Emerging Trends in The Social and Behavioral Sciences · Causation, Theory, and Policy in the Social Sciences

Causation, Theory, and Policy in the Social Sciences

Media

Part of Causation, Theory, and Policy in the Social Sciences

Title: Causation, Theory, and Policy in the Social Sciences
extracted text: Causation, Theory, and Policy in the
Social Sciences
MARK C. STAFFORD and DANIEL P. MEARS

Abstract
Despite a penchant for constructing and testing causal theories, social scientists frequently avoid explicit discussion of causal issues. Illustrating with criminological
literature, we argue that attention to particular causal issues will improve theory and
theory testing and provide a more systematic basis for identifying policy applications. Our argument centers on a discussion of: (i) causal versus spurious effects, (ii)
independent versus shared causes, (iii) reversible versus irreversible causes, including symmetric versus asymmetric causes, (iv) basic versus superficial causes, and (v)
causal heterogeneity among different populations, units of comparisons, including
spatial units, and types of behavior. We further suggest how researchers and policymakers can benefit from consideration of causal issues.

INTRODUCTION
Explanations in the social sciences almost invariably are causal explanations.
While social scientists have debated how to define, model, and demonstrate
causation, key issues still have gone unaddressed. This essay addresses some
of these key issues through the lens of criminological literature, with an eye
to implications for theory and policy in all of the social sciences.
Many criminological theories focus on the causes of crime. Although
some crime theorists appear to eschew the term cause, they substitute other
terms, such as influences, leads to, affects, determines, structures, prevents,
creates, depends on, brings about, increases (or decreases), shapes, results in, is
due to, produces, generates, and forces (DiCristina, 1995; Glenn, 1989). It is
important that criminological theories are causal because noncausal or
covariational theories lack policy applications. One of the principal reasons
to construct causal theories in the social sciences, whether about crime or
another outcome, is to apply them, that is, to use them to identify effective
intervention policies for individuals or populations (Freedman, 1997; Hart
& Honore, 1985; Marini & Singer, 1988; Sampson, Winship, & Knight, 2013).
Emerging Trends in the Social and Behavioral Sciences. Edited by Robert Scott and Stephen Kosslyn.
© 2015 John Wiley & Sons, Inc. ISBN 978-1-118-90077-2.

1

2

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

In the case of crime, the idea is that if X causes crime, then we may be able
to intervene to change X, thereby reducing, if not eliminating, crime.
Criminological theories often involve multiple causes, as do other theories
in the social sciences (Cartwright, 2004; Ragin, 2000). There is recognition
of multiple causes in theories outside the social sciences as well, including
the experimental sciences. However, in theory testing in the experimental
sciences, the effects of particular causes can be separated from the effects
of other causes through randomization. For example, if W, X, and Y are
purported to be causes of Z, the independent causal effects of W can be
estimated by randomly assigning cases to values of W, manipulating W,
and then observing the values of Z. Randomization will ensure that the
effects of X and Y on Z are controlled, leaving only the causal effects of W
on Z. Randomization is often impossible in the social sciences, and criminology is no exception. For example, if a theory attributes lawbreaking by
juveniles to such causes as child abuse, parents’ divorce, and school failure
(as does Agnew, 1992), it would be morally unacceptable to randomly
assign juveniles to abusive and nonabusive families in order to estimate
the independent causal effects of child abuse (Glenn, 1989). Without the
possibility of randomization, tests of criminological theories usually rely on
statistical controls (or partialing) of variables with multiple regression or
similar statistical techniques (also see Farrington & Welsh, 2007). However,
applied statisticians have long questioned the use of these techniques for
drawing causal inferences because of the need to make strong a priori theoretical assumptions that cannot be checked (Clogg & Haritou, 1997; Glenn,
1989; Ragin, 2000; Smith, 1990). Among the more important unchecked
assumptions for testing criminological theories, use of these techniques
assumes that the causal effects of a set of variables are (i) uncorrelated with
the effects of left-out causes (omitted-variable bias) and (ii) the same across
all cases (causal homogeneity).
This essay examines these kinds of assumptions and other issues about
theories of crime causation. This is done by considering: (i) causal versus
spurious effects, (ii) independent versus shared causes, (iii) reversible versus
irreversible causes, including symmetric versus asymmetric causes (iv) basic
versus superficial causes, and (v) causal heterogeneity among different populations, units of comparisons, including spatial units, and types of crime.
While the causal issues considered here lend themselves mainly to quantitative considerations, causal inferences are no less problematic in qualitative
research.
FOUNDATIONAL RESEARCH
There is no accepted definition of “cause,” and it is widely purported that
any attempted definition is destined to fail. We have no desire to join that

Causation, Theory, and Policy in the Social Sciences

3

definitional debate. For the purposes of this essay, Hart’s (Hart & Honore,
1985, p. 29) definition will suffice: “a cause is essentially something which
interferes with or intervenes in the course of events which would normally
take place” so as to change that course of events (also see Freedman, 1997,
p. 116; Stinchcombe, 2005, p. 255). As such, a “cause” is different from both a
necessary and sufficient condition. Suppose that a fire results when someone
drops a lit cigarette on combustible material. Although oxygen in the air is a
necessary condition for a fire, normally we would not think of it as causing
the fire; instead, we would think of the lit cigarette as the cause (Hart & Honore, 1985). Now suppose that a man shoots and kills another man. Although
deprivation of blood cells of oxygen is a sufficient cause of the man’s death,
we would not think of it as the cause; instead, we would think of the shooting
as causing the death. Although deprivation of blood cells of oxygen is a sufficient condition for death, we are more interested in the “cause of death under
circumstances which call for an explanation” (Hart & Honore, 1985, p. 39).
CUTTING-EDGE RESEARCH
CAUSAL VERSUS SPURIOUS EFFECTS
Criminologists have been more likely to consider “causal versus spurious
effects” than any other causal issue, the central question being whether an
independent variable, X, actually causes crime, net of other independent variables that might cause both X and crime (Hirschi & Selvin, 1966). For the sake
of illustration, consider a claim that lack of religion causes delinquency. Johnson, Li, Larson, and McCullough (2000, pp. 37–38) caution that multivariate
analyses are necessary to draw “acceptable” causal inferences about the relationship between religion and delinquency. Such cautions reflect a belief that
the relationship between any independent variable and crime might be spuriously attributable to other variables. Such a belief is ostensibly why many
researchers control for such demographic variables as age, race, gender, and
socioeconomic status (SES) in analyzing criminological data (Glenn, 1989,
p. 130). Such demographic variables could be causally related to crime and
its covariates.
In the case of the religion-delinquency relationship, it is believed that such
variables as work and parental and peer influences may be sources of spuriousness (Benda & Corwyn, 1997; Evans, Cullen, Dunaway, & Burton, 1995).
For example, religion and delinquency might be related only because both
variables are caused by involvement of parents in the lives of their children.
If this is the case, religion should not be significantly related to delinquency
when parental involvement is included with religion in a multivariate statistical analysis. If religion continues to be significantly related to delinquency

4

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

when parental involvement is controlled, it is conventional to conclude that
there is a causal relationship between religion and delinquency.
Although conventional, such a conclusion may be wrong because of
“omitted-variable bias” (Clogg & Haritou, 1997; Freedman, 2004, 2006;
Sobel, 2005). Even if a statistical analysis controls for a wide range of variables to detect spuriousness, it is possible that the true source of spuriousness
is omitted from the analysis, perhaps because it is unknown. There is no
statistical technique, however sophisticated, that can identify an unknown
source of spuriousness, since that is a theoretical, not a statistical issue. The
more general issue is summarized by Clogg and Haritou (1997, p. 106): If it
can be known with certainty that a model about the relationship between X
and Y is “causally” right when Z is included and “causally” wrong when Z is
omitted, “then of course the casual effect can be identified.” The problem is
that this can never be known with reasonable certainty about any purported
causal relationship in nonexperimental research. The solution to omitted
variable bias is not inclusion of more independent variables in a multivariate
analysis. As Clarke (2005, p. 346) has noted:
unless a researcher knows the remaining omitted variable, and furthermore
knows the relationship of that variable with the newly included variable, she
cannot know the effect that the newly included variable will have on the bias of
a coefficient of interest. The newly included variable may decrease the bias, but
it is just as likely to increase the bias. In short, we cannot know the effect on the
bias of including an additional control variable unless we know the complete
and true specification.

