Brief: Do trust levels predict sexual activity among the young?

I'm becoming more convinced that the decline in trust levels during the late 1980s may have a lot to do with the decline starting a few years later in all sorts of risky behavior. The logic is simple: the more (or less) trusting you are of others, the more (or less) risk you're willing to take in social life, and also exploiters will have a more (or less) easy time finding "suckers." We looked before at homicide rates. Now let's turn to sexual activity among young people.

For trust levels, I use the General Social Survey's question on whether you think others can be trusted, and restricted respondents to those from 18 to 25 years old. The GSS doesn't survey minors, so I used this age range for "young people." The time series for trust doesn't seem to vary wildly based on which age range you restrict it to, so it seems OK to use trust levels among young adults as a proxy for trust levels among adolescents. For risky sexual activity, I use the pregnancy rate among females ages 15 to 17 (here), which are yearly from 1972 to 2006; and the percent of high schoolers who have had sex before age 13 (here), which are bi-yearly from 1991 to 2007. The trust data are from 1972 to 2008, and coverage is usually bi-yearly or more frequent.

Here are the results. Trust levels are in blue and sexual activity measures in red.



Although the relationship is not perfect, there's a close match overall, and the timing looks like the trust level changes first, followed by a change in sexual activity. That fits with trust being the cause and risky behavior the effect. The correlation across years between trust and the pregnancy rate is +0.47, and between trust and the early sex rate it is +0.44. The Youth Risk Behavior Survey that I got the early sex rate data from has three other measures of teenage sexual activity, and the relationships are similar but not quite as strong.

The correlations with trust are: +0.15 for percent of high schoolers who'd had sex at least once in the past 3 months; +0.21 for the percent who've had 4+ partners in their life; and +0.23 for the percent who've ever had sex. * I also created an index of young sexual activity, which is just the sum of the fraction of students responding positively to each of the 4 questions. This is the expected number of "yes" answers that a high schooler would give while reading off the check-list of risky sex behaviors. The correlation between this index and trust is +0.25.

In general, I think these correlations are weaker just because there are fewer data -- 1991 to 2007, every other year -- than in the case of pregnancy rates. I suspect that the pre-1991 picture for the 4 behaviors surveyed in the YRBS would have looked highly similar to the teen pregnancy rate, so if we had those data, they would probably make the pattern even stronger. In any case, it seems clear that, while not the entire story, how trusting people are of one another plays a substantial role in how willing young people are to engage in risky sexual behavior. That's not too surprising if we think of trust as a form of insurance, but it's still something that's been completely overlooked as far as I know from the social science lit on these two topics that have previously been studied independently of each other.

If I can find good data on cross-national differences in number of lifetime sex partners, or other measure of promiscuity, I might use the national trust level data from the World Values Survey and turn this into a fuller post. It depends on how easy the former data are to get. Informally, though, my impression is that the low-trust countries are more sexually conservative, while the high-trust ones are more sexually liberal.

* The sexual activity correlations paired the sex activity variable and trust variable for the same year if possible (for 1991 and 1993), but since the data use different sets of alternate years after that, I paired the sex variable in a certain year with the trust variable in the following year.

GSS variables used: trust, age, year

Brief: Trust and crime

I've been thinking more about trust and its effects on all areas of social interaction and culture. It's at the root, really: if you don't trust others, you will try to get by on your own. Sociability requires a certain level of trust, and in a social species like ours there are gains to be had by interacting with others -- especially in a modern market economy where you can outsource so many things to others in the market instead of doing them yourself. You don't make your own shoes, grow your own food, manufacture your own computer, or gather your own news about the President. Sounds great -- why would you possibly withdraw from that sociability?

If you trust others, you're making yourself vulnerable to exploitation. The more trusting people there are in a population, the easier it will be for exploiters to thrive, and that will drive up their numbers. The exploiters may become so numerous, and their effects so offensive, that people start to withhold their trust lest they become the next statistic. That state-of-mind could show up in behavior by simply not venturing out into the public sphere where they'd be vulnerable, hiding out in the safer personal sphere.

