Association of height and violent criminality: results from a Swedish total population study

Abstract

Background: Violent criminality is at least moderately heritable, but the mechanisms behind this remain largely unexplained. Height, a highly heritable trait, may be involved but no study has estimated the effect of height on crime while simultaneously accounting for important demographic, biological and other heritable confounders.

Methods: We linked nationwide, longitudinal registers for 760 000 men who underwent mandatory military conscription from 1980 through 1992 in Sweden, to assess the association between height and being convicted of a violent crime. We used Cox proportional hazard modelling and controlled for three types of potential confounders: physical characteristics, childhood demographics and general cognitive ability (intelligence).

Results: In unadjusted analyses, height had a moderate negative relationship to violent crime; the shortest of men were twice as likely to be convicted of a violent crime as the tallest. However, when simultaneously controlling for all measured confounders, height was weakly and positively related to violent crime. Intelligence had the individually strongest mitigating effect on the height-crime relationship.

Conclusions: Although shorter stature was associated with increased risk of violent offending, our analyses strongly suggested that this relationship was explained by intelligence and other confounding factors. Hence, it is unlikely that height, a highly heritable physical characteristic, accounts for much of the unexplained heritability of violent criminality.

Violence, body height, stature

Key Messages

• Height is unlikely to contribute to risk prediction for violent crime.

• Height is not responsible for a large portion of the unexplained heritability of violent criminality.

• Anthropometric characteristics, childhood demographics, and intelligence do appear to be important in explaining violent criminality.

Introduction

Adult height is associated with a variety of outcomes, such as higher education,¹ lower suicidality² and better psychiatric and somatic health.^3–6 Yet, the relationship between height and interpersonal violence remains unclear.Further, considering the high heritability of height (at least 80%),^7–10 it maybe a mediator of the previously detected, but only partially explained, heritability of violent criminal offending.^11,12 If the genetic factors influencing height were the same as those influencing crime, we would be one step closer to determining which genes may be responsible for violent behaviour. Yet, there are a number of competing hypotheses that could explain any association between height and crime, having little or nothing to do with the genetic properties of height. Considering the public health concern of violent crime,^13–15 it is of considerable importance to explore its aetiology to form appropriate intervention strategies.

Early scholars such as Sheldon¹⁶ and Lombroso¹⁷ argued that height may independently affect crime because a taller person is more physically capable of inflicting violence, especially upon a relatively short person. This notion, however, contradicts several important empirical results on relationships between height and other factors that affect violent criminality.

Many researchers have used height as a measure of human welfare,¹⁸ since shorter adult height may indicate childhood disadvantage.^19,20 Childhood disadvantage, in turn, usually predicts negative adult outcomes,²¹ including a higher propensity for violent criminality, and may confound detected links between adult height and crime.

Along with the high overall heritability of height, some 200 genetic loci have been identified.²² Many of these genes are likely to affect growth more generally and may influence adult outcomes through other pathways such as general cognitive ability or intelligence. Intelligence is associated with adult height,^23–25 possibly by its relatively strong correlation to brain volume.²⁶ Intelligence is also a relatively strong predictor of adult education¹ and income, and a moderate to strong negative predictor of criminal behaviour.^27–30

Many previous studies on the height-crime relationship did not simultaneously adjust for confounding from these factors.^31–35 This raises questions about the validity of significant outcomes indicating a positive effect of height. Nevertheless, height may still be a causal risk factor for crime. Case and Paxson²⁴ addressed height and economic outcomes in longitudinal samples from the USA and UK. After childhood cognitive ability was accounted for, the positive effect of height was attenuated but persisted. Evidence from Sweden suggested that taller men were more likely to obtain a higher education than shorter men, even after adjusting for cognitive ability.¹ Some researchers have suggested that shorter people may face discrimination, which could lead to these results.^1,2

We used a large sample of all Swedish men who underwent compulsory military conscription during 1980–92, to estimate the magnitude and direction of the association of adult height with violent criminality. By linking nationwide longitudinal registers, we were able to adjust for several potential confounders of the height-crime relationship: muscular strength and body mass (BMI), childhood sociodemographic factors, and general cognitive ability (intelligence) at 18–19 years of age.

Methods

We linked six of Sweden’s longitudinal, nationwide administrative population registers. We used the Conscription Register as our sampling frame. Conscription was mandatory for Swedish men until 2007. Estimates from the 1990s show that over 95% of the young male population participated in conscription; those who did not were typically excluded because of severe somatic disorders or mental retardation.³⁶ The Conscription Register provided information on age, weight, height, cognitive performance and isometric muscle strength. We extracted data for all men conscripted 1980 through 1992, approximately 760 000 men (born approximately 1959 to 1975). We excluded those: of 22 years of age or older at conscription; born outside Sweden; with invalid values on any studied variables; or with BMI values <15 or >39. With these restrictions the resulting sample was 713 877 men.