If there is any bias, it is possible to reach wrong conclusions about the variables that should be targeted to reduce crime.
INDEPENDENT VERSUS SHARED CAUSES
An issue closely related to “causal versus spurious effects” is “independent
versus shared causes.” Researchers often attempt to identify variables that
are independently causally related to crime, that is, variables that do not
share causal effects with other variables (for a general discussion of this issue
in the social sciences, see Glenn, 1989, p. 133). To illustrate, consider a simple theory that crime is caused by both race and SES. The theory is “simple” because it does not address how race and SES cause crime, and there
are many possibilities, such as family socialization practices (Farrington &
Welsh, 2007, p. 79). However, because race and SES are likely to be at least
moderately associated, a statistical analysis may show that neither is independently related to crime even if both are actually causes of crime. Sekhon
(2004, p. 24) gives an example outside of criminology of a regression analysis

Causation, Theory, and Policy in the Social Sciences

5

of the relationship between race and uncounted election ballots in the 2000
US Presidential election, which shows how shared causes can be more rather
than less important than independent causes:
If we were able to estimate a regression model … , showing that there was no
relationship between the race of a voter and his or her probability of casting
uncounted ballots when … controlling for a long list of covariates, it would
be unclear what we had found … Before any regression is estimated, we
know that if we measure enough variables well, the race variable itself … will
be insignificant. However, in a world where being black is highly correlated
with socioeconomic variables, it is not clear what we learn about the causality
of ballot problems from a showing that the race coefficient … can be made
insignificant.

Similarly, Marini and Singer (1988, pp. 356–357) also illustrate how
researchers may be interested only in the:
disjunctive plurality of causes that may produce an effect … If an individual
is identified as having high susceptibility to several causes of death and dies
shortly thereafter, this information offers some explanation of why the individual died but does not single out the actual cause of death. It may be irrelevant
to know which of several possible causes produces an effect.

Many criminological theorists ignore the issue of independent and shared
causes, choosing instead to let researchers disentangle it, but theories of crime
causation would benefit from explicitly recognizing that most causes of crime
are probably shared. For example, in his general strain theory, Agnew (1992)
argues that delinquency is caused by anger and other negative emotions that
are, in turn, caused by negative life experiences, such as child abuse, failure
in school, divorce of parents, and loss of a girlfriend/boyfriend. It may be
interesting to learn which negative life experiences are most strongly, independently related to delinquency. However, many negative life experiences
covary as when abused children also fail in school and experience romantic difficulties, and the covariation renders them no less important causes of
delinquency (also see Farrington & Welsh, 2007, p. 22). Moreover, in the case
of policy applications, it would be incredulous to address only those negative life experiences that are independently causally related to delinquency
and ignore the rest.
There is a tendency among criminologists to assume that all (or virtually
all) independent variables in a multivariate analysis are causally related to
crime. For example, if W, X, and Y are entered as independent variables in
a multivariate analysis with crime as the dependent variable, there is a tendency to assume that all three independent variables are causes of crime.

6

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

However, assume that only W and X are causally related to crime and that
Y only covaries (is not causally related) with crime. In such a case, it makes
little sense to control for Y when estimating the effects of W and X on crime.
The problem is worse if Y not only covaries with crime but also with W and X
because controlling for Y could lead to underestimation of the causal effects
of W and X on crime, which could produce incorrect policy decisions about
the variables that need to be manipulated to reduce crime. Farrington and
Welsh (2007, p. 96) argue that in the absence of systematic knowledge about
which variables cause crime and which variables only covary with it that
interventions should involve a “blunderbuss approach” that targets multiple
variables.
REVERSIBLE VERSUS IRREVERSIBLE CAUSES
Causes lend themselves to effective intervention policies only if they are
reversible, and many causes of crime may be irreversible. According to
Gotttfredson and Hirschi (1990), crime is caused by low self-control, coupled
with criminal opportunity. Their theory is that low self-control forms in early
childhood as a function of ineffective parenting and remains stable after
that. For people with low self-control, intervention should not decrease their
propensity to commit crime. Similarly, Moffitt (1993) argues that permanent
neuropsychological impairments, which can be inherited or caused by such
factors as maternal alcohol and other drug use, poor prenatal nutrition,
and brain injury, cause persistent offending across the life course. These
life-course persistent offenders have poor “verbal skills … and … [weak]
self-control” and cannot be rehabilitated (Moffitt, Lynam, & Silva, 1994,
p. 280).
There are different types of irreversible causes (Lieberson, 1985). The
Gottfredson–Hirschi and Moffitt theories point to causes of crime that are
irreversible because they are unalterable. However, causes of crime may be
alterable and still have irreversible effects. According to Lieberson (1985),
most researchers assume that if an increase in X causes an increase in Y, then
a decrease in X should cause a decrease in Y. However, that is true only of
symmetric causes. Asymmetric causes are such that an increase in X causes
an increase in Y but a decrease in X does not cause a decrease in Y. There
are many examples of asymmetric causes outside of criminology. Among
populations or for any population over time, the incidence of lung cancer
increases when many people smoke, and it decreases when many people
stop smoking. Hence, smoking is a symmetric cause of cancer at the population level. However, at the individual level, smoking is an asymmetric cause
of lung cancer because smoking cessation will not cure cancer (though it
may reduce the likelihood of lung cancer among smokers who have not yet

Causation, Theory, and Policy in the Social Sciences

7

acquired it). Hence, among individuals, the cause of lung cancer is different
from what causes its cure (Hart & Honore, 1985, p. 36). Similarly, Uggen and
Piliavin (1998) argue that the reasons why people initiate criminal behavior
probably are different from the reasons they may later desist from it (also
see Kazemian, 2007; Rosenfeld, 2011). Moreover, they argue that it may be
easier to translate causes of desistance into effective intervention policies:
Our ability to isolate the true causal effect of critical etiological factors such as
parents, schools, and neighborhoods is constrained by our inability to manipulate the selection mechanisms guiding their allocation. For both social scientific
research and for policy purposes, manipulation of these factors is unacceptably
invasive in a democratic society. The researcher conducting a desistance study
has a more legitimate and expansive license to intervene in the lives of participants (Uggen & Piliavin, 1998, pp. 1412–1413).

BASIC VERSUS SUPERFICIAL CAUSES
According to Lieberson (1985), many social scientific theories have focused
on superficial rather than basic causes, partly because of a reliance on studying variation. Speaking of the classic image of Sir Isaac Newton sitting under
the proverbial apple tree, Lieberson (1985) argues that social scientists probably would identify something other than gravity as the cause of the apple’s
fall to the ground because gravity is not a variable quantity in earthly situations. Viewed this way, a basic cause is akin to Aristotle’s formal cause,
which involves the very essence of a thing (Marini & Singer, 1988, p. 363).
Theories of crime causation have sometimes posited basic causes. One of the
best examples is Merton’s (1957) theory of anomie, which states that crime in
the United States is caused by a combination of a basic and a superficial cause.
The basic cause is adherence to the American Dream, which according to him,
is a goal universally shared by people in the United States, and the superficial
cause consists of opportunities to achieve the American Dream, which some
people have more than others. There is little wonder why researchers have
focused on the superficial cause more than the basic cause because, according to Merton (1957), the American Dream is a constant that falls outside the
scope of conventional research methodologies.
It may seem from the foregoing example that only superficial causes are
variables. However, it is likely that many basic causes of crime are variables
and, hence, do fall within the scope of conventional research methodologies,
at least in principle. The qualification is important because there may be serious difficulties with incorporating variable basic causes in testing theories of
crime causation, even though it may be possible to do so “in principle.” An
example comes from Sampson and Laub’s (1993) life-course theory in which

