THEORY AND RESEARCH

EXAMPLE BOX
3
Three Elements of Causality

I read that several politicians visited a Catholic school eliminate the alternative explanation that children
in Chicago that had a record of being much more in the two types of schools had different family cir-
successful than public schools in educating children. cumstances that affect learning and that this caused
The next day, the politicians called a news conference learning differences. If the family circumstances (e.g.,
and advocated new laws and the redirection of tax parents’ education and income, family religious belief
money to Catholic schools. As a person who wants and intensity of belief, two-parent versus single-
children to get a good education, I was interested in parent households, degree of parental interest in
the story, but as a social scientist, I critically evaluated child’s education) are the same for children who
it. The politicians’ theory said Catholic schools cause attend both types of schools, then the politicians are
more learning than public schools. They had two ele- on the right track. The focus, then, is on what Catholic
ments of causality: temporal order (first the children schools are doing that improves learning. If the family
attended a Catholic school, then learning improved) circumstances are very different, then the politicians
and association (those attending Catholic schools per- are making a big mistake. Unfortunately, politicians
formed better than those attending public school). are rarely trained in social research and most make
Social researchers know this is not enough informa- quick, high-publicity decisions without the careful
tion. They first try to eliminate alternative explanations reasoning or the patience for precise empirical inves-
and then try to understand the causal mechanism (i.e., tigation. Fortunately, sociologist James S. Coleman
what happens in Catholic schools that helps students and others have studied this issue (see Coleman and
learn more). For example, the politicians failed to Hoffer, 1987).

causal relations—that is, a mutual causal relation- (represented by lighter and darker shaded figures)
ship or simultaneous causality. For example, study- and income level? Some people mistake association
ing a lot can cause a student to get good grades, but for true causality. For example, when I was in col-
getting good grades also motivates the student lege, I got high grades on the exams I took on Fridays
to continue to study. Theories often have recipro- but low grades on those I took on Mondays. Thus,
cal or feedback relationships, but these are difficult an association existed between the day of the week
to test. Some researchers call unidirectional rela- and the exam grade. This association did not mean
tions nonrecursive and reciprocal-effect relations that the day of the week caused the exam grade.
recursive.5 Instead, the reason for the association was that I
2. An association means that two phenomena worked 20 hours each weekend and was very tired
occur together in a patterned way or appear to act on Mondays. If you cannot find an association, a
together. People often confuse the word correlation causal relationship is very unlikely. This is why you
with association. Correlation has a specific techni- want to find correlations and other measures of asso-
cal meaning and there are certain statistical require- ciation. Yet just because you find an association does
ments for it. Association is the more general idea. not mean you have causality. It is a necessary but not
The correlation coefficient is a statistical measure a sufficient condition. In other words, you need it for
that indicates the strength of association, but there causality, but it is not enough alone.
are other ways to measure an association. Some-
times researchers call association concomitant vari-
Association The co-occurrence of two events, char-
ation because two variables vary together. Figure 2 acteristics, or factors so that when one happens or
depicts 38 people from a lower-income neighbor- is present, the other one is likely to happen or be
hood and 35 people from an upper-income neigh- present as well.
borhood. Can you see an association between race

75
 THEORY AND RESEARCH

Lower Income Upper Income

FIGURE 2 Association of Income and Race

To show causality, an association does not have we build controls into the study design itself to elim-
to be perfect (i.e., every time one variable is present, inate alternative causes and isolate the experimental
the other is also). In the example involving exam situation from the influence of all variables except
grades and days of the week, there is an association if the main causal variable. Nonexperimental research
on ten Fridays I got seven As, two Bs, and one C, eliminates alternatives by identifying possible alter-
whereas my exam grades on ten Mondays were six native causes and measuring them. This is common
Ds, two Cs, and two Bs. An association exists, but the in survey research. Once we measure potential alter-
days of the week and the exam grades are not perfectly natives, we use statistical techniques to learn whether
associated. The race and income-level association the causal variable or something else operates on the
shown in Figure 2 is also an imperfect association. effect variable.
3. Eliminating alternatives means that we must 4. Specifying the mechanism in a causal rela-
show that the effect is due to the causal variable, not tionship means that when we create a causal expla-
to something else. It is also called no spuriousness nation, we must have more than two variables that
because an apparent causal relationship that is actu- are correlated, which is “a satisfactory explanation
ally due to an alternative but unrecognized cause is requires that we also specify the social ‘cogs and
called a spurious relationship. While we can observe wheels’” (Hedstrom and Swedberg, 1998:7). We go
temporal order and associations, we cannot empiri- beyond saying that an independent and dependent
cally eliminate all logical alternatives. Eliminating variable are linked, as if the connection were through
possible alternatives is an ideal. This means we can a “black box” of unknown processes. A full causal
demonstrate this only indirectly or rule out the more explanation identifies a causal relationship and
obvious alternative explanations. In an experiment, specifies a causal mechanism.
Let us say we find a strong association between
a person’s social class and her health. We may state
Causal mechanism The part of a causal explanation our “theory” as high-class people live longer and
that specifies the process by which the primary inde-
pendent variable(s) influence the primary dependent
get sick less often than low-class people. However,
variable(s). it is not enough to say that a person’s social class
causes health outcomes. We must also explain why

