Because most philosophies that frown on reproduction don't survive.

Tuesday, June 19, 2012

Is Availability of Porn the Key Predictive Factor in Incidence of Rape?

Commenter Joel got off on a tangent on a post last week, insisting that pornography was good for society because it correlated with decreased incidence of rape. He quoted a Scientific American piece from last year which said in part:
Perhaps the most serious accusation against pornography is that it incites sexual aggression. But not only do rape statistics suggest otherwise, some experts believe the consumption of pornography may actually reduce the desire to rape by offering a safe, private outlet for deviant sexual desires.

“Rates of rapes and sexual assault in the U.S. are at their lowest levels since the 1960s,” says Christopher J. Ferguson, a professor of psychology and criminal justice at Texas A&M International University. The same goes for other countries: as access to pornography grew in once restrictive Japan, China and Denmark in the past 40 years, rape statistics plummeted. Within the U.S., the states with the least Internet access between 1980 and 2000—and therefore the least access to Internet pornography—experienced a 53 percent increase in rape incidence*, whereas the states with the most access experienced a 27 percent drop in the number of reported rapes, according to a paper published in 2006 by Anthony D’Amato, a law professor at Northwestern University.

From this Joel concluded, "This doesn't mean that porn is good for any particular man, but all women should be greatful to live in a society where porn is widely available."

Obviously, even if there is a social correlation between porn availability and decreased rates of rape, this doesn't mean that use or creation of porn is moral. Lots of highly immoral activities may happen to correlate with (or even cause) decreases in other immoral activities, and the fact that one of these is immoral doesn't change the immoral status of the other.

However, the whole set of claims sounded fishy to me. "Lowest levels since the 1960s" seems like one of those statement which might, while technically true, still mask a huge difference, rather like "worst economic downturn since the Great Depression". Similarly, measuring statistics based on "the states with the least Internet access between 1980 and 2000—and therefore the least access to Internet pornography" seemed incredibly vague. The internet wasn't even particularly useful for pornography before the advent of the World Wide Web in the early '90s, so the 1980 to 2000 time frame intentionally included a lot of irrelevant time, and measuring the states with the least internet access was likely to simply get you the poorest and most rural states.

Further, I just found the whole proposed causal mechanism fishy. It sounded like the sort of thing where someone fished for a some correlations that worked just a little bit, but they were probably really just seeing some wider trend. My going hypothesis was the the rate of rape would mirror the rate of other violent crimes such as murder. If the rate of rape deviated from the rate of other violent crimes a lot, it would suggest that rape had different causal mechanisms that other forms of violent crime. If it rose and fell in a similar pattern (and I knew that violent crime as a whole had been falling since peaking in the early '90s) that would suggest that rape was just another, particularly nasty, form of social violence.

A little searching around led me to the Uniform Crime Reporting Statistics database maintained by the FBI. From there I pulled data on the rate per 100,000 of population of the set of violent crimes the database tracks: murder, aggravated assault, forcible rape, and robbery. The some total of these is the Violent Crime Rate. The following chart shows those four constituent rates since 1960.

[Click Image for Full Size Graph]

As you can see, the pattern match is very strong. The lowest correlation is between murder and rape, a 54% correlation. Robbery has a 84% correlation and aggravated assault has a 96% correlation. The overall violent crime rate has a 97% correlation with the rate of forcible rape. With correlations that high, if you ask me to tell you what the rate of rape is in the US in any given year, I'm not going to ask you, "Gee, how available was porn that year?" No, I'll ask you, "What was the overall rate of violent crime?" That is a far, far more predictive indicator than anything vague association with porn availability. And predictability is what science is all about. All the rest of what we're hearing is hand waving and self justification.

I then ran averages for each decade. In the 1960s the rape rate was 12.3 per 100,000 of population.

1970s, 26.0
1980s, 36.5
1990s, 38.3
2000s, 32.2

So Prof. Ferguson's statement is actually false, the rape rate is not at its lowest since the 1960s (it was lower in the '70s than it was in 2010 at 27.5) and the current rape rate is more than 2x the rate for the 1960s, when pornography was unquestionably much less available than now. The only way that this pseudo correlation comes to be is that the web has only existed since the early '90s and by coincidence all forms of violent crime have been on a steady decline since the early '90s. Unless one wants to claim that burglary and aggravated assault rates are being driven down by the availability of internet porn, we don't have much of a causal case to make here.