8

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

they argue that certain life experiences, such as marriage or employment, will
cause offenders to desist from crime or at least reduce their offending. However, suppose that both marriage and desistance from crime are causal effects
of a “desire to change.” The problem is not that marriage and desistance from
crime are spuriously attributable to a “desire to change.” The problem is that
a “desire to change” is very difficult to measure and, hence, to incorporate in
tests of life-course theory even though it is a basic cause and marriage is only
a superficial cause. In such a case, an intervention focusing on marital issues
may fail to reduce crime because it ignores a basic cause.
CAUSAL HETEROGENEITY
In an ideal situation, the causes of crime would be the same across all populations. If X causes crime in one population, say among US residents, then
X would cause crime in all other populations. However, the situation is far
from ideal. The causes of crime may differ among populations because the
relationship between any independent variable and crime will be affected by
the joint distribution of other variables related to crime, and this joint distribution can vary among populations. If so, a causal relationship between
an independent variable and crime likewise will vary among populations
(Fagan, 2013, pp. 628–632).
Something similar is true of the causal relationship between an independent variable and crime over time. For example, unemployment rates and
crime rates are produced by different stochastic or probabilistic processes,
and the result is that the two rates do not “track” well together. The causal
relationship, if any, is likely to be complex, “perhaps with changes in unemployment affecting changes in crime in a nonlinear way, or with structural
breaks (meaning that the causal relationship changes over time)” (Bushway,
2011, pp. 194–195).
Heterogeneity in the causes of crime also may involve different units of
comparison (Bhrolchain & Dyson, 2007). For example, researchers have consistently found either no or a weak negative association among individuals
between SES and delinquency (Tittle, Villemez, & Smith, 1978). However,
there also is considerable evidence of a strong positive relationship among US
territorial units, such as cities and metropolitan areas, between crime and economic deprivation, reflected in such variables as the percent of families that
live below the poverty line and income inequality. Parker’s (1989) research on
city-variation in homicide rates is relevant here, as is Blau and Blau’s (1982)
research that revealed a strong positive relationship among US metropolitan
areas between income inequality and rates of violent crime.
Even for the same population, the causes of crime may differ from one unit
of comparison to the next. It may be useful to consider an analogy about

Causation, Theory, and Policy in the Social Sciences

9

the washing of hands. Even if there is a causal relationship among individuals in a particular country between hand washing and disease, there may be
no causal relationship between the two variables among cities in the same
country. Variation in disease among individuals might not be affected by
factors affecting city-level variation, such as clean water and adequate nutrition. It is not that the one of the causal relationships is right and the other is
wrong. They are just different, and that difference should bedevil theorists,
researchers, and policymakers.
Little (2011, p. 288) offers a criminological example. If church membership
cause a young person to refrain from committing crime, “then we ought to
find at the macro-level that a higher index of church membership will be associated with [cause] a lower crime rate.” However, research is likely to produce
contrary findings because of the many disparate causes of macro-level variation in crime rates.
Heterogeneity in the causes of crime may be even more complex than
this, involving disaggregation by type of crime. To illustrate, Parker (1989)
reported in a study of US cities that different types of homicide may have
different causes. His multivariate analysis included four independent
variables: (i) a poverty index, (ii) income inequality, (iii) a dummy variable
for southern region, and (iv) percent black. Neither income inequality nor
the southern-region dummy variable was significantly positively related
to variation in any of the homicide types. However, the poverty index was
significantly positively related to nonrobbery felony homicides, primary
nonintimate (friends and acquaintances) homicides, and family-intimate
homicides, and percent black was significantly positively related to robbery
homicides and primary non-intimate homicides.
Another example comes by Chamlin and Cochran (1998). They found that
both increases and decreases in oil prices significantly affected the level
of commercial burglaries, but not residential burglaries in Oklahoma City.
They also found evidence of asymmetric causation. Specifically, while a
decrease in oil prices caused a slight increase in commercial burglaries, there
was a substantial decrease in commercial burglaries when oil prices were
increasing—the absolute value of percent change in commercial burglaries
was 10 times greater during the period of oil-price increases compared to
the period of oil-price decreases.
In thinking about causal heterogeneity among types of crime, it bears
emphasizing that there also may be causal heterogeneity within types of
crime, arising from at least two sources. First, different researchers may
measure a given type of crime (e.g., violence) differently, using, for example,
different question wording in surveys. Second, researchers may use different
response options. They may, for example, ask respondents whether they

10

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

have committed any crime, how much crime they have committed, the
timing between crime events, and so on.
Such heterogeneity points to the likelihood that researchers are measuring
different concepts, all of which may relate to crime, but nonetheless reflect
important differences. By way of analogy, there are many types of depression with the differences depending, among other things, on the frequency,
duration, and intensity of the symptoms. In short, not all depression is the
same, and the causes of any given type may vary. Put differently, identifying
types of depression—or any other behavior—is indicated in no small part
because the causes of each may vary, leading to different treatment or policy
applications.
Therefore, it is with criminal behavior; it may be that people who commit
any crime versus no crime fundamentally differ from each other in important ways. However, those differences are not necessarily the same as those
that distinguish repeat versus one-time offenders. In addition, they are not
necessarily the same as those that distinguish offenders who follow different
trajectories.
KEY ISSUES FOR FUTURE RESEARCH
Theorists, researchers, and policymakers need to be aware of the possibility of drawing unwarranted conclusions about causation. However, there is
no need for despair. Even given the limitations of nonexperimental research,
evidence for causation can be convincing, if not conclusive, when generated from diverse studies, both quantitative and qualitative. As Glenn (1989,
p. 123) states: “certainty may be an illusive goal never to be reached, but the
cumulative evidence from studies conducted with different methods may
often bring us … close to certainty.”
There are other equally, if not more, daunting issues that theorists,
researchers, and policymakers need to consider about the causes of crime.
The complexity of crime (and perhaps all human behavior) requires a
complex treatment of causation, including but not limited to the possibility
that crime may involve independent and shared causes, reversible and
irreversible causes, and basic and superficial causes. The alternative to these
more complex views of causation is likely to be ineffective intervention
strategies.
Finally, there is considerable evidence that the causes of crime may be heterogeneous rather than homogeneous, with the heterogeneity dependent on
type of population, units of comparison, and types of crime. An ideal theory would apply to all populations, units of comparison, and types of crime.
However, there is no existing theory that achieves that ideal. At this time, all
theories of crime must be considered partial, and researchers should continue
to search for ways of integrating and applying them.

