a confounding factor. A final difficulty in this area is that often you
o can't randomize who is in a class. The solution is to use the classroom
as a unit.
When performing statistical tests Campbell and Stanley say that
both assignment to groups and sampling must be random. It is acceptable
to do a t-test on gain score or use pretest scores as a covariate.

The second design is the Solomon Four-Group design diagrammed below:

ROO, X MO,

RO O; Cy
K X Os
K O,

I£ 032201, 0220,, 05206, 05203 inference is much stronger.(A 2x2 ANOVA can
be used for analyzing this design). The main effect of testing and the
interaction of testing and X are both determinable, However other problems

remain and this method reguires a high expenditure of effort that may not

be worth it.
e Design 6 is shown below:
X O,
R O, |

This design recognižes that a pretest is not essential,  Sometimes it is

e hard to admin ster er have am ofuke

recommended that this design be used as a pretest would. However where
pretest scores are already available the first design should be used
since the statistical tests for it are more powerful, 'The availability
of pretest scores also makes the examination of the interaction of X and
the pretest ability level possible. For statistical analysis the t-test
is often used,

If there are two or more treatment variables, each at several levels
this complicates the analysis.  Interaction effects can be very important
especially for generalizations.  Sometimes the main effect is nonexistant
but interaction is very important. Nesting means that certain interactions
cannot be measured. Ex: teachers cannot be at all schools so teacher school

interaction cannot be measured,

A S k a kk ok A a ia PE VA ke,