Skip to main content
NCBI home page
JAMA Network logoLink to JAMA Network
. 2018 Sep 5;75(11):1182–1188. doi: 10.1001/jamapsychiatry.2018.2337

Academic Achievement and Drug Abuse Risk Assessed Using Instrumental Variable Analysis and Co-relative Designs

Kenneth S Kendler 1,2,✉, Henrik Ohlsson 3, Abigail A Fagan 4, Paul Lichtenstein 5, Jan Sundquist 4,6,7,8, Kristina Sundquist 4,6,7,8
PMCID: PMC6237552  NIHMSID: NIHMS993940  PMID: 30193341

Key Points

Question

To what extent is the known association between poor academic achievement and risk of drug abuse influenced by causal processes?

Findings

Instrumental variable and co-relative analysis designs, implemented in large population-based Swedish samples with a total of 934 462 participants at a mean follow-up of 19 years, suggested that the association between academic achievement and drug abuse may be causal.

Meaning

These results provide empirical support for efforts to improve academic achievement as a means to reduce risk of drug abuse.

Abstract

Importance

Low academic achievement (AA) in childhood and adolescence is associated with increased substance use. Empirical evidence, using longitudinal epidemiologic data, may provide support for interventions to improve AA as a means to reduce risk of drug abuse (DA).

Objective

To clarify the nature of the association between adolescent AA and risk of DA by using instrumental variable and co-relative analysis designs.

Design, Setting, and Participants

This study, assessing nationwide data from individuals born in Sweden between 1971 and 1982, used instrumental variable and co-relative analyses of the association between AA and DA. The instrument was month of birth. Co-relative analyses were conducted in pairs of cousins (263 222 pairs), full siblings (154 295), and monozygotic twins (1623) discordant for AA, with raw results fitted to a genetic model. The AA-DA association was modeled using Cox regression. Data analysis was conducted from October 2017 to January 2018.

Exposures

Academic achievement assessed at 16 years of age (for instrumental variable analyses), and estimated discordance in AA in pairs of monozygotic twins (for co-relative analyses).

Main Outcomes and Measures

Drug abuse registration in national medical, criminal, or pharmacy registries.

Results

This instrumental variable analysis included 934 462 participants (478 341 males and 456 121 females) with a mean (SD) age of 34.7 (4.3) years at a mean follow-up of 19 years. Earlier month of birth was associated with a linear effect on AA, with the regression coefficient per month equaling −0.0225 SDs (95% CI, −0.0231 to −0.0219). Controlling for AA, month of birth had no association with risk of DA (hazard ratio [HR], 1.000; 95% CI, 0.997-1.004). Lower AA had a significant association with risk of subsequent DA registration (HR per SD, 2.33; 95% CI, 2.30-2.35). Instrumental variable analysis produced a substantial but modestly attenuated association (HR, 2.04; 95% CI, 1.75-2.33). Controlling for modest associations between month of birth and parental educational status and DA risk reduced the association to a HR of 1.92 (95% CI, 1.67-2.22). The genetic model applied to the results of co-relative analyses fitted the observed data well and estimated the AA-DA association in monozygotic twins discordant for AA to equal a HR of 1.79 (95% CI, 1.64-1.92).

Conclusions and Relevance

Two different methodological approaches with divergent assumptions both produced results consistent with the hypothesis that the significant association observed between AA at 16 years of age and risk of DA into middle adulthood may be causal. These results provide empirical support for efforts to improve AA as a means to reduce risk of DA.

To clarify the nature of the association between adolescent academic achievement and drug abuse into middle adulthood, this study uses 2 methods for assessing associations in observational studies, instrumental variable (with month of birth as the instrument) and co-relative analyses, and long-term follow-up to assess data from 934 462 individuals born in Sweden between 1971 and 1982.

Introduction

Poor academic achievement (AA) in adolescence is associated with an increased risk of drug use and subsequent drug abuse (DA).1,2,3,4,5,6 Students who succeed academically tend to develop positive attachments to school, facilitating their commitment to prosocial lifestyles that reduce risk of DA.7 Those who lack this bond are more prone to deviant behaviors, including DA.8

The association between AA and DA is of interest given the prevalence and adverse consequences of drug abuse. In the United States, 7.4 million people aged 12 years or older reported symptoms of a drug use disorder in 2016.9 The National Center on Addiction and Substance Abuse10 estimated that substance abuse and addiction cost the US government 467 billion dollars in 2005. If poor AA contributes causally to risk of DA, then interventions that improve AA should reduce DA risk. Indeed, a range of individual and school-based interventions for children and adolescents that have shown an impact on AA also resulted in lower drug use.11,12,13,14 However, nearly all of these studies examined only substance use and follow-ups were typically short, making it difficult to confidently conclude from the current literature whether improving AA will have a long-term influence on risk of DA.