Joel later asked:
But your statement that the more porn => less rape correlation is "very loose" requires pushback. The fact is, those two trends have been observed together all over the world: whenever porn availability goes up, rape goes down. Everywhere. So:

1) You are actually telling us that this is coincidence? Really? It happens again and again, and you still say, "Coincidence"?

2) There's a simple and straightforward explanation as to why this is in fact causation: The safety valve effect of porn is real.


So tell me: what evidence would it take to convince you of this? Consider your answer carefully, because some researchers are probably looking into it right now, and you may rest assured that I will notify you when their results are published.
Given the shoddy nature of the "research" which Joel has quoted thus far, I rather doubt that there's any convincing research being done on the topic that would support Joel's pet claim, but I'll toss out a couple of minimum requirements to suggest that the availability of internet pornography has anything like a real effect on the incidence of rape:

  • Get a large population sample at a young age and track their usage of pornography through regular surveys.  Show that sex crimes are committed significantly more often by those who do not use porn than those who do.  For bonus points, show that this is the result of lack of access rather than preference.  
  • Track two regions which have shown similar trends in rape rate over the last 50+ years.  Significantly decrease the availability of pornography in one of these two regions, and then measure a statistically significant increase in the incidence of rape in that region relative to the other one, while their trends in other violent crimes remain the same.
  • Do the reverse of the previous test: show two regions in which porn has been highly unavailable, and and in which their crime statistics have been highly similar.  Show that liberalizing porn availability in one of these regions correlates with a decrease in incidence of rape in that region relative to the other, while all other forms of violent crime remain on similar trends.
  • Show that the previous two are repeatable in multiple areas.
  • Show that the rate of rape deviates from the rate of other violent crimes in a way that can be explained by some sort of annual data on porn consumption.

Hit those criteria, and I'd think maybe a decent case is starting to be built.

I'll close by knocking the ball into the other court. The fact that the correlation between overall violent crime and rape is so incredibly high got me thinking: Is the rate of rape currently higher or lower than you would otherwise expect? I averaged out the overall violent crimes rates and rape rates from 1960 to 2010 and found that the ratio between these is 0.06. I then used that relationship to predict where the rape rate "ought" to be each year based on the violent crime rate. The result is that rape has actually declined more slowly than you would expect since 1996, based on the overall violent crime rate. it's actually a pretty steady "over trend" pattern for the last 10 years.



Can't imagine why that would be... Wait a minute. Those lines cross in 1999 and rape has been more common than the violent crime rate would suggest ever since (by an average of 9%). It hasn't been the case that internet pornography has been far more available in the years since 1999 than it was in the years before, has it?

(For the record: No, I don't think that's a particularly strong case. But it certainly cuts against the opposite contention which Joel is trying to make.)

UPDATE: In the comments, Joel has expressed grave concern that I used the FBI's Uniform Crime Reporting Statistics, which are based on crimes reported to law enforcement agencies (with appropriate adjustments to deal with apparent over or under reporting.) He avers it would be better to look at the National Crime Victimization Survey, which is based on an annual survey in which the Justice Department contacts 40,000 households and asks all members of those households over 12 what crimes they have been victims of that year. (The purpose of this methodology is to capture crimes which people don't report, for whatever reason.)

I don't think it's correct, as Joel does, to call these the "real" crime statistics, since the real number of crimes is fundamentally unknowable. Both the number of crimes reported to police and the number of crimes reported to DOJ surveyors are going to be subject to necessary sampling errors, and I'm not sure it's possible to say that one set of data actually reflects the true state of crime in the US better than the other. While it's certainly true that for a wide variety of reasons, women who are victims of rape may not report the crime to the police, it's also the case that in the survey some people will necessarily be missed (people living in highly marginal circumstances will be harder to contact), will not want to talk, or will be unavailable to talk (the NCVS only accepts data from the victim herself, which is why it reports no homicide data, this also means that cases of rape followed by murder will necessarily be unreported.)

