Consider the general linear model, y = Xbeta+e. This model may be of full or less than full

rank.

(a) Explain the purpose of the quantile-quantile diagnostic plot of standardised residuals of a linear model.

(b) If you have a model with one design variable where you know that there is an exponential relationship between the response and design variables, what transforma

tion(s) should you use? (c) [2 marks] Explain how using Akaike's Information Criterion as a goodness-of-fit measure

prevents overfitting.

(d) Compare and contrast the full rank and less than full rank linear models in relation to design variable types and estimability.

(e) On data with two confounding factors, under what circumstances should one

prefer a complete block design to a Latin square design, and vice versa?