Furthermore, although the positive effects from intervention research suggest a causal relationship between AA and DA, this association could arise from several confounders15 not always accounted for in these studies. For example, familial factors likely increase risk of both poor AA and DA.16,17 In addition, reverse causation cannot be ruled out because substance use or abuse during the school years is associated with poor AA.18,19,20,21

To address the nature of the association between AA and long-term risk of DA, we conducted instrumental variable (IV) analyses22,23 (Figure 1). Such an analysis requires the identification of an “instrument,” that is, a variable that influences the risk factor of interest—here AA—but has no direct influence on the key outcome—here DA.

Figure 1. Model for an Instrumental Variable, in This Case Month of Birth.

Figure 1.

Open in a new tab

The key feature of such a variable is that it is associated with an independent variable—here, academic achievement—which in turn contributes to the dependent variable—here, risk of drug abuse. Furthermore, the instrumental variable has no association with the dependent variable except as mediated through the independent variable.

Our instrument, chosen for its empirical association with AA, is month of birth. Many Western countries, including Sweden, have a cutoff date for enrollment in school such that in any class of students nearly all of them come from a single birth year and differ in age by as much as 12 months. Studies in several European countries12,24,25,26,27,28 have shown that within the same year of birth and school class, because of age-related cognitive maturation, older students generally perform better academically than younger ones.

In the present study, we performed an IV analysis in which the independent variable was AA assessed at 16 years of age in a complete population cohort in Sweden with a follow-up period ranging from 15 to 20 years. The dependent variable was DA as assessed from population-wide medical, criminal, and pharmacy registries. By comparing the raw association between AA and DA to that obtained from the IV analysis, we can gain insight into the proportion of the observed association between the 2 variables that may be causal. We also conducted a co-relative analysis29,30 of the association between AA and DA, a different approach to assessing possible causal relationships in observational data. The co-relative analyses estimate, within monozygotic twins, the association between differences in AA and differences in risk of DA, thereby controlling for genetic effects and the impact of rearing in the same household and community.31

Methods

Sample

We analyzed information about individuals from Swedish population-based registers with national coverage linked using each person’s unique identification number replaced by a serial number to preserve confidentiality and anonymity. All procedures contributing to this work complied with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.32 We secured ethical approval for this study from the Regional Ethical Review Board of Lund University, Malmö, Sweden, which also waived the need for obtaining informed patient consent.

The National School Registry contains AA data (ie, grade point average) for all students in grade 9 (usually age 16) from 1988 to 1997. Education is mandatory for all children in Sweden between 7 and 16 years of age. Students are incentivized to perform well because those scoring well are likely to gain admission to desirable secondary schools. We standardized this grade score for each year and sex (mean 0 and SD 1), calling it AA. From 1988 to 1997, scores were expressed on a scale from 1 and 5 (higher numbers indicate higher AA), and students were assessed by a peer-referencing system.33 Under this system, grades had minimal inflation over time and were normally distributed. For a definition of DA, see eAppendix in the Supplement.

The database included all individuals born in Sweden between 1972 and 1981 who had not died or emigrated prior to 16 years of age and were registered in the National School Register the year they turned 15, 16, or 17 years old. The database also included their first year (if any) of registration for DA. Individuals with DA registered at age 16 years or earlier were excluded from the analyses.

Although the school grade tested for AA was mostly composed of individuals who turned 16 years old between January 1 and December 31, for approximately 3% of individuals, their AA was instead registered the year they turned 15 or 17 years old (eTable 1 in the Supplement). The month of birth for these individuals was not randomly distributed: more individuals born in January were registered with AA the year they turned 15 years old; more individuals born in December were registered with AA the year they turned 17 years old. Furthermore, individuals registered with AA at age 15 years had a mean AA of 0.74 SDs, whereas individuals registered at age 17 years had an AA mean of −0.91 SDs. Because the association between month of birth and AA was central to our further analyses, we modeled this nonrandomness. Using all individuals regardless of age at AA, we fitted a regression model estimating AA based on month of birth and age at registration. In these analyses, we used, for individuals registered the year they turned 15 or 17 years old, their estimated value, and for individuals with AA the year they turned 16 years old, their true value. We also found similar results using analyses that included only individuals whose AA was assessed the year they turned 16 years old (eTable 3 in the Supplement).

