Operationally defining homicide

Santos: What I study is not just homicide, it’s what we call ‘intentional homicide.’ So, pretty much, the biggest challenge of the type of research I do is that I do research not on a country level; I do research on a global scale. So, I’m trying to understand the global drivers; things that drive crime. Not just United States. Not just in Brazil, not just, you know, in Africa. Not just in Mississippi. But around the world. And the biggest challenge of doing research on crime on a global scale is that you need definitions that are standardized. You need [that] crimes should be defined the same way across the world. And that’s very complicated when you’re talking about a property crime, like burglary or theft, because the way people enforce – and the way people define – burglary, or theft, or rape, or sex crimes or, you know, fraud: they can be very different between places around this planet. But when you’re talking about homicide, we have a lot of consensus. And it’s not me, it’s the United Nations and the World Health Organization who uses these definitions.

And what is a homicide? A homicide is the killing of one person by another person with the intent to kill, or to cause a very serious harm. And there is another element there. So first is: it differentiates homicides [are] from suicides. Why? Because suicide is also a killing, but by you to you, right? Against yourself. A homicide, in contrast, is one person to another, with the intent to cause that harm. Which differentiates it from, let’s say, a manslaughter or non-intentional homicide, where somebody kills someone by accident. And the third element in the definition is that the homicide cannot be legal; it has to be illegal. And what do we mean by a ‘legal homicide’? First, the homicide in self-defense: you can kill someone because you believe your life is being threatened. And the second time of legal killing are legal killings done by military personnel, right? You can kill people because you’re fighting in a war. And those are also not counted in our statistics as homicide.

[ Back to topics ]

The association between age and homicide

Watkins: The world’s population of young people peaked in the 1960s and 70s, then – from 1980 onward – all regions except Africa experienced a decline in their population of people 15 to 29 years of age. Also during this time, the average global life expectancy increased from 43 to 66 years. Since homicide rates declined concurrently with these worldwide trends, Mateus was curious whether global aging might be related to the international decline in homicide, as he explains next.

Santos: Most of the world has been getting a lot safer since the 1990s, and an international homicide decline exists. And that makes you wonder what … like, look: if you have these global trends, how can you have domestic causes? How do domestic causes explain a global trend? And that led me to take a step back and to think, “What causes could explain this global homicide decline?” And the one that I found most support for was this global aging of the population. So, the World War Two created this huge generation of Baby Boomers, which was this disproportionately big cohort of people. And these people, they they reached, let’s say, late adolescence and early adulthood during the 1960s, 1970s. You know, that time in the United States history when things were very unstable, things were very messy? That is the same time when you had a lot of young people in the population, and the institutions – the social institutions – could not absorb them very well. [So] they interacted with themselves much more than they interacted with adult people people. People had one two three siblings. Nowadays, most people have how many siblings? One? Maybe none, right?

And what we show is that the whole world is getting a lot older very fast. First, because this Baby Boom generation is aging. And also because people are living much longer now, and that is causing a huge decline in the size of the young population relative to the older people. And, consequently, because [of] two reasons: first, because these young people have much more supervision; and, second, because young people are the age group who commit most violent crimes. [So] what we’re seeing is a huge decline in violence. We don’t say that’s just because of age; that’s not our point. But we say that this global process of population aging, it’s bringing a huge contribution to that process for multiple reasons. And, of course, it brings challenges as well, such as many less young people to support a much bigger cohort of elderly, and things like that as well. But our point is simply that this aging of the population is a huge challenge, yes, of course, but it also has this incredible benefit of making societies more peaceful.

[ Back to topics ]

Within- vs. between-country differences

Leigh: Mateus and his colleagues decided to use long-term longitudinal data to examine if – and how – changes in age-structure corresponds to homicide. So Ryan and I were curious whether the same factors responsible for fluctuations of homicides within countries might also be at play between them as well.

