Genetic and environmental influences on impulsivity: A meta-analysis of twin, family and adoption studies

Abstract

A meta-analysis of twin, family and adoption studies was conducted to estimate the magnitude of genetic and environmental influences on impulsivity. The best fitting model for 41 key studies (58 independent samples from 14 month old infants to adults; N = 27,147) included equal proportions of variance due to genetic (0.50) and non-shared environmental (0.50) influences, with genetic effects being both additive (0.38) and non-additive (0.12). Shared environmental effects were unimportant in explaining individual differences in impulsivity. Age, sex, and study design (twin vs. adoption) were all significant moderators of the magnitude of genetic and environmental influences on impulsivity. The relative contribution of genetic effects (broad sense heritability) and unique environmental effects were also found to be important throughout development from childhood to adulthood. Total genetic effects were found to be important for all ages, but appeared to be strongest in children. Analyses also demonstrated that genetic effects appeared to be stronger in males than in females. Method of assessment (laboratory tasks vs. questionnaires), however, was not a significant moderator of the genetic and environmental influences on impulsivity. These results provide a structured synthesis of existing behavior genetic studies on impulsivity by providing a clearer understanding of the relative genetic and environmental contributions in impulsive traits through various stages of development.

Impulsivity is one of the most researched behavioral traits, and has captured the interest of researchers and clinicians spanning the fields of psychology, neuroscience, neurogenetics and psychiatry (Congdon & Canli, 2008; Evenden, 1999; Eysenck, 1967; Eysenck & Eysenck, 1977; Costa & McCrae, 1992; Cloninger, Svrakic, & Przybeck, 1993; Tellegen, 1982). Indeed, impulsivity is of key importance for psychopathology, including antisocial and aggressive behaviors, pathological gambling and attention deficit hyperactivity disorder (ADHD). Thus, researchers have been motivated to attain a deeper understanding of this multidimensional construct and its underlying genetic and environmental etiology. Although numerous twin and adoption studies have investigated the roots of individual differences in impulsivity, it has been difficult to draw clear conclusions given the fact that impulsivity has been described and measured in various ways in the literature. Current framework considers impulsivity to be comprised of several distinct dimensions including a sense of urgency, lack of planning, lack of persistence (or perseverance), and sensation seeking (Whiteside & Lynam, 2001; Fischer, Smith & Cyders, 2008). However, there is yet to exist a comprehensive review of genetic and environmental influences on impulsivity and how these vary across the different facets (or dimensions) and definitions. Twin and family/adoption studies have the distinct ability to unravel genetic and environmental influences and to estimate their simultaneous contributions to individual differences in impulsivity. Thus, the purpose of the present review is to synthesize and consolidate the existing behavior genetic literature on impulsivity and its various methods of assessment and definition. Understanding the genetic and environmental etiology of impulsivity may also help further the understanding of this important multidimensional construct.

To provide a context for the current review, we begin with a summary of the various constructs and definitions of impulsivity as used in clinical and research settings, and its role in personality theory and psychopathology. A review of potential moderators of the genetic and environmental influences in impulsivity measures across studies is also provided as background to the meta-analysis.

Impulsivity: Definitions and Implications

Recent work has demonstrated that ‘impulsivity’ embodies a range of traits that are only moderately (co)related (Smith, Fischer, Cyders, Annus, Spillane, & McCarthy, 2007; Whiteside & Lynam, 2001; Cyders & Smith, 2007). One favorable attempt to identify and define the specific constructs rooted within impulsivity was a factor analysis of various measures of impulsivity by Whiteside and Lynam (2001). Instead of the non-operational (uni)- dimensional approach and definition to impulsive traits, Whiteside and Lynam (2001) identified four (etiologically) distinct constructs of impulsivity namely, (1) urgency or negative urgency, (2) lack of planning, (3) lack of perseverance, and (4) sensation seeking. To date, this four-factor structure of impulsivity has been confirmed on several, independent samples (Lynam & Miller, 2004; Smith et al., 2007).

Based on several lines of research, personality theorists have drawn several conclusions regarding impulsivity including, the fact that there seems to be a fundamental or underlying personality foundation in impulsive behaviors and traits. Personality researchers have also recognized and demonstrated numerous different representations of what that personality root or foundation is; and lastly, researchers have noted that various constructs or traits have been included under the broader realm of ‘impulsivity’ (or impulsive behaviors) (Barratt, 1993; Depue & Collins, 1999; Evenden, 1999; Eysenck et al. 1985; Fischer et al., 2008; McCarthy & Smith, 1995; Petry, 2001; Smith et al., 2007; Tellegen 1982; Whiteside & Lynam, 2001; Zuckerman, 1994).

The role of impulsivity in personality has stemmed from benchmark research conducted by Eysenck et al. (1985), Tellegen (1982) and Barratt (1983). In their research, these personality theorists captured different manifestations of impulsivity using both narrow (psychoticism, venturesomeness, sensation seeking) and broad (extraversion and constraint) definitions employing three, four and five-factor models of personality (Cloninger, Svrakic, & Przybeck, 1993; Costa & McCrae, 1992; Eysenck et al., 1985; Patton, Stanford, & Barratt, 1995; Tellegen, 1982; Whiteside & Lynam, 2001). The proper placement of impulsivity within these different personality models is one example of how personality theorists have struggled with the intricacy of this multidimensional construct. Additionally, some three-factor models including those formulated by Costa and McCrae’s (1992), Tellegen’s (1982) and Cloninger (1993) have even considered impulsivity as a compilation of different factors or components (Evenden, 1999). For example, Cloninger defined impulsivity as a combination of low harm avoidance and high novelty seeking (Cloninger et al, 1993). As a result, there has been a lack of consensus about how to properly define and structure impulsivity as a fundamental trait (Congdon & Canli, 2008).

Impulsivity has also played a significant role in theories of antisocial and criminal behavior (Whiteside & Lynam, 2001). For example, individuals who score high on personality measures of impulsivity have been found to be more antisocial and commit more recidivistic criminal acts compared to those who score low (White, et al., 1994). It is possible that deficits in impulse control may produce antisocial behavior by interfering with an individual’s ability to control their behavior as well as to think of the future consequences of their antisocial acts (Moffitt, 1993; White et al., 1994). Lack of impulse-control may manifest itself in numerous ways in a wide range of traits and behaviors, including risk taking (Barratt & Patton 1983), sensation seeking (Zuckerman, 1979), novelty seeking (Cloninger, 1993), and aggression (Coccaro, 1989, 1998), which are all related to antisocial and criminal behavior.

In addition to its substantial role in the realm of personality and antisocial behavior, impulsivity has played a significant role in numerous psychiatric disorders in both children and adults (American Psychiatric Association, 2004). In its extreme form, impulsivity has been associated with a wide range of mental disorders, one example being attention deficit hyperactivity disorder (ADHD; Avila et al., 2004; Barkley, 1997). ADHD is characterized by pervasive and impairing symptoms of inattention, hyperactivity and impulsivity. Specifically, ADHD encompasses a pattern of behavior in which a child has difficulty waiting for turns, intrudes in on or interrupts conversations, and/or blurts out answers before questions have been completed. Other disorders and behaviors where impulsivity is a major component include conduct disorder (Plutchik & Van Praag, 1995), borderline personality disorder (Conrod, Pihl, Stewart, & Dongier, 2000; Fossati, Barratt, Carretta et al., 2004; Links, Heslegrave, & van Reekum, 1999), substance use and abuse (Sher, Bartholow, & Wood, 2000), suicidal behavior (Dougherty, Mathias, Marsh, Moeller, & Swann, 2004; Esposito & Spirito, 2003; Swann, Dougherty, Pazzaglia, Pham, & Moeller, 2004), and various impulse control disorders e.g., pathological gambling, kleptomania, and pyromania. Impulsivity has been one of the defining characteristics of these disorders, which has affected a large percentage of the general population (Congdon & Canli, 2008).

Even though impulsivity is an important component in both normal varying personality, as well as pathological behaviors, it has not been consistently examined across phenotypic or behavioral genetic studies. Thus, there is a need to synthesize the existing behavior genetic literature on impulsivity with the aim of further elucidating the contributions of genetic and environmental influences on impulsivity.

