In West Side Story Stephen Sondheim set out the theories of juvenile delinquency with more clarity, and certainly more brevity, than the academics who had dreamed them up. A prominent theory in sociological circles is that crime arises from poverty and consequently that the alleviation of poverty by paying social benefits should diminish criminality.
The link between poverty and crime has been demonstrated repeatedly, and recently confirmed for USA and Norway. Repetition of a correlation impacts academic and public opinion. However, as we are wearily cognizant of, correlation is not causation, though in ordinary life it damn well implies it. Correlation is a necessary feature of causation, but not a sufficient proof. The quip should be altered to: correlation is not always causation, but it helps.
This link has been investigated, in a different way, by a gang of sociologists led by Amir Sariaslan (ex-Uppsala) and his colleagues at the great Karolinska in Sweden, the country of Volvo, Saab (RIP), Bofors guns, Primus stoves, interminable Bergman movies, winter candles on the streets of gamla gatan, pacificism, social welfare, and obsessional scandinavian epidemiology. The latter has proved a redeeming feature.
Childhood family income, adolescent violent criminality and substance misuse:quasi-experimental total population study. Amir Sariaslan, Henrik Larsson, Brian D’Onofrio, Niklas Langstrom and Paul Lichtenstein. British Journal of Psychiatry.
Published online ahead of print August 21, 2014, doi:10.1192/bjp.bp.113.136200
Children of parents in the lowest income quintile experienced a seven-fold increased hazard rate (HR) of being convicted of violent criminality compared with peers in the highest quintile (HR = 6.78, 95% CI 6.23–7.38). This association was entirely accounted for by unobserved familial risk factors (HR = 0.95, 95% CI 0.44–2.03). Similar pattern of effects was found for substance misuse.
The authors point out:
Behavioural genetic investigations have found that the liabilities for both violent offending and substance misuse are substantially influenced by shared genetic and, to a lesser extent, family environmental factors.7,8
7 Frisell T, Lichtenstein P, Langstrom N. Violent crime runs in families: a total
population study of 12.5 million individuals. Psychol Med 2011; 41: 97–105.
8 Kendler KS, Sundquist K, Ohlsson H, Palme r K, Maes H, Winkleby MA, et al.
Genetic and familial environmental influences on the risk for drug abuse:
a national Swedish adoption study. Arch Gen Psychiatry 2012; 69: 690–7.
We linked data from nine Swedish, longitudinal, total-population registers maintained by governmental agencies. The linkage was possible through the unique 10-digit civic registration number assigned to all Swedish citizens at birth and to immigrants upon arrival to the country.
The final sample (omitting multiple-births, death, severe handicap and emigrants) consisted of 88.6% of the targeted population (n = 526 167). The
sample included 262 267 cousins and 216 424 siblings nested within 114 671 extended and 105 470 nuclear families.
We calculated mean disposable family income (net sum of wage earnings, welfare and retirement benefits, etc.) of both biological parents for each offspring and year between 1990 and 2008. Income measures were inflation-adjusted to 1990 values according to the consumer price index provided by Statistics Sweden.
Gender, birth year and birth order were included in all models. We also adjusted for highest parental education and parental ages at the time of the first-born child, and parental history of ever being admitted to hospital for a mental disorder.
Violent crime was defined as a conviction for homicide, assault, robbery, threats and violence against an officer, gross violation of a person’s/woman’s integrity, unlawful threats, unlawful coercion, kidnapping, illegal confinement, arson, intimidation,
or sexual offences (rape, indecent assault, indecent exposure or child molestation, but excluding prostitution, hiring of prostitutes or possession of child pornography).
The participants entered the study at their fifteenth birthday and were subsequently followed up for a median time of 3.5 years. The maximum follow-up time was 6 years.
This is a short time to pick up the full flowering of criminal careers, so perhaps should be considered and under-estimate, or purely a measure of juvenile delinquency and not of life time criminality (which usually lasts until middle age).
Readers will know that I cast a particularly baleful eye over all “corrections” and “adjustments” but in this paper the techniques are transparent, and have an intrinsic justification. The data allows them to compare siblings with cousins, and intact nuclear families with more scattered ones: two natural experiments which allow contrasts of shared genes and experience. Crafty. That is my summary, but here is their explanation in detail:
We fitted two separate models for the entire sample (n = 526 167) that gradually adjusted for observed confounding variables. Model I adjusted for gender, birth year and birth order, whereas Model II also adjusted for highest parental education, parental ages at the time of the first-born child and parental history of admission to hospital for a mental disorder.