76
 THEORY AND RESEARCH

and how social class does this. In short, we should Sometimes we state theories as a lawlike gen-
describe exactly what it is about social class that eralization: When X occurs, Y will occur. However,
makes the health outcomes happen. We may believe such “theories” are not a full explanation (Elster
that higher class provides people with more social 1998). They need the causal mechanism. The mech-
resources (knowledge, social connections, leisure anism is often more specific than a general law, but
time, flexible schedule) that enables them to eat it is more general than a specific instance. In a full
healthy food, experience less stress, engage in phys- explanation, the mechanism may be an arrangement
ical exercise, and so forth, which produce better of opportunities or individual desires, which are
health. Social resources are the mechanism that con- more general than a particular opportunity or one
nects class and outcomes (resources include “being desire but less general than a lawlike statement.
in the know,” “knowing the right people,” and hav- Mechanisms add complexity. Instead of a simple law
ing access to opportunities). (if B then R), we find in specific situations that if B
Seeing the mechanism of a full causal expla- sometimes R but at other times P or D. The mecha-
nation may be difficult, especially in the natural sci- nism explains why B does not always cause R but
ences. We may posit unseen mechanisms among can create other outcomes. Perhaps we believe that
subatomic particles or off in distant galaxies to when economic conditions are bad (B), people rebel
explain what we can observe. As research advances, (R). However, as we study many specific situations,
we observe the outline of a mechanism whose exis- we find this is not always true. Sometimes people
tence we first only predicted in theory. Even if we rebel, but at other times they become passive and
cannot directly observe the mechanism now, we can accept their fate (P) and at still other times they fight
still describe how we think it operates. one another and become self-destructive (D). For a
We can use models of a process that we believe complete explanation, we must include the mecha-
connects inputs with outcomes to clarify mecha- nism that tells us when bad conditions produce each
nisms. In economics, the market is a common of the outcomes.
mechanism; it is a process of making exchanges 5. Outlining the causal chain is a process in
between independent buyers and sellers, each with evaluating each part of the chain. Here is an associ-
desires and resources. The market explains how the ation in a causal theory: A rise in unemployment
supply–demand relationship operates. In sociology, causes child abuse to increase. We want to explain
a commonly used mechanism is Merton’s self- these increases. We explain them as being caused by
fulfilling prophecy. A self-fulfilling prophecy a rise in unemployment. To “explain” increased child
occurs when a definition of a situation stimulates abuse, we must identify its primary cause, but a full
behavior that makes a false definition come true. A explanation also requires specifying how this hap-
“negative feedback” mechanism in a prophecy con- pens (i.e., identify a causal mechanism and put it in
nects people’s beliefs and behaviors at one point in a casual chain). The mechanism in this theory is the
time to later outcomes. A classic example of a self- situation of people losing their jobs. Once they lose
fulfilling prophecy is a run on a bank. A bank may their jobs, they feel a loss of self-worth and increased
be very financially stable, but a false rumor starts stress. As they lose self-worth and experience high
that it will fail. This new definition of the situation, stress, they are more easily frustrated and become
although inaccurate, causes many people to with- angry more quickly. Inner social control weakens,
draw their money quickly. As people withdraw and the pattern of living is disrupted. Highly frus-
large amounts of money, the bank weakens. The trated people with lower inner social control may
weakened bank stimulates even more rumors of express their anger by directing violent acts toward
bank failure. The new rumors in turn stimulate those with whom they have close personal contact
more withdrawals. Eventually, accelerating fear (e.g., friends, spouse, children). This is especially
(false definition of the situation) and withdrawals true if they cannot direct their anger in actions
(behavior based on the definition) cause the bank against its source (e.g., an employer, government
to fail (the false definition becomes true). policy, or “economic forces”). The mechanism is

