Hi, 1) Answer to PQ 7.4(i) says that "Non-informative censoring is present since withdrawals give no information about the future mortality of lives remaining in the investigation". However Page 9 of the Chapter 7 notes says the opposite; Example of informative censoring include...withdrawal of life insurance policies because these are likely to be in better average health than those who don't withdraw. The latter seems most sensible and so presume the Answer to PQ 7.4(i) is a mistake? 2) Answer to Q7.6(iii) includes in the alternative solution on pg 57 that the Kaplan Meier Survival function is '14/15 * 11/13 * 10/11 for ['1 and 11/12'<=t <'4 and 4/12']. Should this be: '14/15 * 11/13 * 9/10 to recognise patient N who is censored at 9 months (in the alternative solution)?
Hi Thanks for pointing these out. 1) The question doesn't make it clear what "withdrawal" means in this sense, and it's possible to concoct scenarios where withdrawal is or isn't informative. Having discussed this with other tutors we think this question would benefit from being reworded. The point raised on page 9 of chapter 7 is however very important, that life insurance surrenders are likely to be informative, particularly when the main benefit offered by the product is a death benefit. 2) The original answer at t = [1+11/12, 4+4/12) has KM survival function 14/15 * 12/14 * 10/11. For the alternative answer, life N's censoring time moves from t=10.5 months to t=9 months, before the two deaths at t=10. This changes the second term in the survival function to 11/13, since we have one fewer observation at t=10, but this has no knock on to subsequent time periods, since all that matters is that the life has been censored before the period starts, and this hasn't changed. Thus the adjusted answer of 14/15 * 11/13 * 10/11 is correct. Hope this helps Dave