Statistical Analysis

First, we used a Cox proportional hazards model to investigate the risk of DA as a function of AA, from year of AA registration until end of follow-up (DA registration, death, emigration, or 2012). The hazard ratio (HR) represented the increased risk of DA per 1 SD decrease in AA. This is the crude association between AA and DA used for comparison. The proportionality assumption was checked in all models, and no meaningful violations were noted.

We then used an IV approach to control for unmeasured confounding using month of birth as an instrument. Below, we empirically evaluate the suitability of month of birth as an instrument. We used a 2-stage regression model adapted to a Cox regression framework. The first step was a linear regression of month of birth on AA. The predicted values were thereafter used in a Cox regression model as an exposure variable (instead of the actual AA value). Because the unobserved and observed confounders should be equally distributed among the estimated values of AA, the resulting HRs for AA should be controlled for unmeasured and measured confounding. To obtain 95% CIs, we used a nonparametric bootstrap with 1000 replications.

Methodological concerns have been raised regarding the use of month of birth as an instrument because of potential systematic differences among children born at different times of the year.34 Indeed, as is given in eTable 2 in the Supplement, our sample showed a small but significant decrease in DA among parents with children born in March and April (χ211 = 119.9; P < .001), and a significantly increased parental educational level for parents of children born in March, April, May, September, and October (F11/934 450 = 8.1; P < .001). As these results might violate a key IV assumption (that month of birth has no direct influence on risk of DA), we conducted sensitivity analyses by adding controls for parental DA and educational level in both the first and second stage of the IV analysis to examine their influence on our model results.

In the second analysis, we compared the results from the IV approach with those from a co-relative design, wherein we examined whether the regression results (ie, the crude association between AA and DA) reflected confounding by familial risk factors. From the Swedish Multi-Generation and Twin Registers,35 we identified all monozygotic (MZ) twin, full-sibling, and cousin pairs. Using stratified Cox proportional hazards models, with a stratum for each relative pair, we refitted the analysis (ie, the risk of DA as a function of AA). The HR was then adjusted for unmeasured genetic and environmental factors shared within the relative pair. The MZ twins share 100% of their genes and their rearing environment, suggesting that the HR for MZ twins is controlled for all possible confounding by genes and shared environment. Full siblings and cousins share, respectively, on average 50% and 12.5% of their genes identical by descent. Finally, we combined all 4 samples (ie, population, twins, siblings, and cousins) into 1 data set in which we performed 2 analyses. The first allowed for all parameters in each sample to be independent (ie, similar to 4 separate analyses). In the second analysis, we modeled the association between AA and DA with 2 parameters: 1 main effect and 1 as a linear function of the genetic resemblance (ie, 0 for the population; 0.125, for the cousins; 0.5 for the siblings; and 1 for the MZ twins). The HR for the second parameter gives an indication of the size of the familial confounding. If the second model fitted the data well, as indexed by the Akaike Information Criterion (AIC),36 we also obtained an improved estimation of the association among all relatives, but especially for MZ twins, in which the data were sparse. All statistical analyses were performed from October 2017 to January 2018 using SAS, version 9.4 (SAS Institute Inc).37

Results

IV Analysis

Included in our IV analyses were 934 462 participants (478 341 males and 456 121 females) with AA scores at a mean (SD) age of 16.0 (0.3) years and a mean year of birth of 1976 (2.9 years). Their mean (SD) age at follow-up was 34.7 (4.3) (range, 31-40) years. Of these participants, 33 259 (3.6%) were subsequently registered for DA at a mean (SD) age of 26.7 (5.2) years. The mean (SD) AA of those with a subsequent DA registration was −0.83 (1.0) SDs below the population mean.

The ability of an IV to provide information about the nature of the association between the independent and dependent variables requires that it should both meaningfully predict the independent variable (here, low AA) and show minimal association with the dependent variable (here, DA) except that mediated through the independent variable (Figure 1). Aside from an anomaly in January births, there was a clear monotonic association between our adjusted estimate of month of birth and AA (Figure 2). Modeled as a linear effect, the regression coefficient was estimated at −0.0225 (95% CI, −0.0231 to −0.0219), meaning that being 1 month younger reduced AA by a mean of 2.25% of an SD.

Figure 2. Association Between Month of Birth and Academic Achievement Assessed at 16 Years of Age in Our Sample of 934 462 Swedish Adolescents.

Figure 2.

Open in a new tab

A linear model estimated that being 1 month younger reduced academic achievement in this sample by a mean of 2.25% of a standard deviation (SD). These results included a statistical correction for the 3% of the sample assessed at 15 or 17 years of age and for the slight but statistically significant variation in parental educational level and risk of drug abuse as a function of offspring month of birth. The y-axis depicts SD units.