But then when there are far fewer trusting people, that dries up the resource that exploiters had been thriving on. Now there are lots of them competing to exploit a shrinking number of trusting people. So that will drive down the numbers of exploiters. Eventually people will realize how safe things have become and extend their trust once again, which will in turn drive up the numbers of exploiters as before, and the cycle repeats.

This variant on a model of how hosts and parasites, or predators and prey, interact in ecology suggests looking at data on how trusting people have been over time, and what the crime rate has been like. Before I showed that people rationally respond to changes in the homicide rate by becoming more afraid when it's going up and less afraid when it declines, with about a 2-year lag. So we know that perceptions are affected by crime levels -- but could crime levels be affected by perceptions, i.e. how trustworthy you think other people are?

The General Social Survey asks respondents whether other people can be trusted, cannot be trusted, or that it depends. Here is a plot over time of the percent of respondents who said that other people can be trusted (in black), together with the homicide rate (in red):


Clearly the relationship is less stark than it was for fear resonding to crime, where the correlation within a year was +0.74. Still, the correlation here is +0.53, meaning that indeed higher crime rates are associated with higher levels of trust. And as in the case of fear and crime, that understates the strength of the relationship since there is a time lag. In particular, trust levels started falling steadily before the homicide rate did so. So the predicted relationship is there: crime rates are high throughout the '70s and '80s, and at some point in the mid-late-'80s, people have had enough and start trusting others less and less. With far fewer trusting people to exploit, criminals start scaling back or dying off: the homicide rate peaks in 1991 and starts falling steadily afterward.

We will have to wait a few more decades to see if the rest of the story pans out -- whether with such low crime rates, people will assume it's safe to start trusting people again, and whether those higher levels of trust (if they did happen) would be followed by a rise in the crime rate.

Criminologists and others debate what causes crime to go up or down -- tougher or more lenient punishment by the government, technology that makes it easier or harder to report crimes, etc. I have only a passing familiarity with the range of causes they discuss, but as far as I know, how trusting the average person is does not get much attention, although there may be some minority contingent that does look at it. Perhaps it is as simple as a rise in crime following an increase in the number of people who criminals would consider "suckers," and a decline in crime following a contraction in the numbers of "suckers," just as we see between a host species and a parasite species. The logic is hard to argue against, and the graph above shows that it has some modest empirical support as well.

GSS variables used: trust, year

Brief: Season of birth and signaling anger

The various questions in the General Social Survey that probe people's personalities -- for example, whether you consider yourself outgoing -- don't show strong differences in the responses by birth month. But what if we look at questions that focus more on people's actual behavior? I found one question that asks how much the person agrees with the statement, "When I'm angry, I let people know." Here is the percent who either strongly agree or agree by their birth month:


A majority in all groups signal their anger to others, but there is a very clear seasonal pattern, with this tendency increasing during the fall, winter, and spring birth groups, then falling during the summer group. Moreover, the change is not trivial -- 11.2 percentage points separate the low in August from the peak in April.

This could just be a reflection of genetics rather than environment, whereby anger-signaling mothers pass on their variants for the trait, but where they're more likely to conceive in the later part of summer. Still, I favor an environmental story where babies born into more stressful environments assume unconsciously that life is going to be tough, and so adopt a more defensive "don't mess with me" posture. Most such stress today will not come from parental strife or lack of basic nutrition but from the one source that we still don't have much control over -- pathogens that make us sick. That's why the flu season shows up pretty well in the graph.

There could also be a vitamin D story, where newborns produce less vitamin D during the less sunny months (sunlight hitting the skin is a necessary step), and that this somehow influences their personality or behavioral strategy. But without a clearer mechanism, I wouldn't put much emphasis on this -- why wouldn't the mother's vitamin D level matter, which would tend to go in the opposite direction? In any event, there's something neat to try to explain.

GSS variables used: showangr, zodiac

Just how permissive are the beliefs of non-heterosexuals?

