september 2012

So how did you find the ct3 exams ? I think it had some different parts ie not many probabilities calculations and not many marks on regression

 

i found it 80% ok, but there were one or two I tripped up on....

The convolution question i couldn't get and the one where you had to explain the difference in intervals for a 100sample and a 30sample data batch. I assumed the first was normal the latter was t-distribution. But no idea how they changed when she removed an outlier.

Could go either way.

I've noticed they are changing it about a bit in more recent exams?

 

I think the prescence or lack of an outlier affects the width of the interval.

 

well what I believe is that removing an outlier would result in a bigger confidence interval as the presence of an outlier gives a fake smallel interval i guess but don't know if that is correct... it can be the other way

 

Without doing the maths, presence of an outlier always gives wider interval. This is because an outlier makes the variance look big and thus larger width of the CI.

So, removing outliers should give you a shorter interval. In fact this is one of the key reasons why you would want to remove an outlier so that you can get a more precise estimate of the range where the true value of the unknown parameter is likely to lie.

 

thanks guys, but you really need to have seen the question to know what I mean.

it was quite detailed and involved the person noticing an outlier and then hand picking another result to use as a sustitute - unfortunately i can't remember the whole question.

 

yes I think bapan is correct. the question mentions that he replaces the outlier with a value that is not an outlier hence the variance is smaller

 

Convolution Question

Hi,
Any suggestions to the solution of the convolution question?
This is what I did in the exam

By using the formula in core reading, I ended up getting 1 as the solution i.e Fx(infinity)...since the U(0,1) p.d.f would be 1, the pdf of z would be the integral over p.d.f of X which would be 1.....

Any other possible solutions to the question?

 

I think the solutions are up now, aren't they?

I haven't plucked up the courage to look yet - I'll only notice that all the ones I thought I got right were actually wrong! Think i'll wait for results before I check.

 

Nope they will be out only when the results are declared

 

The limits of integral change under a convolution. So, the answer can't be 1 !!

Anyway, I thought of giving it a shot after looking up the question from the website.

Has anyone got an answer PHI(y) - PHI(y-1) where PHI is the CDF of N(0,1) and Y = U + X?

 

Yes!! I was somewhere near to that. I definately didn't get it completely right as I ran out of time and I literally scrawled an answer as quick as I could, but might get a couple of method marks!

 

Damn!!! You are right about the change of limits and I think PHI(y) - PHI(y-1) would be the right answer...
Whoo I hope this doesnt affect my result....