77
 THEORY AND RESEARCH

part of a larger process or causal chain, and it occurs Positive and Negative Causal Relationships.
after the initial cause (unemployment) and before Causal relationships can be positive or negative.
the effect (child abuse). Many people imply a positive relationship between
We can test each part of a causal chain. In addi- the cause and effect variables if they say nothing.
tion to determining whether unemployment rates A positive relationship means that a higher value
and child abuse occur together, we can consider on the cause goes with a higher value on the effect
whether unemployment increases frustration, and or outcome. For example, as the number of years of
frustrated people become violent toward family a person’s schooling increases, the longer the per-
members. A typical research strategy is to divide the son’s life expectancy is. A negative relationship
causal chain into its parts and then to evaluate each means that a higher value on the cause goes with a
part of the chain against the data. lower value on the effect or outcome. For example,
as the number of years of a person’s schooling
Diagrams of Causal Relations. We can express increases, his or her bigotry and prejudice decreases.
causal relationships and theories using words, pic- In diagrams, a plus sign (!) signifies a positive rela-
tures, or both. We often present diagrams of the tionship and a negative sign (–) signifies a negative
causal relations to provide a simple picture of a rela- relationship. Figure 3 presents some samples of
tionship. This makes it easier for others to see the relationships that can be diagrammed. Researchers
causal relation quickly at a glance. Such symbolic would not use a diagram for a very simple relation-
representations supplement verbal descriptions and ship like the one in Figure 3(a) but find it helpful as
are shorthand for conveying complex information. they increase the number and complexity of causal
The simplest diagram is a two-variable model relationships.
as the one in Figure 3(a). We represent variables At times, the impact of a cause on an outcome
using letters, circles, or boxes. The convention is to is mediated or conditioned. This means that the
represent a cause by an X and the effect by a Y. The cause operates under some conditions but not
arrow shows the direction of causality (e.g., from others. For example, early marriage causes divorce
cause to effect). Sometimes we use subscripts when in modern societies that permit individual freedoms
there is more than one cause (e.g., X1, X2). We sym- and allow for legal divorce but not in highly tradi-
bolize relationships by lines with directional arrows. tional societies. A third factor that mediates the
Causal relations are represented by straight lines. basic cause-effect relationship is diagrammed as a
The convention is to use curved lines with arrows on third line with an arrow that intersects the line with
both ends to show an association that does not imply an arrow between the cause and effect (see Example
that a causal relationship goes in one direction. Box 4, Explaining Racial Conflict).

Structural Explanation. In a causal explanation,

Positive relationship An association between two
one or more factors may cause a response in other
concepts or measures so that as one increases, the factors. This is like one ball that rolls and hits others,
other also increases, or when one is present, the other causing them to begin rolling. In contrast, the logic
is also present. of a structural explanation locates a social pro-
Negative relationship An association between two cess, event, or factor within a larger structure. The
concepts or measures so that as one increases, the structure is like a spiderweb, a wheel with spokes,
other decreases, or when one is present, the other is or a machine with interconnected parts. A structural
absent.
explanation explains social life by noting how one
Structural explanation A type of theoretical expla- part fits within the larger structure. A causal expla-
nation about why events occur and how things work
expressed by outlining an overall structure and empha-
sizing locations, interdependences, distances, or rela-
tions among positions in that structure.

78
 THEORY AND RESEARCH

a.
+
Stress Divorce

b.
Stress +
Divorce
–
Resources

c.
+ +
Stress Fighting Divorce

Divorce
d. +
Stress –
–
Well-Adjusted
Children

e.
Stress Divorce
+ +
– Fighting –
– –
Well-Adjusted
Resources
Children

Explanation of relationship in each diagram

a. Level of stress (financial, social, emotional, etc.) is positively associated with the likelihood that a couple will
divorce.
b. Level of stress is positively associated with the likelihood that a couple will divorce, but the amount of resources
(financial, social, emotional, etc.) they possess is negatively associated with it.
c. Level of stress is positively associated with the frequency of fighting by a couple, which is associated with
the likelihood that the couple will divorce.
d. Level of stress is positively associated with the likelihood that a couple will divorce and negatively associated
with the likelihood that the couple will have emotionally well-adjusted children. In addition, the divorce pro-
cess itself has a negative effect on the emotional adjustment of children.
e. Level of stress and amount of resources are negatively associated with each other (i.e., people who tend
to have many resources are less likely to experience or better able to deal with stress). Level of stress is
positively associated with the frequency of fighting by a couple, but the amount of resources is negatively
associated with it. Amount of fighting is positively associated with the likelihood that a couple will divorce.
Both fighting and the divorce itself are negatively associated with the likelihood that the couple will have
emotionally well-adjusted children.

FIGURE 3 Causal Diagrams

79