All that said, there's no reason not to look at this data as well and see if it, like the FBI's data, suggest that the primary predictive factor for the rape rate is the rate of overall violent crime, or if there seems to be some other factor at work. Here's the overall trend chart.


Running the correlation between the sets of annual data for rape and total violent crime, I get a correlation of 89%. There's also a correlation of 91% between robbery and rape and of 91% between aggravated assault and rape. Simple assault shows a correlation of 82% with rape.

This means both that the rate of change in the rape rate, as shown in the NCVS's survey methodology is not unprecedented, and that it one can predict the number of rapes likely to have occurred in a year with a fair degree of accuracy by knowing the amount of overall violent crime. The availability of pornography does not have this predictive value -- or at least, none of these "studies" which attempt to show a connection between the availability of porn and the number of rapes even attempts to put together some sort of annual number which can be used to measure porn consumption or availability, and so any attempt at prediction is impossible.

Really the only thing that Joel's thesis has going for it, according to the NCVS's data is that the incidence of rape took a sharp down turn in 1991 which was three years before the overall violent crime rate took a sharp downturn in 1994. However, since that time the rape and violent crime as a whole have decreased at almost exactly the same rate.

18 comments:

TheOFloinn said...

Statistically, it's not even a correlation. The X and Y are not being measured on the same units, but on large aggregates of units, even entire countries. There is no attempt to show that those not watching the porn are the ones who are raping (or vice versa). You would have to be a social "scientist" [sic] to fall for that. I once did a 90%+ correlation between %women in labor force and %imported automobiles. That doesn't mean women in the labor force leads to importing cars; nor that we can save Detroit by putting women in the kitchen. It just means that two data time series, each undergoing a secular [time-related] trend will always "correlate."
To establish an association, one ought to employ logistic regression on individuals with X=%time spent watching Inet porn and Y=rape/did not rape. Good luck getting that data. Although it might be possible to determine for convicted rapists whether they spend much time on porn and then do the same with a sample of non-rapists, and do a chi-square.
Of course, people who spend inordinate amounts of time in front of computer screens often don't get out of their mother's basement to actually do much of anything in the physical world. Doesn't have to be porn. Could be video games, or memes.

James Chastek said...

I'm fascinated by Joel's suggested mechanism: the "safety valve effect" of porn. The same hydraulic account is given for anger too, but it overlooks the role of habit-formation and the need to increase stimuli. Saying porn relieves violent sexual desire is analogous to saying that smoking relieves the desire for nicotine: it's true, but certainly not the whole truth. The desire to smoke can be answered with the "safety valve effect" of smoking, but the metaphor hides the fact that the way we are reliving the pressure is itself building up pressure in the system, and doing it more often. I think I we translated this into empirical terms and tried to test it, we would get an experiment like the first one Darwin suggests, and best hypothesis to start with is that violent sexual desire will be more common among habitual porn users than among those who do not consume porn. To my mind, this is more an axiom than a hypothesis, but we could test it if we wanted.

Suzanne said...

I would just like to point out that incidents of rape in the 1960s were likely significantly under-reported compared to incidents in the 2000s. Law enforcement often did not take rape accusations seriously at that time, and often let likely perpetrators go free. Just another factor to consider.

Darwin said...

OFloin,

Agreed, the application of the term 'science' to 'social' informing 'social science' is often something of a stretch. The level of work going on in some of the social science papers one sees published is just shockingly low, and those centering around sex are particularly afflicted. Though to be fair, the Scientific American article did use the more correct term of "association" rathe than "correlation". I slipped in using the latter since that's what I was looking at,but obviously, the claim about porn availability and rate of rape isn't even a claim of correlation, just association, since there isn't actually an ordered data series for porn availability.

Suzanne,

I'd agree that rape was probably significantly under reported in the past -- particularly when dealing with the poor and with minorities. I think that, given that Joel's thesis would need pretty much a complete inverse of the data prior to 1990, that the data withstands that, however. And it's probably also the case that other forms of violent crime were also under reported when committed against the same types of people, though perhaps not to quite such a great extent, so that may erase a lot of the effect.