The raw association between month of birth and DA was modest but statistically significant (HR, 1.016; 95% CI, 1.013-1.019). However, in accord with the assumptions of the IV model, when AA was added to the model, the association between month of birth and DA risk disappeared (HR, 1.000; 95% CI, 0.997-1.004).

In our entire sample, lower AA had a significant association with risk of subsequent DA registration (HR per SD, 2.33; 95% CI, 2.30-2.35). Our IV analysis produced a slightly weaker association (HR, 2.04; 95% CI, 1.75-2.33), consistent with modest confounding. To explore the sensitivity of our estimates to potential biases in our month of birth instrument, we reanalyzed the model controlling for parental educational attainment and DA risk. The estimated association between low AA and risk of DA was still robust although slightly further reduced in magnitude (HR, 1.92; 95% CI, 1.67-2.22).

Co-relative Analysis

The samples sizes of participants in these analyses are given in the Table, which also provides the observed HRs between low AA and risk of DA in the entire sample and then in first-cousin, full-sibling, and monozygotic twin pairs discordant for their level of AA. As expected, the associations were progressively weaker for more closely related pairs of relatives and were quite accurately known for the first 3 groups because of their large sample sizes. However, the association within monozygotic twin pairs was estimated imprecisely because of smaller available samples. On the right side of the Table, we apply our genetic model to these 4 estimates. As measured by AIC, this model fit quite well and substantially increased the accuracy of the HR estimation for the AA-DA association in discordant MZ pairs (HR, 1.79; 95% CI, 1.64-1.92).

Table. Association Between Standardized Low Academic Achievement and Drug Abuse in the General Population and in Cousins, Siblings, and Monozygotic Twins Discordant for Level of Achievement.

Sample Sample Size, No.a Drug Abuse, HR (95% CI)
Observed Predicted
Population 934 462 2.33 (2.30-2.35) 2.33 (2.30-2.35)
Cousin 263 222 2.27 (2.22-2.32) 2.27 (2.26-2.28)
Full sibling 154 295 2.04 (1.96-2.13) 2.04 (1.96-2.13)
Monozygotic twin 1623 1.35 (0.68-2.63) 1.79 (1.64-1.92)
AIC36 NA 906 237.68 906 234.17
Open in a new tab

Abbreviations: AIC, Akaike information criterion; HR, hazard ratio; NA, not applicable.

a

Individuals in the general population sample and pairs discordant for their academic achievement scores in the 4 other samples.

Discussion

We sought to clarify the nature of the association between AA in adolescence and risk of subsequent DA by using 2 different methods for inferring possible causality in observational studies. This is important because of the need to find cost-effective ways to prevent DA.13,38

Our 2 methods, with distinct underlying assumptions, produced convergent evidence that a substantial proportion of the observed association between low AA and subsequent risk of DA may be causal. Both methods estimated that an increase of 1 SD of AA at 16 years of age should decrease long-term risk of DA approximately 45%. These results support the provision of services to students to improve their AA as a method of preventing DA. Several sociological theories may explain our findings. According to both social control theory7 and the social development model,39 students who succeed academically will tend to develop strong positive attachments to school, thereby facilitating their commitment to prosocial lifestyles, which in turn reduces risk of DA. By contrast, those who lack this bond are more prone to a range of deviant behaviors, including DA.

Schools are an optimal setting for delivering such programs given their natural focus on AA and their ability to reach most adolescents. Many school-based programs have been shown to improve AA40 although few have examined and found effects on substance use or abuse.11,12,13,14,41 Such interventions may target individual students with academic services or may be comprehensive interventions that change school policies, organization, or climate to provide a more positive learning environment, thereby increasing the student-school emotional bond.

One school-based intervention, the Good Behavior Game, which is a classroom-based behavior management strategy implemented by teachers in elementary school, evaluated its long-term influence on DA.42 When offered with an enhanced academic curriculum, it increased participants’ test scores and high school graduation rates and reduced drug use.43,44 When participants were followed up to 19 to 21 years of age, DA was reduced 37%, with significant results observed only in males.41 Furthermore, the Good Behavior Game was shown to be cost-beneficial.45

Although the Good Behavior Game has been widely delivered, many evidence-based interventions have not been disseminated,46 and these services must be expanded to produce significant increases in AA and then reductions in DA. Most of these programs are designed for elementary and middle schools, and our results suggest that interventions should be developed and tested for high school students.

While causal effects cannot be determined with confidence from observational data alone, some of these concerns can be alleviated by the use of multiple inferential methods47 or “triangulation.”48 As suggested by philosophers of science, we should have more confidence in results that are robust to variations in analytic procedures,49,50 especially when the methods, as they are here, are quite different, both in implementation and theoretical assumptions. As used here, the IV method relied on the natural experiment of age differences within individual school classes, whereas the co-relative design controlled for familial confounders, especially the expected sharing of all genetic risk factors and rearing environment in MZ twin pairs.