Santos: A lot of people know this, actually, and have mentioned … alluded to this in past research. We just connected dots, I believe, but a country is not more violent than another country – or a city, or a society – is not more violent than another society, necessarily, just because it has more young people. It’s also a matter of what these young people are doing, how are they being socialized, education systems, resources, culture. But the main thing that explains, actually, violence rates is inequality in income. Inequality – social inequality in general – is something that has a very strong correlation with the between-country differences in violence rates. But when we talk about within-country change in the violence rates over time, what the aging of the population does is that it causes this change over time within a single country, right? It may not explain completely the differences between-countries, but it explains why a country is getting more or less violent over the course of time. Because, you see, the homicide rate of countries around the world, they are hugely different. If you go to a country [such] as the United States today – and the United States has a homicide rate of about five, so five people are killed for every 100,000 population. That’s 15,000 people: every year 15,000 people every year, they die from homicide in the United States. So, let’s say, the United States homicide rate dropped by 20%, that decline corresponds to what? One per 100,000 people. It’s a huge decline. But, you know, that one is the same as the homicide rate of Japan. Japan nowadays has a homicide rate of 0.9, so the difference between Japan and the United States is, you know, more than 400%. It’s enormous. So when you model that – when you try to identify what explains this – you cannot see what explains change within [countries] when you’re exploring, you know, in between-country differences. So, in short, the factors that explain differences between-countries in terms of their social indicators, including crime, they do not necessarily have to be the same factors that explain change within-countries over time. Because countries are very different between each other in term, you know, of other social indicators. Brazil, for instance, has a homicide rate of more than 30 per 100,000 people, which is six times that of the United States. And what explains that is not just the fact that Brazil and the United States have different proportions of young people in their populations, it’s other factors.

[ Back to topics ]

Lives lost through homicide vs. war

Watkins: While homicide is projected to cause more deaths globally by 2030 than infectious diseases such as tuberculosis, it’s also true that – nearly every year since the turn of the 21st century – more people have been murdered than have lost their lives to war. Mateus helps us understand this seeming paradox after this short break.

ad: Altmetric‘s podcast, available on Soundcloud, Apple Podcasts & Google Podcasts

Watkins: Here again is Mateus Reno Santos.

Santos: So, one of the reasons the population is aging so much is because people are not dying for anything else besides, like, diseases at very old ages. Those, like, things that would kill people at their youngest ages, they don’t kill people anymore. Particular diseases. Or infant mortality is dropping like a rock, so that means that even though homicides are declining, the proportion of participation they have in the killings of young people in particular, it’s going up by a lot. And, actually, the most shocking statistic that – I think, in my opinion – is that homicides, they kill – I don’t remember the exact proportion – that they kill a lot more than wars around the world. These wars, they kill, like, one-seventh … one-tenth … of the number of killings that we have for homicide around the world every year. Homicides, they kill more or less half a million people around the world every year. That’s the estimate we have. The only year when conflicts killed more than that was in 1994, I believe, because of the Rwanda genocide, you know, when we believe between half a million and 1,000,000 people were murdered. And that’s the only instance over the last 30 years. So, pretty much from a public safety perspective, homicides are a very big deal, and we invest many more resources and make sure that they drop even more than they have.

[ Back to topics ]

Global homicide databases

Leigh: Mateus and his colleagues triangulated their data from two sources: the United Nations Office on Drugs and Crime, or UNODC, and the World Health Organization, or WHO. As the data spanned nearly 70 years, Ryan and I were curious what this breadth of data allowed for that’s less available when looking at a more constrained period of time.