Behavioral Genetic Studies of Impulsivity

To date, there have been more than a hundred published twin and family/adoption studies examining the genetic and environmental etiology of impulsivity. Results from the majority of these studies support the notion of a moderate additive genetic factor influencing impulsivity (e.g., Eaves et al., 1977; Hur & Bouchard, 1997). A smaller number of studies have demonstrated that impulsivity is influenced by non-additive or dominant genetic factors as well (Hur & Bouchard, 1997; Pedersen, Plomin, McClearn, & Friberg, 1988; Seroczynski, Bergeman, & Coccaro, 1999).

While yielding valuable results, previous research has examined impulsivity in a variety of different ways employing different, perhaps even unrelated behaviors due to the multidimensionality of this complex construct. In addition, there is no gold standard for defining impulsivity or impulsive traits, which further accentuates the complexity of synthesizing the quantitative results obtained on this construct. Even with impulsivity’s widespread significance in both normal varying personality, as well as pathological behaviors, it has not been consistently examined across phenotypic or behavioral genetic studies. This lack of integrative review thus motivated us to examine the genetic and environmental influences across a wide range of impulsivity measures and to investigate if results vary across them. The purpose of the current project is to synthesize the existing behavior genetic literature on impulsivity with the aim of further elucidating the contributions from genetic and environmental components on this important construct.

Potential Moderators of Genetic and Environmental Influences on Impulsivity

Method of Assessment

There are two major methods used to assess impulsivity: questionnaire measures and laboratory tasks. Self-report questionnaires have been by far the most commonly used technique for measuring impulsivity among young adolescents and adults. The number of questionnaires and surveys developed to measure impulsivity has been extensive to say the least (see Barratt & Patton, 1986; Patton, Stanford, & Barratt, 1995; Eysenck et al., 1985). A few examples include the Barratt Impulsiveness Scale (Barratt, 1985), the Eysenck Impulsiveness Scale (Eysenck et al., 1985), the Zuckerman Sensation Seeking Scale (Zuckerman, 1979), and the Thurstone Temperament Schedule (Thurstone, 1950).

The second method of assessment, laboratory tasks, includes widely employed laboratory measures such as Continuous Performance Tasks (CPTs), the Go/NoGo task, the Iowa Gambling task, the Stop-Signal Paradigm and Delay of Gratification and Delay Discounting task(s). These tasks examine reaction time to responses as well as number of disinhibited or ‘impulsive’ errors committed. These tasks contrast the inhibition and execution of motor responses. Although somewhat distinct from one another, in the Go/NoGo and stop-signal tasks a successful performance requires the inhibition of a predominant response making it somewhat challenging. These tasks have been widely and effectively employed on a wide range of samples including children and adults in both clinical and healthy (normative) settings (Alderson, Rapport et al., 2008; Aron & Poldrack, 2006; Moeller et al., 2005; Rubia, 2001; Rubia, Smith, Brammer, & Taylor, 2003; Tamm, Menon, & Reiss, 2002). Consequently, results may vary across different measures given such a wide range of laboratory tasks that all aim to measure some form of (behavioral) disinhibition and impulsivity.

Method(s) of assessment may influence the outcome of results in behavior genetic studies (McCartney et al., 1990; Rhee & Waldman, 2002) and vary across measures. For instance in certain laboratory measures, external factors such as time of day and ambient temperature may affect variance estimates since these effectively increase measurement error and hence non-shared environmental effects (Boomsma & Gabrielli, 1985). Thus, in the present study, assessment method was used as a moderator comparing the magnitude of genetic and environmental influences in questionnaire measures of impulsivity to that of laboratory tasks of impulsive behaviors.

Age

The influence of genetic and environmental factors may vary across development as well as across various traits (Loehlin, 1992; Plomin, 1986). Previous research has demonstrated that for different phenotypes a pattern of decrease in shared environment and a concomitant increase in heritability and non-shared environmental effects during development, particularly for personality traits and cognitive abilities (Scarr & McCartney, 1983; Loehlin, 1992; Miles and Carey, 1997). A possible explanation to the observed pattern of decreasing shared environmental effects as an individual grows older might be that they actively seek out environmental situations that are more closely matched to the his or her genotype, which in turn might increase the influence of non-shared environmental effects. In other words, as children grow older they are probably less supervised, become more independent and may select environments that are correlated with their phenotypes. Thus far, studies investigating developmental changes in the genetic and environmental influences on personality traits have yielded inconsistent and conflicting results (McCartney, 1990). As a result, it is important to examine whether the influence of genetic and environmental factors in impulsivity varies at different time points across development and age.

Sex

Thus far, twin studies have found no evidence of significant sex differences in the genetic and environmental etiology of psychopathology and normal personality (Jang, 2005; Bouchard & Loehlin, 2001) with the exception of antisocial behavior (Tuvblad, 2006; Heiman, et al., 2004; Moffitt, Caspi, Rutter & Silva, 2001). There is also indication of sex differences in childhood disinhibition with boys showing higher mean disinhibition levels than girls (Gladstone, 2004). Given this sex difference on a phenotypic level, it is also important to examine whether the magnitude of genetic and environmental effects differs in males and females. Earlier behavior genetic studies have been inconsistent regarding sex differences in impulsivity, with some reporting significant sex differences in traits such as control and constraint (Finkel & McGue, 1997), and others reporting no significant differences in sex (McGue et al., 1993). Thus, the present study examined whether sex is a significant moderator of impulsivity by comparing the results for males and females. Because of the unique relationship between opposite sex twins, including prenatal effects arising from a male and female sharing the womb, sex differences were examined for studies with and without opposite sex twins.

Study Design

As twin and family/adoption studies rely on different methodological assumptions, the results from studies using these different study designs were also directly compared in the current analyses. The classical twin design compares the similarity of MZ twins to that of DZ twins and is recognized to be one of the most effective study designs for estimating the relative contribution of genes and environmental influences to human traits (Evans, Gillespie & Martin, 2002). Perhaps one of the most criticized assumptions of the twin design is the equal environment assumption (EEA). In the twin design MZ twins who share all their genes, are compared to DZ twins who on average share half of their genes. If MZ twins are more similar than DZ twins on a particular trait, then that would indicate the importance of genetic effects. However, it has to be assumed that environmentally caused similarity is roughly the same for both types of twins. If this assumption is violated, higher correlations among MZ twins may be due to environmental factors, rather than genetic factors, and heritability estimate will be overestimated (Plomin et al., 2001). Studies that have examined the EEA have generally shown that the assumption is fully justified (see for example, Jacobson et al 2002). The twin design also assumes random mating in the parent generation. Assortative mating tends to increase similarity between DZ twins, thereby biasing heritability estimates downward and the shared environmental estimates upward. For example, twin studies might overestimate the effects of shared environment if assortative mating is not taken into account. However, assortative mating for most personality traits has been found to be low in magnitude (Maes et al., 1998). Further, twins and singletons have generally been found to experience similar rates of psychiatric disorders and behavioral and emotional problems (Gjone & Novik, 1995; Moilanen et al., 1999; Simonoff et al., 1997; van den Oord et al., 1995). There are however, two ways in which twins differ from singletons: (i) lower birth weight and shorter gestational age (Plomin et al., 2001), and (ii) delayed language development (Rutter & Redshaw, 1991). These differences have however, been found to have a minor effect on traits and behavior later in life (Christensen et al., 2006).

As with the twin design, adoption studies also rely on several methodological assumptions. One important assumption in the adoption design is random placement. It is generally assumed that adoptive children are randomly placed into homes. However, that is seldom the case, since adoptive families are usually selected for similarities to biological parents (Evans et al., 2002). Adoptive parents also tend to be in good health and more affluent. Adoption designs are also subject to generalizability problems due to the fact that results from adopted individuals may not be representative of the greater population (Evans et al., 2002). Additionally, parent-offspring adoption studies might underestimate the influence of genetic effects if developmental changes are important to impulsivity, such that different genes may be important at different ages. Moreover, sibling adoption studies might overestimate shared environmental influences; previous adoption studies have found little to no genetic influences in temperament (Plomin, Coon, Carey, Defries, & Faulker, 1991). Therefore, due to the fact that twin and adoption/family studies rely on different methodological assumptions, it is important to investigate study design as a potential moderator.

