8/13/2012




              Confidence Interval of µ                                                                               Student’s t-distribution
                                                 (σ unknown)
                                        X −µ                                                                    • Distribution of a cont. symmetric random variable
        1 − α = P[−tα                            <
                                              < tα           ]
                                 2
                                        S
                                     , ( n −1)
                                                 2
                                                   , ( n −1)                                                      with range (- ∞,∞). Similar appearance to Z except
                                            n                                                                     fat-tailed
                                         S                                 S                                    • has a parameter, ν, known as degrees of
                 = P[ X − t α               < µ < X + tα                      ]
                            2
                              , ( n −1)   n                    2
                                                                 , ( n −1)  n                                     freedom(d.f.)      N (0,1)
  So, 100(1-α)% C.I. for µ is :                                                               S      Standard                                                   → t with the d . f .of the denominator
                                                         X        ± tα                    ×                                                         χ2
                                                                                                     error                                               df
                                                                           2
                                                                               ,( n −1)        n
                        pt. estimate
                                                                      table-value                               • If X1, X2,…,Xn are SRSWR from N(µ, σ2), then
                                                                                                                                                          (X − µ)
                                                                                                                                                              σ
  Valid for sampling from a Normal population;                                                                                      n ( X − µ)               n
                                                                                                                                               =                                   → t with ( n − 1) d . f .
  if sample size n small)                                                                                                               S        (n − 1) S 2
  If n is large, using z-value instead of t –value is ok                                                 5
                                                                                                                                                                   σ 2 ( n − 1)
                                                                                                                                                                                                                      6




     Chi-square (χ2)Distribution                                                                                                       Confidence Interval of σ2
• Distribution of a continuous (skewed) random variable with range                                                                                                         (n − 1) S 2
  (0,∞)                                                                                                                                1 − α = P[ χ 2 α                <                     2
                                                                                                                                                                                          < χα                ]
• has a parameter, ν, known as degrees of freedom(d.f.)
                                                                                                                                                      1− , n −1
                                                                                                                                                        2                     σ2                2
                                                                                                                                                                                                     , n −1


• mean = ν;           variance = 2 ν                                                                                                                (n − 1) S 2                    (n − 1) S 2
• If X1, X2,…,Xn are SRSWR from N(0, 1), then                                                                                                = P[         2
                                                                                                                                                                       <σ 2 <                          ]
                                                                                                                                                      χα                             χ2 α
                             2             2                 2         2                                                                                      , n −1                     1− , n −1
                        X + X + ⋯ + X → χ with n d . f .
                            1              2                 n                                                                                            2                                2
                                                                                                                                                                                                              2   2
                                                                                                                A 90% C.I. (2-sided) of σ2                                               ( 29×1557 , 29×.1708 )
                                                                                                                                                                                            42.
                                                                                                                                                                                                .46
                                                                                                                                                                                                      17
                                                                                                                                                                                                           .46
• If X1, X2,…,Xn are SRSWR from N(µ, σ2), then
                                 n

          (n − 1) S 2       ∑(X              i   − X )2                                                           Valid for sampling from a Normal population if n is small.
                        =    i =1
                                                          → χ 2 with (n − 1)d . f .
             σ   2
                                           σ     2
                                                                                                                  Example: LR 7.49* suppose we want C.I. for σ2
• Ex.: check that     S2    is unbiased for                      σ2   and find S.E. of         S2.                LR Section 11.5. (problem Sc11-7, Problems 11-39 to11-41, 11-45
                                                                                                                  (modify others)
                                                                                                         7                                                                                                            8
                                                                                                                  Aczel Sounderpandyan: Section 6.5




                                                                                                                                            Statistical Hypothesis
                      Confidence Interval
                                                                                                                                            Estimation and TOH
     • Of two sample mean, proportion, variance                                                                                                                                   Parameter

