratio, as it is the top X% divided by the bottom 100%. So all these measures can be called Kuznets ratios.

Kuznets ratios are weakly Lorenz-consistent, since progressive or regressive transfers within each group analyzed will not affect the various indicators.

Ben Jann of the University of Bern created a convenient Stata command, pshare, to calculate the share of income received by any group along the income distribution [9]. The default is the income shares of the five quintiles (20% poorest to 20% richest). However, due to the high level of inequality in electricity consumption indicated by the Gini coefficient, we find it important to disaggregate the top quintile, and see the share of electricity consumed by the top 10%, top 5% and top 1% of households in each municipality.

In addition, in order to calculate the Palma ratio, we use the 40% and 90% cut-offs.

Figure 6 shows how the Gini coefficient (of annual electricity consumption + 1 kWh) compares to the following three Kuznets ratios: Top 5% of electricity consumption share; top 1% share; and the Palma ratio.


Figure 6: Comparing the Gini coefficient with three Kuznets ratios

Source: Authors’ elaboration.


The first two measures, which focus on the top end of the distribution, confirms -and even exaggerates- the finding that the poorest municipalities are the most unequal.

However, the Palma ratio shows a completely different pattern. The extreme outlier in our small sample is Uyuni, with a Palma ratio of 777, followed by Ixiamas, Machacamarca and Patacamaya. In contrast, all the mayor cities of Bolivia have very low levels of inequality by this measure. There is nothing about these results that seems even remotely related to intuition.

In the range of Gini coefficients between 0.4 and 0.63, there seem to be a positive relationship with the Palma ratio, but for higher Gini’s the relationship becomes completely random and unrelated to common sense.


Other inequality measures

There are several standard measures of dispersion reported by statistical packages, but which do not have fancy names nor particularly intuitive interpretations, and thus are not widely used in the inequality literature.

The first three indicators reported by the Stata command inequal (developed by Edward Whitehouse of OECD in Paris), are the Relative Mean Deviation (RMD), the Coefficient of Variation (CV) and the Standard Deviation of Logs (SDL). Figure 7 plots these inequality measures against the Gini for our sample of 25 municipalities.


Figure 7: Comparing RMD, CV, and SDL to the Gini coefficient

Source: Authors’ elaboration.


The first of these measures, RMD, are closely related to the Gini coefficient; the second one, CV, exaggerates the relationship between inequality and poverty, but the third paints a different picture that might possibly correspond better to intuition. According to the bland indicator called “Standard Deviation of logs”, Santa Cruz de la Sierra has high inequality, while Poroma (one of the poorest municipalities in Bolivia) has low inequality, which seem to correspond to intuition.

The same inequal Stata command reports the Mehran, Piesch and Kakwani measures of inequality. They are not well known, nor widely used, and since they all correlate strongly with the Gini coefficient (see Figure 8), we don´t think they provide much additional insights into inequality within Bolivian municipalities.


Figure 8: Comparing Mehran, Piesch and Kakwani to the Gini coefficient

Source: Authors’ elaboration.



In this blog we have investigated the properties of 16 alternative measures to the Gini coefficient of inequality. Most of them agree with the Gini coefficient that poor, rural municipalities are more unequal than rich, urban municipalities.

However, we found two inequality measure that provide different, yet plausible patterns of inequality. Those are the Standard Deviation of logs and the Generalized Entropy Index with a parameter of -1 (strong emphasis on the lower end of the distribution).

Since the results seemed intuitive for our sample of 25 municipalities, we decided to crunch the numbers for all 339 municipalities. Figure 9 shows how Figure 2 would look if we use GE(-1) instead of the Gini coefficient. And Figure 10 shows the same, but using the Standard Deviation of logs measure instead.


Figure 9: Comparing poverty and inequality using GE(-1) for all municipalities in Bolivia

Source: Authors’ elaboration.

Figure 10: Comparing poverty and inequality using Standard Deviation of logs for all municipalities in Bolivia

Source: Authors’ elaboration.