Publication Date

Finally, the potential moderating effect of publication date was tested. That is, we examined whether studies published before 1985 differed from studies published after 1986. This cut-off was chosen because it seemed to be reasonable considering the fact that studies conducted before 1985 used different methods of reporting findings (such as not reporting twin correlations or effect sizes).

Present Study

Although numerous studies have investigated the genetic and environmental influences on impulsivity, there have been no systematic reviews summarizing these results. The present study is therefore the first comprehensive and systematic review of the heritability and environmentality of impulsivity. We employed a meta-analysis to summarize and consolidate results obtained from twin, adoption, and family studies in order to better understand the genetic and environmental etiology of impulsivity and its underlying dimensions. With any systematic review or meta-analyses, it is useful to identify and determine the influence of different study and sample characteristics such as age and sex on the study results (Rosenthal, 1991). Thus, we also examined the influence of potential moderator variables on the genetic and environmental factors underlying impulsivity, including method of assessment, age, sex, study design, and publication date.

Method

Search Strategy

We searched for twin and family/adoption studies of impulsivity in the PsycInfo and Medline/PubMed [http://www.ncbi.nlm.nih.gov/sites/entrez] databases from January 1, 1970 to February 01, 2011. We limited our search to the year 1970 due to differences in assessment methods and low sample sizes in behavior genetic studies conducted before 1970. Search terms used included the following: “impulsivity”, “disinhibition”, “sensation seeking”, “novelty seeking”, “control”, “constraint”, “self-control”, “ADHD”, “neuroticism”, “extraversion”, “psychoticism”, “inhibitory control”, “response inhibition”, “excitement seeking”, “thrill seeking”, “boredom susceptibility”, “Barratt Impulsiveness Scale”, “personality”, “Temperament and Character Inventory”, “Temperament and Personality Questionnaire”, “HEXACO”, “Go/NoGo”, “CPT or continuous performance task”, “delay discounting”, “risk taking”, “Stop Signal”, “behavioral inhibition system”, “adventuresomeness”, “boldness”, “unreliability”, “urgency”, “Stroop Test”, “(lack of) planning”, “EASI”, and paired these terms with “twins”, “heritability”, “siblings”, “genetics”, “adoption”, “family”, “adoptees”, “genes”, “environment”, “environmental”, and “families”. [Appendix A shows the search text used in this process]. Cross-referencing of the retrieved studies was applied to identify any additional, potentially relevant studies of interest. The “Related citations” link, a feature from PubMed was also used. Information about relevant published, unpublished data in the process of being published was also collected and reviewed by contacting researchers directly through the Behavior Genetics Association’s list serve via email. Additionally, close examination of effect and sample sizes for MZ and DZ twin pairs by means of funnel plots revealed little to no publication bias in the current analyses.

Inclusion Criteria

From these searches, an initial pool of 121 studies was found. These studies were carefully reviewed to determine if they met the inclusionary criteria. First, studies must have used genetically informative data. That is, studies must have used a methodology that either compared MZ and DZ twins or compared the association between impulsive traits and behavior in parents and offspring and other immediate relatives. Second, the studies must have measured impulsivity either through laboratory tasks or measures (e.g., CPTs, Stop Signal paradigms and Go/NoGo tasks; Dougherty et al., 2000; 2003), or through questionnaire/survey measures. Third, we broadly defined impulsivity as a personality trait that involves “swift action without forethought or conscious judgment and a tendency to act with less forethought than do most individuals of equal ability and knowledge” (Moeller et al., 2001). Additionally, impulsivity has been defined to encompass traits such as, ‘lack of planning’ and ‘risk taking’ (Eysenck and Eysenck, 1977; Patton et al., 1995). Thus, traits, scales, and items similar to those found in DSM-IV-TR Axis I (clinical) and Axis II (developmental and personality) disorders (e.g., ADHD impulsivity, and antisocial personality disorder; American Psychiatric Association DSM-IV-TR Diagnostic and Statistical Manual of Mental disorders, 2004) were included. Finally, data from each study had to be presented in a form [e.g., twin intraclass or family correlations reporting the Pearson (r-statistic)] that allowed for the calculation of the effect sizes. Only one study conducted by Anokhin et al. (2004) examined psychophysiological methods to study the heritability of impulsivity in young twins (females) was found, and was thus intentionally not included in these analyses. Instead the search focused on questionnaires and laboratory tasks assessing behavioral and motor impulsivity. Published studies were further restricted to those using human subjects and written in English.

As a result of this procedure, a total of 41 studies (34%) out of the initial pool of 121 were included in the present analyses. Tables 1 and 2 display the twin and family/adoption studies (respectively) included for analyses. From the 41 total studies, there were 58 independent samples and 153 data groups used for a total N= 27,147 twins, adoptees and relatives. The number of groups refers to the number of independently analyzed components in the samples, and the number of samples refers to the number of independent studies in the analyses. Additionally, we received and included data and results from three unpublished studies (including the Go/NoGo task from the Southern California Twin study). Ninety to ninety-five percent of the studies included assessed zygosity through questionnaire measures. Therefore, method for determining zygosity was not used as a potential moderator.

Table 1.

