The association between childhood blood lead level and criminal offending

Abstract

Importance

Lead is a neurotoxin with well-documented effects on health. Research suggests lead may be associated with criminal behavior. This association is difficult to disentangle from low socioeconomic status, a factor in both lead exposure and offending.

Objective

To test the hypothesis that higher childhood blood-lead level (BLL) is associated with greater risk of (1) criminal conviction, (2) recidivism (repeat conviction), (3) violent conviction, and (4) variety of self-reported criminal offending, in a setting where BLL was not associated with low socioeconomic status (SES).

Design

Prospective cohort study based on a population-representative 1972–73 birth cohort; the Dunedin Multidisciplinary Health and Development Study observed participants to age 38 years.

Setting

Dunedin, New Zealand

Participants

553 lead-tested members of the Dunedin Study

Exposure

BLL measured at age 11 years.

Main Outcomes and Measures

Official criminal conviction cumulative to age 38 years (data collected in 2013). Single conviction or recidivism. Nonviolent or violent crime conviction. Self-reported variety of crime types at ages 15, 18, 21, 26, 32, and 38 years.

Results

Participants included 553 individuals who had been lead-tested at age 11 years (54% male). Mean (SD) BLL at age 11 years was 11.01 (4.62) μg/dL. Outcome prevalence was: criminal conviction, 28%; recidivism, 16%; violent conviction, 10%. Self-reported offending variety ranged from 0–10 at each assessment and followed the well-established age-crime curve. BLL was a poor predictor of criminal conviction (AUC .58).

Overall, associations between BLL and conviction outcomes were weak. Estimated effect of BLL was lower for recidivism than for single conviction and lower for violent offending than nonviolent offending. Sex-adjusted associations between BLL reached statistical significance for only one of the six self-reported offending outcomes, at age 15 years (r=0.10; 95%CI, 0.01, 0.18; P = .023).

Conclusions and Relevance

This study overcomes past limitations of studies of BLL and crime by studying the association in a place and time where the relation was not confounded by childhood socioeconomic status. Findings failed to support a dose-response association between BLL and consequential criminal offending.

Introduction

Recent discoveries of elevated lead levels in the drinking water of several US cities have prompted researchers to re-examine the issue of long-term effects of lead exposure.^1–4 Childhood lead exposure has been linked to abnormalities in brain structures, such as the hippocampus and cerebral cortex, and to disrupted neuronal development, myelination, and neurochemical processing.^5–7 Lead exposure has been linked to childhood behavioral problems and lasting declines in intelligence.^8–10 One hypothesized behavioral effect of high levels of lead exposure is increased antisocial and criminal behavior.¹¹ Criminal offending carries consequences for offenders’ life outcomes, including employment, family life, and life expectancy.^12–14 Research has found that individuals with elevated childhood blood lead levels (BLL) show increased criminal offending throughout adulthood.^15–19 Aggregate-level research has also linked estimated environmental lead exposure to higher crime rates.^20–27 One of the most intriguing propositions is that declines in American children’s BLLs, which began in the 1970s as lead was gradually removed from the environment, contributed to lower crime rates across the United States as the children became adolescents during the mid-1990s.^11,28,29

Some findings, however, indicate that past studies may have yielded inaccurate estimations of lead’s true associations with health and behavioral outcomes due to confounding by low socioeconomic status.^30,31 New Zealand’s Dunedin Multidisciplinary Health and Development Study (Dunedin Study) offers the opportunity to study the effects of lead in the body on crime, independent of socioeconomic status. In the Dunedin Study, as previously reported, childhood BLLs were representative of New Zealand children’s BLLs, and were unassociated with childhood socioeconomic status.^9,32,33 New Zealand has a reputation as a “green” country, but during the Dunedin Study cohort’s childhood in the 1970’s and 1980’s there was evidence of the association between leaded gasoline and elevated BLLs.³⁴ The measure of BLL used in the present article has provided evidence of associations in the Dunedin cohort between higher childhood lead levels and lower midlife IQ and socio-economic attainment .⁹ Using the same BLL measure, the present study tested the hypothesis that lead is positively associated with crime.

