My Math Forum Expected mean squared error

 June 18th, 2014, 04:05 AM #1 Newbie   Joined: Jun 2014 From: Italy Posts: 6 Thanks: 0 Expected mean squared error Could anyone help me to solve this exercise? Suppose the estimation errors e_i are independent and identically distributed random variables that takes values 1 and -1 with an equal probability. X is a uniform random variable that takes on 100 possible values (P(X=x)=1/100 , x=1,2,..100). Since x is a number that can take on different values with different probabilities, we should talk about the expected MSE, where the expectation E x is the averaging procedure over the probability distribution of x. Calculate E_AE and E_EA with the formulas below for N=100 and tell if the relationship between E_AE and E_EA hold. Please show me how to solve this exercise
June 18th, 2014, 12:17 PM   #2
Global Moderator

Joined: May 2007

Posts: 6,754
Thanks: 695

Quote:
 Calculate E_AE and E_EA with the formulas below for N=100 and tell if the relationship between E_AE and E_EA hold.
What relationship?

Since the e_i are independent and have mean 0, the only terms that survive in E_EA are the $\displaystyle e_i^2$ terms, so the expressions are almost the same, the difference being a factor of 1/N.

 June 18th, 2014, 01:55 PM #3 Newbie   Joined: Jun 2014 From: Italy Posts: 6 Thanks: 0 I try to explain better. We are talking about the problem of averaging a crowd. we say the number that a crowd of N people wants to guess is x, and each person in the crowd makes a guess yi : yi = x+ei (x) i.e. the true value plus some error ei ... then the book explains that we are interested to compare the following two quantities: 1.the average of expected mean-squared error (E_AE) 2.the expected mean-squared error of the average (E_EA) Since x is a number that can take on different values with different probabilities, we should talk about the expected MSE, where the expectation E x is the averaging procedure over the probability distribution of x. if the estimates yi are independent, we have E_EA=(1/N)E_AE, instead if the estimates are completely dependent then the averaged estimate is just the same as each estimate,so the error is the same and we have E_EA=E_AE
 June 19th, 2014, 01:35 PM #4 Global Moderator   Joined: May 2007 Posts: 6,754 Thanks: 695 Mathematically you seem to be correct. I have trouble understanding what point you are trying to make.
 June 20th, 2014, 12:52 PM #5 Newbie   Joined: Jun 2014 From: Italy Posts: 6 Thanks: 0 I simply would understand how to solve the exercise. Calculate E_AE and E_EA with the formulas below for N=100 considering the different probabilities of X. I don't know how to apply these data in the two formulas. At the end I think E_AE and E_EA should have a value. Which is?
 June 21st, 2014, 12:13 PM #6 Global Moderator   Joined: May 2007 Posts: 6,754 Thanks: 695 $\displaystyle E(e_i^2) = 1, E(e_ie_j) = 0$ for all i, and j ≠ i. Net result: $\displaystyle E_{AE}=1, E_{EA} = \frac{1}{N}$

 Tags error, expected, squared

### expected squared error

Click on a term to search for related topics.
 Thread Tools Display Modes Linear Mode

 Similar Threads Thread Thread Starter Forum Replies Last Post jkh1919 Calculus 5 August 4th, 2012 06:30 PM esther Calculus 1 October 25th, 2011 03:57 AM bignick79 Algebra 9 June 29th, 2010 12:59 PM Shocker Advanced Statistics 1 February 8th, 2010 04:47 AM Tuneman Advanced Statistics 4 March 18th, 2009 10:37 AM

 Contact - Home - Forums - Cryptocurrency Forum - Top