Limitations

These results should be interpreted in the context of 6 potentially significant methodological limitations. First, our assessment of DA was limited to data available from Swedish registries. While such administrative data has important advantages (eg, no refusals or reporting biases), it cannot replicate results of interview-based assessments. Our cases are in general probably more severe than those meeting the DSM-V criteria for substance use disorder51 at interview, although the lifetime prevalence of drug abuse or dependence in nearby Norway is only slightly higher than the estimates obtained in Sweden.52 Second, our measure of AA does not perfectly reflect the way this construct is typically defined in the education literature, which includes students’ attitudes about their teachers, their commitment to school, their educational aspirations, and their level of truancy.4,53,54 Third, we could not assess potential mediators of the association between AA and DA, such as the tendency for low-achieving students to associate with deviant peers.2,8,21

Fourth, our instrument, month of birth, was not without limitations. A portion (3%) of the sample population was not tested at age 16. We statistically corrected for this anomaly and also presented the results when those cases were excluded, obtaining similar results. Furthermore, month of birth was weakly associated with parental educational level and risk of DA. The results did not appreciably differ when we included these variables as covariates. Although imperfect, our IV analyses with month of birth are unlikely to be substantially biased.

Fifth, our co-relative design does not control for environmental confounders specific to individuals that could influence both AA and risk of AD. Finally, while IV analyses should protect against the impact of reverse causation (ie, prior drug use predicting both poor AA and DA risk),55,56 we evaluated this bias by reanalyzing our data including varying buffer periods in which we censored DA registrations soon after the assessment of AA because early drug use should be associated with early DA registration. Buffer periods of up to 8 years produced no meaningful changes in the causal AA-DA associations from our IV or co-relative analyses (eTable 4 in the Supplement).

Conclusions

Two statistical methods, IV and co-relative analyses, based on divergent assumptions both suggested that the association between AA at age 16 years and subsequent risk of DA for a follow-up period of 15 to 20 years may be causal. Consistent with intervention studies,11,12,13,14,41 most of which had much shorter follow-ups and softer outcomes (eg, substance use), these results suggest that programs that improve AA in adolescence should result in meaningful reductions in long-term DA risk.

Supplement.

eAppendix.

eTable 1. Number of Individuals Born 1972 to 1981 and Information About School Achievement Based on Month of Birth

eTable 2. Information on Parental History of Drug Abuse and Level of Education by Month of Birth of Their Proband Offspring

eTable 3. Analyses Including Only Individuals Whose Academic Achievement (AA) Was Assessed the Year They Turned 16

eTable 4. Evaluation of Possible Reverse Causation (Early Onset Drug Use Predicting Both Poor Academic Achievement (AA) and High Risk for Drug Abuse) by Censoring Drug Abuse Registrations Over Varying Buffer Periods From AA Assessment to Ages 18 to 26

Click here for additional data file. (99.2KB, pdf)