In addition to testing whether BLL is related to criminal conviction, the present study tested three additional hypotheses about crime. First, based on results of past research,¹⁹ we tested the hypothesis that BLL is related to a greater likelihood of recidivistic versus one-time criminal conviction. Many people who experience a criminal conviction never recidivate (have a repeat conviction from a new court case); some, however, recidivate.³⁵ Recidivating offenders may be etiologically different from one-time offenders and are a concern of the criminal justice system.^36,37 Second, research has also found a specific link between lead and violence,^27,38 leading us to test the hypothesis that BLL is related to a greater likelihood of conviction for violent versus nonviolent offenses. Third, we tested the hypothesis that higher BLL is related to more self-reported offending (offenses committed as reported by a respondent, regardless of criminal conviction). It is well-established that more self-reported offending predicts increased risk of criminal conviction, but the vast majority of offenses go undetected.³⁹ Self-reports of offending may provide a more complete picture of the social burden imposed by an individual’s offending, whether or not it resulted in an official record.

Methods

Sample

Participants are members of the Dunedin Multidisciplinary Health and Development Study, a longitudinal investigation of health and behavior. The full cohort (N=1,037; 91% of eligible births; 52% male) were all of the individuals born between April 1972 and March 1973 in Dunedin, New Zealand, who were eligible based on residence in the province and who participated in the first assessment at age 3. The cohort represented the full range of socioeconomic status in the general population of New Zealand's South Island.⁴⁰ On adult health, the cohort matches the New Zealand National Health and Nutrition Survey on key health indicators (e.g. body mass index, smoking, doctor visits).⁴⁰ The cohort is primarily white(<7% self-identify as having non-Caucasian ancestry) matching the demographics of the South Island.⁴⁰ Assessments were carried out at birth and ages 3, 5, 7, 9, 11, 13, 15, 18, 21, 26, 32, and most recently at age 38 years, when 95% of the 1,007 participants still alive took part. Written informed consent was obtained from participants and protocols were approved by the ethical review boards of the participating universities.

Blood lead level

Approximately 30 ml of venous blood was collected from each 11-year-old who participated in the assessment (carried out at the Research Unit) and who freely agreed to give blood; 579 of the 803 children (72%) in attendance agreed to give blood. An additional 122 children, aged 11 years and tending to live outside city limits, were assessed in their schools, where blood could not be drawn. Whole blood samples were analyzed through graphite furnace atomic absorption spectrophotometry. BLL is reported in micrograms per deciliter (1 μg/dL=0.0483μmol/l). Details on the method of blood collection, division, storage, quality assurance and analysis procedures have been described previously.^32,33,41

Criminal offending

Official conviction records were obtained through a search of the central computer system of the New Zealand Police, which provides details of all New Zealand convictions and Australian convictions communicated to the New Zealand Police. Searches for all convictions occurring from the age from which conviction was permissible (14 years), were conducted after each assessment at ages 21, 26, 32, and 38 (search completed in 2013). Conviction was coded 1 for convicted participants and 0 for unconvicted participants. Subgroups were one-time offenders versus those who recidivated (had a repeat conviction from a new court case) and participants only convicted of nonviolent crime versus those convicted of a violent crime.

A self-reported offending interview was administered at ages 15, 18, 21, 26, 32, and 38, using a one-year retrospective window at each wave. Four types of offenses were assessed. Property offenses included items such as vandalism, breaking and entering, motor-vehicle theft, embezzlement, shoplifting, and frauds. Rule offenses included items such as reckless driving, public drunkenness, soliciting or selling sex, giving false information on a loan or job application, and disobeying court orders. Drug-related offenses included using and selling various types of illicit drugs. Violent offenses included items about simple and aggravated assault, gang fighting, robbery, arson, and forced sex. We used offense items to create a variety score of self-reported offending. Variety scores are calculated by summing one point for a “yes” response to each different offense. Variety scores typically correlate with frequency,⁴² but the variety score is preferred over frequency because variety is less skewed, does not overweight trivial offenses, and is less affected by recall errors (e.g. “Have you shoplifted?” is more accurately recalled than “How many times have you shoplifted?”).^43,44 The variety scores ranged from 0 to 30 offense types; higher numbers indicated greater crime involvement. To eliminate potential null findings due to skew in self-reported offending we winsorized self-reported offending variety scores at 10, the ≥ 90^th percentile self-reported offending variety score for each age. Results using the full variety scales did not substantively differ from results using the winsorized scales (see eTable 1 in the Supplement). Self-reported offending across interviews at ages 15, 18, 21, 26, 32, and 38 followed the universal “age-crime curve”^36,45: the mean variety of self-reported offending rose from age 15 years to its peak at ages 18 and 21 years, and declined thereafter.

