3.1 Why transform?

Upside: make life easy both practically (problems may evaporate, e.g. outliers become less severe) and theoretically (normal theory results, t-tests, p-values are credible)

Downside: may be hard to interpret

Rationale:

Symmetry - ``middle'' well defined

Easier to compare with normal (ie heavy tailed).

Methodology may require symmetry (ie normal theory)

Facilitates comparisons between observations that are on the same scale but far apart, (ie changes in Microsoft sales and changes in Apple's).

May be more interpretable - aid in decision making. Unit costs rather than total costs.

May put data onto a more useful scale, ie transform proportions with a logit transform.

Can make comparisons easier by stabilizing variance

Can transform to obtain additivity (ie Cobb-Douglas)

Interaction may only be present due to modeling on the wrong scale, so that transformation erases the need for interaction.