1
play

1 Method of Moments Examples of Method of Moments 1 n - PDF document

Aug 05, 2023 •196 likes •240 views

What Are Parameters? Why Do We Care? In real world, dont know true parameters Consider some probability distributions: = p Ber(p) But, we do get to observe data Poi( l ) = l o E.g., number of times coin comes up

  1. What Are Parameters? Why Do We Care? • In real world, don’t know “true” parameters • Consider some probability distributions:  = p  Ber(p)  But, we do get to observe data  Poi( l )  = l o E.g., number of times coin comes up heads, lifetimes of disk  = (p 1 , p 2 , ..., p m )  Multinomial(p 1 , p 2 , ..., p m ) drives produced, number of visitors to web site per day, etc.  = (a, b)  Need to estimate model parameters from data  Uni(a, b)  “Estimator” is random variable estimating parameter  Normal( m ,  2 )  = ( m ,  2 ) • Want “point estimate” of parameter  Etc.  Single value for parameter as opposed to distribution • Call these “parametric models” • Estimate of parameters allows: • Given model, parameters yield actual distribution  Better understanding of process producing data  Usually refer to parameters of distribution as   Future predictions based on model  Note that  that can be a vector of parameters  Simulation of processes Recall Sample Mean Sampling Distribution • Consider n I.I.D. random variables X 1 , X 2 , ... X n • Note that sample mean X is random variable  X i have distribution F with E[X i ] = m and Var(X i ) =  2  “Sampling distribution of mean” is the distribution of the random variable X  We call sequence of X i a sample from distribution F  Central Limit Theorem tells us sampling distribution of n X    m  Recall sample mean: where X i E [ X ] X is approximately normal when sample size, n , is n  i 1 large  2   Recall variance of sample mean: Var ( X ) o Rule of thumb for “large” n : n > 30, but larger is better (> 100) n o Can use CLT to make inference about sample mean  Clearly, sample mean X is a random variable Confidence Interval for Mean Example of Confidence Interval • Consider I.I.D. random variables X 1 , X 2 , ... • Idle CPUs are the bane of our existence  X i have distribution F with E[X i ] = m and Var(X i ) =  2  Large (unnamed) company wants to estimate average  1 n 2  number of idle hours per CPU  2  n ( X X )  Let    X X Var ( X ) 2 i S i  n n n 1  1   225 computers are monitored for idle hours i i 1  For large n , 100(1 – a )% confidence interval is: 2     Say hrs., hrs 2 ., so hrs. 4 . 1 X 11 . 6 S 16 . 81 S   S S      Estimate m , mean idle hrs./CPU, with 90% conf. interval , X z X z a a / 2 / 2   n n a  a     0 . 10 , / 2 0 . 05 , ( z ) 0 . 95 , z 1 . 645 a a  z   a / 2 / 2 where ( ) 1 ( / 2 )   a / 2 S S     a  a     X z , X z 0 . 05 , / 2 0 . 025 , ( ) 0 . 975 , 1 . 96 o E.g.: z z a a a a  / 2 / 2  / 2 / 2 n n  Meaning: 100(1 – a )% of time that confidence interval is   4 . 1 4 . 1        computed from sample, true m would be in interval 11 . 6 1 . 645 , 11 . 6 1 . 645 11 . 15 , 12 . 05   225 225 o Not : or m is 100(1 – a )% likely to be in this particular interval  90% of time that such an interval computed, true m is in it X 1