                                                                                                                                                                mean               proportion Standard
                                                                                      σ 12
                     µ1 − µ 2                        ,   π1 − π 2                ,       2
                                                                                                                                                                                              deviation
                                                                                      σ2                                                   Single
                                                                                                                       Population




                                                                                                                                           Two


                                                                                                                                           Multiple
                                                                                                         9




                                                                                                                                                                                                                                 1
8/13/2012




                   Clear-Tone Radios                                                                      Clear-Tone Radios:
                                                                                    H0: π ≤ 0.1 vs. H1: π > 0.1
 Shankar had to decide whether the proportion of defective                          C.R. is X ≥c (c possibly 3,2,1,0)
   items in a box exceeds 10% or not based on his                                   Compute the possible Prob(type I error) for various c
   inspection of 3 items randomly picked.                                                        c      P ( T y p e     I   e r r o r )

                                                                                                 3      .0 0 0 7

                                                                                                 2
We have settled the problems as:                                                                        .0 2 5 7

                                                                                                 1      .2 7 3 4

Shankar should accept the lot if
                                                                                   Exercise: Compute the Prob[Type II error] for the different C.R.’s.
no more than 1 radio is defective out of 3.
                                                                                   Note that it depends on the true π (defective proportion). So, to get
                                                                                   an idea plot the values for certain π > 0.1
                                                                                    Power of a test           =          1 - Probability[Type II error]
                                                                            11                                =          P[rejecting H0 when H1 is true] 12




                              Problem                                                         Basic elements of TOH
The telescope manufacturer wants its telescopes to have standard                    • Hypothesis -- assertion about a parameter the truth of
deviations in resolutions to be significantly below 2 when focusing                   which is to be inferred based on sample/relevant statistic
on objects 500 light-years away. When a new telescope is used to                    • Null hypothesis (H0) vs. Alternative Hypothesis (H1)
focus on an object 500 light-years away 30 times, the (sample)                      • Two kinds of error (Type I error and Type II error)
standard deviation turns out to be 1.46.
                                                                                                 D e c i s io n      D o not         R e je ct H 0
Should this telescope be sold?                                                                 T ru th             r eje c t H 0

                                                                                               H 0 is              C o rrec t        T ype I
                                                                                               co rrect            d e c is i o n    e rro r

                                                                                               H 1 is             T yp e II          C o rre ct
                                                                            13                 co rrect           erro r             d e c is i o n      14




 • The two errors are NOT typically equi-important and this
   decides which one should be the null hypothesis
 • A test is usually specified by fixing the critical region (CR)
   (you reject H0 if the sample falls here) or equivalently the
   acceptance region which is its complement.
 • Given a test (equivalently, a CR or the cutoff point), one
   would like to compute the probabilities of Type I (and II) error
   to gauge the consequences of wrong decisions
 • With n fixed, P(type I error) increases with decrease in P(type
   II error) and vice versa.
 • The usual approach is
    – Decide on acceptable level (called level of significance), α, for P(type I
      error)
    – determine the cutoff so that P(type I error) = α