Effect Sizes (r) and Ns for Twin Studies Included in the Meta-Analysis

Study (Author, Year)	Measure	Assessment Method	Age Group	Sex	MZ r (N)	DZ r (N)
Ando et al. 2004	Novelty Seeking	TCI	Young Adults	M+F	.23 (414)	.12 (203)
Anokhin et al. 2010	Disinhibition	Delay Discounting	Adolescents	M	.82 (63)	.14 (50)
				F	.46 (66)	−.14 (31)
				M-F	--	.17 (56)
Blonigen et al. 2003	Impulsive-nonconformity	PPI	Adults	M+F	.51 (89)	−.05 (47)
Buss & Plomin 1975	Impulsivity	EASI	Children	M	.84 (38)	.05 (33)
				F	.71 (43)	.59 (24)
Carey et al. 1978	Self-control	CPI	Adults	M+F	.52 (692)	.30 (503)
Carey & Rice 1983	Impulsivity	DPQ	Adults	M	.52 (79)	−.24 (34)
				F	.37 (152)	.15 (72)
DiLalla et al. 1994	Behavioral inhibition	Laboratory observation	Infants	M+F	.82 (90)	.47 (67)
Gagne et al. 2010	Inhibitory Control	Laboratory observation	Infants	M+F	.38 (266)	.16 (322)
Groot et al. 2004	Errors of commission	Go/NoGo	Children	M	.48 (52)	.49 (36)
				F	.35 (73)	.26 (35)
				M-F	--	.05 (35)
Heath et al. 1994	Novelty Seeking	TPQ	Adults	M	.35 (396)	.06 (222)
				F	.42 (933)	.14 (532)
				M-F	--	.07 (564)
Heiman et al 2004	Novelty Seeking	JTCI	Children	M	.50 (132)	.22 (106)
				F	.35 (140)	.18 (94)
				M-F	--	.03 (98)
	Novelty Seeking	TPQ		M	.25 (67)	.12 (52)
				F	.37 (87)	−.27 (43)
				M-F	--	.16 (66)
Heiman et al 2003	Novelty Seeking	TCI	Adults (older)	F	.42 (308)	.37 (131)
			Adults (younger)	F	.47 (269)	.12 (141)
Heiser et al. 2006	Errors of commission	CPT	Children	M+F	.47 (17)	.88 (12)
Hur et al. 1997*	Sensation Seeking/Impulsivity	SSS	Adults	M+F	.50 (57)*	.10 (49)*
Isen et al. 2010	Novelty Seeking	JTCI	Children	M	.18 (121)	.37 (77)
				F	.22 (127)	.05 (94)
				M-F	--	.19 (132)
Jang et al. 2002	Impulsivity	NEO-PI	Adults	M+F	.37 (253)	.26 (207)
				M+F	.37 (526)	.22 (269)
Johnson et al. 2004	Impulsivity	PRF	Adults	M+F	.45 (183)	.32 (64)
Koopmans et al. 1995	Sensation Seeking/Disinhibition (impulsivity)	SSS	Children/Adolescents	M	.59 (260)	.42 (230)
				F	.58 (348)	.45 (304)
				M-F	--	.30 (449)
Larsson et al. 2006	Impulsive Behaviors	YPI	Adolescents	M	.57 (384)	.20 (326)
				F	.57 (449)	.31 (365)
				M-F	--	.22 (328)
Loehlin et al. 1976	Impulsive Behaviors	TTS	Adults	M+F	.39 (45)	−.11 (34)
				M+F	.52 (102)	.02 (119)
				M+F	.35 (44)	--
Matheny et al. (1989)	Behavioral inhibition	Lab observation	Infants	M+F	.71 (33)	.25 (32)
McGue et al. 1993	Control	MPQ	Adults	M+F	.19 (48)	.05 (35)
Mustanski et al. 2003	Excitement Seeking	Zuckerman’s Scale	Adolescents	M	.82 (239)	.14 (315)
				F	.46 (412)	−.14 (354)
Neale et al. 1989	Impulsivity	EASI	Infants	M	.49 (106)	.12 (129)
				F	.76 (113)	.18 (83)
				M-F	--	.20 (108)
Owen 1970	Inhibition-withdrawal	Lab task	Children	M+F	.34 (18)	.16 (24)
Pedersen et al. 1988*	Impulsivity	KSP	Adults	M+F	.45 (141)	.09 (202)
				M+F	.40 (92)*	.15 (212)*
Plomin et al. (1975)	Reflection-impulsivity	MFFT	Children	M+F	.60 (32)	.61 (22)
Robinson et al. 1992	Inhibition	Lab observation	Infants	M	.61 (76)	.34 (90)
				F	.52 (108)	.16 (82)
Saudino et al. 1999	Impulsivity	KSP	Adults	M+F	.48 (79)	−.06 (51)
Schachar et al. 2011	Inhibition	Stop Signal	Children	M+F	.35 (55)	−.02 (76)
Seroczynski et al. 1999	Impulsivity	BIS	Adults	M+F	.43 (182)	.17 (118)
				M+F	--	.32 (93)
Sherman et al. 1997	Impulsivity	DICA-R	Children	M+F	.92 (194)	.32 (93)
Stevenson et al. 1985	Impulsivity	EASI	Infants	M	−.19 (22)	.34 (28)
				F	.16 (19)	−.19 (17)
			Children	M	.22 (59)	−.11 (78)
				F	.37 (74)	−.02 (55)
			Children	M	.14 (25)	−.31 (23)
				F	−.04 (20)	−.25 (13)
Stoel et al. 2006	Sensation Seeking/Disinhibition (impulsivity)	SSS	Children	M	.59 (376)	.40 (302)
				F	.52 (702)	.31 (426)
				M-F	--	.28 (572)
Takahashi et al 2007	Behavioral Inhibition	BIS	Young Adults	M+F	.31 (85)	.12 (32)
Thapar et al. 2000	Impulsivity	ADHD	Children	M+F	.73 (728)	.27 (1183)
Vierriko et al. 2004	Impulsivity	ADHD	Children	M	.73 (260)	.37 (290)
				F	.76 (296)	.10 (277)
				M-F	--	.33 (513)
Vicken et al. 2005	sensation seeking	MMPI	Young Adults	M	.47 (104)	.33 (95)
				F	.55 (161)	.26 (107)
Young et al. 2009	Novelty Seeking	TCI	Adolescents	M+F	.39 (159)	.19 (134)

Note.

^*= twins reared apart (including in Hur et al., 1997 and Pedersen et al. 1988) - MZ=monozygotic twins reared together; DZ=Dizygotic twins reared together; DZ-OS= dizygotic opposite sex twin pairs; r = Pearson or Intraclass twin correlations; N = number of twins pairs/participants; M = males (male twins); F = females (female twins); M +F = study included or combined males and females together; M-F = DZ opposite sex male- female twins; ADHD=attention deficit hyperactivity disorder; SSS= sensation seeking scale; BIS= Barratt Impulsiveness Scale; BIS = Behavioral Inhibition System; EASI=extraversion, agreeableness, sociability, impulsivity; KSP= Karolinska Scales of Personality; YPI= Youth Psychopathy Inventory; TTS= Thurstone Temperament Schedule; MPQ=Multidimensional Personality Questionnaire; CPT= continuous performance test; FFT= matching familiar figures test; CPI= California Psychological Inventory; PPI = Psychopathic Personality Inventory; PRF = Personality Research Form; DPQ= Differential Personality Questionnaire; MMPI = Minnesota Multiphasic Personality Inventory; (J)TCI = (Junior) Temperament and Character Inventory; NEO-PI= Neuroticism, extraversion, openness, psychoticism, impulsivity

Table 2.

Effect Sizes (r) and Ns for Family/Adoption Studies Included in the Meta-Analysis

Study (Author, Year)	Operationalization/Measure	Assessment Method	Age Group	Sex	P-O r (N)	Bio-Sibs r (N)	Adoptive-Sibs r (N)	M-O (Adoptive) r (N)	Adoptive Relatives*
Dougherty et al. 2003	Impulsivity	BIS	Children/Adolescents	M+F	−.38 (28)	--	--	--	--
Loehlin et al. 1976	Impulsivity	TTS	Adults	M+F	.10 (117)	.07 (102)	−.02 (143)	.04 (541)	--
				M+F	.09 (149)
Plomin et al. 1991	Impulsivity	EASI	Children	M+F	−.01(235)	--	.03 (65)	--	−.02 (222)
									.09 (222)
									.12 (77)

Note.

^*= adoptive relatives including adoptive mother – biological child, or adopted child –adoptive mother, or adoptive siblings; P-O= parent offspring pairs; A-B= adopted child biological parent; A-A= adopted child adoptive mother; adoptive sibs= adoptive siblings; bio sibs= biological siblings; M-O= adoptive mother – offspring; r = Pearson or Intraclass twin correlations; N = number of participants (family pairs); BIS= Barratt Impulsiveness Scale; BIS= Behavioral Inhibition System; EASI=extraversion, agreeableness, sociability, impulsivity; TTS= Thurstone Temperament Schedule

Reasons for Exclusion

One justification for exclusion from the meta-analysis was non-independent sampling. Some authors published the same data or same exact cohort of data in two or three different sources. Thus, studies that included analyses on a subset of a larger sample already part of the meta-analysis were excluded from the analyses (e.g., Blonigen et al. 2006; Blonigen et al., 2005; Finkel et al., 1997; Holmes et al., 2002; Emde et al. 1992). For these instances, only one of the studies was considered for the meta-analysis. Another issue of non-independent sampling was the fact that some authors of a single publication examined more than one dependent measure of impulsivity (for example, both a questionnaire and a laboratory task) in their sample. A third issue was the fact that several authors (in different publications) examined different dependent measures in the same sample. Forth, several publications used the follow-up data of the same sample. Lastly, studies were excluded if they reported on a behavior that was indirectly related to impulsivity or that could (partly) be a result of impulsive behavior e.g., promiscuous or risky sexual behaviors, behaviors related to recklessness (driving without a seatbelt) or substance use/abuse. These constructs fell outside the scope of the current meta-analysis.

Several different options for dealing with non-independent samples have been proposed (Rosenthal, 1991) including: (1) choosing the best dependent measure, (2) averaging the effect sizes of the dependent measures, or (3) running the meta-analyses once for each dependent measure. The third option was not chosen due to practical issues because there were a large number of effect sizes from non-independent samples. For the present meta-analysis, the first and second options were chosen.

Data Analyses

Determination of Effect Sizes

All of the adoption and twin studies included in the present study used a continuous variable as the measure of impulsivity and reported either intraclass or Pearson product-moment correlations, which were the effect sizes (r) used in the current analyses. These effect sizes were input and analyzed in a model-fitting program (Mx; Neale, 1992; 2003) that estimates the relative contribution of genetic and environmental influences and tests alternative genetic or etiologic models.