References

  • 1.Gauffin K, Vinnerljung B, Fridell M, Hesse M, Hjern A. Childhood socio-economic status, school failure and drug abuse: a Swedish national cohort study. Addiction. 2013;108(8):1441-1449. doi: 10.1111/add.12169 [DOI] [PubMed] [Google Scholar]
  • 2.Schulenberg J, Bachman JG, O’Malley PM, Johnston LD. High school educational success and subsequent substance use: a panel analysis following adolescents into young adulthood. J Health Soc Behav. 1994;35(1):45-62. doi: 10.2307/2137334 [DOI] [PubMed] [Google Scholar]
  • 3.Spooner C. Causes and correlates of adolescent drug abuse and implications for treatment. Drug Alcohol Rev. 1999;18(4):453-475. doi: 10.1080/09595239996329 [DOI] [Google Scholar]
  • 4.Fothergill KE, Ensminger ME, Green KM, Crum RM, Robertson J, Juon HS. The impact of early school behavior and educational achievement on adult drug use disorders: a prospective study. Drug Alcohol Depend. 2008;92(1-3):191-199. doi: 10.1016/j.drugalcdep.2007.08.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Hawkins JD, Catalano RF, Miller JY. Risk and protective factors for alcohol and other drug problems in adolescence and early adulthood: implications for substance abuse prevention. Psychol Bull. 1992;112(1):64-105. doi: 10.1037/0033-2909.112.1.64 [DOI] [PubMed] [Google Scholar]
  • 6.Henry KL, Knight KE, Thornberry TP. School disengagement as a predictor of dropout, delinquency, and problem substance use during adolescence and early adulthood. J Youth Adolesc. 2012;41(2):156-166. doi: 10.1007/s10964-011-9665-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Hirschi T. Causes of Delinquency. Berkeley: University of California Press; 1969. [Google Scholar]
  • 8.Catalano RF, Hawkins JD. The Social Development Model: a theory of antisocial behavior In: Hawkins JD, ed. Delinquency and Crime: Current Theories. New York, NY: Cambridge University Press; 1996:149-197. [Google Scholar]
  • 9.Substance Abuse and Mental Health Services Administration Key substance use and mental health indicators in the United States: Results from the 2016 National Survey on Drug Use and Health. Rockville, MD: Center for Behavioral Health Statistics and Quality, Substance Abuse and Mental Health Services Administration; 2017. [Google Scholar]
  • 10.The National Center on Addiction and Substance Abuse Shoveling Up II: The Impact of Substance Abuse on Federal, State, and Local Budgets. New York, NY: Columbia University; 2009. [Google Scholar]
  • 11.Fletcher A, Bonell C, Sorhaindo A, Strange V. How might schools influence young people’s drug use? development of theory from qualitative case-study research. J Adolesc Health. 2009;45(2):126-132. doi: 10.1016/j.jadohealth.2008.12.021 [DOI] [PubMed] [Google Scholar]
  • 12.Eggert LL, Thompson EA, Herting JR, Nicholas LJ, Dicker BG. Preventing adolescent drug abuse and high school dropout through an intensive school-based social network development program. Am J Health Promot. 1994;8(3):202-215. doi: 10.4278/0890-1171-8.3.202 [DOI] [PubMed] [Google Scholar]
  • 13.National Research Council and Institute of Medicine Preventing Mental, Emotional, and Behavioral Disorders Among Young People: Progress and Possibilities. Washington, DC: National Academies Press; 2009. [PubMed] [Google Scholar]
  • 14.Lewis KM, Bavarian N, Snyder FJ, et al. Direct and mediated effects of a social-emotional and character development program on adolescent substance use. Int J Emot Educ. 2012;4(1):56-78. [PMC free article] [PubMed] [Google Scholar]
  • 15.National Research Council and Institute of Medicine Education and Delinquency: Summary of a Workshop. Panel on Juvenile Crime: Prevention, Treatment, and Control. Washington, DC: National Academy Press; 2000. [Google Scholar]
  • 16.Moss HB, Vanyukov M, Majumder PP, Kirisci L, Tarter RE. Prepubertal sons of substance abusers: influences of parental and familial substance abuse on behavioral disposition, IQ, and school achievement. Addict Behav. 1995;20(3):345-358. doi: 10.1016/0306-4603(94)00077-C [DOI] [PubMed] [Google Scholar]
  • 17.Barnard M, McKeganey N. The impact of parental problem drug use on children: what is the problem and what can be done to help? Addiction. 2004;99(5):552-559. doi: 10.1111/j.1360-0443.2003.00664.x [DOI] [PubMed] [Google Scholar]
  • 18.Cox RG, Zhang L, Johnson WD, Bender DR. Academic performance and substance use: findings from a state survey of public high school students. J Sch Health. 2007;77(3):109-115. doi: 10.1111/j.1746-1561.2007.00179.x [DOI] [PubMed] [Google Scholar]
  • 19.Allison S, Roeger L, Reinfeld-Kirkman N. Does school bullying affect adult health? population survey of health-related quality of life and past victimization. Aust N Z J Psychiatry. 2009;43(12):1163-1170. doi: 10.3109/00048670903270399 [DOI] [PubMed] [Google Scholar]