To assess the effects also of unobserved genetic and environmental factors, we fitted stratified Cox regression models to cousin (n = 262 267) and sibling (n = 216 424) samples with extended or nuclear family as stratum, respectively. The stratified
models allow for the estimation of heterogeneous baseline hazard rates across families and thus capture unobserved familial factors. This also implies that exposure comparisons are made within families. Model III was fitted to the cousin sample and adjusted for observed confounders and unobserved within extended-family factors. Model IV was fitted on the sibling sample and accounted for unobserved nuclear family factors and for gender, birth year and birth order.
Cousin and sibling correlations on the exposure variable were calculated based on a varying-intercepts, mixed-effects model where the intercepts are allowed to vary across families.
The magnitude of the variation was expressed as an intra-class correlation (ICC). The ICC measures the degree to which observations are similar to one another within clusters; in this case cousins and siblings nested within extended and nuclear family clusters. The measure ranges between 0 and 1, where the latter implies that cousins and siblings have identical exposure values within families.
As you can see, each model picks away at what would otherwise be seen as a purely economic cause of criminality and drug abuse. Model II which adjusts for parental education and mental illness has a big effect.
In an unusual departure, The Economist devoted an article to this paper, which suggests that they are beginning to wake up to the human factors in economics. Admittedly, they sub-titled it A disturbing study of the link between incomes and criminal behaviour, suggesting they were disturbed. Here are The Economist’s conclusions:
That suggests two, not mutually exclusive, possibilities. One is that a family’s culture, once established, is “sticky”—that you can, to put it crudely, take the kid out of the neighbourhood, but not the neighbourhood out of the kid. Given, for example, children’s propensity to emulate elder siblings whom they admire, that sounds perfectly plausible. The other possibility is that genes which predispose to criminal behaviour (several studies suggest such genes exist) are more common at the bottom of society than at the top, perhaps because the lack of impulse-control they engender also tends to reduce someone’s earning capacity.
Neither of these conclusions is likely to be welcome to social reformers. The first suggests that merely topping up people’s incomes, though it may well be a good idea for other reasons, will not by itself address questions of bad behaviour. The second raises the possibility that the problem of intergenerational poverty may be self-reinforcing, particularly in rich countries like Sweden where the winnowing effects of education and the need for high levels of skill in many jobs will favour those who can control their behaviour, and not those who rely on too many chemical crutches to get them through the day.
This is only one study, of course. Such conclusions will need to be tested by others. But if they are confirmed, the fact that they are uncomfortable will be no excuse for ignoring them.
What The Economist might have said is: Since this is a total population study of five birth cohorts and is the largest by far in the literature, it has high credibility, and the result will stand until another study of equal quality finds otherwise.
I think The Economist calling these results disturbing is commendable. They admit that they did not expect or like the potential implications of this study, which I think is far better than pretending they are totally objective in their evaluation.
ReplyDeleteInstead capturing the entire screen for your screenshots, you should use the Snipping Tool, which enables you to select an area of the screen to be captured:
ReplyDeleteThank you for this advice, which I will implement in future.
ReplyDeleteThis comment has been removed by the author.
Delete"Correlation is a necessary feature of causation"
ReplyDeleteDepending on how one interprets the sentence, it is wrong or misleading. Where there is causation, it may be masked by other factors, resulting in no (substantial) correlation.
I can see the theoretical argument, but can you give an actual example?
ReplyDeleteYou can probably find some, but it is rare. I can imagine this example. Back when smoking was cool (very long ago), higher SES people did it more. Since higher SES people live longer for a variety of reasons, running a correlation with smoking would yield a positive correlation with longevity, while it still being true that smoking causes death.
ReplyDeleteAny real life example of a dangerous habit first picked up by social elites would yield this result. Surely there are such examples in history.
Actually, for completeness sake, on the "smoking causes death" thing, randomized controlled trials on smokers found that quitting does nothing to improve their longevity. So there's that...