Studies of gay and bisexual men show that they are more sexually permissive in their behavior, and likewise for bisexual women compared to straight or lesbian women. But sexual behavior is always constrained by the simple fact that it takes two. Why don't we look at their beliefs on sexually permissive behavior, which might give us a purer view of the differences between heterosexuals and non-heterosexuals.

(I collapse gay and bisexual males into a single non-heterosexual group, but preserve the three-part division among females, because that's what the findings on arousal argue for. Men are aroused either by male or female erotic images, but not by both; whereas some women truly are aroused by both.)

We will treat "sexual permissiveness" as a personality trait or preference that's too tough to measure precisely, but that we can investigate using La Griffe du Lion's method of thresholds. We imagine sexual permissiveness as a continuous trait -- some people score really low, others really high, and everything in between. At some point toward the high-end of this spectrum, we set a threshold and ask what percent of the various groups score above that threshold? We can then use this to figure out how far apart the average scores of the various groups are, assuming the trait to be normally distributed. Knowing that, we can also predict what ratio of one group to another we would expect at some extreme value of the trait -- far out into the right tail.

To identify a clearly high-end threshold for sexual permissiveness, consider what your beliefs are about two 14 to 16 year-olds having sex. The General Social Survey asks just such a question: is it always wrong, almost always wrong, sometimes wrong, or not wrong at all? As you look at higher values of sexual permissiveness, at some point you reach the value where the person is permissive enough to think that teenage sex is OK (maybe with provisos) rather than always wrong. Here is a picture to help see what I mean:


Along the spectrum of permissiveness, at the low end we find people who believe that oral sex is OK -- people whose value is to the left of that point are not even that permissive -- while farther to the right, we find people who are permissive enough to believe that teenage sex is OK. Farther to the right still, we find people with ever more permissive beliefs, perhaps going so far as to condone public unprotected sex. A group of people will have a distribution of values along this spectrum, and we'll assume it's normal for convenience and because most personality traits are made to be normal too. We also assume the various groups have the same variance.

By finding out what percent of some group lies to the right of our threshold point, we can work backward to determine what z-score this threshold represents for the group. Doing this for two groups, we can find the difference between the means of each. For example, if 16% of group A exceeds the threshold, while just 2.5% of group B exceeds the threshold, that implies that the threshold is +1 S.D. above A's mean and +2 S.D. above B's mean. That means that there is a difference of 2 - 1 = 1 S.D. between the two group's means, favoring A.

Furthermore, knowing how far apart the two groups are on average, we can extrapolate what the A-to-B ratio would be at an even further extreme value that we haven't even observed. Obviously there will be more A's than B's (if the two groups are the same size), but the question is by how much. These extreme value predictions are useful because we find extremes more fascinating than averages, and it may be harder to get honest answers out of people as we ask more and more extreme questions.

Now back to sexual permissiveness. The GSS allows two ways to figure out who is heterosexual or not. One is straightforward and asks what sex your sex partners have been over the last 5 years: only male, male and female, or only female. They also ask how many male (or female) partners you've had since age 18. If a male answers 0 partners, he's straight; if he answers 1 or more, he's not. You can do the same for females; bisexuals have to answer 1 or more partners for both questions. The sample sizes are a bit larger using the second method, but the results are very similar. So I'll show the graphs for both methods, but I'll only use the second one in applying the method of thresholds.

First, here are the graphs for male heterosexuals vs. non-heterosexuals, for the first and then second method:



Straight males are more likely to say teenage sex is always wrong, while gay males are more permissive -- although a bare majority of them still think it's always wrong. Next, the differences among females for the first and then second method:



Notice first that females overall are less permissive than males -- no surprise there. Lesbians are more permissive than straight females, but not nearly by the same gap as the one separating gay and straight males. However, there is a yawning chasm separating bisexual females not only from the other female groups but even from both male groups! This result about their beliefs is consistent with their sexual behavior, as surveys (including the GSS) routinely show that bisexual females lead over-sexed lives. They have a general "wild child" mindset, so it shouldn't be surprising that those who think teenage sex is always wrong are in the minority among bisexual women.