  • 20.Horwood LJ, Fergusson DM, Hayatbakhsh MR, et al. Cannabis use and educational achievement: findings from three Australasian cohort studies. Drug Alcohol Depend. 2010;110(3):247-253. doi: 10.1016/j.drugalcdep.2010.03.008 [DOI] [PubMed] [Google Scholar]
  • 21.Hirschfield PJ. Schools and crime. Annu Rev of Criminol. 2018;1:149-169. doi: 10.1146/annurev-criminol-032317-092358 [DOI] [Google Scholar]
  • 22.Tchetgen Tchetgen EJ, Walter S, Vansteelandt S, Martinussen T, Glymour M. Instrumental variable estimation in a survival context. Epidemiology. 2015;26(3):402-410. doi: 10.1097/EDE.0000000000000262 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Boef AG, Souverein PC, Vandenbroucke JP, et al. Instrumental variable analysis as a complementary analysis in studies of adverse effects: venous thromboembolism and second-generation versus third-generation oral contraceptives. Pharmacoepidemiol Drug Saf. 2016;25(3):317-324. doi: 10.1002/pds.3956 [DOI] [PubMed] [Google Scholar]
  • 24.Solli IF. Left behind by birth month. Educ Econ. 2017;25(4):323-346. doi: 10.1080/09645292.2017.1287881 [DOI] [Google Scholar]
  • 25.Strøm B. Student achievement and birthday effects. schooling and human capital in the global economy: revisiting the equity-efficiency quandary. Presented at the CESifo-Harvard University/PEPG Conference on Schooling and Human Capital in the Global Economy: Revisiting the Equity-Efficiency Quandary; September 3-4, 2004; Munich, Germany. [Google Scholar]
  • 26.Jürges H, Schneider K. Why young boys stumble: early tracking, age and gender bias in the German school system. Ger Econ Rev. 2011;12(4):371-394. doi: 10.1111/j.1468-0475.2011.00533.x [DOI] [Google Scholar]
  • 27.Russell RJ, Startup MJ. Month of birth and academic achievement. Pers Individ Dif. 1986;7(6):839-846. doi: 10.1016/0191-8869(86)90082-6 [DOI] [Google Scholar]
  • 28.Crawford C, Dearden L, Greaves E Does when you are born matter? the impact of month of birth on children’s cognitive and non-cognitive skills in England. Institute for Fiscal Studies website. IFS Briefing Note: BN 122. 11-1-2011. https://www.ifs.org.uk/publications/5736. Published November 1, 2011. Accessed July 30, 2018.
  • 29.Windle M, Windle RC. A prospective study of alcohol use among middle-aged adults and marital partner influences on drinking. J Stud Alcohol Drugs. 2014;75(4):546-556. doi: 10.15288/jsad.2014.75.546 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Kendler KS, Lönn SL, Sundquist J, Sundquist K. Smoking and schizophrenia in population cohorts of Swedish women and men: a prospective co-relative control study. Am J Psychiatry. 2015;172(11):1092-1100. doi: 10.1176/appi.ajp.2015.15010126 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Kendler KS, Neale MC, MacLean CJ, Heath AC, Eaves LJ, Kessler RC. Smoking and major depression: a causal analysis. Arch Gen Psychiatry. 1993;50(1):36-43. doi: 10.1001/archpsyc.1993.01820130038007 [DOI] [PubMed] [Google Scholar]
  • 32.World Medical Association World Medical Association Declaration of Helsinki: ethical principles for medical research involving human subjects. JAMA. 2013;310(20):2191-2194. doi: 10.1001/jama.2013.281053 [DOI] [PubMed] [Google Scholar]
  • 33.Wikimedia Foundation. Academic grading in Sweden. https://en.wikipedia.org/wiki/Academic_grading_in_Sweden. Accessed May 27, 2015.
  • 34.Buckles KS, Hungerman DM. Season of birth and later outcomes: old questions, new answers. Rev Econ Stat. 2013;95(3):711-724. doi: 10.1162/REST_a_00314 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Lichtenstein P, Sullivan PF, Cnattingius S, et al. The Swedish Twin Registry in the third millennium: an update. Twin Res Hum Genet. 2006;9(6):875-882. doi: 10.1375/twin.9.6.875 [DOI] [PubMed] [Google Scholar]
  • 36.Akaike H. Factor analysis and AIC. Psychometrika. 1987;52(3):317-332. doi: 10.1007/BF02294359 [DOI] [Google Scholar]
  • 37.SAS Institute Inc. Base SAS 9.4 Procedures Guide: Statistical Procedures, Second Edition https://support.sas.com/documentation/cdl/en/procstat/66703/PDF/default/procstat.pdf. Published 2013. Accessed October 2017.
  • 38.US Department of Health and Human Service Facing addiction in America: the surgeon general’s report on alcohol, drugs, and health. https://addiction.surgeongeneral.gov. Published 2016. Accessed October 2017. [PubMed]
  • 39.Cranford JA. DSM-IV alcohol dependence and marital dissolution: evidence from the National Epidemiologic Survey on Alcohol and Related Conditions. J Stud Alcohol Drugs. 2014;75(3):520-529. doi: 10.15288/jsad.2014.75.520 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Institute of Education Sciences. What works clearinghouse. https://ies.ed.gov/ncee/wwc/FWW. Accessed February 28, 2018.
