Abstract
Researchers are often interested in estimating the causal effect of some treatment on individual criminality. For example, two recent relatively prominent papers have attempted to estimate the respective direct effects of marriage and gang participation on individual criminal activity. One difficulty to overcome is that the treatment is often largely the product of individual choice. This issue can cloud causal interpretations of correlations between the treatment and criminality since those choosing the treatment (e.g. marriage or gang membership) may have differed in their criminality from those who did not even in the absence of the treatment. To overcome this potential for selection bias researchers have often used various forms of individual fixed-effects estimators. While such fixed-effects estimators may be an improvement on basic cross-sectional methods, they are still quite limited when it comes to uncovering a true causal effect of the treatment on individual criminality because they may fail to account for the possibility of dynamic selection. Using data from the NSLY97, I show that such dynamic selection can potentially be quite large when it comes to criminality, and may even be exacerbated when using more advanced fixed-effects methods such as Inverse Probability of Treatment Weighting (IPTW). Therefore substantial care must be taken when it comes to interpreting the results arising from fixed-effects methods.
Similar content being viewed by others
Notes
Poisson regressions are used given the dependant variable in each specification is a count variable. Negative Binomial specifications could also be used in such contexts, with an analogous discussion to the one presented here.
Thanks to an anonymous reader for pointing this paper and selection out.
Numerous papers have argued that smoking can act as a gateway drug to harder drugs (Torabi et al. 1993; Chen et al. 2002; Fleming et al. 1989). However, this claim is often supported by showing that smoking is simply correlated with a higher likelihood of harder drug use later on. Clearly, this association is subject to the critique that is the subject of this paper.
However, standard errors are clustered by individual, meaning the standard errors are adjusted for the fact that observations for the same individual are not statistically independent across panel periods. In terms of the equations above, the residual r ij terms are not assumed to be independent, rather they are allowed to have an arbitrary correlation within individual (but are independent across individuals).
Gang status in a given panel year was determined by whether the respondent answered yes to the question “Have you been a member of a gang since last interviewed?”
Smoking status in a given panel year was determined by whether the respondent answered yes to the question “Have you smoked a cigarette since last interviewed?”
Technically, these estimates are called a conditional fixed-effects estimates, but will be referred to simply as a fixed-effects estimator here for brevity. Coefficients on the other included time varying regressors are also omitted for brevity. Note that the number of observations and number of persons shown at the bottom of Table 3 indicate that individuals for whom the dependant variable (i.e. crime count for each category) is zero in all panel years do not contribute to the estimates.
McKinnish (2008) has argued that fixed-effects estimators can also suffer from bias toward zero due to a type of measurement error attenuation bias in instances where the treatment in question has both transitory and longer-term components, but where only the longer-term component is expected to have a substantial impact on the outcome of interest. Given the treatments of interest here—gang participation, marriage, and smoking—this concern is relatively minor in this analysis.
All specifications also include each individual’s age and age squared, along with indicators for panel year. Gordon et al. (2004) actually use negative binomial regressions. Poisson regressions are used here since there is some variation in each individual’s “exposure” in each panel year, as the time between interviews varied between individuals. It is more straightforward to incorporate such variable exposure in Poisson regressions. However, all results in this paper are qualitatively equivalent using negative binomial specifications as done by Gordon et al. (2004).
Moreover, if anything, gang status has a stronger connection to criminality using the NSLY97 data than the delinquency data used by Gordon et al. (2004). This likely has to do with the fact that the oldest youth in the data set used by Gordon et al. (2004) were 16, while the youth in the NLSY97 data used here range from 15 years old (the youngest cohort in the first panel year) up to 24 years old (the oldest cohort in the 2005 panel). Given criminality generally rises during these ages, it should not necessarily be unexpected to find a stronger association between gang status and criminality in this data set covering a larger age range.
This Hierarchical Linear Model method was introduced by Bryk and Raudenbush (1992) and has been used in a very similar context to this (but without IPTW weights) by Horney, Osgood, and Marshall (1995).
Hierarchcial Linear Modeling software (HLM 6.0) must be used to estimate this empirical model when period specific individual weights are used, as standard fixed-effects estimation routines require individuals to have the same weight in each panel year.