(I've looked at responses to this teenage sex question across all sorts of demographic groups, and aside from GSS respondents who are youngsters themselves, I haven't found a single group where the majority think it is OK, let alone one where one estimate puts them at over 60%. Very wild indeed.)

Comparing gay to straight males, we find a difference in means of 0.31 S.D. favoring gay males. Comparing lesbian to straight females, there is a 0.17 S.D. gap favoring lesbians. And comparing bisexual to straight females, there is a 0.68 S.D. gap favoring bisexuals. To make this more intuitive, let's pretend we were talking about height instead of sexual permissiveness. It's as if gay men were 0.92 inches "taller" than straight men on average; as if lesbians were 0.51 inches "taller" than straight women on average; and as if bisexual women were 2 inches "taller" than straight women on average. You might notice the gay-straight gap in the real world, probably not the lesbian-straight gap, but almost certainly you'd notice the bisexual-straight gap.

Now suppose we wanted to predict how much one group would predominate at an even more extreme value along the permissiveness spectrum. That might be holding the belief that it's OK to have unprotected sex in public -- or actually having done so. These extremes are harder to investigate directly because the events that would tip us off to people being there are incredibly rare, and asking people if they've ever had unprotected sex in public may not give us very reliable answers. So we turn to the method of thresholds and ask what ratio of non-heterosexuals to heterosexuals we predict to find at, say, the value of 3 S.D. above the heterosexual average.

Obviously this won't tell us what to expect in the real world since the straight and non-straight populations are very different in size. Still, we'll pretend they were the same size in order to highlight how different the various groups are. In a moment, I'll re-adjust these ratios to reflect the groups' real sizes in order to get better real-world predictions.

At the extreme value of 3 S.D. above the straight male mean, we expect a gay-to-straight ratio of 2.6 to 1. At the value of 3 S.D. above the straight female mean, we expect a lesbian-to-straight ratio of 1.7 to 1, and a bisexual-to-straight ratio of 7.6 to 1! Clearly the minds of heterosexuals and non-heterosexuals are not the same on average.

Correcting for the huge differences in population size between straights and non-straights, I've used the GSS to figure out what percent of the male and female population are gay, straight, or bisexual. For males, 96% are straight and 4% gay. For females, 96.4% are straight, 2% are lesbian, and 1.6% are bisexual. In the real world, then, we expect 9.2 straight males for every gay male at the extreme; 28.1 straight females for every lesbian; and 8 straight females for every bisexual female. The larger point remains, though: non-heterosexuals, especially bisexual women, are more permissive in the beliefs -- and so presumably in their actions -- than heterosexuals.

As an analogy, consider the fact that Ashkenazi Jewish people score much higher (about 1 S.D.) on IQ tests than other Europeans. Still, because Jews make up such a small fraction of the overall population (about 2%), it will still be the case that non-Jewish Europeans will outnumber Jews at extreme IQ levels. The first fact tells us that a representative person from the two groups will be pretty different, perhaps due to their ancestors facing different evolutionary selection pressures or due to having different hormone levels or whatever. The second fact tells us that we shouldn't let the first fact lead us to expect a majority of the unusual group at unusual levels, unless the two groups are similar in size.

GSS variables used: teensex, sex, sexsex5, nummen, numwomen

Brief: Is the "culture of fear" irrational?

We hear a lot about how paranoid Americans are about certain things -- people in the middle of nowhere fearing that they could be the next target of a terrorist attack, consumers suspicious of everything they eat because they heard a news story about it causing cancer, and so on.

Of course, we could be overreacting to the magnitude of the problem, as when we panic about a scenario that has a 1 in a trillion chance of occurring but that sounds disastrous if it did happen. It's not clear, though, what the "appropriate" level of concern should be for a disaster of a given magnitude and chance of happening. So the charge of irrationality is harder to level using this argument about a single event.

