Title: Statistical Control Applet
Project director: Carl W. Roberts
Abstract:
Understanding correlations and partial correlations requires that
students see these numbers as "transparent" (i.e., as windows
into specific patterns among their data). What better way to develop
this understanding than by having students create data that correspond
to specific sets of correlations? Trial and error is important here.
Students may think they know how men's and women's educations and
incomes may differ such that an overall positive association would
"disappear" when controlling for
gender. Weakening faulty presuppositions thus becomes just as important
as confirming correct ones. Yet creating and analyzing data sets is
so time-consuming that in their frustrating pursuit of "the right
answer," students commonly lose sight of the ultimate objective
(namely, seeing statistics as windows into their data). The applet
developed in this project dramatically reduces the "busy work"
in this exercise, by automatically generating and analyzing data while
students are freed to concentrate on real-time creation of those data
patterns they imagine to correspond to specific correlations among
a set of variables.
Project description:
The applet's user interface has four parts: two over which the user
has control, and two in which information is displayed. User control
is over the placement of data points between two axes, and over the
selection of an inkwell (representing one of a control variable's
two levels). The information displayed consist of the data matrix
that correspond to the user-specified data points and four correlations
calculated from these data. The largest area is bounded at left and
bottom respectively with Y and X axes. Below and to the left of this
area are two "inkwells" of different colors. The user is
then able to place single data points on the area between the axes
by "dipping" the computer's pointer into an inkwell and
placing a dot-of-that-color onto the area. The vertical axis (Y) should
represent to the student the range of values taken by the dependent
variable (e.g., income); the horizontal axis (X) should represent
the range of values taken by the independent variable (namely, a measure,
like
education, thought of as having some causal influence on the dependent
variable). A third variable (a.k.a. a control variable) is introduced
via the inkwells. This inkwell-variable can represent gender (male
vs. female), race (African-American vs. white), or another 2-level
variable. To the right of the space between the axes, the user can
see the data matrix that is being built as data points are added between
the axes. Zeros in the 3rd column of the data matrix correspond to
data points associated with the upper inkwell; ones correspond to
those associated with the lower one. Finally, only once two dots of
each color are placed between the axes do four correlations appear
below the X axisthree
zero-order correlations (i.e., between each pair of variables) plus
the partial correlation between X and Y while controlling for the
inkwell-variable.
Applet URL:
http://statwebserver.stat.iastate.edu/croberts/applets/partialnp.html