Causation, Theory, and Policy in the Social Sciences

11

REFERENCES
Agnew, R. (1992). Foundation for a general strain theory of crime and delinquency.
Criminology, 30(1), 47–87.
Benda, B. B., & Corwyn, R. F. (1997). Religion and delinquency: The relationship after
considering family and peer influences. Journal for the Scientific Study of Religion,
36(1), 81–92.
Bhrolchain, M. N., & Dyson, T. (2007). On causation in demography: Issues and illustrations. Population and Development Review, 33(1), 1–36.
Blau, J. R., & Blau, P. M. (1982). The cost of inequality: Metropolitan structure and
violent crime. American Sociological Review, 47(1), 114–129.
Bushway, S. D. (2011). Labor markets and crime. In J. Q. Wilson & J. Petersilia (Eds.),
Crime and public policy (pp. 183–209). New York, NY: Oxford University Press.
Cartwright, N. (2004). Causation: One word, many things. Philosophy of Science, 71(5),
805–819.
Chamlin, M. B., & Cochran, J. K. (1998). Causality, economic conditions, and burglary.
Criminology, 36(2), 425–440.
Clarke, K. A. (2005). The phantom menace: Omitted variable bias in econometric
research. Conflict Management and Peace Science, 22(4), 341–352.
Clogg, C. C., & Haritou, A. (1997). The regression method of causal inference and a
dilemma confronting this method. In V. R. McKim & S. P. Turner (Eds.), Causality
in crisis? Statistical methods and the search for causal knowledge in the social sciences
(pp. 83–112). Notre Dame: University of Notre Dame Press.
DiCristina, B. (1995). Method in criminology: A philosophical primer. New York, NY:
Harrow and Heston.
Evans, T. D., Cullen, F. T., Dunaway, R. G., & Burton, V. S. (1995). Religion and crime
re-examined: The impact of religion, secular controls, and social ecology on adult
criminality. Criminology, 33(2), 195–224.
Fagan, A. A. (2013). Family-focused interventions to prevent juvenile delinquency:
A case where science and policy can find common ground. Criminology and Public
Policy, 12(4), 617–650.
Farrington, D., & Welsh, B. C. (2007). Saving children from a life of crime: Early risk factors
and effective interventions. New York, NY: Oxford University Press.
Freedman, D. A. (1997). From association to causation via regression. In V. R. McKim
& S. P. Turner (Eds.), Causality in crisis? Statistical methods and the search for causal
knowledge in the social sciences (pp. 113–161). Notre Dame, IN: University of Notre
Dame Press.
Freedman, D. A. (2004). Graphical models for causation and the identification problem. Evaluation Review, 28(4), 267–293.
Freedman, D. A. (2006). Statistical models for causation: What inferential leverage
do they provide? Evaluation Review, 30(6), 691–713.
Glenn, N. D. (1989). What we know, what we say we know: Discrepancies between
warranted and unwarranted conclusions. In H. Eulau (Ed.), Crossroads of social science: The ICPSR 25th anniversary volume (pp. 119–140). New York, NY: Agathon
Press.

12

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

Gotttfredson, M. R., & Hirschi, T. (1990). A general theory of crime. Stanford, CA: Stanford University Press.
Hart, H. L. A., & Honore, T. (1985). Causation in the law (2nd ed.). Oxford, England:
Clarendon Press.
Hirschi, T., & Selvin, H. C. (1966). False criteria of causality in delinquency research.
Social Problems, 13(3), 254–268.
Johnson, B. R., Li, S. D., Larson, D. B., & McCullough, M. (2000). A systematic review
of the religiosity and delinquency literature. Journal of Contemporary Criminal Justice, 16(1), 32–52.
Kazemian, L. (2007). Desistance from crime: Theoretical, empirical, methodological,
and policy considerations. Journal of Contemporary Criminal Justice, 23(1), 5–27.
Lieberson, S. (1985). Making it count: The improvement of social research and theory.
Berkeley: University of California Press.
Little, D. (2011). Causal mechanism in the social realm. In P. M. Illari & J. Willamson
(Eds.), Causality in the sciences (pp. 273–295). New York, NY: Oxford University
Press.
Marini, M. M., & Singer, B. (1988). Causality in the social sciences. In C. C. Clogg
(Ed.), Sociological methodology (pp. 347–409). Washington, DC: American Sociological Association.
Merton, R. K. (1957). Social theory and social structure (rev. and enl. ed.). New York, NY:
Free Press.
Moffitt, T. E. (1993). Adolescence-limited and life-course persistent antisocial behavior: A developmental taxonomy. Psychological Review, 100(4), 674–701.
Moffitt, T. E., Lynam, D. R., & Silva, P. A. (1994). Neuropsychological tests predicting
persistent male delinquency. Criminology, 32(2), 277–300.
Parker, R. N. (1989). Poverty, subculture of violence, and type of homicide. Social
Forces, 67(4), 983–1007.
Ragin, C. C. (2000). Fuzzy-set social science. Chicago, IL: University of Chicago Press.
Rosenfeld, R. (2011). Changing crime rates. In J. Q. Wilson & J. Petersilia (Eds.), Crime
and public policy (pp. 559–588). New York, NY: Oxford University Press.
Sampson, R. J., & Laub, J. H. (1993). Crime in the making: Pathways and turning points
through life. Cambridge, MA: Harvard University Press.
Sampson, R. J., Winship, C., & Knight, C. (2013). Overview of: ‘Translating causal
chains: Principles and strategies for policy-relevant criminology.’ Criminology and
Public Policy, 12(4), 585–616.
Sekhon, J. S. (2004). Quality meets quantity: Case studies, conditional probability
and counterfactuals. Working paper. Retrieved from http://sekhon.berkeley.
edu/papers/QualityQuantity.pdf
Smith, H. L. (1990). Specification problems in experimental and nonexperimental
social research. In C. C. Clogg (Ed.), Sociological methodology (pp. 59–91). Washington, DC: American Sociological Association.
Sobel, M. E. (2005). The scientific model of causality. Sociological Methodology, 35,
99–133.
Stinchcombe, A. L. (2005). The logic of social research. Chicago, IL: University of
Chicago Press.

Causation, Theory, and Policy in the Social Sciences

13

Tittle, C. R., Villemez, W. J., & Smith, D. A. (1978). The myth of social class and criminality: An empirical assessment of the empirical evidence. American Sociological
Review, 43(5), 643–656.
Uggen, C., & Piliavin, I. (1998). Asymmetrical causation and criminal desistance. The
Journal of Criminal Law and Criminology, 88(4), 1399–1422.

FURTHER READING
Berk, R. A. (1987). Causal inference as a prediction problem. In D. M. Gottfredson &
M. Tonry (Eds.), Crime and justice: A review of research (pp. 183–200). Chicago, IL:
University of Chicago Press.
Farrington, D. P. (2003). Development and life-course criminology. Criminology, 41(2),
221–256.
Illari, P. M., Russo, F., & Williamson, J. (2011). Causality in the sciences. New York, NY:
Oxford University Press.
Pearl, J. (2000). Causality: Models, reasoning, and inference. New York, NY: Cambridge
University Press.
Sampson, R. J., Laub, J. H., & Wimer, C. (2006). Does marriage reduce crime? A
counterfactual approach to within-individual causal effects. Criminology, 44(3),
465–508.

MARK C. STAFFORD SHORT BIOGRAPHY
Mark C. Stafford is a Professor in the School of Criminal Justice at Texas State
University, San Marcos, where he is currently Doctoral Program Director. He
has been a faculty member at Washington State University and the University
of Texas at Austin, where he was one of the founders of the Center for Criminology and Criminal Justice Research. He has been a Postdoctoral Fellow at
the Center for Advanced Study in the Behavioral Sciences at Stanford University and also at the University of Colorado, Boulder. He was an IEAT-FORD
Chair of Criminality, Violence, and Public Policy in the Institute of Interdisciplinary Advanced Studies at Federal University of Minas Gerais, Belo Horizonte, Brazil. He has published extensively on deterrence and rational-choice
behavior, victimization and fear of crime, gangs, and causes of crime and
juvenile delinquency.
DANIEL P. MEARS SHORT BIOGRAPHY
Daniel P. Mears is the Mark C. Stafford Professor of Criminology at Florida
State University’s College of Criminology and Criminal Justice. He conducts
research on crime and justice theory and policy. His work has appeared in
Criminology, the Journal of Research in Crime and Delinquency, and other
crime and policy journals and in a book, American Criminal Justice Policy

14

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

(Cambridge University Press), which received the Academy of Criminal Justice Sciences outstanding book award.
RELATED ESSAYS
The Role of Data in Research and Policy (Sociology), Barbara A. Anderson
Expertise (Sociology), Gil Eyal
The Evidence-Based Practice Movement (Sociology), Edward W. Gondolf
Why Do States Sign Alliances? (Political Science), Brett Ashley Leeds
Why Do Governments Abuse Human Rights? (Political Science), Will H.
Moore and Ryan M. Welch
The Social Science of Sustainability (Political Science), Johannes Urpelainen
Why Do States Pursue Nuclear Weapons (or Not) (Political Science), Wilfred
Wan and Etel Solingen
Translational Sociology (Sociology), Elaine Wethington; Causation, Theory, and Policy in the
Social Sciences
MARK C. STAFFORD and DANIEL P. MEARS

Abstract
Despite a penchant for constructing and testing causal theories, social scientists frequently avoid explicit discussion of causal issues. Illustrating with criminological
literature, we argue that attention to particular causal issues will improve theory and
theory testing and provide a more systematic basis for identifying policy applications. Our argument centers on a discussion of: (i) causal versus spurious effects, (ii)
independent versus shared causes, (iii) reversible versus irreversible causes, including symmetric versus asymmetric causes, (iv) basic versus superficial causes, and (v)
causal heterogeneity among different populations, units of comparisons, including
spatial units, and types of behavior. We further suggest how researchers and policymakers can benefit from consideration of causal issues.