Model Fitting Analyses

Structural equation modeling was used to perform the genetic model-fitting analyses in Mx (Neale et al., 2003). The traditional quantitative genetic model is based on the notion that the observed variance in a trait is a linear function of genetic (G), shared (C), and non-shared (E) environmental variances. The traditional approach divides the phenotypic variance (V_p) into genetic (V_G) and environmental (V_E) influences [V_p=V_G +V_E]. Environmental factors may be divided further into common or shared variances (V_C) and unique or non-shared influences (V_U). Additionally, the degree of additive (A) and non-additive genetic influences (D) comprise the amount of variance in the liability of a trait (in this case, impulsivity) that is due to genetic differences among individuals. If genetic influences are additive, this means that the effects of alleles from different loci are independent and add up to influence the liability for a different trait. If genetic influences are non-additive, then the alleles interact with one another to influence the liability for a trait. Shared environmental effects (C) demonstrate what is common and shared among family members making them more similar to one another, while non-shared environmental influences (E) denote the amount of variance that is due to environmental influences that are experienced uniquely and make members of the same family different from one another (Neale & Cardon, 1992). Appendix B outlines the expected correlations among relatives.

Alternative or comparison models, which contain different sets of causal influences, are directly compared in behavior genetic analyses. These models may include some or all of the types of influences described above: additive genetic (A), shared environmental (C), non-additive genetic (D), and non-shared environmental influences (E). In the present meta-analysis, an ACE model, ADE model, an AE model, and a CE model were all compared. Since both (C) and (D) rely on the same information (the difference between MZ and DZ twin correlations), it is impossible to estimate them simultaneously (in an ACDE model) using data from only twin pairs reared together (please see Posthuma et al., 2003 for a full review). However, if other types of data are included in the analyses such as, correlations from adoption studies (adoptive and/or biological parents), then this provides another source of valuable information for the estimation of shared environmental effects and the ACDE model can be tested. As in previous meta-analyses of behavior genetic studies (e.g., Rhee & Waldman, 2002), the ACDE model was tested when all of the data from both twin and adoption studies were included in the meta-analysis.

Data from both twin pairs reared together and apart were included. The effect sizes and the number of participants (Ns) from each included study were entered in separate groups in the structural equation modeling program Mx (Neale, 2003).

Analyses of All Data

The analyses were conducted first on all of the data including both twin and adoption studies. The ACDE, ACE, ADE, AE, and CE models were all compared. The fit of each model, as well as the fit for the competing or comparison models, was assessed using both the chi-square statistic and the Akaike Information Criterion (AIC), a fit index that reflects both the fit of the model and its parsimony (Akaike, 1987). The model with the lowest AIC and the lowest chi-square value relative to its degrees of freedom was considered to be the best fitting model.

Assessment of Possible Outlier Studies

Certain studies may be outliers or exert undue influence on the results due to the large sample sizes or the inclusion of several different samples, or using measures of questionable validity. For the present study, the data was analyzed both with and without these “outlier” studies. The results were not significantly altered when outlier studies were removed from the analyses. These additional results are available upon request.

Assessment of Potential Moderators

The extent to which genetic and environmental effects on impulsivity varied across key study variables was examined through additional analyses. Separate analyses were conducted to evaluate several moderators: (1) method of assessment (laboratory tasks vs. questionnaire/survey measures); (2) age (infants, children, adolescents, adults); (3) sex (pairs of male, female, and opposite sex relatives); (4) study design (twin vs. adoption studies); (5) publication date (studies published from 1970–1985 and those published from 1986–2011). Potential moderation was tested by contrasting the fit of a model in which the parameter estimates are constrained to be equal across levels of the variable in question to the fit of the model in which parameter estimates were free to vary across levels on the same variable. If the fit of the two models was significantly different (according to difference in chi-squares) this indicated that the moderator effect on genetic and environmental parameter estimates was significant. It should be noted that it is possible that a non-significant result may have been due to lack of power and little variability in the levels of the moderator.

Subscale Analyses

Recent studies suggest that impulsivity is comprised of four distinct dimensions including a sense of urgency, lack of planning, lack of persistence (or perseverance) and sensation seeking (Whiteside & Lynam, 2001; Fischer, Smith & Cyders, 2008). Additional analyses were thus carried out to examine specifically the genetic and environmental effects contributing to each of these dimensions. Each impulsivity measure (or trait) from each included study was examined and assigned to one of these four aspects of impulsivity. However, due to the fact that most authors report effect sizes for total number of items from full (dimension) scales (e.g., novelty seeking), these subscale analyses were carried out according to a more general, or broad-sense definition of impulsivity. For example, ‘novelty seeking’, or ‘sensation seeking’ were included as full scales because authors presented the effect sizes for the full scales in their studies and not just the impulsivity items or sub-scales incorporated into those higher order dimension. Ample studies were found for three out of the four categories including lack of planning, sensation seeking, and lack of persistence (or perseverance) [Appendix C lists the studies included in each of these three categories]. Results for the category of urgency did not yield meaningful results, since only two studies could be categorized under this construct and are thus not presented.

Results

Analyses of All Data

The results from analyses including all samples that met inclusion criteria (41 studies; 58 independent samples; 153 data groups; 27,147 participants) are presented in Table 3. Briefly, the number of samples refers to the number of independent studies in the analyses, and the number of groups refers to the number of independently analyzed components (i.e., types of relatives) in the samples. The ADE model was the best fitting model compared to the other models as illustrated by the lowest AIC value (see Table 3), with moderate to high additive genetic and unique environmental effects [A = .38, D = .12, E = .50]. It is noteworthy that the effect of shared family environment (C) on impulsivity was negligible even in the full ACDE model, suggesting family resemblance is generally not explained by their shared experiences.

Table 3.

Best-fitting Models for Twin and Family/Adoption Studies and Impulsivity Subscale Analyses

Model	χ2	df	p-value	AIC	A	C	D	E
Analyses of All Data
ACDE	1271.14	440	<.001	391.14	.38	.00	.12	.50
ACE	1291.59	441	<.001	409.59	.49	.00	--	.51
ADE	1271.14	441	<.001	389.14	.38	--	.12	.50
AE	1291.59	442	<.001	407.59	.49	--	--	.51
CE	2225.55	442	<.001	1341.55	--	.34	--	.66
Analyses of Twin Studies Only
ACE	1233.30	408	<.001	417.30	.49	.00	--	.51
ADE	1212.92	408	<.001	396.92	.34	--	.16	.50
AE	1233.30	409	<.001	415.30	.49	--	--	.51
CE	2029.23	409	<.001	1211.23	--	.36	--	.64
Subscale Analyses Lack of planning
ACDE	346.10	236	<.001	−125.89	.29	.00	.12	.59
ACE	357.60	237	<.001	−116.40	.40	.00	--	.60
ADE	346.10	237	<.001	−127.89	.29	--	.12	.59
AE	357.60	238	<.001	−118.40	.40	--	--	.60
CE	589.12	238	<.001	113.12	--	.24	--	.76
Sensation Seeking ACDE	356.88	119	<.001	118.88	.29	.02	.16	.53
ACE	362.78	120	<.001	122.78	.47	--	.00	.53
ADE	356.926	120	<.001	116.926	.34	--	.13	.53
AE	362.776	121	<.001	120.78	.47	--	--	.53
CE	655.875	121	<.001	413.875	--	.34	--	.66
Lack of Perseverance
ACE	251.46	30	<.001	191.46	.67	.00	--	.33
ADE	225.72	30	<.001	165.72	.37	--	.32	.31
AE	251.46	31	<.001	186.46	.67	--	--	.33
CE	756.49	31	<.001	649.49	--	.44	--	.56

Note. AIC= Akaike Information Criterion; A= additive genetic effects; C= shared environmental effects; E= nonshared environmental effects; D=nonadditive genetic effects. Best fitting model is in bold.

Study design: Twin vs. family/adoption studies

The chi square difference test revealed a significant difference in genetic and environmental influences on impulsivity when estimated in twin versus adoption studies Δχ² (4, N= 18,259) = 53.11, p<.001. We therefore conducted additional analyses to investigate the influence of genetic and environmental in twin studies. The best fitting model based on the AIC value in twins (142 groups, 38 studies, 39 samples, N= 18,259) was the ADE model (A = .34, D = .16, E = .50) (see Table 3). However, these results should be interpreted carefully because there was a preponderance of twin studies (38) compared to family/adoption studies (3).