References
Acierno R, Kilpatrick DG, Resnick H, Saunders B, De Arellano M, Best C (2000) Assault, PTSD, family substance use, and depression as risk factors for cigarette use in youth: findings from the national survey of adolescents. J Trauma Stress 13(3):381–396
Acri JB, Grunberg NE (1992) A psychophysical task to quantify smoking cessation-induced irritability: the reactive irritability scale (RIS). Addict Behav 17:587–601
Apel R, Bushway S, Paternoster R, Brame R, Sweeten G (2008) Using state child labor laws to identify the causal effect of youth employment on deviant behavior and academic achievement. J Quant Criminol 24:337–362
Ashenfelter O (1978) Estimating the effect of training programs on earnings. Rev Econ Stat 60:47–57
Berntson GG, Beattie MS, Walker JM (1976) Effects of nicotinic and muscarinic compounds on biting attack in the cat. Pharmacol Biochem Behav 5:235–239
Bloklund AAJ, Nieuwbeerta P (2005) The effects of life circumstances on longitudinal trajectories of offending. Criminology 43:1203–1240
Bryk AS, Raudenbush SW (1992) Hierarchical linear models: application and data analysis methods. Sage, Newbury Park
Burt RD, Dinh KT, Peterson AV Jr, Sarason IG (2000) Predicting adolescent smoking: a prospective study of personality variables. Prev Med 30:115–125
Bushway S, Brame R, Paternoster R (1999) Assessing stability and change in criminal offending: a comparison of random effects, semiparametric, and fixed effects modeling strategies. J Quant Criminol 15(1):23–61
Chaiken JM, Chaiken MR (1990) Drugs and predatory crime. In: Tonry M, Wilson JQ (eds) Drugs and crime. University of Chicago Press, Chicago, IL
Chen X, Unger JB, Palmer P, Weiner MC, Johnson CA, Wong MM, Austin G (2002) Prior cigarette smoking initiation predicting current alcohol use: evidence for a gateway drug effect among California adolescents from eleven ethnic groups. Addict Behav 27(5):799–817
Cherek DR (1981) Effects of smoking different doses of nicotine on human aggressive behavior. Psychopharmacology 75:339–345
Driscoll P, Baettig K (1981) Selective inhibition by nicotine of shock-induced fighting in the rat. Pharmacol Biochem Behav 14:175–179
Farrington DP (1988) Studying changes within individuals: the causes of offending. In: Ruter M (ed) Studies of psychosocial risk: the power of longitudinal data. Cambridge University Press, New York
Farrington DP, West DP (1995) Effects of marriage, separation, and children on offending by adult males. In: Blau ZS, Hagan J (eds) Current perspectives on aging and the life cycle, volume 4: delinquency and disrepute in the life course. JAI, Greenwich, CT
Farrington DP, Gallagher B, Morley L, Ledger RJ, West DJ (1986) Unemployment, school leaving, and crime. Br J Criminol 26:335–356
Fleming R, Leventhal H, Glynn K, Ershler J (1989) The role of cigarettes in the initiation and progression of early substance use. Addict Behav 14:261–272
French MT, McGeary KA, Chitwood D, McCoy CB, Inciardi JA, McBride D (2000) Chronic drug use and crime. Subst Abuse 21(2):95–109
Furgusson DM, Swain-Campell NR, John Horwood L (2000) Deviant peer affiliations, crime and substance use: a fixed effects regression analysis. August, Journal of Abnormal Child Psychology
Giordano PC, Cernkovich SA, Rudolph JL (2002) Gender crime, and desistance: toward a theory of cognitive transformation. Am J Sociol 107:990–1064
Gordon RA, Lahey BB, Kawai E, Loeber R, Stouthamer-Loeber M, Farrington D (2004) Anti-social behavior and youth gang membership: selection and socialization. Criminology 42(1):55–87
Grogger J (1995) The effects of arrests on the employment and earnings of young men. Q J Econ 110(1):51–71
Halaby CN (2004) Panel models in sociological research: theory into practice. Annu Rev Sociol 30:507–544
Hausman J, Taylor W (1981) Panel data and unobervable individual effects. Econometrica 49:1377–1398
Haynie DL, Giordano PC, Manning WD, Longmore MA (2005) Adolescent romantic relationships and delinquency involvement. Criminology 43(1):177–210
Heckman JJ, Smith JA (1999) The pre-programme earnings dip and the determinants of participation in a social programme. Implications for simple programme evaluation strategies. Econ J 109:313–348
Hernan MA, Brumback B, Robins JM (2000) Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology 11:561–570
Hirschi T, Gottfredson M (1995) Control theory and the life-course perspective. Stud Crime Crime Prev 4(2):131–142
Hoffman JP, Xu J (2002) School activities, community service, and delinquency. Crime Delinq 48:568–591
Horney J, Wayne Osgood D, Marshall IH (1995) Criminal careers in the short-term: intra-individual variability in crime and its relation to local life circumstances. Am Sociol Rev 60:655–673