INTRODUCTION
Explanations in the social sciences almost invariably are causal explanations.
While social scientists have debated how to define, model, and demonstrate
causation, key issues still have gone unaddressed. This essay addresses some
of these key issues through the lens of criminological literature, with an eye
to implications for theory and policy in all of the social sciences.
Many criminological theories focus on the causes of crime. Although
some crime theorists appear to eschew the term cause, they substitute other
terms, such as influences, leads to, affects, determines, structures, prevents,
creates, depends on, brings about, increases (or decreases), shapes, results in, is
due to, produces, generates, and forces (DiCristina, 1995; Glenn, 1989). It is
important that criminological theories are causal because noncausal or
covariational theories lack policy applications. One of the principal reasons
to construct causal theories in the social sciences, whether about crime or
another outcome, is to apply them, that is, to use them to identify effective
intervention policies for individuals or populations (Freedman, 1997; Hart
& Honore, 1985; Marini & Singer, 1988; Sampson, Winship, & Knight, 2013).
Emerging Trends in the Social and Behavioral Sciences. Edited by Robert Scott and Stephen Kosslyn.
© 2015 John Wiley & Sons, Inc. ISBN 978-1-118-90077-2.

1

2

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

In the case of crime, the idea is that if X causes crime, then we may be able
to intervene to change X, thereby reducing, if not eliminating, crime.
Criminological theories often involve multiple causes, as do other theories
in the social sciences (Cartwright, 2004; Ragin, 2000). There is recognition
of multiple causes in theories outside the social sciences as well, including
the experimental sciences. However, in theory testing in the experimental
sciences, the effects of particular causes can be separated from the effects
of other causes through randomization. For example, if W, X, and Y are
purported to be causes of Z, the independent causal effects of W can be
estimated by randomly assigning cases to values of W, manipulating W,
and then observing the values of Z. Randomization will ensure that the
effects of X and Y on Z are controlled, leaving only the causal effects of W
on Z. Randomization is often impossible in the social sciences, and criminology is no exception. For example, if a theory attributes lawbreaking by
juveniles to such causes as child abuse, parents’ divorce, and school failure
(as does Agnew, 1992), it would be morally unacceptable to randomly
assign juveniles to abusive and nonabusive families in order to estimate
the independent causal effects of child abuse (Glenn, 1989). Without the
possibility of randomization, tests of criminological theories usually rely on
statistical controls (or partialing) of variables with multiple regression or
similar statistical techniques (also see Farrington & Welsh, 2007). However,
applied statisticians have long questioned the use of these techniques for
drawing causal inferences because of the need to make strong a priori theoretical assumptions that cannot be checked (Clogg & Haritou, 1997; Glenn,
1989; Ragin, 2000; Smith, 1990). Among the more important unchecked
assumptions for testing criminological theories, use of these techniques
assumes that the causal effects of a set of variables are (i) uncorrelated with
the effects of left-out causes (omitted-variable bias) and (ii) the same across
all cases (causal homogeneity).
This essay examines these kinds of assumptions and other issues about
theories of crime causation. This is done by considering: (i) causal versus
spurious effects, (ii) independent versus shared causes, (iii) reversible versus
irreversible causes, including symmetric versus asymmetric causes (iv) basic
versus superficial causes, and (v) causal heterogeneity among different populations, units of comparisons, including spatial units, and types of crime.
While the causal issues considered here lend themselves mainly to quantitative considerations, causal inferences are no less problematic in qualitative
research.
FOUNDATIONAL RESEARCH
There is no accepted definition of “cause,” and it is widely purported that
any attempted definition is destined to fail. We have no desire to join that

Causation, Theory, and Policy in the Social Sciences

3

definitional debate. For the purposes of this essay, Hart’s (Hart & Honore,
1985, p. 29) definition will suffice: “a cause is essentially something which
interferes with or intervenes in the course of events which would normally
take place” so as to change that course of events (also see Freedman, 1997,
p. 116; Stinchcombe, 2005, p. 255). As such, a “cause” is different from both a
necessary and sufficient condition. Suppose that a fire results when someone
drops a lit cigarette on combustible material. Although oxygen in the air is a
necessary condition for a fire, normally we would not think of it as causing
the fire; instead, we would think of the lit cigarette as the cause (Hart & Honore, 1985). Now suppose that a man shoots and kills another man. Although
deprivation of blood cells of oxygen is a sufficient cause of the man’s death,
we would not think of it as the cause; instead, we would think of the shooting
as causing the death. Although deprivation of blood cells of oxygen is a sufficient condition for death, we are more interested in the “cause of death under
circumstances which call for an explanation” (Hart & Honore, 1985, p. 39).
CUTTING-EDGE RESEARCH
CAUSAL VERSUS SPURIOUS EFFECTS
Criminologists have been more likely to consider “causal versus spurious
effects” than any other causal issue, the central question being whether an
independent variable, X, actually causes crime, net of other independent variables that might cause both X and crime (Hirschi & Selvin, 1966). For the sake
of illustration, consider a claim that lack of religion causes delinquency. Johnson, Li, Larson, and McCullough (2000, pp. 37–38) caution that multivariate
analyses are necessary to draw “acceptable” causal inferences about the relationship between religion and delinquency. Such cautions reflect a belief that
the relationship between any independent variable and crime might be spuriously attributable to other variables. Such a belief is ostensibly why many
researchers control for such demographic variables as age, race, gender, and
socioeconomic status (SES) in analyzing criminological data (Glenn, 1989,
p. 130). Such demographic variables could be causally related to crime and
its covariates.
In the case of the religion-delinquency relationship, it is believed that such
variables as work and parental and peer influences may be sources of spuriousness (Benda & Corwyn, 1997; Evans, Cullen, Dunaway, & Burton, 1995).
For example, religion and delinquency might be related only because both
variables are caused by involvement of parents in the lives of their children.
If this is the case, religion should not be significantly related to delinquency
when parental involvement is included with religion in a multivariate statistical analysis. If religion continues to be significantly related to delinquency

4

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

when parental involvement is controlled, it is conventional to conclude that
there is a causal relationship between religion and delinquency.
Although conventional, such a conclusion may be wrong because of
“omitted-variable bias” (Clogg & Haritou, 1997; Freedman, 2004, 2006;
Sobel, 2005). Even if a statistical analysis controls for a wide range of variables to detect spuriousness, it is possible that the true source of spuriousness
is omitted from the analysis, perhaps because it is unknown. There is no
statistical technique, however sophisticated, that can identify an unknown
source of spuriousness, since that is a theoretical, not a statistical issue. The
more general issue is summarized by Clogg and Haritou (1997, p. 106): If it
can be known with certainty that a model about the relationship between X
and Y is “causally” right when Z is included and “causally” wrong when Z is
omitted, “then of course the casual effect can be identified.” The problem is
that this can never be known with reasonable certainty about any purported
causal relationship in nonexperimental research. The solution to omitted
variable bias is not inclusion of more independent variables in a multivariate
analysis. As Clarke (2005, p. 346) has noted:
unless a researcher knows the remaining omitted variable, and furthermore
knows the relationship of that variable with the newly included variable, she
cannot know the effect that the newly included variable will have on the bias of
a coefficient of interest. The newly included variable may decrease the bias, but
it is just as likely to increase the bias. In short, we cannot know the effect on the
bias of including an additional control variable unless we know the complete
and true specification.