TheOFloinn said...

It's still the case that two data series will always correlate when they are both trending during the same time frame, even when there is no causal connection at all between them, or even a lurking variable affecting both. There have been positive correlations between hemlines and stock prices, between Pacific salmon and sunspots, and between the size of the universe and the size of my suits. But if I lose weight, the universe will not implode.

Scientific theories are underdetermined; but they are also overinterpreted.

mary said...

Darwin...you are just so darn awesome!

TheOFloinn said...

Here's a comment from a statistician on another paper in the campaign:

http://wmbriggs.com/blog/?p=3357

August said...

What about the relationship between violent crime and the availability of violent movies?
The 'porn reduces rape' argument can be abstracted out to include other violent crimes. Torture, murder- maybe if they can sit on the couch and watch it they get lazy and don't act it out.
Personally, I think the largest factor is that the U.S. incarcerates way too many people- so many of those who would do the crimes are already in jail.

Anonymous said...

Darwin, the DoJ says that rape is the most under reported crime. Your total reliance on reported rape, and your complete inattention to the easy-to-find figures regarding actual rape, is disappointing. In particular, your charts are pretty much worthless, and your denial of Prof. Ferguson's statement is hard to excuse.

A more accurate chart, with highly debatable (but at least fact-based) conclusions, can be found here:

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=913013


Joel

Darwin said...

Joel,

You have got to be kidding me. It's one thing to be taken in by a fluff piece in a pop-science magazine, it's another to be willfully ignorant and to cite such an incredibly badly written paper. I mean, seriously, this guy tries to derive a confidence interval based on eight pieces of data. It's no surprise this paper does not appear to have actually made it into a journal. Further, he first makes a big deal about the decline starting back in the 70s because of porn magazines and movies, then proceeds to base his incredibly sketchy attempt at statistical analysis on the availability of internet -- ignoring the magazines and movies he claims to have been such a powerful force. He totally fails to control for overall crime levels, poverty, etc. And he looks at only 8 states when he could very easily have looked at all 50. (My guess is he probably did but discarded that when it didn't fit his thesis. There's no reason to arbitrarily select only the top four and bottom four in relation to internet access unless that's the only way he could get the result he wanted.)

I did in fact find what you call the "actual rape" figures first, but since you cited an article which claimed to compare the 1960s with the present, I needed a data set which went back at least to the 60s, which the FBI data does. Further, there are a fair number of concerns about the interview approach as well. You act as if the survey approach is sure to be more accurate than a crime statistics approach, but clearly in both cases one relies on whether someone has the opportunity to report (contacts the police or is contacted by a surveyor) and on whether the victim feels able to respond honestly. I certainly agree that the FBI figures under-report the number of actual forcible rapes, but neither is it necessarily appropriate to immediately assume that the under-reporting was far higher in the past, or that the survey based information is to be blindly trusted above the FBI data set.

At a minimum, the analysis I did here is more fact based than Prof. D’Amato's in that I looked at a full data sequence of government collected data on rape incidence and compared that to a similar full ordered set of annual data on another set of potentially predictive factors. D'Amato doesn't come close to doing that, and it's unclear whether that's because he's being disingenuous or because he's so utterly clueless about what is required to produce a valid statistical correlation.

Brandon said...

A more accurate chart, with highly debatable (but at least fact-based) conclusions)

Well, that was a waste of time to look at. So that no one else has to waste theirs, you can actually get the chart that D'Amato plagiarized, and an explanation of how it was obtained, here:

http://bjs.ojp.usdoj.gov/content/glance/rape.cfm

The data spreadsheet, which is the thing that is actually needed in this context (in part because it allows comparison with violent crime rates with this method of collecting information), is here:

http://bjs.ojp.usdoj.gov/content/glance/tables/viortrdtab.cfm

To say that D'Amato's conclusions are 'highly debatable' is a bit of an understatement; although I liked his airy and unsupported dismissal of better education for boys and girls, shifts in patterns of behavior associated with drug use, and incarceration as all "minor factors" that "cannot begin to explain such a sharp decline," in comparison to internet usage.