DeleteThanks, but did it actually lead to that correlation, smoking correlated with longevity? I am simply repeating the challenge Michael Woodley made at the London Conference on Intelligence. I know there are some spurious correlations, but if things are correlated, one of those variables will generally be the cause. I am not saying that there have not been instances of this, but that currently I do not have a list of them, if there are any.
ReplyDelete"However, as we are wearily cognizant of, correlation is not causation, though in ordinary life it damn well implies it. Correlation is a necessary feature of causation, but not a sufficient proof. The quip should be altered to: correlation is not always causation, but it helps."
ReplyDeleteThe quip was fine until it morphed on the internet into,
'Correlation does not imply causation.';
thereby reversing the meaning of the phrase.
Nice trick. A bit like watching Stephen Sackur interview someone who doesn't understand the double-negative.
I suggest the best way out of the confusion is to take the bull by the horns, flooding therapy; spread the phrase,
'Correlation does imply causation.'
No doubt there would be a wave of does-doesn't arguments in the blogosphere but at least some people would come out of it, Vocab + 1.
If anyone infers from that either a disdain for comprehensive schooling or admiration for the works of Jane Austen, those would not be correlations too far.
However, I do not intend to imply that all people are capable of objective evaluation of things out there, like correlations. Many people seem to be stuck in a subjective mindset; understanding the external world only in so far as it 'speaks to' them personally. That could be why the phrase morphed in the first place.
"Thanks, but did it actually lead to that correlation"
ReplyDeleteWhat do we know about the 'slippery slope' [thought to word to deed]?
How strong is the correlation between verbal abuse and/or indecent exposure and actual bodily harm?
Is their a linear relationship with age or are there life-cycle flash points, and/or environmental thresholds? The role of authority could be one factor, conformity to group norms another.
What I said in London was that whilst it is true that correlation does not necessarily equate to causation, all causally related variables will be correlated. Thus correlation is always necessary (but not in and of itself sufficient) for establishing causation. The claim that 'correlation does not equal causation' is therefore meaningless when used to counter the results of correlative studies in which specific causal inferences are being made, as the inferred pattern of causation necessarily supervenes upon correlation amongst variables. Whether the variables being considered are in actuality causally associated as per the inference is another matter entirely. The correct critique of such findings therefore is from mediation, i.e. the idea that a given correlation might be spurious owing to the presence of 'hidden' variables that are generating the apparent correlation. A famous example is yam production and national IQ, which across countries correlate negatively. It would be wrong to say that yam production somehow inhibits IQ, as the association will in fact turn out to be mediated by something like temperature and latitude. These variables are in turn proxies for historical and ecological trends that make the sort of countries that yield fewer yams the sort of countries that are typically populated by higher ability people, and vice versa. The causation in this case is via additional variables, which cause the covariance between the two variables of interest, without there being a direct effect of one on the other. Properly constructed multivariate models can use these patterns of mediation to infer the likelihood of causation going in one direction or another. Thus it is possible to actually test causal inference amongst a population of correlated variables. By far the best way of doing this is to compare the fits of models containing specific theoretically prescribed patterns of causal inference against (preferably many) alternative theoretically plausible models, in which alternative patterns of causation are inferred (Figueredo & Gorsuch, 2007). Sir William Gemmell Cochran termed this “Fisher’s Dictum‟:
ReplyDelete“About 20 years ago, when asked in a meeting what can be done in observational studies to clarify the step from association to causation, Sir Ronald Fisher replied; `Make your theories elaborate.' The reply puzzled me at first, since by Occam's razor, the advice usually given is to make theories as simple as is consistent with known data. What Sir Ronald meant, as subsequent discussion showed, was that when constructing a causal hypothesis one should envisage as many different consequences of its truth as possible, and plan observational studies to discover whether each of these consequences is found to hold. (Cochran, 1965, §5).
Ref.
Cochran, W. G. (1965). The planning of observational studies of human populations
(with Discussion). Journal of the Royal Statistical Society. Series A, 128, 134–155.
Figueredo, A. J., & Gorsuch, R. L. (2007). Assortative mating in the jewel wasp. 2.
Sequential cononical analysis as an exploratory form of path analysis. Journal of
the Arizona-Nevada Academy of Science, 39, 59-64.