If there is any bias, it is possible to reach wrong conclusions about the variables that should be targeted to reduce crime.
INDEPENDENT VERSUS SHARED CAUSES
An issue closely related to “causal versus spurious effects” is “independent
versus shared causes.” Researchers often attempt to identify variables that
are independently causally related to crime, that is, variables that do not
share causal effects with other variables (for a general discussion of this issue
in the social sciences, see Glenn, 1989, p. 133). To illustrate, consider a simple theory that crime is caused by both race and SES. The theory is “simple” because it does not address how race and SES cause crime, and there
are many possibilities, such as family socialization practices (Farrington &
Welsh, 2007, p. 79). However, because race and SES are likely to be at least
moderately associated, a statistical analysis may show that neither is independently related to crime even if both are actually causes of crime. Sekhon
(2004, p. 24) gives an example outside of criminology of a regression analysis

Causation, Theory, and Policy in the Social Sciences

5

of the relationship between race and uncounted election ballots in the 2000
US Presidential election, which shows how shared causes can be more rather
than less important than independent causes:
If we were able to estimate a regression model … , showing that there was no
relationship between the race of a voter and his or her probability of casting
uncounted ballots when … controlling for a long list of covariates, it would
be unclear what we had found … Before any regression is estimated, we
know that if we measure enough variables well, the race variable itself … will
be insignificant. However, in a world where being black is highly correlated
with socioeconomic variables, it is not clear what we learn about the causality
of ballot problems from a showing that the race coefficient … can be made
insignificant.

Similarly, Marini and Singer (1988, pp. 356–357) also illustrate how
researchers may be interested only in the:
disjunctive plurality of causes that may produce an effect … If an individual
is identified as having high susceptibility to several causes of death and dies
shortly thereafter, this information offers some explanation of why the individual died but does not single out the actual cause of death. It may be irrelevant
to know which of several possible causes produces an effect.

Many criminological theorists ignore the issue of independent and shared
causes, choosing instead to let researchers disentangle it, but theories of crime
causation would benefit from explicitly recognizing that most causes of crime
are probably shared. For example, in his general strain theory, Agnew (1992)
argues that delinquency is caused by anger and other negative emotions that
are, in turn, caused by negative life experiences, such as child abuse, failure
in school, divorce of parents, and loss of a girlfriend/boyfriend. It may be
interesting to learn which negative life experiences are most strongly, independently related to delinquency. However, many negative life experiences
covary as when abused children also fail in school and experience romantic difficulties, and the covariation renders them no less important causes of
delinquency (also see Farrington & Welsh, 2007, p. 22). Moreover, in the case
of policy applications, it would be incredulous to address only those negative life experiences that are independently causally related to delinquency
and ignore the rest.
There is a tendency among criminologists to assume that all (or virtually
all) independent variables in a multivariate analysis are causally related to
crime. For example, if W, X, and Y are entered as independent variables in
a multivariate analysis with crime as the dependent variable, there is a tendency to assume that all three independent variables are causes of crime.

6

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

However, assume that only W and X are causally related to crime and that
Y only covaries (is not causally related) with crime. In such a case, it makes
little sense to control for Y when estimating the effects of W and X on crime.
The problem is worse if Y not only covaries with crime but also with W and X
because controlling for Y could lead to underestimation of the causal effects
of W and X on crime, which could produce incorrect policy decisions about
the variables that need to be manipulated to reduce crime. Farrington and
Welsh (2007, p. 96) argue that in the absence of systematic knowledge about
which variables cause crime and which variables only covary with it that
interventions should involve a “blunderbuss approach” that targets multiple
variables.
REVERSIBLE VERSUS IRREVERSIBLE CAUSES
Causes lend themselves to effective intervention policies only if they are
reversible, and many causes of crime may be irreversible. According to
Gotttfredson and Hirschi (1990), crime is caused by low self-control, coupled
with criminal opportunity. Their theory is that low self-control forms in early
childhood as a function of ineffective parenting and remains stable after
that. For people with low self-control, intervention should not decrease their
propensity to commit crime. Similarly, Moffitt (1993) argues that permanent
neuropsychological impairments, which can be inherited or caused by such
factors as maternal alcohol and other drug use, poor prenatal nutrition,
and brain injury, cause persistent offending across the life course. These
life-course persistent offenders have poor “verbal skills … and … [weak]
self-control” and cannot be rehabilitated (Moffitt, Lynam, & Silva, 1994,
p. 280).
There are different types of irreversible causes (Lieberson, 1985). The
Gottfredson–Hirschi and Moffitt theories point to causes of crime that are
irreversible because they are unalterable. However, causes of crime may be
alterable and still have irreversible effects. According to Lieberson (1985),
most researchers assume that if an increase in X causes an increase in Y, then
a decrease in X should cause a decrease in Y. However, that is true only of
symmetric causes. Asymmetric causes are such that an increase in X causes
an increase in Y but a decrease in X does not cause a decrease in Y. There
are many examples of asymmetric causes outside of criminology. Among
populations or for any population over time, the incidence of lung cancer
increases when many people smoke, and it decreases when many people
stop smoking. Hence, smoking is a symmetric cause of cancer at the population level. However, at the individual level, smoking is an asymmetric cause
of lung cancer because smoking cessation will not cure cancer (though it
may reduce the likelihood of lung cancer among smokers who have not yet

Causation, Theory, and Policy in the Social Sciences

7

acquired it). Hence, among individuals, the cause of lung cancer is different
from what causes its cure (Hart & Honore, 1985, p. 36). Similarly, Uggen and
Piliavin (1998) argue that the reasons why people initiate criminal behavior
probably are different from the reasons they may later desist from it (also
see Kazemian, 2007; Rosenfeld, 2011). Moreover, they argue that it may be
easier to translate causes of desistance into effective intervention policies:
Our ability to isolate the true causal effect of critical etiological factors such as
parents, schools, and neighborhoods is constrained by our inability to manipulate the selection mechanisms guiding their allocation. For both social scientific
research and for policy purposes, manipulation of these factors is unacceptably
invasive in a democratic society. The researcher conducting a desistance study
has a more legitimate and expansive license to intervene in the lives of participants (Uggen & Piliavin, 1998, pp. 1412–1413).

BASIC VERSUS SUPERFICIAL CAUSES
According to Lieberson (1985), many social scientific theories have focused
on superficial rather than basic causes, partly because of a reliance on studying variation. Speaking of the classic image of Sir Isaac Newton sitting under
the proverbial apple tree, Lieberson (1985) argues that social scientists probably would identify something other than gravity as the cause of the apple’s
fall to the ground because gravity is not a variable quantity in earthly situations. Viewed this way, a basic cause is akin to Aristotle’s formal cause,
which involves the very essence of a thing (Marini & Singer, 1988, p. 363).
Theories of crime causation have sometimes posited basic causes. One of the
best examples is Merton’s (1957) theory of anomie, which states that crime in
the United States is caused by a combination of a basic and a superficial cause.
The basic cause is adherence to the American Dream, which according to him,
is a goal universally shared by people in the United States, and the superficial
cause consists of opportunities to achieve the American Dream, which some
people have more than others. There is little wonder why researchers have
focused on the superficial cause more than the basic cause because, according to Merton (1957), the American Dream is a constant that falls outside the
scope of conventional research methodologies.
It may seem from the foregoing example that only superficial causes are
variables. However, it is likely that many basic causes of crime are variables
and, hence, do fall within the scope of conventional research methodologies,
at least in principle. The qualification is important because there may be serious difficulties with incorporating variable basic causes in testing theories of
crime causation, even though it may be possible to do so “in principle.” An
example comes from Sampson and Laub’s (1993) life-course theory in which