We used the Cause of Death and Migration Registers to track whether subjects were deceased or had migrated, respectively; subjects not appearing in these registers were assumed to be able to be convicted of an offence in Sweden. The Cause of Death Register contains information on all deaths among registered Swedish citizens and residents, even if they died while abroad. The Migration Register provides data on all immigration and emigration for registered citizens and residents of Sweden, and birthplace information. The Crime Register supplied information on the type of crime and offence date for all court convictions from 1973 to 2009 (less than 0.1% of convictions were missing personal identification numbers and were thus excluded from this study). It is estimated that 1% of these convictions were overturned on appeal.³⁷ In Sweden, the age of criminal responsibility is 15 years; offences committed before this age are unrecorded. In this register, guilty convictions comprise all convictions regardless of the reason (e.g. insanity) or the sentencing outcome (e.g. sentences to forensic psychiatric treatment). Also, plea-bargaining is not allowed in Sweden.

We used the Multi-Generation Register to link biological and adopted children to their parents. To be included in the register, parents must have lived in Sweden after 1947, when national personal identification numbers were introduced. Nearly all of those born in Sweden after 1967 are connected to their parents.³⁷ Using this linkage, we connected children to their parents’ socioeconomic status according to the 1970 and 1975 National Censuses.

Outcome

This was defined as the first conviction for a violent crime. Our definition of violence included homicide, assault, threatening behaviour, coercion, kidnapping and false imprisonment, intimidation, robbery and violence and threats against a peace officer, as well as sexual crimes. Attempted and aggravated offences were included whenever applicable.

Exposure

The exposure was height measured in standing centimetres.

Confounders

The first set of confounders was additional anthropometric characteristics of strength and fitness. According to early criminologists, these features could account for the ability to commit acts of violence.^38,39 Isometric muscle strength was measured using the IsoKai machine, which primarily captures leg, back and shoulder strength.⁴⁰ The original metric was in newtons, which was then converted into nine categories, with higher values indicating greater strength. The body mass index (BMI) simultaneously considers height and weight (kilograms divided by the squared height in metres). BMI was used as a fitness measure.

The second set of confounders, childhood sociodemographic characteristics, may be seen as confounders of the height-crime relationship resulting from the link between height and childhood disadvantage. We included the decile of taxed income of the head of household from the 1970 or 1975 Censuses (whichever appeared first in the data), occupation type for the head of household categorized in 10 categories according to Statistics Sweden’s socioeconomic index, a dichotomous variable indicating whether the head of the household was a single mother, maternal age at birth, parity of mother and the index man’s number of maternal full and half-siblings at the end of follow-up.

Third, as a measure of general cognitive ability (intelligence), we used the conscript’s performance on the compulsory Swedish Enlistment Battery (SEB80). SEB80, subdivided into sections aimed at capturing different aspects of cognitive ability, has been shown to better reflect overall general cognitive ability than individual aspects of intelligence.⁴¹

Analytic strategy

Following descriptive data analysis, we used Cox proportional hazard regression which yields estimates of the relative hazard of violent crime at any given age covered by the data. The effect of height on violent crime was treated as linear and analyses were conducted with subjects with complete information on all variables. No methods for imputation were used. The unit of time was age in years, beginning at age 15. Individuals who tentatively entered the sample after age 15 were excluded due to lack of possible criminal conviction data (left-censoring). All models were adjusted for birth year to address potential cohort effects, as some evidence from Norway found a declining correlation between height and intelligence over time.^42,43 All individuals in a given birth cohort were subject to the same time at risk. The mean time at risk was approximately 27 years across both offenders and non-offenders, and the oldest individuals were followed until the age of 50. We followed individuals until their first criminal conviction. When the individual was not convicted, he was followed until the end of the follow-up period, migration or death. There is a possibility of data inaccuracies as a result of people not reporting moves abroad. However, we assumed that emigration and any inaccuracies in that variable did not bias the results on our key variable, height.

