Tranaformations

Question 1

Random sampling

Answer 1

Resample data

Question 2

Independent observations

Answer 2

Factor out non-independence by adding a new explanatory factor
‘Subject’

Question 3

Homogeneity of variance

Answer 3

Often violated by observations that cannot take negative values (length, weight, volume, counts etc)

Solution: log transformation of response (regression: of x as well)

Question 4

Norma distribution of resdidual variation

Answer 4

Violated bybobservations in form of percentaces or counts

Solution: arcsine root tranaformation of proportions or logictic regression on proportions

Square root trasformation on counts, use a glm whixh can assume poisson e

Question 5

Regression

Answer 5

1. when reaponse is exponential -> ln explanatory variable
2. when the variables are of different dimensions suggesting a power function -> logging both axes
3. respomse may saturate -> inverse them
4. response is cyclic -> use circular functions such as sin or cos