8

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

they argue that certain life experiences, such as marriage or employment, will
cause offenders to desist from crime or at least reduce their offending. However, suppose that both marriage and desistance from crime are causal effects
of a “desire to change.” The problem is not that marriage and desistance from
crime are spuriously attributable to a “desire to change.” The problem is that
a “desire to change” is very difficult to measure and, hence, to incorporate in
tests of life-course theory even though it is a basic cause and marriage is only
a superficial cause. In such a case, an intervention focusing on marital issues
may fail to reduce crime because it ignores a basic cause.
CAUSAL HETEROGENEITY
In an ideal situation, the causes of crime would be the same across all populations. If X causes crime in one population, say among US residents, then
X would cause crime in all other populations. However, the situation is far
from ideal. The causes of crime may differ among populations because the
relationship between any independent variable and crime will be affected by
the joint distribution of other variables related to crime, and this joint distribution can vary among populations. If so, a causal relationship between
an independent variable and crime likewise will vary among populations
(Fagan, 2013, pp. 628–632).
Something similar is true of the causal relationship between an independent variable and crime over time. For example, unemployment rates and
crime rates are produced by different stochastic or probabilistic processes,
and the result is that the two rates do not “track” well together. The causal
relationship, if any, is likely to be complex, “perhaps with changes in unemployment affecting changes in crime in a nonlinear way, or with structural
breaks (meaning that the causal relationship changes over time)” (Bushway,
2011, pp. 194–195).
Heterogeneity in the causes of crime also may involve different units of
comparison (Bhrolchain & Dyson, 2007). For example, researchers have consistently found either no or a weak negative association among individuals
between SES and delinquency (Tittle, Villemez, & Smith, 1978). However,
there also is considerable evidence of a strong positive relationship among US
territorial units, such as cities and metropolitan areas, between crime and economic deprivation, reflected in such variables as the percent of families that
live below the poverty line and income inequality. Parker’s (1989) research on
city-variation in homicide rates is relevant here, as is Blau and Blau’s (1982)
research that revealed a strong positive relationship among US metropolitan
areas between income inequality and rates of violent crime.
Even for the same population, the causes of crime may differ from one unit
of comparison to the next. It may be useful to consider an analogy about

Causation, Theory, and Policy in the Social Sciences

9

the washing of hands. Even if there is a causal relationship among individuals in a particular country between hand washing and disease, there may be
no causal relationship between the two variables among cities in the same
country. Variation in disease among individuals might not be affected by
factors affecting city-level variation, such as clean water and adequate nutrition. It is not that the one of the causal relationships is right and the other is
wrong. They are just different, and that difference should bedevil theorists,
researchers, and policymakers.
Little (2011, p. 288) offers a criminological example. If church membership
cause a young person to refrain from committing crime, “then we ought to
find at the macro-level that a higher index of church membership will be associated with [cause] a lower crime rate.” However, research is likely to produce
contrary findings because of the many disparate causes of macro-level variation in crime rates.
Heterogeneity in the causes of crime may be even more complex than
this, involving disaggregation by type of crime. To illustrate, Parker (1989)
reported in a study of US cities that different types of homicide may have
different causes. His multivariate analysis included four independent
variables: (i) a poverty index, (ii) income inequality, (iii) a dummy variable
for southern region, and (iv) percent black. Neither income inequality nor
the southern-region dummy variable was significantly positively related
to variation in any of the homicide types. However, the poverty index was
significantly positively related to nonrobbery felony homicides, primary
nonintimate (friends and acquaintances) homicides, and family-intimate
homicides, and percent black was significantly positively related to robbery
homicides and primary non-intimate homicides.
Another example comes by Chamlin and Cochran (1998). They found that
both increases and decreases in oil prices significantly affected the level
of commercial burglaries, but not residential burglaries in Oklahoma City.
They also found evidence of asymmetric causation. Specifically, while a
decrease in oil prices caused a slight increase in commercial burglaries, there
was a substantial decrease in commercial burglaries when oil prices were
increasing—the absolute value of percent change in commercial burglaries
was 10 times greater during the period of oil-price increases compared to
the period of oil-price decreases.
In thinking about causal heterogeneity among types of crime, it bears
emphasizing that there also may be causal heterogeneity within types of
crime, arising from at least two sources. First, different researchers may
measure a given type of crime (e.g., violence) differently, using, for example,
different question wording in surveys. Second, researchers may use different
response options. They may, for example, ask respondents whether they

10

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

have committed any crime, how much crime they have committed, the
timing between crime events, and so on.
Such heterogeneity points to the likelihood that researchers are measuring
different concepts, all of which may relate to crime, but nonetheless reflect
important differences. By way of analogy, there are many types of depression with the differences depending, among other things, on the frequency,
duration, and intensity of the symptoms. In short, not all depression is the
same, and the causes of any given type may vary. Put differently, identifying
types of depression—or any other behavior—is indicated in no small part
because the causes of each may vary, leading to different treatment or policy
applications.
Therefore, it is with criminal behavior; it may be that people who commit
any crime versus no crime fundamentally differ from each other in important ways. However, those differences are not necessarily the same as those
that distinguish repeat versus one-time offenders. In addition, they are not
necessarily the same as those that distinguish offenders who follow different
trajectories.
KEY ISSUES FOR FUTURE RESEARCH
Theorists, researchers, and policymakers need to be aware of the possibility of drawing unwarranted conclusions about causation. However, there is
no need for despair. Even given the limitations of nonexperimental research,
evidence for causation can be convincing, if not conclusive, when generated from diverse studies, both quantitative and qualitative. As Glenn (1989,
p. 123) states: “certainty may be an illusive goal never to be reached, but the
cumulative evidence from studies conducted with different methods may
often bring us … close to certainty.”
There are other equally, if not more, daunting issues that theorists,
researchers, and policymakers need to consider about the causes of crime.
The complexity of crime (and perhaps all human behavior) requires a
complex treatment of causation, including but not limited to the possibility
that crime may involve independent and shared causes, reversible and
irreversible causes, and basic and superficial causes. The alternative to these
more complex views of causation is likely to be ineffective intervention
strategies.
Finally, there is considerable evidence that the causes of crime may be heterogeneous rather than homogeneous, with the heterogeneity dependent on
type of population, units of comparison, and types of crime. An ideal theory would apply to all populations, units of comparison, and types of crime.
However, there is no existing theory that achieves that ideal. At this time, all
theories of crime must be considered partial, and researchers should continue
to search for ways of integrating and applying them.

