Hallam Hurt and colleagues have kindly sent me more of their recent publications, one of which has some relevance to their reported claim that “poverty causes more problems than crack cocaine”.
Handzel, Brodsky, Betancourt and Hurt “Longitudinal Follow-up of Poor Inner-city Youth Between Ages 8 and 18: Intentions Versus Reality” Pediatrics; originally published online February 20, 2012.
It looks at African American children’s dreams about their future, and the reality of the actual outcomes, and attributes the greatest effect to exposure to violence and poor home environments. The paper makes sad reading. At age 9 the children had upbeat expectations, in line with the American dream, the founding fathers effect of Protestant self-improvement and general good natured optimism. No low expectations here. No lack of self-esteem. By age 18 reality had intervened, and half of those cheerful childish expectations had been dashed.
At age 9, 94% of participants felt it unlikely they would try marijuana; 93% felt they were unlikely to get arrested; 92% felt they were likely to attend college or trade school; 81% did not know one could become pregnant with first-time sex (this last one is ignorance, not over-optimistic self perceptions). Age 18 33% had used drugs, 33% had been arrested, 19% had dropped out of school, and 20% had become parents. 56% experienced at least one of those events. No relationship was found between childhood intentions and the eventual documented outcomes. So, although the popular mantra is “you can become whatever you want to be” the chances of that coming true are distressingly low.
The authors conclude their paper thus:
“During childhood, our cohort of poor, inner-city children were idealistic regarding
their future. However, by the ages of 16 to 18, more than 50% of these teens already had at least 1 TAE. Factors most strongly associated with TAEs were greater exposure to violence and a poorer home environment, both early childhood environmental factors.”
My reading is that their conclusion is partly right as far as it goes, but the actual results suggest that their environmental interpretation is far from being the only one possible. For example, let us take the statement that 56% had experienced a “trajectory altering event”, the odds of such an event increasing with greater exposure to violence and a poorer home environment. The phrase “trajectory altering event” suggests an external force like a meteorite, but on examination this turns out to be things that the teenagers have got themselves into: using drugs, getting arrested, dropping out of school, and failing to use contraception when they were in no position to raise a family. Crucially, it is NOT an independent risk factor. It is consequence of their failure to see the deleterious consequences of their own actions. So, the factors are “strongly associated” but are very unlikely to be causal, and are more likely to be associated effects of other primary causes.
What do the authors think has caused this horrible reverse in fortunes? The authors identify violence and poorer home environments as increasing the risk of a “trajectory altering event”. Below is their Table 1 showing the difference between those who had a good outcome and those who did not, or as they describe it, “no” and “some” trajectory adverse events:
The P values are given, but let us rank them by effect size (difference in means as a ratio of the standard deviation of those without TAEs): exposure to violence 1.0, home environment 1.02, IQ 0.41 and the others seem too small or given as categories.
The authors then do a logistic regression, and come up with the following odds ratios.
Gender 10.21 sig
Caregiver using drugs 3.66 non sig
Exposure to violence 1.14 sig
Full Scale IQ 0.96 non-sig
Home environment 0.77 sig
As regards discussing findings in terms of p values, this is the issue to which Gigerenzer has drawn attention. We researchers tend to use p values in a slavish fashion, as part of a ritual. Ranking findings by effect size (rather than computing fluke probability) is often more illustrative of the features of the sample in question. Researchers need to tell us about their samples in plain language, and then give the basic findings in terms of effect sizes. Even a simple correlation matrix would be some help (and in that instance the p values could be mentioned). After that, if it really seems worth while, they could try structural equation modelling.
The fact that boys are one order of magnitude more likely than girls to get into trouble is well known. “Caregiver using drugs” at 3.66 fails the significance test, yet suggests an effect might be found in larger samples, and is certainly a putative causal variable. The violence measure at age 7 does not distinguish between violence in the home and violence outside it, but it would have been worth while giving those scores separately. Violence in the home tends to be particularly damaging, because it denies the child any place of safety.
Now let us look at the language being used. “Exposure to violence” is neutral, but suggests the children were walking down the road and some external event occurred. Might we be dealing with violent families? We need more data, but it is possible that this is what is happening. How would you say, in modern parlance, that a child was being brought up by a family in which the parents were on drugs and paid little attention to the child and were violent to each other and/or to the child? Problem family? Dangerous family? Careless family? All these make it clear that parents were failing to care for children in their charge, if not in their love. “Vulnerable family” gives a clue, without giving away too much. Some researchers think it would be better to try to distract readers from this reality, and say that all problems could be traced to damp houses with peeling wall paper. “Poor home environment”? That should cover a multitude of sins. It put the problem outside of individual decision making.
Looking at the paper as a whole, on the positive side, the measures of childhood obtained from a longitudinal study, and are not retrospective. Less happily, there has been a high drop-out rate, particularly among boys. The sample size of 79 is small and so constrained by the selection criteria that it is hard to draw strong conclusions. We need comparison groups and larger samples to have confidence in the findings. For example, one could look at other poor children, different ethnic groups, including white children of similar IQs in order to tease out possible causes. Other samples of white children seem to show more congruence between childhood expectations and behaviour. In general, arguing from within small samples tends to be error prone when dealing with large questions.
Let me be clear. I am in favour of clinical samples being followed long term. The authors are trying to pick out potential life events which have re-directed the trajectory of children’s lives. That is a worthy thing to do. My gripe is that they haven’t said very much about the possibility that bad parenting is a major factor. In this low intelligence group even a 5.2 IQ point difference has had an effect on adult outcomes, in that the slightly higher intelligence children are less likely to mess up their lives. To their credit the authors discuss this in the context of early intervention studies, of which I am perhaps inordinately fond.
I would simply like the research to be described in a way that gave equal place to nature and nurture. These authors started their study long before collecting genetic samples was the norm, but they seem particular keen to show that the key factors are external and “environmental” whereas they are actually mostly about things which are happening within the family. In summary, this paper does not buttress the claim that poverty is responsible for anything. However, it certainly backs the obvious point that parents should not take drugs and should not expose children to violence. That is a finding of sorts.