TheOFloinn said...

It is always educational when economists and lawyers hold forth on statistics.
Consider that the reported rates show the selfsame pattern for several different categories of crime, rising, peaking, and falling in synch. If I saw this happening on four different packaging machines (as I did once upon a time) I would immediately suspect what we call a common cause, that is, a cause that is common to all four machines (or crime types). In the case of the Kwiklock packaging machines that turned out to be the lot of carton flats from which all four machines were drawing.
It is difficult to suppose that the availability of internet porn accounts for the matching declines in murder&negligent manslaughter, aggravated assault, and burglary

Anonymous said...

Darwin, well duh. D'Amato is an ideologue who grinds his axe hard in this paper. But the chart is accurate, and none of your figures touch it. Real rape, as opposed to police-reported rape, shows a completely different trend from the ones you based your posting on. Could you please revise your posting based on real data, so we can have a real debate?

Joel

Paul Zummo said...

Wow, Joel angrily banging his fists on the table and yelling "NO! My stats are better!" sure has me convinced. Good thing porn has no long-term impacts on people's behavior and attitude.

Anonymous said...

Paul, Google "under report rape" and spend five minutes educating yourself. Darwin's entire posting is based on figures that are off by, at best, a factor of two, as every law enforcement agency in the US acknowledges. And it's not like better data is hard to find, as Brandon helpfully noted above.

It's Darwin's blog, of course, and if he's comfortable having misleading data on his front page that's totally his business. But I figure part of the purpose of comments on a blog is to enable people to point these things out.

Joel

TheOFloinn said...

This Joel dude seems to be unclear on the nature of statistics. The UCR system tallies reported crimes, adjusted for estimated under-reporting. Not all robberies are reported, for example. The series in its current sense goes back to 1958. (Prior to that, it included statutory rape without use of force.) "The UCR Program compiles data from monthly law enforcement reports or individual crime incident records transmitted directly to the FBI or to centralized state agencies that then report to the FBI. The Program thoroughly examines each report it receives for reasonableness, accuracy, and deviations that may indicate errors. Large variations in crime levels may indicate modified records procedures, incomplete reporting, or changes in a jurisdiction’s boundaries. To identify any unusual fluctuations in an agency’s crime counts, the Program compares monthly reports to previous submissions of the agency and with those for similar agencies." Because different jurisdictions may define crimes differently, the reporting agencies are required to apply a uniform crime definition, to weed out unsubstantiated claims, and so on.

The NCVS goes back to 1973 (which means it is unable to shed light on crime rates earlier than that) and is garnered from a stratified, multi-stage cluster sample of about n=40,000. The primary sample unit is a county or a large metropolitan area. Large units are completely covered; smaller units are sampled. The sample size was reduced in 1996. "No attempt is made to validate the information against police records or any other source." Thus, while NCVS includes crimes not reported to the police, it also includes incidents which the respondent believes were crimes: drunken seduction, esp. during the 70s, might be reported as rape; sharp dealing might be reported as robbery. Because it is a sample, NCVS is subject to sampling uncertainty.
Data for both series are available from the Statistical Abstracts of the United States. One may use this data to make comparative graphs of UCR and NCVS over the same time frames for rape and for other crimes, and study whether the pattern is the same between reporting methodologies and across different kinds of crime.

Rebekka said...

I don't know about China and Japan, but in Denmark prostitution is legal but not regulated as it is in Holland. Even if you don't consider trafficking, sex slavery, drug abuse, etc, prostitutes as a population endure a massive amount of sexual assault and abuse (and I don't mean what they're paid for). So merely remarking that the incidence of rape has fallen since porn became readily available is not worth much.

Darwin said...

I've added an update dealing with the NCVS data. Short answer: the correlation is almost as strong between total violent crime and rape in this survey as in the FBI's UCR data. Violent crime in general continues to have a very strong predictive power in relation to rape, while none of the porn advocates have managed to even come up with a porn metric which would allow them to calculate any correlation that might exist. However, since violent crime itself is so predictive, I don't see much room for a "porn effect".