Santos: To measure homicide, I used two main datasets. The first one was a global homicide database that I developed as a consultant in partnership with a huge team, with great people, at the United Nations Office on Drugs and Crime in Vienna, Austria. And we pretty much did a global search of reliable data on homicide since 1990 until 2016. And it was a lot of heavy data search – and a lot of verification work, documentation work, and a lot of consultation with countries – in order to find great counts that are comparable on an international scale. And the United Nations have been collecting these data with countries for 30 years now. So, for 30 years, every year the United Nations sends out a survey asking countries, “Hey, how many homicides did you have? Are these actually homicides? Where did you get this data from?” The United Nations’ hope is that the data they collect come from criminal justice organizations. But, in contrast, we have a second source. And that’s the richness of homicide statistic, actually. It’s that you have two completely independent sources of data on homicides. They are a crime and as a crime they are recorded by criminal justice organizations, and as a public health concern they also produce a medical record. And some of those causes of death, they are classified in the World Health Organization mortality database. They are related to homicides. And, fortunately, the definition of homicide in the UNODC data, it’s very consistent with the definition of homicides in the WHO data, which means that I was able to use both sources to validate one another, and to produce this very long-term data trends since 1950 all the way to 2016. So, over more than 70 years for 26 countries, which is a lot. And that huge trend was crucial to me because changing demographics, they take decades to unfold. So, if you just have a five-year window of data, or if you just have ten years of data, you might not be able to see enough variance. [You] may not be able to see enough variation in the age composition. And if you cannot see variation, you cannot see the covariance – the variation in the homicide trend – so you may not be able to see a relationship that actually exists because you’re looking too closely.

[ Back to topics ]

The representativeness of the two datasets

Watkins: While the UN dataset tracked 126 of the world’s countries with more than 1,000,000 residents in 2015, the data only capture homicides since 1990. Conversely, the WHO Mortality Database captured homicide data since 1950, but for a much smaller set of countries. So Doug and I followed up by asking Mateus about the extent to which the data were representative of all the world’s nations … or of only a few.

Santos: What I thought in the beginning was that I would only be able to see the effect of age if I looked at a large enough number of years. Ideally, I wanted data since the 1940s, 1950s. Why? Because I knew that Second World War caused a Baby Boom generation. And I knew that the Baby Boom generation caused variance in the proportion of the population between 15 to 29 years of age, so it caused variance in the age structure of the population. And by taking a look at that variance, I could see the covariance. On the other side, how many countries have great data since 1950? Very few countries. How many? Twenty-six. According to my criteria, twenty-six only. And that’s a very small number. And I wanted to explore if the homicide, the decline was actually this global event. And if it was, I wanted to see whether it had a global explanation.

The reason why I used the UNODC data that only goes from 1990, but it covers 126 countries, and these countries had more than 90% of the world’s population. So, using that second database I would really be able to make assertions, right? To tell a story about the whole world. And what I found, actually, I found that the effect of age depends on homicide level. So, I found that even looking at that data since 1990, I was able to see an effect of age for the safest countries in the world. Why? Because nothing else was interfering with the homicide trend, but I saw that as you looked at that effect, you know, further down the homicide rate distribution. It got harder and harder to see the effect of age. So my interpretation of that variation is that – the main reason I saw such a huge effect in what I call the long-series sample – is because the long-series sample is composed of much safer countries to begin with. You know, the safe democracy is stable. They have great data. They have the money, and the resources, and have had these resources since the 1950s. And the reason that [the] effect of age is not as clear in the – what I call-high covered sample – is because it has a lot of countries which are very dangerous; very violent. Which have a lot of criminogenic forces, and which have a lot of instability. And since I cannot control for that instability, those criminogenic forces interfere with the homicide trade much more.

[ Back to topics ]

Adjusting for potential confounds

Leigh: As Mateus mentioned earlier, just as a population’s average age relates to its homicide rate, so too does a country’s income inequality and its relative safety and security. But since there could be many other factors that explain the associations between these variables, we were curious how potential confounds were adjusted for in the project.

Santos: It’s very hard to find the social indicator – to find an indicator at the country level – that was available, in this case for 56 years. So we had to be very limited in the number of control variables we included in our models. Because if we included control variables that had no data over the long-term, that meant that we would lose data. That we would have, maybe, attrition; we would have many more incomplete observations. So, the control variables we ended up using are the percent of the population who are male [and] the GINI index, which is inequality. We took that from an incredible project; it’s called the Standardized World Income Inequality Database. And we use the GDP per capita of countries [to], you know, summarize economic development as a measure of economic development. And the percent of the population that is urban.