The first model provided unadjusted associations between height and violent crime without controlling for covariates. Models 2–4 considered height along with each of the three classes of control variables separately. The fifth and final model included all control variables simultaneously. Finally, we included a sensitivity analysis to test for non-linearity in the height-crime relationship. All models were estimated using the statistical software R⁴⁴ and the survival package.⁴⁵

Results

Descriptive information is presented in Table 1. On average, convicted violent offenders were about 0.78 centimetres shorter than non-violent men, were slightly weaker and had marginally higher BMIs. They also experienced lower parental household income, had parents who were less likely to be employed in the technical sector and more likely to be employed in industry and were more likely to have been raised by a younger and single mother, have a higher birth order and more siblings. Convicted violent criminals also had about 1.3 Stanine units lower total IQ; 89% of the sample had complete data on all variables and were used for remaining analyses (n = 632 965 men). The Pearson correlation between height and IQ was 0.13 (95% confidence interval: 0.13–0.14).

Table 1

Descriptive statistics for all men born 1960–82 in Sweden who underwent mandatory conscription by violent criminal conviction status

Variable	Convicted of violent crime
	Yes		No
	Mean (SD)	N	Mean (SD)	N
Height (cm)	178.64 (6.52)	49439	179.42 (6.53)	664438
Muscular strength	5.72 (1.60)	47548	6.18 (1.65)	642436
Missing	NA	1891	NA	22002
BMI	22.19 (3.05)	49439	21.81 (2.82)	664438
Parental incomea	5.13 (2.51)	48596	6.04 (2.34)	654985
Missing	NA	843	NA	9453
Childhood SESb (%)
Agriculture, forestry (self-employed)	1.74	859	3.70	24586
Agriculture, forestry (labourer)	2.15	1064	2.00	13270
Industry, commerce, transport or service trades (self-employed)	5.35	2647	4.94	32856
Independent professions: medical doctor, attorney (self-employed)	0.40	198	0.64	4243
Director (employed)	1.09	537	2.23	14802
Technical, humanitarian, office or commercial trades	24.43	12080	37.25	247495
Industry or transport	47.07	23273	38.97	258915
Service trades	3.41	1688	1.77	11790
Military	0.64	314	0.98	6483
Non-identified occupation	0.13	65	0.08	513
Missing	13.58	6714	7.45	49485
Single mother household (%)
No	71.02	35114	84.02	558262
Yes	28.04	13861	15.07	100108
Missing	0.94	464	0.91	6068
Maternal age at birth of index persona	25.17 (5.55)	49406	26.69 (5.40)	664054
Missing	NA	33	NA	384
Birth ordera	2.03 (1.23)	49406	1.89 (1.07)	664054
Missing	NA	33	NA	384
Number of maternal siblingsa	1.99 (1.41)	49406	1.69 (1.15)	664054
Missing	NA	33	NA	384
IQ (9-level Stanine score)	3.93 (1.72)	49223	5.24 (1.89)	663051
Missing	NA	216	NA	1387
Age at conscription (years) (%)
17	0.08	41	0.12	771
18	84.98	42011	86.98	577948
19	11.76	5813	11.81	78478
20	2.30	1135	0.87	5796
21	0.89	439	0.22	1445
Birth year (%)
1959	0.09	45	0.02	116
1960	0.04	20	0.01	74
1961	1.12	556	0.70	4637
1962	8.23	4067	6.92	45957
1963	8.46	4183	7.41	49202
1964	8.93	4415	7.91	52573
1965	8.44	4173	7.66	50892
1966	8.08	3995	7.68	51046
1967	7.26	3590	7.00	46543
1968	6.61	3270	6.76	44906
1969	6.53	3226	7.04	46758
1970	6.69	3307	7.10	47186
1971	6.73	3326	7.37	48974
1972	6.27	3102	7.22	47947
1973	6.04	2987	6.97	46296
1974	6.01	2972	6.85	45521
1975	4.46	2205	5.39	35810

NA, not applicable.

^aTable displays mean values. Categorical values entered into models.

^bDefined as highest occupation of parental head of household in National Censuses 1970 or 1975.

Models 1–5 (Table 2) display Cox proportional hazard modelling results with hazard ratios adjusted to represent 10-cm increments in height. Schoenfeld residuals indicated that the proportional hazard assumption was not violated for the height variable. Model 1 indicated a 15% reduction in the hazard of a violent crime conviction per 10-cm increase in height. Model 2 included additional anthropometric variables. This model had a significantly improved fit over the first and showed an 11% reduction in the violent crime hazard per 10-cm increase in height. The third model included sociodemographic variables and height. Height remained a significant predictor of violent crime, and a 10-cm increase was related to a 7% reduction in the hazard of a violent crime conviction, even lower than when controlling for the additional anthropometric covariates (Model 2). Model 4 included intelligence and found a 2% decrease in the hazard for violent crime conviction for each 10-cm increase in height. The fifth, final model considered all covariates simultaneously; the effect estimate for height changed direction, from negative to positive. A 10-cm increase in height was now related to a 3% increase in the hazard of conviction for a violent crime.