Subscale Analyses

Similarly, subscale analyses revealed that ADE models were also the best-fitting for the distinct constructs of lack of planning, lack of persistence, and sensation seeking. Additive and non-additive genetic effects accounted for an overall 51–69% of the total variance, while non-shared environmental effects contributed to the remainder of the variance in impulsive behaviors within each category of subscales (31–49%). Lack of persistence exhibited the highest broad-sense heritability estimate with both additive and non-additive genetic effects explaining over two-thirds of the total variance in lack of persistence (see Table 3).

Assessment of Potential Moderators

Table 4 displays the results obtained from the analyses examining assessment method, age, publication date, study design and sex as moderators of the magnitude of the genetic and environmental influences on impulsivity. Briefly, significant differences in genetic and environmental influences were found across age, sex, and study design (twin vs. adoption studies), but not for method of assessment or publication date. Results for each of these potential moderators are discussed below.

Table 4.

Moderator Analyses

Moderator Variables	χ2	df	p-value	AIC	A	C	D	E
Assessment Method
Parameters free	1270.36	434	<.001	402.36
Questionnaires					.38	.00	.11	.51
Laboratory Measures					.19	.04	.28	.49
Parameters constrained	1271.14	438	<.001	395.14	.38	.00	.12	.50
Difference in χ2	0.78	4	0.95
Age (4-groups)
Parameters free	1094.75	426	<.001	242.75
Infants					.28	.00	.25	.47
Children					.47	.00	.12	.41
Adolescents			\		.31	.02	.23	.46
Adults					.31	.00	.10	.59
Parameters constrained	1271.14	438	<.001	395.14	.38	.00	.12	.50
Difference in χ2	176.4	12	<.001
Age (3-groups without infants)
Parameters free	989.69	373	<.001	243.69
Children					.47	.00	.12	.41
Adolescents					.31	.00	.23	.46
Adults					.31	.00	.07	.59
Parameters constrained	1163.89	381	<.001	401.89	.38	.00	.11	.51
Difference in χ2	174.20	8	<.001
Sex (with Opposite Sex twins)
Parameters free	956.70	254	<.001	448.70
Males					.45	.00	.19	.36
Females					.09	.05	.35	.51
Parameters constrained	1211.51	258	<.001	695.507	.08	.01	.50	.41
Difference in χ2	254.81	4	<.001
Sex (without Opposite Sex twins)
Parameters free	640.28	221	<.001	198.28
Males					.41	.00	.12	.47
Females					.23	.00	.26	.51
Parameters constrained	651.72	225	<.001	201.72	.29	.01	.20	.50
Difference in χ2	11.44	4	0.02
Twin vs. Family/Adoption Studies
Parameters free	1218.03	434	<.001	350.03
Twin					.14	.07	.29	.50
Family/Adoption					.00	.04	.00	.96
Parameters constrained	1271.14	438	<.001	395.14	.38	.00	.12	.50
Difference in χ2	53.11	4	<.001
Publication Date
Parameters free	1266.86	434	<.001	398.86
1970–1985					.28	.00	.18	.54
1986–2011					.385	.00	.115	.50
Parameters constrained	1271.14	438	<.001	395.14	.38	.00	.12	.50
Difference in χ2	4.28	4	.37

Note. AIC= Akaike Information Criterion; A= additive genetic effects; C=shared environmental effects; E= nonshared environmental effects; D=nonadditive genetic effects.

Assessment Method

The chi-square difference test was non-significant for assessment method Δχ² (4, N=27,147) = 0.78, p = 0.95, see Table 4, indicating were no differences in the magnitude of the genetic and environmental influences on impulsivity when measured through laboratory tasks (10 study samples; N= 1919 participants) or questionnaire measures (44 study samples; N= 25,494). Even though no significant differences were detected, further analyses were carried out to examine the genetic and environmental effects in laboratory vs. questionnaire measures. The ADE model was the best fitting model for tasks performed in the laboratory with additive and non-additive genetic effects accounting for 32% and 19% of the variance, respectively, while non-shared or unique environmental effects accounted for 49% of the variance (A= .32, D = .19, E= .49). The ADE model was also the best fitting model for questionnaire measures with equally moderate to high influences from both additive and non-additive genetic effects and non-shared environmental effects (A= .38, D= .12, E= .50).

Age

The chi-square difference test indicated that age was a significant moderator and that the magnitude of genetic and environmental influences on impulsivity in infants (6 study samples), children (18 study samples), adolescents (8 study samples), and adults (22 study samples) are significantly different from each other [Δχ² (12, N=27,147) = 176.4, p<.001 for the 4 group comparison and Δχ² (8, N= 25,356) = 174.20, p<.001 for the 3 group comparison], see Table 4. We therefore conducted additional analyses to investigate the influence of genetic and environmental effects within each age group. As seen in Figure 1, the ADE model was the best-fitting model for each of the four age groups (in Figure 1, N represents to the number of twin pairs), although the magnitude of effects varied across ages. Both additive and non-additive genetic effects as well as non-shared environmental effects were each important in infants (A= .28, D= .25, E= .47), in children (A= .47, D= .12, E = .41), in adolescents (A= .31, D = .23 E= .46), and in adults (A= .31, D= .10, E= .59) (see Figure 1). The magnitude of broad sense heritability (additive and non-additive genetic effects combined) ranged from (h²_B=.41) to (h²_B=.59) throughout infancy and adulthood. Figure 1 illustrates the genetic and environmental effects across the four age groups. However, this age trend should be interpreted carefully for several reasons including (1) the fact that both twin and family/adoption studies were included in these particular age analyses, (2) these analyses include both laboratory and survey measures of impulsivity, and (3) because the studies included in the meta-analyses examined a wide range of ages (these possible confounds will be discussed in further detail in the discussion). For the purposes of simplification, age was used as a categorical variable (infants, children, adolescents, adults). That is, age is not represented on an interval scale in Figure 1.

Figure 1. Age trends in genetic and environmental influences in impulsivity.

Note: A = non-additive genetic variance effects; D = additive genetic variance effects; E = unique environmental variance effects.

Sex

The chi square difference test revealed significant sex differences in genetic and environmental influences on impulsivity when considering all studies, including the DZ opposite sex twins [Δχ² (4, N= 18,259) = 254.81, p<.001]. The best fitting model when males, females, and opposite sex twins were included was the ACDE model (males: A= .45, C= 0, D= .19, E= .36; females: A = .09, C = .045, D = .35, E = .51). We also conducted analyses to investigate sex differences in the influence of genetic and environmental effects while excluding opposite- sex twins. When the DZ opposite sex twin pairs were dropped from the analyses, these sex differences remained significant Δχ² (4, N= 13,841) = 11.44, p=.02, Table 4. However, the best fitting model when only same-sex male and female twins were included was the ADE model with moderate influences from both additive and non-additive genetic effects as well as non-shared environmental effects (males: A= .41, D = .12, E= .47; females: A = .23, D = .26, E = .51). The parameter estimates for both genetic and environmental effects are somewhat different in models with and without opposite sex twins, particularly for the girls.

Publication Date

There were no significant differences between older studies (published from 1970 – 1985) and more recent studies (published from 1986–2011) [Δχ² (4, N=27,147) = 4.28, p = 0.37; see Table 4], with similar heritability estimates in new and older studies. In our analyses, the best-fitting model for older studies was an AE model with (A = .45; E =.55), while the best fitting model for the more recent studies was an ADE model with (A = .38; D = .12; E = .50). Although an additional chi square difference test indicated a significant difference between published and unpublished studies [Δχ² = 78.97, df =4, p<.01)], these results should be interpreted with caution because there were 41 published studies compared to 3 unpublished studies.

Discussion

The present study is the first to systematically examine the heritability of impulsivity across studies of infants, children, adolescents and adults, analyzing over 27,000 individuals (or pairs of relatives) in 41 key studies. When all data from both twin and adoption/family studies that met inclusion criteria were analyzed, the overall results indicated important genetic and unique environmental influences on impulsivity. The magnitudes of the non-additive genetic influences were estimated in addition to shared environmental effects, although the best fitting model was an ADE model. Accordingly, there were strong additive genetic (A= .38), and non-shared environmental effects (E = .50), and some non-additive genetic effects (D= .12) on impulsivity. Shared environmental effects were negligible and non-significant, so that twin and family resemblance for impulsive responding appears to be explained entirely by heritable factors.