Something we also did is that we used a modeling strategy, a method, that’s called fixed effects regression. And a property of fixed effects models is that they tend to be very conservative. Why? Because, first, they focus on estimating what drives change: what is the effect of each one of the variables included in the model on the change in the homicide rate. The model is not concerned with the difference between countries, the model is concerned with the change within countries over time. So, what caused that variation over time? The second – as an added benefit – the model, it controls for all characteristics of each country which are stable over time. Anything about a country that doesn’t change – never changes over time – it’s controlled for by this fixed effects model, even if that variable is not explicitly added to our model. Which makes it a very conservative modelling strategy.

Still, the model is susceptible to bias, especially to other factors of countries that can change over time. So our findings can also be a result of selection – it can be driven by something else. But that’s the whole thing about science, it can always be something else. You never have the final answer, right? And I think that’s important to say is, like, it can always be something else. But what, right? That’s the whole point is to keep the discussion moving forward. And it goes to the whole idea of falsifiability, right? If we have definitive answers then we’re not, you know, it’s not science anymore, we’re doing something else.

[ Back to topics ]

Other criminogenic forces

Watkins: While demographic variables can be controlled for in a statistical analysis, conditions that produce higher crime rates – such as those related to political, social, and economic distress – might also account for the decrease in homicide among an aging population. So Doug and I asked Mateus how these variables were accounted for.

Santos: One of the things we identified in previous research – not just us but also previous studies – is that the homicide decline is not entirely global. This is true for the whole world, and this is true also in the United States. In the United States, not all counties experienced the decline is in crime [to] the same extent. Actually, there is a pattern to that. We found that the safest countries were experiencing, or are experiencing, homicide declines at a much faster rate than the most violent countries. Other people also found that same pattern in the United States. The safest counties are experiencing violent crime declines at a much faster rate – at [a] much faster pace – than the most violent counties.

Problem is, why? Why aren’t the most violent countries around the world experiencing homicide declines? We hypothesized that the most violent countries in the world were not participating in the global homicide decline because other factors, other criminogenic forces – such as organized crime, drug trade, economic and political instability – were driving the crime trend. Were driving their homicide trend. And because these factors were driving the homicide trend, and because these factors were so impactful, we were not being able to see the effect of age. So, these countries are not able to enjoy the effect of the aging of the populations simply because there’s so much else going on. [There’s] too much instability. So the challenge is: how do you test for that? How do you test that the effect of age gets smaller as these criminogenic forces grow stronger? And our strategy was to evaluate the effect of age for countries according to their level of violence. We tested if the effect of age was the same for the safest countries in the world and for the most violent countries in the world. And what we found is that the effect of age composition is much, much, much stronger. Very strong. Actually it’s extremely strong for the most safe countries in the world. So, pretty much, if there is nothing else going on in your society, then you can clearly feel that pacifying effect of the aging of your population. And we use the quantile regression to do that. The quantile regression enables us to evaluate the differences in the effect of age across the distribution of the homicide rate. And it was a very interesting result: as homicide rates grew higher the effect of age declined.

[ Back to topics ]

Imputing missing data

Leigh: With data spanning nearly 70 years across multiple countries, inevitably, some data points were missing in the UNODC and WHO’s materials. However, statistical techniques allow for “imputing” missing data with substitute values derived from the data that are available. Mateus explains next how he and his team applied this technique in their study.

Santos: The UNODC data has the better data, because the UNODC spends a lot more resources on homicides themselves. WHO data is very noisy, especially when you’re looking at long-term data. Sometimes, if you look over the longer past, you’re gonna see very noisy numbers. Sometimes you’re gonna see duplicates, and things of that sort. So you have to impute data for the years for which you have no data for each country. So what we did was we tried to take advantage of the overlapping years between the WHO and UNODC in order to speak about the quality of both data sources. So if there is a bigger than 30% difference between the homicide rate in both sources, then we did not combine. And, let’s say you have only WHO data until 1980 and the UNODC data only starts in 1990: you have ten years without data. Then we did not select that country. So, pretty much I used the same method I developed in partnership with the United Nations.