Potential confounders

Analyses adjusted for sex. As in prior studies, cohort boys had more criminal offending and slightly higher childhood BLL than girls (Figure 1).^9,44 Control for sex was indicated because a male excess of both BLL and offending could yield spurious associations in the cohort. In addition, higher BLLs and higher rates of crime are usually more likely to be found among children who come from families with low socioeconomic status.^30,45 However, we report in Figure 1 the null association between childhood socioeconomic status and BLL in this cohort (this null association was previously reported for the sub-sample of the Dunedin Study having IQ data⁹).

Figure 1. Distribution of blood lead level.

Panel A shows distribution of BLL by sex. Mean blood lead level is lower for females (10.51 μg/dL) compared to males (11.44 μg/dL; P=.018). Panels B-D show box and whisker plots and histograms depict the distribution of childhood blood lead levels for participants from families of low–, middle–, and high–socioeconomic status based on the 6-point Elley-Irving scale (categories 1 and 2 [low status], 3 and 4 [middle status], 5 and 6 [high status]), which codes Study members’ parents’ occupations and their associated income and education levels. There was no significant socioeconomic gradient in lead exposure in the cohort children (r < -.001; 95%CI, -.08-.08; P =.99). High blood-lead levels were observed in all status groups. The box and whisker plots show the median (vertical black line inside the box), lower and upper interquartile ranges (IQRs; left and right borders of the box), lower-bound value equal to the 25th percentile minus 150% of the IQR and upper-bound value equal to the 75th percentile plus 150% of the IQR (whiskers), and outlier observations (circles). The histogram interval bins represent whole integers of blood-lead level.

Statistical analysis

The analytic sample for this study was N=553. Starting with the 579 participants with age-11 blood lead data, we excluded 26 individuals who did not consent to criminal record searching or did not survive to the age at which we first obtained consent for searches (21 years). To test whether the subset of the cohort who were lead-tested was representative of the un-tested cohort who met inclusion criteria, we compared the analytic sample (N=553) to the sub-sample without BLL data, N=414. Lead-tested participants did not differ significantly (P >.05) from untested children on official criminal conviction, self-reported offending, sex, or childhood socioeconomic status (SES). Because rural children tended to be seen at their schools at age 11 (where lead could not be tested), more lead-tested children lived in town, and lead-tested children had 2 points higher childhood IQ (see eTable 2 in the Supplement).

Study members with BLL > 23 μg/dL (n=10) were referred by the Study for medical evaluation, following the recommended clinical protocol at that time, and may have been treated.⁴¹ We have retained these Study members in our analysis as their removal did not substantively alter results.

We calculated descriptive statistics for the complete analytic sample and also by BLL (in approximately 5 μg/dL bins). We analyzed the predictive accuracy of BLL for conviction using receiver operating characteristic (ROC) curve and area under the curve (AUC). As a comparison, we included the ROC and AUC for male sex and a model which combined male sex and BLL. The ROC curve plots the predicted true positive (sensitivity) and false positive (1-specificity) rates for conviction. The AUC is an indicator of predictive accuracy. An AUC value of 0.5 indicates that model variables are no better at predicting the outcome than random chance. An AUC value of 1 indicates that model variables perfectly predict the outcome.

We tested the associations between childhood BLL and measures of offending using regression models (see Table). Using logistic regression models, we estimated the odds ratio of criminal conviction by BLL. Using multinomial logistic regression models, we estimated the odds ratio of being a one-time offender or recidivating versus a non-offender by BLL, and the odds ratio of being a non-violent offender or a violent offender versus a non-offender by BLL. BLL was analyzed as a continuous measure; however, it is presented in 5-μg/dL units in the Table, the current reference level for clinical attention and therefore a measure that is meaningful to clinicians and policymakers. Moreover, 5 μg/dL represents approximately 1 standard deviation of BLL in the cohort. We estimated the amount of self-reported offending at each assessment age by BLL using ordinary least squares regression. For self-reported offending, all model variables were standardized so that the results could be interpreted as correlation coefficients (r). All models were adjusted for the effect of sex; adjusted and unadjusted results are presented in the Table. Model fit statistics and tests of model fit comparing adjusted to unadjusted models can be found in eTable 3 in the Supplement.