But we also get comparisons of risk between events wrong, for example when we fear traveling by airplane more than traveling by car, even though planes are safer. Here the case for irrationality is straightforward: for a given level of disaster (say, breaking your arm, dying, or whatever), we should panic more about the more probable ways that it can occur. The plane vs. car example makes us look irrational.

Still, there's another way we could measure how sensible our response is, only instead of comparing two sources of danger at the same point in time, comparing the same source of danger at different points in time. That is, for a given level of disaster, any change in the probability of it happening over time should cause us to adjust our level of concern accordingly. If dying in a plane crash becomes less and less likely over time, people should become less and less afraid of flying. When I looked into this before, I found that the NYT's coverage of murder and rape has become increasing out-of-touch with reality: while the crime statistics show the murder and rape rates falling after the early 1990s, the NYT devoted more and more of its articles to these crimes. So at least at the Newspaper of Record, they were responding irrationally to danger.

But what about the average American -- maybe the NYT responds in the opposite way that we might expect because when violent crime is high, people see and hear about plenty of awful things outside of the media, so that writing tons of articles about murder and rape wouldn't draw in lots more readers. In contrast, when the society becomes safer and safer, an article about murder or rape is suddenly shocking -- just when you thought things were safe! -- and so draws more readers, who start to doubt their declining concern about violence.

The General Social Survey asks people whether there's any area within a mile of their house where they're afraid to go out at night. Here is a plot of the percent of people who say that they are afraid to go out at night, along with the homicide rate for that year:


Clearly there is a tight fit between people's perception of danger and the reality underlying that fear. The Spearman rank correlation between the two within a given year is +0.74. That's assuming that people respond very quickly to changes in violence; it might be even higher because there appears to be somewhat of a lag between a change in the homicide rate and an appropriate change in the level of fear. For example, the homicide rate starts to decline steadily after a peak in 1991, but people's fear doesn't peak until two years later, when it too steadily declines. That makes sense: even if you read the crime statistics, those don't come out until two years later. To respond right away, you'd have to be involved in the collection and analysis of those data. The delay is more likely due to people hearing through word-of-mouth that things are getting better -- or not getting negative word-of-mouth reports -- and that it takes awhile for this information to spread throughout people's social network.

Putting all of the data together presents a mixed picture on how rational or irrational our response to the risk of danger is. But here is one solid piece that average people -- not those with an incentive to misrepresent reality, either in a more negative or more positive way -- do respond rationally to risk.

GSS variables used: fear, year

Brief: Do people vote selfishly when it comes to vices?

In Bryan Caplan's eye-opening book The Myth of the Rational Voter, he devotes some time to the literature showing that voters do not vote their narrow self interests. The well-to-do favor social safety net programs, men are typically more pro-choice, and so on. In general, people claim to and do vote for what they believe will make society better off. He says that the exceptions are personal vices, such as smokers being much more against smoking bans than are non-smokers. I went to the General Social Survey to see what other examples I could dig up on this pattern.

First:


Those who saw an x-rated movie last year, compared to those who didn't, are much more likely to want to keep pornography legal for adults, rather than ban it altogether.

Second:


Among married people, those who have cheated on their spouse are much more likely to want easier divorce laws. The idea is that if we had stricter laws, their vice could be more harshly punished, say by having to pay massive damages in divorce court if it were uncovered.

Next:


The more sex partners a woman has had in the past year, the more willing she is to support abortion for any reason whatsoever. The idea is that for such women, abortion is one form of birth control, and lacking this method of last resort would constrain their ability to indulge their vice of sleeping with a variety of men.

Finally, an exception to the "vice leads to selfish voting" rule:


These two graphs show that illegal drug users feel the same way as non-drug users about how well our drug policy is doing, rather than view it as too harsh or unjust, as we might have expected from the previous three cases. This case is different in that the vice is illegal, while the other three vices are perfectly legal. Obviously they could not vote selfishly because there are no "legalize it" pieces of legislation on the table.