  • 41.Kellam SG, Brown CH, Poduska JM, et al. Effects of a universal classroom behavior management program in first and second grades on young adult behavioral, psychiatric, and social outcomes. Drug Alcohol Depend. 2008;95(suppl 1):S5-S28. doi: 10.1016/j.drugalcdep.2008.01.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Fletcher A, Bonell C, Hargreaves J. School effects on young people’s drug use: a systematic review of intervention and observational studies. J Adolesc Health. 2008;42(3):209-220. doi: 10.1016/j.jadohealth.2007.09.020 [DOI] [PubMed] [Google Scholar]
  • 43.Bradshaw CP, Zmuda JH, Kellam SG, Ialongo NS. Longitudinal impact of two universal preventive interventions in first grade on educational outcomes in high school. J Educ Psychol. 2009;101(4):926-937. doi: 10.1037/a0016586 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Furr-Holden CD, Ialongo NS, Anthony JC, Petras H, Kellam SG. Developmentally inspired drug prevention: middle school outcomes in a school-based randomized prevention trial. Drug Alcohol Depend. 2004;73(2):149-158. doi: 10.1016/j.drugalcdep.2003.10.002 [DOI] [PubMed] [Google Scholar]
  • 45.Washington State Institute for Public Policy Benefit-Cost Results. Olympia: Washington State Institute for Public Policy; 2017. [Google Scholar]
  • 46.Catalano RF, Fagan AA, Gavin LE, et al. Worldwide application of prevention science in adolescent health. Lancet. 2012;379(9826):1653-1664. doi: 10.1016/S0140-6736(12)60238-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Kendler KS, Gardner CO. Dependent stressful life events and prior depressive episodes in the prediction of major depression: the problem of causal inference in psychiatric epidemiology. Arch Gen Psychiatry. 2010;67(11):1120-1127. doi: 10.1001/archgenpsychiatry.2010.136 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Munafò MR, Davey Smith G. Robust research needs many lines of evidence. Nature. 2018;553(7689):399-401. doi: 10.1038/d41586-018-01023-3 [DOI] [PubMed] [Google Scholar]
  • 49.Goodman SN, Fanelli D, Ioannidis JP. What does research reproducibility mean? Sci Transl Med. 2016;8(341):341ps12. doi: 10.1126/scitranslmed.aaf5027 [DOI] [PubMed] [Google Scholar]
  • 50.Lipton P. Testing hypotheses: prediction and prejudice. Science. 2005;307(5707):219-221. doi: 10.1126/science.1103024 [DOI] [PubMed] [Google Scholar]
  • 51.American Psychiatric Association Diagnostic and Statistical Manual of Mental Disorders. 5th ed Washington, DC: American Psychiatric Association; 2013. [Google Scholar]
  • 52.Kringlen E, Torgersen S, Cramer V. A Norwegian psychiatric epidemiological study. Am J Psychiatry. 2001;158(7):1091-1098. doi: 10.1176/appi.ajp.158.7.1091 [DOI] [PubMed] [Google Scholar]
  • 53.Bryant AL, Schulenberg JE, O’Malley PM, Bachman JG, Johnston LD. How academic achievement, attitudes, and behaviors relate to the course of substance use during adolescence: a 6-year, multiwave national longitudinal study. J Res Adolesc. 2003;13(3):361-397. doi: 10.1111/1532-7795.1303005 [DOI] [Google Scholar]
  • 54.Fleming CB, Catalano RF, Haggerty KP, Abbott RD. Relationships between level and change in family, school, and peer factors during two periods of adolescence and problem behavior at age 19. J Youth Adolesc. 2010;39(6):670-682. doi: 10.1007/s10964-010-9526-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.UNU-MERIT. Techniques for dealing with reverse causality between institutions and economic performance. https://www.merit.unu.edu/publications/working-papers/abstract/?id=4718. Accessed February 2018.
  • 56.Yu P. Chapter 7: endogeneity and instrumental variables. http://web.hku.hk/~pingyu/6005/LN/LN7_Endogeneity%20and%20Instrumental%20Variables.pdf. Accessed May 31, 2018.

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement.

eAppendix.

eTable 1. Number of Individuals Born 1972 to 1981 and Information About School Achievement Based on Month of Birth

eTable 2. Information on Parental History of Drug Abuse and Level of Education by Month of Birth of Their Proband Offspring

eTable 3. Analyses Including Only Individuals Whose Academic Achievement (AA) Was Assessed the Year They Turned 16

eTable 4. Evaluation of Possible Reverse Causation (Early Onset Drug Use Predicting Both Poor Academic Achievement (AA) and High Risk for Drug Abuse) by Censoring Drug Abuse Registrations Over Varying Buffer Periods From AA Assessment to Ages 18 to 26

Click here for additional data file. (99.2KB, pdf)

Articles from JAMA Psychiatry are provided here courtesy of American Medical Association

RESOURCES