Hsiao C (1986) Chapter 3: Simple regression with variable intercepts. Analysis of panel data. University of Cambridge Press, Cambridge UK
Hutchinson RR, Emley GB (1973) Effects of nicotine on avoidance, conditioned suppression and aggression response measures in animals and man. In: Dunn WL (ed) Smoking behavior: motives and incentives. Winston, Washington
Imbens G, Angrist J (1994) Identification and estimation of local average treatment effects. Econometrica 62(2):467–475
Jamner LD, Shapiro D, Jarvik ME (1999) Nicotine reduces the frequency of anger reports in smokers and nonsmokers with high but not low hostility: an ambulatory study. Exp Clin Psychopharmacol 7:454–463
Kassel JD, Paronis CA, Stoud LR (2003) Smoking, stress, and negative affect: correlation, causation, and context across stages of smoking. Psychol Bull 129(2):270–304
King RD, Massoglia M, Macmillan R (2007) The context of marriage and crime: gender, the propensity to marry, and offending in early adulthood. Criminology 45:33–65
Koval JJ, Pederson LL (1999) Stress-coping and other psychosocial risk factors: a model for smoking in grade 6 students. Addict Behav 24(2):207–218
Laub JH, Sampson RJ (1993) Turning points in the life course: why change matters to the study of crime. Criminology 31:301–325
Lloyd D, Taylor J (2004) Stress as a distal predictor of heavy smoking initiation in young people. Paper presented at the Annual meeting of the American Sociological Association
McKinnish T (2008) Panel data models and transitory fluctuations in the explanatory variable. Adv Econom 21:335–358
Nagin DS, Farrington DP (1992a) The stability of criminal potential from childhood to adulthood. Criminology 30:235–260
Nagin DS, Farrington DP (1992b) The onset and persistence of offending. Criminology 30:501–523
Osgood DW, Wilson JK, O’Malley PM, Bachman JG, Johnston LD (1996) Routine activities and individual deviant behavior. Am Sociol Rev 61:635–655
Robins JM, Hernan MA, Brumback B (2000) Marginal structural models and causal inference in epidemiology. Epidemiology 11(5):550–560
Rodgers RJ (1979) Effects of nicotine, mecamylamine, and hexamethonium on shock-induced fighting, pain reactivity, and locomotor behavior in rats. Psychopharmacology 66:93–98
Sampson RJ, Laub JH (1990) Crime and deviance over the life course: the salience of adult social bonds. Am Sociol Rev 55:609–627
Sampson RJ, Laub JH (1993) Crime in the making: pathways and turning points through life. Harvard University Press, Cambridge
Sampson RJ, Laub JH, Wimer C (2006) Does marriage reduce crime? A counterfactual approach to within-individual causal effects. Criminology 44(3):465–508
Schechter MD, Rand MJ (1974) Effects of acute deprivation of smoking on aggression and hostility. Psychopharmacologia 35:19–28
Siqueira L, Diab M, Bodian C, Rolnitzky L (2000) Adolescents becoming smokers: the roles of stress and coping methods. J Adolesc Health 2000(27):399–408
Solon G (1989) The value of panel data in economic research. In: Kasprzyk D, Duncan G, Kalton G, Singh MP (eds) Panel surveys. Wiley, New York, pp 486–496
Stewart L, Livson N (1966) Smoking and rebelliousness: a longitudinal study from childhood to maturity. J Consult Psychol 30(3):225–229
Stolzenberg L, D’Alessio SJ (2007) The effect of divorce on domestic crime. Crime Delinq 53:281–302
Torabi M, Bailey WJ, Maid-Jabbari M (1993) Cigarette smoking as a predictor of alcohol and other drug use by children and adolescents: evidence of the ‘gateway drug’ effect. J Sch Health 63(7):302–306
Turner CF, Kim L, Rogers SM, Lindberg LD, Pleck JH, Sonenstein FL (1998) Adolescent sexual behavior, drug use, and violence: increased reporting with computer survey technology. Science 280:867–873
Warr M (1998) Life-course transitions and desistance from crime. Criminology 36:183–216
Wills TA, Sandy JM, Yaeger AM (2002) Stress and smoking in adolescence: a test of directional hypotheses. Health Psychol 21(2):122–130
Winship C, Morgan SL (1999) The estimation of causal effects from observational data. Annu Rev Sociol 25:659–706
Wooldridge JM (2002) Econometric analysis of cross section and panel data. MIT, Cambridge
Acknowledgments
The author thanks Eric Helland, Shawn Bushway, Ray Paternoster, Brian Johnson, and Serkan Ozbeklik, as well as the comments of three anonymous referees and the editors Alex Piquero and James Lynch for encouragement and helpful suggestions on earlier drafts of this project.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Bjerk, D. How Much Can We Trust Causal Interpretations of Fixed-Effects Estimators in the Context of Criminality?. J Quant Criminol 25, 391–417 (2009). https://doi.org/10.1007/s10940-009-9073-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10940-009-9073-y