It is interesting to note that the two measures GE(-1) and Standard Deviation of logs agree on many of the most unequal municipalities (e.g. Rurrenabaque, Reyes, Cobija, Santa Cruz de la Sierra, Puerto Suarez), while they completely disagree with the Gini coefficient about the high level of inequality in small, poor, completely rural municipalities. Both are in reasonable agreement concerning the relative rankings of the large urban municipalities, with all agreeing that Cochabamba is the most equal of the 10 capital cities (+ El Alto) and Cobija the most unequal. The correlation between the two is 0.83, which is high, but not so high as to be redundant.

Of the 17 different ways to measure inequality that we have tried out in this blog, GE(-1) and Standard Deviation of logs seem to correspond best to intuition about inequality within Bolivian municipalities. The Gini coefficient, and all the other measures that correlate strongly with the Gini coefficient, do not seem to correspond to an intuitive perception of inequality, at least not for very poor municipalities.

It is clear that the optimal choice of inequality measures depends greatly on the distribution of the data, and that the Gini coefficient is not automatically the best choice.

After having processed millions of data points for 300+ municipalities in 17 different ways, we conclude that the best indicators to include in our Municipal Atlas of the SDGs in Bolivia under SDG 10 are: 1) Inequality Index 1: Standard Deviation of log household electricity consumption, and 2) Inequality Index 2: Generalized Entropy (α = -1) of household electricity consumption.

We do recognize, however, that we have engaged in what could be called “method-mining” in order to obtain results that correspond better to our intuition and expectations. Thus, in order not to give too much weight to these electricity-consumption-based inequality measures, we recommend to include other measures of inequality under SDG 10. Next week we will explore within-municipality inequality measures based on education levels.



[1] World Development Indicators, Gross National Income per capita, Atlas method, 2018.

[2] Andersen, L. E., B. Branisa & F. Calderón (2019) “Estimaciones del PIB per cápita y de la actividad económica a nivel municipal en Bolivia en base a datos de consumo de electricidad.” Investigación ganadora presentada al Centro de Investigaciones Sociales (CIS) de la Vicepresidencia del Estado Plurinacional de Bolivia. Mayo.

[3] In the latest Human Development Report 2019, which analyzes inequality, James Foster and Nora Lustig provide a useful overview to help decide between alternative inequality measures. See Spotlight 3.2, pp. 136-138.

[4] See Francisco Ferreira’s blog “In defense of the Gini coefficient”:

[5] See

[6] We also considered the possibility of adding more than 1 kWh per year to each observation, because all households receive a transfers from the sun amounting to at least a couple of lightbulbs for 12 hours per day, and perhaps a whole lot more. In cold regions, the sun also provides free heat, but in hot regions, that may not be perceived as a benefit, but rather as a disservice. Thus, it is virtually impossible to take into account these varying contributions from the sun, and any values we might chose would be arbitrary. Researchers face the same dilemma when calculating income inequality, or wealth inequality, as there are so many public services or environmental services that contribute to the wellbeing of families, but are extremely difficult to quantify and include in each family’s income/wealth. It is major research problem that we do not pretend to be able to solve here, so in this blog we will just add 1 kWh to each observation in order to overcome the technical problem of zeros and be able to actually compare all the different inequality statistics.

[7] See

[8] See

[9] See


* SDSN Bolivia.

The viewpoints expressed in the blog are the responsibility of the authors and do not necessarily reflect the position of their institutions. These posts are part of the project “Atlas of the SDGs in Bolivia at the municipal level” that is currently carried out by the Sustainable Development Solutions Network (SDSN) in Bolivia.

What is the collective transportation situation in Bolivia?

The world population has been increasing greatly, growing from 2.6 billion in 1950 to 7.3 billion by 2015. The world population is estimated to reach 8.5 billion by 2030 [1]. Due to rural-urban migration, this growth is accompanied by an exponential population increase in cities. In 1960 the population residing in cities was 1.0 billion (33.6% of the total population), however, in 2018 that number had increased to 4.2 billion (55.3%) [2]. This rapid urbanization is putting pressure on the fresh water supply, wastewater treatment, living environment, and public health.

Traffic jams and air contamination are important problems that afflict cities, causing not only a loss of efficiency in economic activities as a result of time wasted stuck in traffic, but also serious environmental and health consequences.