The effects of potential moderators on the magnitude of the genetic and environmental influences on impulsivity were also examined. Age, sex, study design and year of publication accounted for significant differences in the genetic and environmental influences on impulsivity. In addition, the results from published studies were significantly different from those in unpublished studies, although this result should be interpreted cautiously considering the fact that there were only three unpublished studies included in the current analyses compared to forty-one published ones.

Assessment method as a moderator

Researchers studying personality and temperament have found that parent reports tend to yield DZ correlations that are very low or even negative. This may be the result of parents’ exaggerating the differences between their DZ twins, which has been described as a rater contrast effect (Loehlin, 1992). One example of such a finding emerged from the MacArthur longitudinal twin study (Emde et al., 1992). No resemblance of DZ twins on measures of behavioral inhibition and shyness was found using parent reports, but significant DZ resemblance was found using observational measures of the same constructs. Plomin’s (1981) review of twin studies examining personality concluded that objectively assessed behavior yielded lower heritabilities than self-reports and parent reports.

Surprisingly in the present meta-analyses, assessment method (i.e., questionnaire vs. laboratory task) was not a significant moderator of the genetic and environmental influences on impulsivity. Most questionnaires and structured interview measures have low or negligible correlations with one another (Monahan & Steadman, 1994), and have low order and often, insigni cant correlations with non-questionnaire measures of impulsivity (Barratt & Patton, 1983). Furthermore, methods of assessment have been known to influence the outcome of results in behavior genetic studies (McCartney et al., 1990; Rhee & Waldman, 2002) and vary across measures.

Age as a moderator

Consistent with behavior genetic research, which has generally found that heritability estimates of human behavioral traits increase with age, the present study found significant moderating effects of age (Scarr & McCartney, 1983; Miles & Carey, 1997; Tuvblad, 2006). Interestingly, our results indicated that the distribution in broad-sense heritability [both additive (A) and non-additive (D) genetic effects] varied with age – with heritability being slightly higher in studies examining impulsive behaviors during childhood and slightly lower during adulthood (see Plomin et al., 2001; Loehlin, 1992; Hay, Bennett, McStephen, Rooney & Levy, 2004). This may be an artifact of the number of studies or kinds of measures included in these particular analyses. It may also be due to the wide range in ages included in these analyses. Other reasons for the varying distribution of broad-sense heritability throughout infancy to adulthood may be due to confounding in (moderating) effects including, assessment method, sex, twin vs. family or adoption studies, or a combination of these effects with age. For example, analyses for children and adults included both twin and family/adoption studies (with 18 and 22 study samples included in children and adult analyses, respectively), whereas analyses for infants (which included only 6 study samples) and adolescents (which included 8 study samples) did not – this may have contributed to the varying distribution of both A and D effects observed in the results. Another example of a possible confounding effect may be due to the varying number of laboratory vs. survey measures in each age group - with infant analyses including 4 laboratory studies (vs. 2 survey); analyses for children including 5 laboratory studies (vs. 13 survey studies); analyses for adolescents including only 1 study using a laboratory measure (vs. 7 survey) and adult analyses including only survey measures (and no laboratory tasks). In addition to these confounds, there was a preponderance of female twins included in the adult analyses, whereas a more level distribution of males and females were included in the other age groups. Moreover, analyses for infants and children included both older (1970–1985) and more recent (1986–2011) studies, whereas analyses for adolescents and adults predominantly contained only the more recent studies. A recent study conducted by Young et al (2009) demonstrated that behavioral disinhibition was highly heritable at age 12, but variance components for genetic effects became ‘more balanced’ during later adolescence (Young, Friedman, Miyake, Willcutt, Corley, Haberstick & Hewitt, 2009). Externalizing behaviors (impulsivity) have also been reported to have strong genetic effects during childhood and early adolescence (Hay et al., 2004).

Including non-twins and adoptive siblings in our meta-analyses allows us to assess the non-additive nature of impulsivity. The distribution of non-additive genetic effects on impulsivity varied across age groups – this may be due to numerous factors including dominance effects or epistasis (Keller, Coventry, Heath, & Martin, 2005). Previous studies investigating personality measures including novelty seeking and harm avoidance in twins and siblings have found that non-additive genetic effects contribute to variance in personality traits (Keller et al., 2005).

Sex as a moderator

No matter how impulsivity is assessed or operationalized, studies have shown that it is more prevalent in males than females (e.g., Bezdjian et al., 2009; Moffitt et al., 2003). Given this sex difference in prevalence, it is important to consider whether the magnitude of genetic and environmental influences differs in males and females. Therefore, the present meta-analysis examined whether sex is a significant moderator of the results of behavior genetic studies of impulsivity by comparing the results for males, females, and both sexes (i.e., studies reporting results for a combined sample of males and females or studies reporting results for opposite-sex twin pairs).

In the present meta-analysis, significant sex differences were found when both including and excluding the DZ opposite sex pairs from the analyses. The best fitting model when males, females, and opposite sex twins were included was the ACDE model (with 64% broad sense heritability in males, and 44% broad sense heritability in females). The best fitting model when only same-sex male and female twins were included was the ADE model (with a broad sense heritability of 53% in males and 49% in females). Interestingly, genetic influences were slightly stronger in males than in females. For both males and females, shared environmental influences were of negligible importance.

These results fall in line with results reported by Carey and Rice (1983) and Finkel and McGue (1997), in which significant sex differences were found for personality traits. Previous behavior genetic studies examining personality traits; however, have not found significant differences in sex (McGue et al. 1993).

Study design as a moderator

Significant differences in genetic and environmental effects across study design (i.e., twin versus adoption and family studies) were also found, which illustrate the importance of including both twin and family studies in these analyses. However, these results should be interpreted with caution because the analyses included very few adoption studies. Examining twin and adoption studies simultaneously allows us to estimate both common environmental effects as well as nonadditive genetic effects. The best fitting model overall was the ADE model, suggesting that there were no significant shared environmental influences on impulsivity. However, the number of studies included in the analyses might serve as a limitation. There were more twin studies available to include than adoption and family studies, which might have significantly altered the results and findings.

Subscale Analyses of Impulsivity

In addition to the overall synthesis of 41 studies examining impulsivity (or impulsive traits), we undertook a more comprehensive and contemporary approach to examining the genetic and environmental effects in the separate subtypes of this multidimensional construct. Results demonstrated that the genetic and environmental effects influencing the different sub-dimensions of impulsivity (including sensation seeking, lack of planning, and lack of persistence or perseverance) are similar to one another. Additive and non-additive genetic effects for the subscales of sensation seeking, lack of planning and lack of persistence range between 51–69%, while non-shared environmental effects accounted for the remainder of the contributing variance in these (impulsive) traits. Interestingly, lack of persistence displayed the highest broad-sense heritability estimate of nearly 70%, whereas genetic effects contributing to sensation seeking and lack of planning were closer to 40–50%. Even though impulsivity is a multidimensional construct that embodies several layers and distinct facets, the genetic and environmental influences on the different sub-traits or dimensions of impulsivity seem to have at least some similarities in the magnitude of genetic and environmental effects. In the future, this finding can be expanded to include the genetic and environmental covariance between the different dimensions of impulsivity with other risky behaviors including substance use.

Limitations

Examining impulsivity in the most comprehensive manner including, animal and molecular studies was outside of the scope of this project. The present study instead focused only on biometric twin and family studies of impulsivity. All of the studies included in this meta-analysis reported the Pearson or intraclass correlation in their publications, which may lead to some methodological limitations in the meta-analysis (since raw data could not be directly examined and analyzed). Given that only correlations were analyzed, there was no way to compare the variances of the different types of relatives (MZ twins, DZ twins, twins vs. adoptees) or across sex or age. This may be an important consideration because there may be differences in the variances of impulsivity measures across different relatives. Also, the present meta-analysis focused mainly on questionnaires and lab measures. Only one study has looked at the heritability of impulsivity using the p300 during a Go/NoGo task (Anokhin et al., 2004), which was left out of these particular analyses. Another limitation in the present meta-analysis is the fact that there might be confounding among the moderator variables especially among age, sex, assessment method, and type of study. Confounding among moderator variables or interaction effects among moderator variables was not directly examined in this present study.

