
Advanced Statistics Advanced Probability and Statistics Math Forum 
 LinkBack  Thread Tools  Display Modes 
May 13th, 2016, 01:01 PM  #1 
Senior Member Joined: Feb 2015 From: london Posts: 121 Thanks: 0  Markov Chains
There are 3 test, J,K,L. One of these tests is taken each day. A test is randomly chosen each day, and it can either have a good (G) or bad (B) score. The probability that the score is good given a test is P(J) = 3/10 P(K) = 3/5 P(L) = 9/10 If the test score is bad, we will always use test J tomorrow. If the test score is good, the test tomorrow could be either J,K or L, each with 1/3 probability. There are two questions:  1. Calculate approx proportion of total number of days, the results are good if this testing procedure has been running for a very long time. I calculated this to be 3/7 by solving for $\displaystyle \pi_{G}$ using the following: $\displaystyle [\pi_{G} \pi_{B}] = [\pi_{G} \pi_{B}] [\begin{smallmatrix} 0.6&0.4\\ 0.3&0.7 \end{smallmatrix}] $ 2. The testing procedure has been running for a very long time, calculate the proportions of time in which tests J, K, L were used. I dont really know where to begin for this question, I cant see any example in my notes. There is a hint in the question which is the following: Consider the fact that for i ={ J,K,L} P ( Test i used on day n) = P (Test i used on day n and results are G on day n1) + P (Test i used on day n and results are B on day n1) Then apply appropriate conditioning and limiting operations Any ideas? 
May 13th, 2016, 06:54 PM  #2 
Senior Member Joined: Oct 2013 From: New York, USA Posts: 639 Thanks: 85 
Part of the question is circular because the good or bad score influences what the next test is, but without knowing what percent of the time each test was, given, how can you calculator the probability of a good score? Test J is most likely, and Test K and Test L are equally likely. Therefore I'm going to start with made up percentages of 50% for Test J, 25% for Test K, and 25% for Test L. This results in a probability of a good score being 0.525 (52.5%). That results in Test J being given 65% of the time, Test K being given 17.5% of the game, and Test L being given 17.5% of the time. That completes iteration 1 (is that term mathematically appropriate here?). I'm not going to show my arithmetic, but I'll state my formulas. When the percent of good scores is known, call it the variable a, and make a variable b that = 1  a. The probability of Test J being given is b + a/3. The probabilities for Test K and Test L are each a/3. The probability of a good score when the percent of the time each test is given is known = P(J)*probability of Test J being given + P(K)*probability of Test K being given + P(L)*probability of Test L being given. For iteration 2, use the 65%, 17.5%, and 17.5% probabilities to calculate the probability of a good score. This time that probability is .4575 (45.75%). That results in Test J being given 69.5% of the time, Test K being given 15.25% of the game, and Test L being given 15.25% of the time. If you're correct and the probability of a good score is 3/7 (42.9% to the nearest tenth of a percent), doing iterations will make the probability eventually get very close to 42.9% and not change much. I did 2 iterations, and my guess is that fewer than 10 will be necessary before they produce very close results. Some people could probability write a computer program to do this quickly. I don't like or remember matrices, so I can't comment on your 3/7 probability.

May 14th, 2016, 06:23 AM  #3 
Senior Member Joined: Oct 2013 From: New York, USA Posts: 639 Thanks: 85 
I did more work: Iteration 3: P(good score) = .43725 P(next test is J) = .7085 P(next test is K) = .14575 P(next test is L) = .14575 Iteration 4: P(good score) = .431175 P(next test is J) = .71255 P(next test is K) = .143725 P(next test is L) = .143725 Iteration 5: P(good score) = .4293525 P(next test is J) = .713765 P(next test is K) = .1431175 P(next test is L) = .1431175 Iteration 6: P(good score) = .42880575 P(next test is J) = .7141925 P(next test is K) = .14293525 P(next test is L) = .14293525 I'll stop here. It seems like P(good score) = 3/7, P(next test is J) = 5/7, P(next test is K) = 1/7, and P(next test is L) = 1/7. Given P(good score) = 3/7, it's easy to know that P(bad score) = 4/7. P(next test is J) = 4/7 + 1/7. I should have thought of that last night. 
May 14th, 2016, 08:50 AM  #4 
Senior Member Joined: Feb 2015 From: london Posts: 121 Thanks: 0 
Thanks EvanJ for all the info. So if I understand your second reply correctly, we don't need to use iterations to calculate the proportion of time that each test is used if we know what the P(good score) is ( which we do). We can instead just use P(next test is X) = P(Bad score) * P(next test is X) + P(Good score)*P(next test is X) i.e P(J) = 4/7 * 1 + 3/7 * 1/3 = 5/7 P(K) = 4/7 * 0 + 3/7 * 1/3 = 1/7 P(L) = 4/7 * 0 + 3/7 * 1/3 = 1/7 To check this result we can add up 5/7 + 1/7 + 1/7 = 1. 
May 14th, 2016, 04:06 PM  #5  
Senior Member Joined: Oct 2013 From: New York, USA Posts: 639 Thanks: 85  Quote:
 
May 27th, 2016, 01:32 AM  #6 
Senior Member Joined: Aug 2012 Posts: 229 Thanks: 3 
Hey calypso. Just out of curiosity for these "long time" questions  have you covered the nature of a stationary distribution and the theory behind obtaining it? 

Tags 
chains, markov 
Thread Tools  
Display Modes  

Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Markov Chains  inequality  Algebra  11  November 26th, 2009 03:12 AM 
Help about markov chains?  aldors  Advanced Statistics  1  November 17th, 2009 09:22 AM 
Markov chains  Katherine  Advanced Statistics  2  May 27th, 2009 01:50 PM 
markov chains  aptx4869  Advanced Statistics  0  April 30th, 2007 04:57 PM 
Help about markov chains?  aldors  Algebra  0  December 31st, 1969 04:00 PM 