  2. Method of Moments Examples of Method of Moments   1 n   • Recall: n- th moment of distribution for variable X: • Recall the sample mean: ˆ X X m E [ X ] i 1 n  i 1 m  n [ ] E X  This is method of moments estimator for E[X] n • Consider I.I.D. random variables X 1 , X 2 , ..., X n • Method of moments estimator for variance 1 n   X i have distribution F ˆ  2  Estimate second moment: m X 2 i n  1 n 1 n 1 n 1      i 2 2 ˆ  ˆ  2 ˆ  k Var ( ) [ ] ( [ ])  Let  X E X E X m X m X ... m X 1 i 2 i k i n n n    i 1 i 1 i 1   ˆ ˆ 2  Estimate: Var ( ) ( ) X m m ˆ m are called the “sample moments” 2 1   i n 2  2   ( X X ) 1 n 1 n 1 n            2 2 2 2 1 i o Estimates of the moments of distribution based on data X X X X i i i   n n n n    i 1 i 1 i 1 • Method of moments estimators  Recall sample variance:  Estimate model parameters by equating “true”  n     2 2 2 2 2 ( X X )  n ( X X )  n ( X 2 X X X ) n     i  ˆ  ˆ  ˆ 2 i i i i 1 ( ( ) 2 ) S m m moments to sample moments: m m     2 1 n 1 n 1 n 1 n 1 i i   i 1 i 1 Small Samples = Problems Estimator Bias ˆ    • What is difference between sample variance and • Bias of estimator: [ ] E MOM estimate for variance?  When bias = 0, we call the estimator “unbiased”  Imagine you have a sample of size n = 1  A biased estimator is not necessarily a bad thing 1 n   What is sample variance?   Sample mean is unbiased estimator X X i n  2    n ( X X ) i 1  2 i undefined S  2  n ( X X )  n 1   Sample variance is unbiased estimator 2 i  S i 1  n 1  I.e., don’t really know variability of data  i 1 n  1 S  MOM estimator of variance = is biased 2  What is MOM estimate of variance? n   1 n 2  2 2  2 o Asymptotically less biased as n   ( X X ) ( X X )     1 i 1 i i i i 0 1 n  For large n , either sample variance or MOM estimate  I.e., have complete certainty about distribution! of variance is fine. o There is no variance Estimator Consistency Method of Moments with Bernoulli ˆ • Estimator “consistent”: for e > 0     e  • Consider I.I.D. random variables X 1 , X 2 , ..., X n lim P (| | ) 1   n  X i ~ Ber( p )  As we get more data, estimate should deviate from true value by at most a small amount • Estimate p  This is actually known as “weak” consistency 1 n       ˆ ˆ p E [ X ] m X X p  Note similarity to weak law of large numbers: i 1 i n  i 1  m  e  lim (| | ) 0 P X  Can use estimate of p for X ~ Bin( n , p )   n  Equivalently:  If you know what n is, you don’t need to estimate that  m  e  lim (| | ) 1 P X   n  Establishes sample mean as consistent estimate for m  Generally, MOM estimates are consistent 2

  3. Method of Moments with Poisson Method of Moments with Normal • Consider I.I.D. random variables X 1 , X 2 , ..., X n • Consider I.I.D. random variables X 1 , X 2 , ..., X n  X i ~ Poi( l )  X i ~ N( m ,  2 ) • Estimate l • Estimate m 1 n 1 n  ˆ  l      l m      m ˆ ˆ ˆ [ ] [ ] E X m X X E X m X X 1 1 i i i i n n   1 1 i i  But note that for Poisson, l = Var(X i ) as well! • Now estimate  2  Could also use method of moments to estimate:   m  2 ˆ ˆ 2 ( ) m 2 1   n 2  2  ( ) n 2  X X   2 n n n ( X X ) ˆ 1  1  1  l  2  2  ˆ  ˆ 2  i  l [ ] [ ] ( ) i 1   2   m ˆ  2    i E X E X m m 2 2 i 1 X X X 1 i 2 1  i  i n n n n n    i 1 i 1 i 1  Usually, use first moment estimate  More generally, use the one that’s easiest to compute Method of Moments with Uniform • Consider I.I.D. random variables X 1 , X 2 , ..., X n  X i ~ Uni( a , b )  Estimate mean: 1 n  m  ˆ   m ˆ m X 1 i n  i 1  Estimate variance:   n  2 2 ( X X )       2 ˆ ˆ 2 i ˆ 2 ( ) i 1 m m 2 1 n   2 a b ( b a ) m    2  For Uni( a , b ), know that: and 2 12  Solve (two equations, two unknowns): o Set b = 2 m – a , substitute into formula for  2 and solve: ˆ ˆ    ˆ    ˆ 3 and 3 a X b X 3

Recommend

More recommend