A few people are still deeply mystified as to why you do a 2 group test on the period 
differences to assess the treatment effect (and vice versa). First step is to remember 
that this is the way it is because it is a 'cross-over' trial. For an intuitive 
explanation consider that for group 1, the period difference (1-2) is treatment A 
result - Treatment B result). For group 2 the period difference is 
Treatment B res - Treatment A res; if there is no substantial difference 
between A and B then it won't matter if you take A from B or B from A, 
i.e. the period differences for the two groups will be much the same, 
hence it can be assessed by a two-sample t-test. Likewise, 
treatment A - tratment B invovles period 1- period 2 for group 1 and vv for group 2.

With cross-over designs it is always sensible to draw a diagram even if it is not specifically asked for. Visuall assessment of carryover is important sinc ethe available statistical test severely lacks power because it is an intersubject comparison whilst the otehr tests for period and treatment are intra-subject. If a diagram is specifically asked for then credit is reserved for title, labels and 'neatness' --- this credit would be lest without these elements.

A common fault (especially amongst the UGs) was not give the answer in the context of the question, e.g. saying 'Not significant'. or ''p>0.05', or 'treatment not significant' [and nothing else], instead of a proper answer such as  “no worthwhile evidence that the treatment affects the rate of urea synthesis” or “There is weak evidence that the new treatment increases the rate of urea production compared with the standard”.   Failure to give an answer in these sort of terms loses substantial credit (typically more than 10% of the available marks for the question).

I expected people to use the SPSS output to give the relevant p-values etc, 
rather than calculate them --- impressed by those who typed out the full 
details of the calculations though . If you're given computer output 
in an exam then you are allowed to use it --- spending time 
recalculating standard deviations from raw values 
etc is a catastrophic waste.

A p-value of 0.059 seems to me to be not bad evidence, 
so need to say at least somethng like 'some evidence' 
rather than a bald 'Not Significant', (what's 0.009 
between friends?).

I did not expect people to use logistic regression 
to do question 2 --- especially not Minitab, 
MH chi-squared tests are fine but ok to do 'for practice'.

I've already commented re relative risks and odds ratios.

I liked the exact interpretation of Q3 by somebody who 
said that there is clear evidence that the cream 
reduces the odds of complaining about pain.

For those who attempted question 4, especially the final part, it
is necessary to review the answer in common sense terms. Answers 
given of a few thousand or indeed of a few tens of thousands are 
clearly implausible --- and it was emphasised that the question
was based on a real experiment. Incidentally, it was astonishng
that so few people had traced the original article --- googling
spoons bmj takes you directly to it, clues provided in the question.
Especially noting that many of you (MMAth and MSAc) had spent the 
previous few days  practising searching medical literature for the 
assessed assignment.





Nick