Future Directions

For this particular meta-analysis, impulsivity was broadly defined to incorporate all aspects, definitions and dimensions of impulsivity in order to fully understand the genetic and environmental etiology of impulsive behaviors. Future studies may focus on different aspects or dimensions of impulsivity more closely and may also want to include results from molecular genetic studies, or take a more narrow approach to defining impulsive behaviors. Even though a broad definition of impulsivity was employed, subscale analyses indicated that the contributing genetic and environmental effects were more or less similar in magnitude. Future studies should test for differences in the operationalization of impulsivity by examining the moderating effects for different measures of impulsivity, and how each particular measure operationalizes impulsivity. Those analyses were not conducted in this particular study because there were too few of each instrument/definition to investigate any moderating effects at that level.

Conclusions

In conclusion, the current study found substantial genetic and unique environmental effects on impulsivity with no common environmental influences. Data from large, methodologically sound twin and adoption studies have also demonstrated that traits including impulsivity are significantly heritable (Goldman, & Fishbein, 2000). Furthermore, the heritability of self-reported personality traits related to impulsiveness and irritability in twins reared together and apart showed heritability rates that ranged from 20% to 62%. Other twin studies have also indicated a strong genetic heritability for impulsivity. Genes may therefore modulate behaviors that involve impulse control, which can lead to manifestations such as conduct disorder, antisocial personality disorder, ADHD and alcoholism.

However, heritability is a group statistic. Therefore, the precise extent to which a trait is genetically influenced (or if a trait is measurably heritable) usually has poor predictive value on any individual case. Many traits may be strongly genetically influenced in particular families, but demonstrate nearly zero heritability in a population sample. Also, highly heritable behavioral traits may be strongly determined by environmental factors and individual influences (Goldman & Fishbein, 2000). High heritability in traits in no way implies that environmental intervention will be futile in modifying the trait. Genetic findings may be necessary to formulate effective interventions.

Thus, the results from the present meta-analysis demonstrated that approximately half of the variance in impulsivity was explained by genetic influences (additive genetic and non-additive genetic factors) and the remaining portion was explained by non-shared environmental influences. These results help increase our understanding of this important and complex construct by shedding some light on the role genetic and environmental influences play on the development of impulsive behaviors. Indeed, more longitudinal studies on impulsivity are needed - the present study provides a bit more insight into the genetic and environmental contributions influencing impulsive behaviors.

Research Highlights.

• Meta-analyses conducted on 58 independent samples from 41 independent studies examining the heritability of impulsivity using twin, family and adoption studies (conducted from 1970–2011)

• Genetic and environmental effects on impulsivity were examined in infants, children, adolescents, and adults. Overall results demonstrated close to equal proportions of variance due to both genetic and non-shared environmental effects

• Age, sex, and study design were all significant moderators of the magnitude of genetic and environmental influences on impulsivity

• Total genetic effects were important for all ages, but appeared to be strongest in children.

• Total genetic effects appeared to be stronger in males than in females.

• The relative contribution from genetic and environmental effects were also examined for distinct subscales for impulsivity including, lack of planning, lack of persistence, and sensation seeking

Appendix A

Terms used in PsycInfo and Medline/PubMed Searches

Impulsivity	Temperament and Personality Questionnaire
Disinhibition	Excitement Seeking
Sensation seeking	Go/NoGo
Novelty seeking	Inhibitory Control
Control	Response Inhibition
Constraint	Thrill Seeking
Self-control	Boredom Susceptibility
ADHD	Barratt Impulsiveness Scale
Neuroticism	Personality
Extraversion	Temperament and Character Inventory
Psychoticism	Delay Discounting
Twins	Unreliability
Heritability	EASI
Genetics	Urgency
Adoptees	Risk Taking
Adoption	Stop Signal
Genes	Continuous Performance Task
Environmental	Behavioral Inhibition System
Environment	HEXACO
Siblings	Adventuresomeness
Family	Boldness
Families	Stroop Test

Appendix B

Expected Correlations for Twin, Family, and Adoption Studies

Relationship	Expected Correlations
Twin Studies
MZ twin pairs reared together	1*A + 1*C + 1*D
MZ twin pairs reared apart	1*A + 1*D
DZ twin pairs reared together	.5*A + 1*C + .25*D
DZ twin pairs reared apart	.5*A + .25*D
Family Studies
Parent-Offspring	.5*A + 1*C
Adoption Studies
Adoptee-adoptive parent	1*C
Adoptee-biological parent	.5*A
Biological child-biological parent	.5*A + 1*C
Adoptive siblings	1*C
Biological siblings	.5*A + 1*C + .25*D

Note: A = non-additive genetic variance effects; C = common or shared environmental variance effects; D = additive genetic variance effects

Appendix C

Studies included in the categorical subscale analyses for impulsivity

Non (or Lack of) Planning	Task/Measure
Anokhin et al 2010	Delay Discounting
Buss & Plomin, 1975	EASI (Buss & Plomin, 1975)
Carey & Rice 1983	DPQ (Tellegen, 1979)
DiLalla et al 1994	Inhibitory control (lab)
Dougherty et al 2003	BIS (Patton et al., 1995)
Gagne et al 2010	Inhibitory control (lab)
Groot et al 2004	Go/NoGo
Heiser et al 2006	CPT
Johnson et al 2004	PRF (Jackson, 1999)
Loehlin et al 1986	TTS (Thurstone, 1950)
Matheny et al 1989	Behavioral inhibition
McGue et al 1993	MPQ (Control; Tellegen, 1982)
Neale et al 1989	EASI (Buss & Plomin, 1975)
Owen et al 1970	Inhibition (lab)
Pedersen et al 1988	KSP (Karolinska Scales of Personality)
Plomin et al 1991	EASI (Buss & Plomin, 1975)
Plomin et al 1975	MFFT (Kagan et al. 1963)
Robinson et al 1992	Inhibition (lab)
Saudino et al 1999	KSP (Karolinska Scales of Personality)
Schachar et al 2011	Stop-Signal
Serocynzski et al 1999	BIS (Patton et al., 1995)
Stevenson et al 1985	EASI (Buss & Plomin, 1975)
Takahashi et al 2007	BIS (Behavioral Inhibition System; Carver & White, 1994)
Sensation Seeking
Ando et al 2004	TCI (novelty seeking) (Cloninger et al., 1991)
Heath et al 1994	TPQ (Cloninger et al., 1991)
Heiman et al 2004	JTCI/TPQ (Cloninger et al., 1991)
Heiman et al 2003	TCI (Cloninger et al., 1991)
Hur et al 1997	SSS (Zuckerman)
Isen et al 2009	JTCI (Cloninger et al., 1991)
Koopmans et al 1995	SSS (Zuckerman, 1994)
Mustanski et al 2003	SSS (Zuckerman, 1994)
Stoel et al 2006	SSS (Zuckerman, 1994)
Vicken et al 2005	MMPI (Hathaway & McKinley, 1942)
Young et al 2009	TCI (Cloninger et al., 1991)
Lack of Perseverance
Blonigen et al 2003	PPI (Lilienfeld & Andrews, 1996)
Larsson et al 2006	YPI (Andershed et al., 2006)
Sherman et al 1997	DICA-R (Herjanic & Reich, 1982)
Thapar et al 2000	ADHD (impulsive items/behaviors)
Vierikko et al 2004	ADHD

Note: All Studies predominantly used community and population based samples including twin registries. ADHD=attention deficit hyperactivity disorder; SSS= sensation seeking scale; BIS= Barratt Impulsiveness Scale; BIS = Behavioral Inhibition System; EASI=extraversion, agreeableness, sociability, impulsivity; KSP= Karolinska Scales of Personality; YPI= Youth Psychopathy Inventory; TTS= Thurstone Temperament Schedule; MPQ=Multidimensional Personality Questionnaire; CPT= continuous performance test; MFFT= matching familiar figures test; CPI= California Psychological Inventory; DPQ= Differential Personality Questionnaire; PPI = Psychopathic Personality Inventory; PRF = Personality Research Form; MMPI = Minnesota Multiphasic Personality Inventory; (J)TCI = (Junior) Temperament and Character Inventory