But even when you just ask them their opinion, they still don't espouse the view that promotes their own self-interest. Perhaps the vices that we criminalize, on average, really are more harmful than those that we don't criminalize -- shooting heroin really is more ruinous than cheating, sleeping around, or watching porn. If that's so, then the heroin addict doesn't view his vice as something that's unobjectionable, and so doesn't view tough drug laws as an untenable constraint on his liberty, in the way that a porn addict would view his own vice and the attempts to criminalize it. He probably recognizes that shooting heroin is something that people should be protected from by making it harder to try out. The porn addict, by contrast, realizes that it never really hurt anyone, so people should not be protected from a false menace.

So, by experiencing how destructive illegal drugs are first-hand, users put on their "do what's best for society" hat and confess that current drug laws aren't as bad as ivory tower detractors might think. The other vices don't appear to destroy society, so those who indulge in them don't think about what's best for society -- you only get into that mindset when you perceive that something is a real problem that needs to be solved.

GSS variables used: pornlaw, xmovie, divlaw, evstray, abany, partners, sex, natdrug, hlth5, evidu

What predicts income dissatisfaction?

Let's test the idea that as people attain higher status, they believe that their income is increasingly not enough to cover their expenses -- after getting that big promotion and moving to a better neighborhood, you've got more income, but it sure seems like life is more expensive, what with the private tutors and sports coaches for your kids, higher home prices, replacing the IKEA furniture with stuff from Crate & Barrel, etc. Life felt more affordable before.

The General Social Survey asks the following question:

Do you feel that the income from your job alone is enough to meet your family's usual monthly expenses and bills?

Note that the wording does not refer to things in the "take only what you need to survive" category. If you pay for tutoring, that's a usual monthly expense. If you go out to McDonald's several times a month, that's a usual monthly expense. So it's referring to whether the person's income is enough to support their broader lifestyle.

If the above idea is right, the percent who answer Yes should decline as we go up the status scale. In fact, the effect should be especially strong since this question was only asked in 2002 and 2006 -- right before the recent boom, and then during the height of the boom. People were furiously maxing out their credit cards, and they were surrounded by other people with lots of cool new stuff that they bought using maxed out credit cards. A feeling of income inadequacy at the higher status levels should have been particularly acute.

Out of curiosity, I also checked to see how income dissatisfaction varied within other demographic groups. Here are the results (click to enlarge):



The lowest income groups are the least satisfied, although once family income hits $20,000, dissatisfaction stays pretty constant. However, the very rich actually feel even more satisfied than other income groups. The same pattern holds for job prestige (socioeconomic index), education level, and self-reported economic class. Income dissatisfaction is basically constant across IQ levels (the lowest levels, 0 - 2, show funny results because of very small sample sizes). So, there is no evidence that higher-status people -- however measured -- are more bitter about their income level than lower or middle-status people. Quite the opposite.

As for age, surprisingly there is no big change from the college years through middle age, although by the time people are in their 60s, they are a bit more satisfied than before -- perhaps due to the falling need for social preening and oneupsmanship by then. Before then, though, more than 50% of each age group feels inadequate -- just unhappy in its own way (not getting a hot car when young vs. not finding a nice home when older).

Also surprisingly, there are are no race differences at all. We might have expected whites to feel more satisfied since they score higher on average for the above measures of social status. I fooled around by restricting the social status variables to the higher end or lower end, and it didn't seem to matter -- blacks and whites were still basically dead even. What they spend their money on may differ, but no racial group feels more or less satisfied than any other.

Finally, there is a gigantic sex difference, dwarfing anything else we've seen. While 57% of men are satisfied with their income, only 34% of women are. We can't attribute this to women wanting to shop more, or for more expensive things -- men have their eyes on super-expensive cars and other auto-related junk, amateur sports equipment, big flashy gadgets and electronics, and so on. And men don't make that much more than women, so it can't be due to differences in income. I think it must be a personality difference rooted in the brain's wiring. In questionnaires, women consistently score much higher on the personality trait Neuroticism than men. And this shows up in the real world, as they suffer more from depression than men. The feelings of anxiety, self-doubt, and impulsiveness that characterize Neuroticism should show up when we ask people if they're satisfied with their income level.