In the UN, people are very concerned – rightfully so – with communicating methods well. So our biggest concern is not just to develop a method that is correct and accurate, it’s also to develop a method that people can understand. And the method we used was a simple moving average. We impute the data using a simple moving average. We tried to use regression analysis to create an imputation, and we did, and it looks good still. It looks fine: there’s no problem while using regression analysis for calculating these regional trends. But, in the end, the cleanest method – the most parsimonious, the simplest method – was that for countries which had missing years, we used the moving average to impute that missing year for that same country. So, pretty much, we imputed the data for 2010 if that was missing using a moving average of all other years with data for that country. And what we did was that we exponentially weighted that moving average, which means that closer years – years that are closer to the imputation – have a much greater weight in the average than years that are further away from the imputation. And that worked extremely well. I had no problem. But before doing that, of course, you collect the data for a lot of countries in a lot of years. So we would have to use that method as little as possible. And one cool thing is that the United Nations has that code – it’s an R markdown code – and I think I even included the code with the paper.

[ Back to topics ]

Implications of the study’s findings

Watkins: Mateus and his colleagues found that a one percentage point increase in the proportion of youth in a population was associated with an increase in the homicide rate of 5.4%. So, a big takeaway from the paper for us was that demographic patterns deserve special attention in explaining homicide trends. We finished our conversation by asking Mateus what he believes the broader implications of their findings are.

Santos: I would predict that the world for the next 50 years is going to continue to grow older, and as a consequence it is going to continue to grow safer. And not just because of that; also because of changes in the habits of the young population: people are spending more time indoors, youth are offending less. One of the biggest implications of this paper is that everything that people have been saying that caused so much change in the homicide trend … a lot of that discourse is probably overstated, right? So what we need to do now, I think, in my interpretation – what I’m trying to do for my research agenda – is to try to look for policies around the world. Is try to find things that actually work. To try to find one of those things that people say “Hey, I did this legislation. I did this policy change. I facilitated economic growth. I did something that caused the homicide rate to change.” And I’d love to take a look at that trend, to try to age-adjust that trend, meaning to account for the effect of age distribution. To account for everything else that could impact a trend, and to find the effect of that policy very clearly. Because so far what I’ve been finding is that criminal justice policy – and legislations in general – they tend to have a much smaller effect than we would like them to have. And mostly because to commit a homicide is to commit a crime, which means it’s to go against the law. So what the law says doesn’t matter that much, right? You’re breaking it anyways. So, criminal justice is extremely important, and without it crime would skyrocket. That effect has a limit, but we don’t know that limit very well. Law-abiding people tend to think that effect is huge. Why? Because we are afraid of the law, we are afraid of the police, we are afraid of going to prison. But a proportion of society don’t have that same understanding.

[ Back to topics ]

Links to article, bonus audio and other materials

Watkins: That was Mateus Rennó Santos discussing his article, “The contribution of age structure to the international homicide decline,” which he published with Alexander Testa, Lauren Porter, and James Lynch on October 9th, 2019 in the open-access journal PLOS One. You’ll find a link to their paper at parsingscience.org/e64, along with bonus audio and other materials we discussed during the episode.

Leigh: Reviewing Parsing Science on Apple Podcasts is a great way to help others discover the show, especially over the holidays. If you haven’t already done so, head over to parsingscience.org/review to learn how. Or, if you have a comment or suggestion for future topics or guests, visit us at parsingscience.org/suggest … or leave us a voice message toll-free at 1-844-XPLORIT. That’s 1-844-975-6748.

[ Back to topics ]

Preview of next episode

Watkins: Next time, in episode 65 of Parsing Science, we’ll be joined by Luke Chang from the Department of Psychological and Brain Sciences at Dartmouth College. He’ll discuss his research into socially-transmitted placebo effects, through which patients can pick up on subtle facial cues that reveal their doctor’s beliefs in how effective a treatment will be.

Luke Chang: We know that the doctors’ expectations also matter, but it’s never really been quantified … the degree to how much they mattered. And so that was basically one of the things we were trying to set out in the study: can we manipulate the doctors’ beliefs, and then showed that gets transferred over a social interaction that affects the patient’s outcome.

Watkins: Doug and I will be taking a break through the end of the year, so we hope that you’ll join us again in January.

[ Back to topics ]