                                                                            15




                                                                                                                                                                     2

Session 13

  • 1.
    8/13/2012 Confidence Interval of µ Student’s t-distribution (σ unknown) X −µ • Distribution of a cont. symmetric random variable 1 − α = P[−tα < < tα ] 2 S , ( n −1) 2 , ( n −1) with range (- ∞,∞). Similar appearance to Z except n fat-tailed S S • has a parameter, ν, known as degrees of = P[ X − t α < µ < X + tα ] 2 , ( n −1) n 2 , ( n −1) n freedom(d.f.) N (0,1) So, 100(1-α)% C.I. for µ is : S Standard → t with the d . f .of the denominator X ± tα × χ2 error df 2 ,( n −1) n pt. estimate table-value • If X1, X2,…,Xn are SRSWR from N(µ, σ2), then (X − µ) σ Valid for sampling from a Normal population; n ( X − µ) n = → t with ( n − 1) d . f . if sample size n small) S (n − 1) S 2 If n is large, using z-value instead of t –value is ok 5 σ 2 ( n − 1) 6 Chi-square (χ2)Distribution Confidence Interval of σ2 • Distribution of a continuous (skewed) random variable with range (n − 1) S 2 (0,∞) 1 − α = P[ χ 2 α < 2 < χα ] • has a parameter, ν, known as degrees of freedom(d.f.) 1− , n −1 2 σ2 2 , n −1 • mean = ν; variance = 2 ν (n − 1) S 2 (n − 1) S 2 • If X1, X2,…,Xn are SRSWR from N(0, 1), then = P[ 2 <σ 2 < ] χα χ2 α 2 2 2 2 , n −1 1− , n −1 X + X + ⋯ + X → χ with n d . f . 1 2 n 2 2 2 2 A 90% C.I. (2-sided) of σ2 ( 29×1557 , 29×.1708 ) 42. .46 17 .46 • If X1, X2,…,Xn are SRSWR from N(µ, σ2), then n (n − 1) S 2 ∑(X i − X )2 Valid for sampling from a Normal population if n is small. = i =1 → χ 2 with (n − 1)d . f . σ 2 σ 2 Example: LR 7.49* suppose we want C.I. for σ2 • Ex.: check that S2 is unbiased for σ2 and find S.E. of S2. LR Section 11.5. (problem Sc11-7, Problems 11-39 to11-41, 11-45 (modify others) 7 8 Aczel Sounderpandyan: Section 6.5 Statistical Hypothesis Confidence Interval Estimation and TOH • Of two sample mean, proportion, variance Parameter mean proportion Standard σ 12 µ1 − µ 2 , π1 − π 2 , 2 deviation σ2 Single Population Two Multiple 9 1
  • 2.
    8/13/2012 Clear-Tone Radios Clear-Tone Radios: H0: π ≤ 0.1 vs. H1: π > 0.1 Shankar had to decide whether the proportion of defective C.R. is X ≥c (c possibly 3,2,1,0) items in a box exceeds 10% or not based on his Compute the possible Prob(type I error) for various c inspection of 3 items randomly picked. c P ( T y p e I e r r o r ) 3 .0 0 0 7 2 We have settled the problems as: .0 2 5 7 1 .2 7 3 4 Shankar should accept the lot if Exercise: Compute the Prob[Type II error] for the different C.R.’s. no more than 1 radio is defective out of 3. Note that it depends on the true π (defective proportion). So, to get an idea plot the values for certain π > 0.1 Power of a test = 1 - Probability[Type II error] 11 = P[rejecting H0 when H1 is true] 12 Problem Basic elements of TOH The telescope manufacturer wants its telescopes to have standard • Hypothesis -- assertion about a parameter the truth of deviations in resolutions to be significantly below 2 when focusing which is to be inferred based on sample/relevant statistic on objects 500 light-years away. When a new telescope is used to • Null hypothesis (H0) vs. Alternative Hypothesis (H1) focus on an object 500 light-years away 30 times, the (sample) • Two kinds of error (Type I error and Type II error) standard deviation turns out to be 1.46. D e c i s io n D o not R e je ct H 0 Should this telescope be sold? T ru th r eje c t H 0 H 0 is C o rrec t T ype I co rrect d e c is i o n e rro r H 1 is T yp e II C o rre ct 13 co rrect erro r d e c is i o n 14 • The two errors are NOT typically equi-important and this decides which one should be the null hypothesis • A test is usually specified by fixing the critical region (CR) (you reject H0 if the sample falls here) or equivalently the acceptance region which is its complement. • Given a test (equivalently, a CR or the cutoff point), one would like to compute the probabilities of Type I (and II) error to gauge the consequences of wrong decisions • With n fixed, P(type I error) increases with decrease in P(type II error) and vice versa. • The usual approach is – Decide on acceptable level (called level of significance), α, for P(type I error) – determine the cutoff so that P(type I error) = α 15 2