Feedback to  those who submitted work for marking will be provided individually. This consists of a list of comonly occuring errors with an indication of which apply to you. A tick means you navigated that potential pitfall ok, a circle and exclamation marks indicates it is a point that needs attention. All of the erros were committed by at least one person.  Below is some general feedback, including expanded versions of the list of common errors. Any grade of gamma+  or under suggests it would be well worth sorting out things you do not understand sooner rather than later --- i.e. within the next week or so and not just before the examinations. Whilst I shall probably be around in the last couple of weeks before the exams you cannot guarantee on being able to book time to sort out things you have forgotten completely.

Now for the general feedback:-->>
In the first question a few people used a random selection of 3/4 and 1/4 for choosing A and B but this does not end up with 24 As and 8 Bs unless you are lucky. 
It could be that it must be exactly 24 As and 8 Bs (e.g. because the packets have already been delivered) so either use 6 blocks of size 4 or randomly permute a 
sequence of 24 As and 8 Bs (see solutions of how to do this in S+ with sample(.)).

Curiously some people did not actually provide a full list though this was specifically requested in the question (which means that in the context of an exam there would be marks specifically allocated to this which will be lost if not provided). Similar comments apply in 2(iii) and for any other items specifically requested in the question.

In the minimisation question several people succeeded in spreading this over several pages of detailed calculations, with a separate table for each of the thirteen steps. Only a one simple table is needed ---- you should appreciate that when the contract 
was costed only twenty minutes of your time was allowed for producing this and if you spend longer then you will not be earning your salary (and you'll run out of 
time in an exam).

Note that in the formulae for sample sizes phi-inv(beta) and phi-inv(alpha/2) are negative, even though they are squared in the formulae, not important maybe in the calculation of n (except it makes the equals sign a blatant lie) but it will give you the wrong answer if you try inverting the formula to get the power. 

In the sample size question a few people ignored that it asked for total number and so did not double the number calculated (after rounding up, never down note). 
Similarly if 40 to 60 in total are observed this means 20 to 30 in each group. 
Some voluntarily extended the question to allow for an arbitrary number of drop-outs (10% or 20% ....) --- this was not requested (nor costed in the contract .... ). 

A few people gave answers on length of pregnancy to five decimal places (when the units are weeks, i.e. apparently accurate to around 6 seconds). This is 
not sensible and looses credit. A couple of people fell into the subtle trap provided by 'twin pregnancies' --- agreed, twin studies often call for a 'paired t-test' --- i.e. a one-sample test on differences --- but here the 
fact that twins are involved is a red herring (but may be why the original paper from which this is taken was published in the first place since twins are 'inherently interesting'). Some people clearly did the question either 
'by hand' or using one or other package. It is worth making sure you can do both, so as to develop skills needed both for written exams and for real life (both might be needed in both situations).