Hi, I think its now OK to post about exams. What do you all think about the CT4 paper? I think it was a good paper except a few odds. Main thing was that there wasn't any tricky Markov question.
Apr 2014 CT4 I thought it was a really long exam....I didn't finish the last question, just did the first part of it, and it was time over... Apart from that I thought most of the questions was fair and doable..How about you guys? what do you think?
I thought the bookwork questions (1-3) were ok for a change. Past exam papers sometimes asked really useless questions where you just have to memorize a lot of bullet points. This was not the case in this exam (except for the stochastic vs deterministic question maybe). I spent a lot of time (probably too much) on the question with the census data/exposure to risk. It was really tricky and I still don't have the full marks I'm sure. For the rest, the exam was doable, but too long. Due to time pressure I made some sloppy mistakes.
Yeah Piddox I think your not the only one to have fallen prey on the exposed to risk question...it took me a while to come out with the central expose to risk...then I got confused when the question was phrase as cal the rate last bday... Spent unnecessary amt of time on that question, but am not sure I'll get all the marks...
How did people go with the question that asked if there was a statistical difference between two mortality rates where both mortality rates were estimated from the data. I think the two rates were both mortality rates due to heart disease but one was from the healthy state and the other was from another state. How did any of you perform that test?
I thought that would of been a two-sample test, with a pooled variance. Problem is, the two variance i caculated for the problem were miles out. I screwed that one up tbh, like the rest, but assume that what it should of been
Yeah that's what I originally thought. But when comparing this statistic to a t-distribution, how many degrees of freedom should it have been on? We're essentially comparing the difference between two Poisson parameters but each of those parameters are estimate's from a sample size of one. Doesn't this then mean that we can't assume normality and thus can't use the 2 sample t-test? Another option would be to consider confidence intervals around the Poisson parameters with the variance given by the CRLB. But I'm not convinced that this was the right way to go around it.
I can't remember the question now, but wouldn't the degrees of freedom been given by the sample size - ie, the number of transitions of interest? And the variance would of been given by mu2/d or mu/v, where mu=d/v, which is the same as the CRLB anyway?