These other demographic findings are pretty neat, but the most robust and important finding is that income dissatisfaction does not increase as you move up the status scale. We may have this misperception because those higher-status people who are not satisfied are the ones most likely to speak up about it, and the ones who the media are most likely to cover because of the schadenfreude value it provides the readers with. No one wants to write or read a news feature on a bunch of people who say, "I'm rich and educated -- and it is all it's cracked up to be!"

GSS variables used: rincblls, race, sex, age, class, sei, wordsum, educ, realinc

Brief: Have we gotten more or less sympathetic since Adam Smith's time?

In his Theory of Moral Sentiments, Adam Smith made the observation that we care less about the disasters that befall others if they are remote and faceless, while we panic at much smaller hardships of our own. But he wrote that before the Industrial Revolution really took off, and so before the peace-and-life-valuing merchant classes genetically replaced the old warring aristocratic class. In the meantime, capital punishment has been widely banned across Europe and its off-shoots, we have laws against cruelty to animals, and we have TV and other media that provide us with vivid daily images of the troubles that beset people in far-off places. So, does his observation still hold up in our more bleeding heart times?

To check this, I searched the NYT for its coverage of a rival first-world country, Japan, and a non-threatening third-world country, Indonesia. If newspapers cover a country out of sympathy for their plight -- and this supply would reflect demand for such coverage -- then there should be declining coverage of Japan during their incredible boom of the 1980s, but increasingly more coverage as they slid into the Lost Decade of the 1990s and even early half of the 2000s. Similarly, coverage of Indonesia should spike during 2004 - 2005 in the wake of the disastrous tsunami, which hit Indonesia much harder than other countries. (This event was pretty close to Smith's hypothetical earthquake that swallowed all of China.) There should also be a spike during their financial crisis of 1997 - 1998, although this was not a deadly catastrophe that could provide the level of gory detail as a tsunami, so this spike should be smaller than that of the tsunami coverage.

Here is the NYT's coverage of these countries:




Counter to the sympathy hypothesis, coverage of Japan shot up during the 1980s, as Americans began to fear more and more that Japan was going to economically take over our country. When Japan slid into a long recession, those fears evaporated, and the supply of alarming stories declined as a result, all the way to the present. This is not to say that there was no demand for sympathy stories about the Japanese recession, only that fear is a stronger driver of coverage than sympathy.

The Indonesia data give some support to the sympathy hypothesis. There is indeed a two-year increase for 2004 - 2005, reversing a previous steady decline. After that moment of sympathy, though, coverage starts to decline again. Moreover, the upward blip during the tsunami is tiny compared to the skyrocketing coverage of 1997 - 1998, when Indonesia was rocked by a financial crisis. This reflects our panic that the Asian Financial Crisis would infect our own economy. Before this threat to us, we paid relatively little -- and consistently little -- attention to Indonesia.

On the whole, the coverage data support Adam Smith's claim, despite our being more emotionally sensitive than people were during his time. However, there is no real paradox here if we view the demand for sympathy stories -- and the coverage that supplies that demand -- as a function of both some baseline concern for others plus a component that responds to actual disasters. Let's say that sympathy is a simple, linear function of disasters:

S = b + r*D

Where S is level of sympathy, b is our baseline level in the absence of news about disasters, D is the level of disasters that we hear about, and r measures how responsive our sympathy is to those disasters.

All that Smith was saying is that r is greater for disasters that are nearer to us than for disasters that are more remote. He made no claim about our baseline level. The genetic and cultural proliferation of the merchant classes -- and the concomitant doing away with public executions, slavery, etc. -- may have increased the baseline without affecting how r responds to disasters at different social distances. This seems like a useful distinction to draw, as it clears up a lot of the confusion about whether we've become more selfish or sensitive in recent centuries.