Table.

Descriptive statistics of study variables and associations of blood lead level with offending outcomes.

Measure	Descriptive statistics N = 553	Descriptive statistics by blood lead level (BLL) category (μg/dL)				Association with BLLa (95% CI)[P value]
		≤ 5 μg/dL N=33 (6%)	6–10 μg/dL N=265 (48%)	11–15 μg/dL N=173 (31%)	> 15 μg/dL N=82 (15%)	Unadjusted	Adjusted for male sexb
Lead
Childhood blood lead level (μg/dL), mean (SD)	11.01 (4.62)
Criminal conviction
No conviction, N (%)	399 (72.2%)	25 (75.8%)	204 (77.0%)	112 (64.7%)	58 (70.7%)	1 - reference	1 - reference
Conviction, N (%)	154 (27.8%)	8 (24.2%)	61 (23.0%)	61 (35.3%)	24 (29.3%)	1.29 (1.06, 1.56) [.013]*	1.23 (1.00, 1.51) [.052]
Conviction subgroup comparison 1
One-time offender, N (%)	68 (12.3%)	2 (6.1%)	29 (10.9%)	26 (15.0%)	11 (13.4%)	1.29 (0.99, 1.68) [.058]	1.25 (0.95, 1.64) [.107]
Recidivistic offender, N (%)	86 (15.5%)	6 (18.2%)	32 (12.1%)	35 (20.2%)	13 (15.9%)	1.28 (1.01, 1.63) [.046]*	1.21 (0.93, 1.57) [.149]
Conviction subgroup comparison 2
Nonviolent offender, N (%)	101 (18.3%)	5 (15.2%)	40 (15.1%)	38 (22.0%)	18 (22.0%)	1.33 (1.06, 1.67) [.013]*	1.28 (1.01, 1.61) [.041]*
Violent offender, N (%)	53 (9.6%)	3 (9.1%)	21 (7.9%)	23 (13.3%)	6 (7.3%)	1.20 (0.89, 1.62) [.241]	1.13 (0.82, 1.55) [.449]
Self-reported offending variety, mean (SD)
Age 15 years	1.99 (2.82)	1.39 (2.34)	1.79 (2.77)	2.17 (2.83)	2.49 (3.04)	0.10 (0.02, 0.19) [.016]*	0.10 (0.01, 0.18) [.023]*
Age 18 years	4.24 (3.15)	3.25 (2.69)	4.06 (3.02)	4.54 (3.33)	4.64 (3.25)	0.09 (0.00, 0.17) [.043]*	0.06 (−0.02, 0.14) [.164]
Age 21 years	4.22 (3.02)	3.82 (2.51)	4.15 (2.92)	4.30 (3.21)	4.49 (3.15)	0.04 (−0.04, 0.13) [.296]	0.01 (−0.06, 0.09) [.748]
Age 26 years	2.83 (2.55)	2.03 (1.49)	2.58 (2.40)	3.27 (2.74)	3.01 (2.78)	0.09 (0.01, 0.17) [.032]*	0.06 (−0.02, 0.13) [.147]
Age 32 years	2.25 (2.19)	1.50 (1.83)	2.03 (1.83)	2.80 (2.54)	2.08 (2.37)	0.08 (−0.01, 0.16) [.076]	0.04 (−0.04, 0.12) [.278]
Age 38 years	1.10 (1.59)	0.74 (0.73)	1.01 (1.49)	1.36 (1.88)	0.99 (1.44)	0.04 (−0.04, 0.12) [.320]	0.02 (−0.06, 0.10) [.659]
Sex
Female, N (%)	255 (46.1%)	17 (51.5%)	137 (51.7%)	68 (39.3%)	33 (40.2%)	reference
Male, N (%)	298 (53.9%)	16 (48.5%)	128 (48.3%)	105 (60.7%)	49 (59.8%)	0.10 (0.02, 0.18) [.018]*

Notes: Full Sample N=553. Wave 15: n= 543, wave 18: n=535, wave 21: n=541, wave 26: n=543, wave 32: n=542, wave 38: n=540.