Table 2

Height at conscription (models in 10-cm increments) and hazard ratio(HR) (confidence interval, CI) of violent criminal convictions among men born 1960–82 in Sweden

HR (95% CI)	Model 1	Model 2	Model 3	Model 4	Model 5
	0.85 (0.84-0.86)	0.89 (0.88-0.91)	0.93 (0.92-0.94)	0.98 (0.96-0.99)	1.03 (1.01-1.05)
Adjusted for	–	Additional anthropometric variables	Childhood sociodemographic variables	Cognitive ability	Anthropometric, sociodemographic, and cognitive ability

The sensitivity analyses showed that the interpretation of the results did not change when testing for a non-linear association between height and crime (data not shown; see full model results of the main analysis as Supplementary data at IJE online).

Discussion

We linked nationwide, longitudinal registers for all 760 000 men who underwent mandatory military conscription in Sweden 1980–92, to test whether increased height was related to a lower risk of violent criminality. In initial unadjusted analyses, the shortest men had an almost doubled risk of violent crime conviction compared with the tallest men. However, the height-crime relationship was attenuated when controlling for three groups of potential confounders: anthropometric characteristics, childhood demographics and general cognitive ability. The effect changed direction, from negative to positive, when all three categories of covariates were simultaneously adjusted for. This means that, given two men of different height but otherwise similar anthropometric characteristics, childhood demographics and general cognitive ability, the taller of the two is more likely to be convicted of a violent crime.

One possible explanation for the positive link between height and violent crime conviction when controlling for covariates is that taller people may be more physically able to threaten to or actually use interpersonal violence. However, this was contradicted by the observation that muscular strength did not behave similarly to height; the negative association between strength and violent crime remained when adjusting for all other covariates. More importantly, the effect of height is very minimal and the results indicated that the confounders were more important for the violent crime outcome.

There are some limitations to this study. First, the outcome used was convictions for violent crime, a legal procedure, not acts of violent behaviour. Acts of violence not handled by the criminal justice system were not captured in our analysis. Estimates from Sweden’s crime victimization survey suggest that, overall, about 30–40% of all violent criminal behaviours against a person are reported.⁴⁶ Sexual offences appear to be the least reported (14% in 2009) and aggravated assaults appear to be the most reported (64% in 2009). We assume that height is randomly distributed throughout unreported acts of violence. Also, in Sweden, rates and resolution of violent crime are similar to those of other EU countries.⁴⁷ The inability to capture all acts of violence should not affect the reliability or generalizability of the results.

Second, since we excluded immigrants from this analysis, it is difficult to say whether these results are generalizable to immigrants, especially if they grew up in less developed countries where height is more likely to be influenced by malnourishment.⁴⁸ In such cases, the economic variables may have a stronger mitigating effect on the height-violent crime association than found here.

Third, there may be problems with the confounders in the analysis. It is, for instance, likely that our measures of childhood socioeconomic factors imperfectly captured early social circumstances due to random measurement error. Without measurement error, the adjusted, positive association of height and violent crime may have been slightly stronger. Also, there is the possibility that we have omitted confounders of the height-crime relationship in our analysis. Such an omission could have masked a true negative association.

Fourth, we adjusted for many covariates although underlying causal associations are not well known. If any of the potential confounders are actually mediators of the possible effect of height on violent crime, we introduced bias by ‘closing’ a causal path, and potentially by creating spurious associations through common causes of the mediator and violent crime.⁴⁹ However, since the childhood demographic variables were measured many years before the conscription testing, they are unlikely to be potential mediators. In contrast, intelligence, BMI, muscular strength and height were measured at the same time point. We are reluctant to suggest that height has a causal influence on either of the other traits, just as we do not believe that either of these traits have a causal influence on height. For instance, if the association between height and intelligence were due to greater brain volume, we should not consider this an effect of height, but rather a latent growth factor being a common cause to both traits. By adjusting for intelligence and the other traits, we hoped to block confounding by such latent factors.

In conclusion, we used population data to evaluate the relationship between height and criminality. In unadjusted analyses, height had a moderate negative relationship to violent crime; the shortest of men born in Sweden were twice as likely to be convicted of a violent crime as the tallest. This association was attenuated and even reversed by adjusting for physical or anthropometric characteristics, childhood demographics and intelligence—which appeared to have the strongest effect. The adjusted positive association of height and violent crime was significant, but marginal. Hence, height is unlikely to contribute to risk prediction for violent crime when other known risk factors are taken into account, and we found no evidence to suggest that height is responsible for more than a negligible portion of the unexplained heritability of violent criminal behaviour.