The Shape of Self-Control

The question

“Self-control” is almost always measured as one number, but that number combines two different kinds of difficulty: problems of impulse and attention, and problems of conduct. Do the two change on the same schedule across childhood and adolescence, or does one speed up while the other holds its pace?

Why it matters

If a single score averages a component that speeds up in adolescence with one that does not, then “self-control improves in adolescence” means something different depending on which problem a young person has. Any study, program, or theory that uses one number averages over a real difference it cannot see, and may aim an intervention at the wrong developmental window for the wrong problem.

What I did

I followed about 5,600 children from the UK Millennium Cohort Study, using parent ratings at ages 5, 7, 11, 14, and 17, and modeled how each component changes over time. To rule out the possibility that the trend reflects only parents rating their children differently as they grow, I compared the parent reports with teachers’ independent ratings at ages 7 and 11.

What I found

Hyperactivity and attention problems fall sharply after age 11, more than five times steeper than conduct problems over the same years, and three quarters of their total improvement between 5 and 17 happens after age 11. Conduct problems keep improving too, at a steady slow rate, with their improvement split about evenly between childhood and adolescence. The difference survives every check I ran, including a split-half replication and the teacher comparison: teachers, who never saw the parents’ answers, recorded change about three times larger than parents did. Both components also keep improving past ages 8 to 10, the age at which classic theory claims self-control is “set.”

What it means

Treating self-control as one trait hides that adolescence is where most of the remaining change sits for one kind of problem, while for the other it offers no more change per year than middle childhood did. This is the foundation the other two papers build on: the two components need to be measured and understood separately.

The question

Gottfredson and Hirschi say parents install self-control early, and it is then fixed. Two parts of that claim can be tested separately. First, does the family set the starting level of self-control problems, and does it also set the later course, or only the level? Second, once development is underway, does the influence run parent to child, as the theory assumes, or child to parent?

Why it matters

If families set the level but not the later course, then what parents do matters early, and something else matters later. If the relationship runs child to parent, then the standard “parenting fixes the child” story is incomplete, and the child must be understood as an active party in their own development.

What I did

Using the same cohort, I tested whether four early family conditions (harsh discipline, household routine, parental closeness, and economic disadvantage) predict the level and the rate of change of self-control problems. I then used a within-family model, following the same children and parents over five waves, to ask which way the parent–child relationship runs.

What I found

All four family conditions predict the age-5 level, with harsh discipline the strongest. Economic disadvantage predicts the level and nothing else: it leaves both later rates of change untouched. That is a claim about disadvantage specifically, not about parenting, which does reach the rate of change (harsh discipline touches both slopes, closeness the childhood slope, household routine the adolescent one). Most importantly, the parent–child relationship runs child to parent: a child’s self-control shapes later closeness far more than closeness shapes later self-control, and that effect grows about fourfold from childhood into adolescence, though the panel’s unequal lags mean part of that growth is a lag-length artifact rather than pure developmental strengthening.

What it means

The family is not a one-way cause. Economic disadvantage sets the level and stops there, while parenting reaches both the level and the later rate of change, and the child becomes the active party driving the relationship. This is the “origin” leg of the dissertation’s argument: what sets the level is not the same as what sets the shape.

The question

Two things are usually run together. First, does a child’s developmental course, meaning how fast their self-control improved and whether that improvement bent at adolescence, predict adult justice contact over and above where they started? Second, is there a distinct type, meaning the small “life-course-persistent” group that criminological theory predicts, or is everyone simply somewhere on one continuous scale?

Why it matters

If the course matters, then watching how a child changes over time is worth more than a single early assessment. If there is no type, then schemes that try to find and flag an at-risk group early are looking for something that does not exist. Both answers affect how resources should be allocated.

What I did

I followed the cohort to age 23 and linked each child’s self-control curve to their self-reported justice contact. The primary outcome is any formal contact; a more serious measure, a caution, arrest, court appearance, or guilty verdict, is carried alongside it, because police stops are largely a matter of policing practice rather than offending in the UK. I then compared two competing descriptions of each young person, their continuous developmental course versus a “type” assignment from a trajectory-class model, and asked which better predicts adult contact in a strict out-of-sample test.

What I found

1. All three features of the curve, the starting level, the rate of improvement, and the adolescent bend, predict adult formal contact about equally (per standard-deviation increase: level 1.29, rate 1.32, bend 1.30). Where a child started tells you no more than how they changed.
2. Most of this does not run through measured teenage delinquency. About 85% reaches adulthood by other routes, so the standard “cascade” is only a partial story.
3. There is no type. The trajectory classes are near-identical in shape, class membership predicts nothing, and in the out-of-sample test the continuous course beats the type, while the type is even worse than simply knowing sex and household disadvantage. The serious outcome is where the course matters most: the rate of improvement predicts a formal sanction (caution or above) more strongly than it predicts any police contact.

What it means

For theory, developmental risk in self-control is continuous rather than categorical: a degree of risk, not a kind of person. For practice, early-identification programs built on finding a “type” should expect to underperform the simpler alternative of measuring how a child’s self-control is changing.