## PaGE 5.0 Documentation - the "study design"

The study design is set when you choose **File New Analysis Type**

### 1 Channel Data

There are three ways to compare two conditions using microarrays.
The most straightforward is to use 1-channel^{1}
arrays and to hybridize some number of
replicate arrays for each condition. This gives a number of intensities
for each array element, in each condition. We refer to this design as a 2-sample design.
### 2 Channel Data

A second strategy, known as
the *reference design*, uses 2-channel data, where the
experimental conditions are hybridized in one channel and a common
reference is hybridized to the other channel. This common reference is
identical for all arrays. In this case the data consist of ratios (or log ratios),
where the denominator is always the gene's intensity in the reference
sample, and the numerator is one or the other experimental condition.
As above, we refer to this design as
a 2-sample design, because it produces two sets of intensities for array element,
corresponding to the two conditions.

The third
possibility is the *direct comparison* design.
2-channel arrays are used, with condition 0 being
hybridized to one channel and condition 1
hybridized to the other. Since we do not separate the channels,
but instead work directly with the ratios,
this requires an analysis which is somewhat different from the 1-channel or reference
design.

### Paired Data

A 2-sample design can also be "paired".
Suppose for example that two anatomical regions are compared in each of $m$
animals. It might be that the expression of gene $G$ in the first region is always
twice as high as the expression in the second, but the exact values
depend strongly on the animal.
In this case it is often better to run a paired analysis. This approach
considers the data as ratios of paired experiments (or differences if
the data are log transformed), analagous to the direct comparison design in which
the experiments are naturally paired.
Therefore this can increase the power of the results if there
is a strong effect in the data. If there is not a strong paired effect, then it will
likely decrease the power to run a paired analysis.

1. We consider Affymetrix arrays as 1-channel.