^aFor conviction, odds ratio from logistic regression of conviction on blood lead level (BLL) in 5 μg/dL units; reference category is no conviction. For conviction subgroups, odds ratios from multinomial logistic regression models of subgroup on BLL in 5 μg/dL units; reference category is no conviction. For self-reported offending variety, standardized β coefficients reported for offending variety regressed on BLL; these estimates can be interpreted as correlation coefficients (r).

^bModel fit was significantly improved by adjusting for all but self-reported offending variety at age 15 years. Details can be found in eTable 3 in the Supplement.

^*p < .05

In sensitivity analyses we subjected the BLL measure to a logarithmic transformation and a correction for hematocrit levels;⁴⁶ the results did not differ substantively from the results reported here (see eTable 4 in the Supplement).

Statistical analyses were conducted in R, version 3.4.0, using the nnet⁴⁷ and pROC⁴⁸ packages.

Results

Childhood BLLs ranged from 4 to 31 μg/dL. The mean (SD) childhood BLL was 11 (4.62) μg/dL (Table 1). There was no significant socioeconomic gradient in lead exposure in the cohort children (r <-.001; 95%CI, -.08-.08; P=.99). High BLLs were observed among children from all socioeconomic status groups (Figure 1).

Of participants, 154 (28%) had a criminal conviction, 68 (12%) were one-time offenders, 86 (16%) had recidivated, 101 (18%) were non-violent offenders, and 53 (10%) were violent offenders.

Criminal conviction was more prevalent and more frequent at higher BLLs (Table). ROC curve AUC (Figure 2), however, showed that BLL alone was a poor predictor of conviction. That is, at no point or threshold did BLL predict conviction with dependable accuracy. The predictive accuracy of male sex exceeded that of BLL, and BLL contributed a minimal increase in predictive accuracy beyond that of male sex.

Figure 2. Predicting crime: blood lead level (BLL) (red), male sex (blue), BLL and male sex (black).

We assessed the ability of BLL to correctly predict criminal conviction. The AUC is an indicator of a model’s predictive accuracy. An AUC value of 0.5 (the diagonal line) indicates poor predictive accuracy; that is, the model variables are no better at predicting the outcome than random chance. Points above the diagonal represent better than random predictive accuracy. An AUC value of 1 indicates perfect predictive accuracy. The figure shows that blood lead level was a worse predictor of criminal conviction in the Dunedin cohort than male sex and added little to the model beyond sex.

Logistic regression models supported the positive association between BLL and conviction, but sex-adjusted results failed to reach statistical significance (see Table). Specifically, each 5 μg/dL higher BLL was significantly associated with a 1.29 increase in the odds of criminal conviction (95%CI, 1.06, 1.56; P=.013). After controlling for sex, the association between BLL and conviction was slightly attenuated and not statistically significant (adjusted odds ratio [aOR] 1.23; 95%CI, 1.00, 1.51; P=.052).

Analysis of recidivism revealed that BLL was not significantly associated with the odds of recidivism (conviction subgroup comparison 1 in the Table). BLL was not significantly associated with being a one-time offender (odds ratio [OR] 1.29; 95%CI, 0.99, 1.68; P=.058; aOR 1.25; 95%CI, 0.95 1.64; P=.107). Initially, comparing offenders who recidivated to non-offenders, each 5 μg/dL higher BLL was significantly associated with a 1.28 increase in the odds of being a recidivating offender (95%CI, 1.01, 1.63; P=.046). However, after controlling for sex, the association between higher BLL and being a recidivating offender was slightly attenuated and not statistically significant (aOR 1.21; 95%CI, 0.93, 1.57; P=.149). In each of the models the estimated odds ratio was slightly lower for being a recidivating offender than for being a one-time offender.

Analysis of violence revealed that BLL was not significantly associated conviction for violence (conviction subgroup comparison 2 in the Table), although it was associated with conviction for nonviolence. Comparing nonviolent offenders to non-offenders, each 5 μg/dL higher BLL was significantly associated with a 1.33 increase in the odds of being a nonviolent offender (95%CI, 1.06, 1.67; P=.013), and this association remained significant after controlling for sex (OR=1.28, 95%CI 1.01, 1.61; P=.041). However, BLL was not significantly associated with being a violent offender and the estimated odds ratio was slightly lower than that for being a nonviolent offender (OR 1.20; 95%CI 0.89, 1.62; P=.241; aOR 1.13; 95%CI 0.82, 1.55; P=.449).

