Social Research Methods: Qualitative and Quantitative Approaches

ANALYSIS OF QUANTITATIVE DATA

MORE THAN TWO VARIABLES

Statistical Control

Demonstrating an association between two vari-
ables is an important first step for understanding
the data. However, it is not sufficient for you to say
that an independent variable causes a dependent
variable. In addition to temporal order and associa-
tion, we must eliminate alternative explanations that
can make the hypothesized relationship spurious.
Experimental researchers do this by choosing a
research design that physically controls potential
alternative explanations for results (i.e., that threaten
internal validity).
In nonexperimental research, we can statisti-
cally control for alternative explanations with con-
trol variables (discussed shortly). We examine the
control variables with multivariate tables and sta-
tistics that help us decide whether a bivariate rela-
tionship might be spurious. We can also show the
relative size of the effect of multiple independent
variables on a dependent variable.
A control variableis a third (or fourth or fifth)
variable that represents an alternative explanation
for a two-variable relationship. It is a “control” in
that is adjusts for, or takes into account, the effects
of variables other than the primary independent and
dependent variable of a hypothesis. For example,

your bivariate table shows that taller teenagers like

baseball more than shorter ones do. But the bivari-

ate relationship between height and attitude toward

baseball might be spurious. Why is this; because

you suspect that teenage males are taller than

females and you suspect that males like baseball

more than females do? To test whether the rela-

tionship is actually due to height, you must control

for gender. By controlling for gender, you are sta-

tistically removing their effect. Once you do this,

you can see whether the bivariate relationship

between height and attitude toward baseball

remains or whether the association between height

and baseball attitude was really due to gender.

You can “control for” a third variable by seeing

whether the bivariate relationship persists within

categories of the control variable. For example, you

control for gender, and the relationship between

height and baseball attitude persists. This means that

tall males and tall females both like baseball more

than short males and short females do. In other

words, the control variable has no effect. When this

is so, the bivariate relationship is not spurious, and

the control variable (suspected alternative explana-

tion) has no effect.

What if the bivariate relationship weakens or

disappears after you control for gender? It means

that tall males are no more likely than short males

to like baseball, and tall females are no more likely

to like baseball than short females. It indicates that

the initial bivariate relationship is spurious and sug-

gests that the third variable (in this case gender),

not height, is the true cause of differences in atti-

tudes toward baseball.

Control variable A “third” factor that shows whether

a bivariate relationship holds up to alternative expla-

nations; can occur before or between other variables.

SUMMARY OF MEASURES OF ASSOCIATION

Greek High
Measure Symbol Type of Data Association Independence

Lambda λ Nominal 1.0 0
Gamma γ Ordinal +1.0, –1.0 0
Tau (Kendall’s) τ Ordinal +1.0, –1.0 0
Rho ρ Interval, ratio +1.0, –1.0 0
Chi-square χ^2 Nominal, ordinal Infinity 0

TABLE 4 continued