The Sustainable Development Goal (SDG) number 11 (Sustainable Cities and Communities) aims to make cities and human settlements inclusive, safe, resilient and sustainable. To achieve this, cities must invest in public transportation. Specifically, Target 11.2 seeks to provide access to safe, affordable, accessible and sustainable transport systems for all by expanding public transportation.

For this reason, in the Municipal Atlas of the SDGs in Bolivia, we include an innovative indicator that measures the supply of collective transportation in each municipality.

The data was obtained from the National Statistical Institute (INE), based on the Municipal Tax Administration Registry Office (RUAT). The data includes the car fleet by municipality and vehicle class between 2003 and 2017.

During the construction of the indicator, we decided to consider only vehicles that are destined for collective transport, that is, that have the capacity to transport 10 or more passengers. That is why only the categories bus, microbus and minibus were included in the construction of the indicator. The analysis considers a potential seating capacity of 54 for buses, 29 for microbuses and 14 for minibuses [3].

After estimating the number of collective transport seats, we divided the result by the population of the respective municipality to have the number of available seats per thousand inhabitants. This analysis allows us to identify the municipalities with the greatest collective transport capacity.

In addition to the above mentioned privately owned buses, we also added the seats available in Mi Teleférico (cable cars operating in La Paz and El Alto as a system of public transportation), as well as in the PumaKatari and ChikiTiti buses (La Paz), and Wayna Bus (El Alto).

In the case of Mi Teleférico, the analysis was different because it has interurban lines between La Paz and El Alto. That is why we consider the number of cabins of each line and the location of each station to determine the seats available in each municipality. For example, for the Red line, with 109 cabins and three stations, we determined that a third of the cabins belong to El Alto and two-thirds to La Paz since the line has only one station in El Alto and two in La Paz. The following table shows the distribution of cabins in both cities according to each station’s location:


Table 1: Station and cabin distribution by Mi Teleférico line

Source: Author’s elaboration based on data from Mi Teleférico.


The analysis allowed us to approximate the total number of seats available per thousand inhabitants of Mi Teleférico in both La Paz and El Alto, considering that each cabin has a maximum transport capacity of ten passengers.

The estimates of seats available for public transport provided by the PumaKatari, ChikiTiti and Wayna Bus were made in the same way as those based on RUAT data. All the existing vehicles in the municipality were considered: 179 PumaKatari buses and 30 ChikiTitis in La Paz and 60 Wayna Buses for El Alto. Likewise, the maximum transport potential of each of these vehicles was also used (61, 55 and 82, respectively) [4] [5].

The following graph shows the results for the ten municipalities with the largest number of seats available per thousand inhabitants, by type of collective transport. The municipality of Achocalla in La Paz is the one with the most available seats per 1000 inhabitants. Most collective transportation seats available come from minibuses.


Graph 1: Top 10 municipalities by number of available seats, by type of collective transport

Source: Author’s elaboration based on SDG 11 results for the Municipal Atlas of the SDGs in Bolivia.


Finally, we use the methodology proposed by the SDSN office in Paris to estimate the thresholds of this indicator. It uses the average +/- half the Standard Deviation to define the ranges in which a municipality is considered to have reached the goal, is close to reaching it, still has significant challenges, or has great challenges. In Bolivia we consider that a municipality has reached the target if there are at least 93 seats available per thousand inhabitants. Despite the fact that this figure seems low, the results indicate that only 12% of Bolivian municipalities have managed to reach this goal, 6% are close to reaching it, while the remaining 82% still have significant/great challenges to reach this goal. Although initiatives such as Mi Teleférico, PumaKatari, ChikiTiti and Wayna Bus have helped increase public transportation supply, there is still a long way to go, especially considering the trends of population growth and migration to cities.




[3] El RUAT define a los autobuses como ómnibuses.




* Alejandra Gonzales Rocabado, Assitant Manager, SDSN Bolivia

The viewpoints expressed in the blog are the responsibility of the authors and do not necessarily reflect the position of their institutions. These posts are part of the project “Atlas of the SDGs in Bolivia at the municipal level” that is currently carried out by the Sustainable Development Solutions Network (SDSN) in Bolivia.