Self-reported offending variety was weakly associated with BLL and only reached statistical significant at ages 15, 18, and 26 years (see Table). After controlling for sex, the association between higher BLL and self-reported offending variety remained weak and was statistically significant only at assessment age 15 years (r=0.10; 95%CI, 0.01, 0.18; P=.023).

Discussion

The present study is, to our knowledge, the only study to examine the association between childhood BLL and criminal offending in a cohort where the association is not intertwined with social class. We found weak associations between BLL and criminal offending. BLL predicted criminal conviction at chance levels and was outperformed by male sex. Contrary to expectations, the estimated effect of BLL was lower for recidivistic than one-time conviction and lower for violent than nonviolent offending. Associations with self-reported offending were weak and, with the exception of age 15 years, not statistically significant. Our results failed to support a consistent dose-response relationship between BLL and criminal offending, measured through official and self-reports, from adolescence to adulthood. This finding does not negate the evidence of detrimental health effects of lead. At-risk children should continue to be tested for lead and efforts to reduce lead exposure should continue.

Our findings of a tenuous association between BLL and offending run contrary to past research, which has shown a dose-response lead-crime relationship.^15–19 Our findings are unlikely due to poor-quality measurement of lead in the blood because past research from the Dunedin Study that used the same measure of blood lead as used in the present study found that higher childhood BLL was associated with declines in intelligence test scores and downward social mobility at 38 years of age.

Our results may be due to the absence of three factors found to be problematic in past studies indicating dose-response effects of BLL. First, past lead-crime association studies, in which high BLL and low socioeconomic status were associated, may not have completely overcome confounding.^30,31 Second, many past lead-crime studies used high-risk or clinical samples, which meta-analysis has shown to yield higher effects of lead on crime.⁴⁹ Third, past longitudinal lead-crime studies have notable attrition, which may upwardly bias estimates if high BLL-low crime or low BLL-high crime individuals disproportionately leave the sample.

Limitations

This study has limitations. First, BLL was only measured at one point, age 11 years. Multiple BLL measures are desirable. Yet, evidence indicates that the mean of BLL measurements across childhood can yield results similar to those from a single, later-childhood measurement, as used here.⁵⁰ Past research analyzing multiple BLL measures from early to middle childhood also finds the strongest effects among BLL measurements most proximal to the outcome, as we found with self-reported offending at age 15 years.^51,52 Second, this study’s findings may not generalize to non-white children. High blood lead levels are prevalent among children in many non-Western countries across the world, and in African American children in the United States.^53,54 There is currently no evidence to suggest that racial or ethnic groups differ on lead metabolism. In the United States, specifically, lead’s effects may be amplified by the stress of racial disparity and discrimination, or by malnutrition, which are often experienced by African American children in the United States.^5,55 Third, the study was also predominantly of children from non-rural areas. Mean BLL in the Dunedin Study, 11 μg/dL, was comparable to BLLs in other developed locations in the 1980’s, but this limits inferences about the effects of lower levels of lead seen in present-day American children.⁵³ Fourth, our crime outcomes were assessed at age 15 years and older; there is evidence that elevated BLLs are associated with childhood and pre-teen problem behaviors.^51,53

Conclusions

This study fails to support a dose-response relationship between BLL and criminal offending in a sample in which there was no association between BLL and childhood socioeconomic status. Previously detected associations between BLL and criminal offending may be due to lead toxicity disproportionately affecting disadvantaged groups. Responses toward lead exposure should focus on health consequences, not potential criminal consequences.

Key Points.

Question

Is childhood lead exposure associated with criminal offending in a setting where the degree of lead exposure was unrelated to socioeconomic status?

Findings

In 553 New Zealanders observed for 4 decades, lead exposure in childhood had a weak relationship with official criminal conviction and self-reported offending from ages 15 to 38 years. Lead exposure did not predict consequential offending outcomes: greater variety of offenses, conviction, recidivism, or violence.

Meaning

Responses toward lead exposure should focus on consequences for health, not potential consequences for crime.