Counted Wrong

Data, young adults, and the criminal justice system

Adult at 18. Two words in a statute, and everything changes: which court hears the case, which facility takes custody, how long a record follows someone. The line is exact. The people crossing it are not.

This page follows 1,354 young people through Pathways to Desistance, a study that interviewed them 11 times over seven years after a serious offense. At every interview they answered the same questions about maturity. The result is a rare thing: a measurement of growing up, taken on the same people, spanning the very line the law cares about.

The Study

Pathways to Desistance enrolled 1,354 young people in the early 2000s, shortly after each was found responsible for a serious offense. It then interviewed them for seven years: a baseline interview plus follow-ups out to 84 months, with ages 14 to 26 observed along the way. The public archive is ICPSR study 29961.

Maturity here means the Psychosocial Maturity Inventory, PSMI for short: a questionnaire scored 1 to 4, with three subscales. Self-reliance covers acting on your own judgment, identity covers how settled your sense of yourself is, and work orientation covers how you handle work and obligations. The overall score averages the three. In this data the measure runs through age 24.

WarningWho Is in These Numbers

Everyone here was enrolled after a serious offense, in the juvenile and adult systems of two large metro areas (Philadelphia and Phoenix). This is not a portrait of all young people. It is the group the justice system actually handles, which is exactly where the line at 18 does its heaviest work.

The Curve

Start with the overall score. Line up all of the interviews by the age the person was when they gave one, and take the mean at each age.

Each point pools every interview given at that age. The shaded band is not a confidence interval: it shows how spread out individual scores are around each mean.

The mean climbs from 3.01 at age 14 to 3.15 at age 18. If maturity finished at 18, the curve would flatten there. It does not. By 24 the mean reaches 3.35 and is still rising at the edge of the data.

Now do the arithmetic the line implies. This cohort gained 0.14 points from 14 to 18. It gained 0.21 points from 18 to 24. Sixty percent of the climb we can observe lands on the adult side of the line.

Most of the growth comes after the line.

TipTeaching Prompt: The Zoomed Axis

The y axis above runs from about 2.5 to 3.9, not the full 1 to 4. On the full scale the climb would look flatter and the shaded band would still dwarf it. Which version tells the truth? What is each one trying to persuade you of?

Four Measures, One Shape

The three subscales could each have told a different story. They almost do: identity barely moves before 18, gaining 0.04 points from 14 to 18, then climbs five times that after. Work orientation starts lowest and keeps climbing longest. But the headline repeats in every panel: more than half of each measure’s observed growth lands after 18, and nothing flattens at the line.

All four panels share one y axis, so slopes are comparable. Growth after 18, as a share of each measure’s total: identity 83 percent, overall 60 percent, work orientation 54 percent, self-reliance 53 percent.

NoteTeaching Prompt: One Score or Four

The overall PSMI averages three subscales that grow on different schedules. When does a composite clarify, and when does it bury the interesting part? The Hip-Hop Periodic Table teaches the same lesson from the other direction.

The Spread Inside Every Age

Means are tidy. People are not. The next chart drops the averaging: each column is an age, and each cell shows the share of that age’s interviews landing in one half-point score bin.

The dashed line sits where the 17 and 18 columns meet. Shares are of the cells shown. Blank cells are not zeros: any cell holding fewer than 10 interviews was suppressed before this page ever saw the data.

Read any column bottom to top and the range is wide. At 18, scores spread from 2 to 4, and the spread within that single age (a standard deviation of 0.47) is bigger than the entire average climb from 14 to 24 (0.34 points). Two people on either side of their 18th birthday can be counted differently by every system that uses the line, while scoring identically on the thing the line is supposed to mark.

ImportantTeaching Prompt: Means and People

A judge never meets a mean. If you had to design a rule that respected both the average curve and the spread around it, what would it use besides a birthday?

Growing Up Inside the Study

The study did not sample teenagers and adults separately. It watched one group cross the line. At the first interview, 92 percent of the cohort was under 18. Four years in, no one was.

The under 18 line stops after month 36 because later waves had no one under 18 left to count. At month 84 the plum line dips a hair below 100 percent: one person had aged past 25, and cells that small are suppressed.

The cohort’s mean age moves from 16.0 at the first interview to 23.0 at the last. Every curve above comes from the same people aging through the window the law splits in half.

What Gets Counted Wrong

None of this says 18 is meaningless. Courts, custody, and records have to draw lines somewhere, and this data cannot say where. What it says is narrower and harder to ignore: whatever 18 marks, it is not the end of growing up. The average person in this cohort did most of that growing after the law stopped calling them a child, and the gap between any two people of the same age was wider than the gap the line is supposed to mark.

A line someone drew, treated as a fact of nature, will count some people wrong. This study kept measuring long enough to show it. Most of the growth comes after the line.

How Old Is Old?, this site’s companion piece, walks the same problem at the other end of the lifespan.

Where the Numbers Come From

Pathways to Desistance data is restricted access. The person-level records live on one analyst’s machine under a data use agreement, and they never touch this website or its repositories. What this page renders is aggregates only, produced by a small script whose whole job is refusing to output anything person-shaped.

• Every table behind this page is a group summary: means, standard deviations, and binned counts.
• Every published cell covers at least 10 interviews. Smaller cells were suppressed at the source, which is why some heatmap cells are blank.
• The unit is the interview, not the person. Someone interviewed twice while 18 counts twice in the age 18 column. In total, 1,354 people gave 14,894 interviews, and more than 12,000 of those carry a maturity score.
• Ages pool the panel: the age 18 column holds everyone who was 18 at any of their interviews, whichever wave that was.

NoteData Ethics: Why Cells Disappear

Ten is not a magic number. It is a promise: no published cell small enough to point at a person. Suppression trades a little completeness for a lot of privacy, and the blank cells above are that trade made visible.

Clarity is kindness, so here are the exact numbers behind the curves.

All seven published tables are downloadable: psmi_by_age (the means and spreads behind the curves), age_score_counts_maturity, age_score_counts_self_reliance, age_score_counts_identity and age_score_counts_work_orient (the binned counts behind the heatmap), ea_share_by_wave (the cohort crossing the line), and mean_age_by_wave (the cohort’s aging).

Data: Pathways to Desistance (ICPSR 29961), person-level access restricted. This page renders only the published aggregate tables above; the analysis code lives in a private repository.