Causation, Theory, and Policy in the Social Sciences

11

REFERENCES
Agnew, R. (1992). Foundation for a general strain theory of crime and delinquency.
Criminology, 30(1), 47–87.
Benda, B. B., & Corwyn, R. F. (1997). Religion and delinquency: The relationship after
considering family and peer influences. Journal for the Scientific Study of Religion,
36(1), 81–92.
Bhrolchain, M. N., & Dyson, T. (2007). On causation in demography: Issues and illustrations. Population and Development Review, 33(1), 1–36.
Blau, J. R., & Blau, P. M. (1982). The cost of inequality: Metropolitan structure and
violent crime. American Sociological Review, 47(1), 114–129.
Bushway, S. D. (2011). Labor markets and crime. In J. Q. Wilson & J. Petersilia (Eds.),
Crime and public policy (pp. 183–209). New York, NY: Oxford University Press.
Cartwright, N. (2004). Causation: One word, many things. Philosophy of Science, 71(5),
805–819.
Chamlin, M. B., & Cochran, J. K. (1998). Causality, economic conditions, and burglary.
Criminology, 36(2), 425–440.
Clarke, K. A. (2005). The phantom menace: Omitted variable bias in econometric
research. Conflict Management and Peace Science, 22(4), 341–352.
Clogg, C. C., & Haritou, A. (1997). The regression method of causal inference and a
dilemma confronting this method. In V. R. McKim & S. P. Turner (Eds.), Causality
in crisis? Statistical methods and the search for causal knowledge in the social sciences
(pp. 83–112). Notre Dame: University of Notre Dame Press.
DiCristina, B. (1995). Method in criminology: A philosophical primer. New York, NY:
Harrow and Heston.
Evans, T. D., Cullen, F. T., Dunaway, R. G., & Burton, V. S. (1995). Religion and crime
re-examined: The impact of religion, secular controls, and social ecology on adult
criminality. Criminology, 33(2), 195–224.
Fagan, A. A. (2013). Family-focused interventions to prevent juvenile delinquency:
A case where science and policy can find common ground. Criminology and Public
Policy, 12(4), 617–650.
Farrington, D., & Welsh, B. C. (2007). Saving children from a life of crime: Early risk factors
and effective interventions. New York, NY: Oxford University Press.
Freedman, D. A. (1997). From association to causation via regression. In V. R. McKim
& S. P. Turner (Eds.), Causality in crisis? Statistical methods and the search for causal
knowledge in the social sciences (pp. 113–161). Notre Dame, IN: University of Notre
Dame Press.
Freedman, D. A. (2004). Graphical models for causation and the identification problem. Evaluation Review, 28(4), 267–293.
Freedman, D. A. (2006). Statistical models for causation: What inferential leverage
do they provide? Evaluation Review, 30(6), 691–713.
Glenn, N. D. (1989). What we know, what we say we know: Discrepancies between
warranted and unwarranted conclusions. In H. Eulau (Ed.), Crossroads of social science: The ICPSR 25th anniversary volume (pp. 119–140). New York, NY: Agathon
Press.

12

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

Gotttfredson, M. R., & Hirschi, T. (1990). A general theory of crime. Stanford, CA: Stanford University Press.
Hart, H. L. A., & Honore, T. (1985). Causation in the law (2nd ed.). Oxford, England:
Clarendon Press.
Hirschi, T., & Selvin, H. C. (1966). False criteria of causality in delinquency research.
Social Problems, 13(3), 254–268.
Johnson, B. R., Li, S. D., Larson, D. B., & McCullough, M. (2000). A systematic review
of the religiosity and delinquency literature. Journal of Contemporary Criminal Justice, 16(1), 32–52.
Kazemian, L. (2007). Desistance from crime: Theoretical, empirical, methodological,
and policy considerations. Journal of Contemporary Criminal Justice, 23(1), 5–27.
Lieberson, S. (1985). Making it count: The improvement of social research and theory.
Berkeley: University of California Press.
Little, D. (2011). Causal mechanism in the social realm. In P. M. Illari & J. Willamson
(Eds.), Causality in the sciences (pp. 273–295). New York, NY: Oxford University
Press.
Marini, M. M., & Singer, B. (1988). Causality in the social sciences. In C. C. Clogg
(Ed.), Sociological methodology (pp. 347–409). Washington, DC: American Sociological Association.
Merton, R. K. (1957). Social theory and social structure (rev. and enl. ed.). New York, NY:
Free Press.
Moffitt, T. E. (1993). Adolescence-limited and life-course persistent antisocial behavior: A developmental taxonomy. Psychological Review, 100(4), 674–701.
Moffitt, T. E., Lynam, D. R., & Silva, P. A. (1994). Neuropsychological tests predicting
persistent male delinquency. Criminology, 32(2), 277–300.
Parker, R. N. (1989). Poverty, subculture of violence, and type of homicide. Social
Forces, 67(4), 983–1007.
Ragin, C. C. (2000). Fuzzy-set social science. Chicago, IL: University of Chicago Press.
Rosenfeld, R. (2011). Changing crime rates. In J. Q. Wilson & J. Petersilia (Eds.), Crime
and public policy (pp. 559–588). New York, NY: Oxford University Press.
Sampson, R. J., & Laub, J. H. (1993). Crime in the making: Pathways and turning points
through life. Cambridge, MA: Harvard University Press.
Sampson, R. J., Winship, C., & Knight, C. (2013). Overview of: ‘Translating causal
chains: Principles and strategies for policy-relevant criminology.’ Criminology and
Public Policy, 12(4), 585–616.
Sekhon, J. S. (2004). Quality meets quantity: Case studies, conditional probability
and counterfactuals. Working paper. Retrieved from http://sekhon.berkeley.
edu/papers/QualityQuantity.pdf
Smith, H. L. (1990). Specification problems in experimental and nonexperimental
social research. In C. C. Clogg (Ed.), Sociological methodology (pp. 59–91). Washington, DC: American Sociological Association.
Sobel, M. E. (2005). The scientific model of causality. Sociological Methodology, 35,
99–133.
Stinchcombe, A. L. (2005). The logic of social research. Chicago, IL: University of
Chicago Press.

Causation, Theory, and Policy in the Social Sciences

13

Tittle, C. R., Villemez, W. J., & Smith, D. A. (1978). The myth of social class and criminality: An empirical assessment of the empirical evidence. American Sociological
Review, 43(5), 643–656.
Uggen, C., & Piliavin, I. (1998). Asymmetrical causation and criminal desistance. The
Journal of Criminal Law and Criminology, 88(4), 1399–1422.

FURTHER READING
Berk, R. A. (1987). Causal inference as a prediction problem. In D. M. Gottfredson &
M. Tonry (Eds.), Crime and justice: A review of research (pp. 183–200). Chicago, IL:
University of Chicago Press.
Farrington, D. P. (2003). Development and life-course criminology. Criminology, 41(2),
221–256.
Illari, P. M., Russo, F., & Williamson, J. (2011). Causality in the sciences. New York, NY:
Oxford University Press.
Pearl, J. (2000). Causality: Models, reasoning, and inference. New York, NY: Cambridge
University Press.
Sampson, R. J., Laub, J. H., & Wimer, C. (2006). Does marriage reduce crime? A
counterfactual approach to within-individual causal effects. Criminology, 44(3),
465–508.

MARK C. STAFFORD SHORT BIOGRAPHY
Mark C. Stafford is a Professor in the School of Criminal Justice at Texas State
University, San Marcos, where he is currently Doctoral Program Director. He
has been a faculty member at Washington State University and the University
of Texas at Austin, where he was one of the founders of the Center for Criminology and Criminal Justice Research. He has been a Postdoctoral Fellow at
the Center for Advanced Study in the Behavioral Sciences at Stanford University and also at the University of Colorado, Boulder. He was an IEAT-FORD
Chair of Criminality, Violence, and Public Policy in the Institute of Interdisciplinary Advanced Studies at Federal University of Minas Gerais, Belo Horizonte, Brazil. He has published extensively on deterrence and rational-choice
behavior, victimization and fear of crime, gangs, and causes of crime and
juvenile delinquency.
DANIEL P. MEARS SHORT BIOGRAPHY
Daniel P. Mears is the Mark C. Stafford Professor of Criminology at Florida
State University’s College of Criminology and Criminal Justice. He conducts
research on crime and justice theory and policy. His work has appeared in
Criminology, the Journal of Research in Crime and Delinquency, and other
crime and policy journals and in a book, American Criminal Justice Policy

14

EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES

(Cambridge University Press), which received the Academy of Criminal Justice Sciences outstanding book award.
RELATED ESSAYS
The Role of Data in Research and Policy (Sociology), Barbara A. Anderson
Expertise (Sociology), Gil Eyal
The Evidence-Based Practice Movement (Sociology), Edward W. Gondolf
Why Do States Sign Alliances? (Political Science), Brett Ashley Leeds
Why Do Governments Abuse Human Rights? (Political Science), Will H.
Moore and Ryan M. Welch
The Social Science of Sustainability (Political Science), Johannes Urpelainen
Why Do States Pursue Nuclear Weapons (or Not) (Political Science), Wilfred
Wan and Etel Solingen
Translational Sociology (Sociology), Elaine Wethington