‘Statistical Inference -IT (Sequential Analysis) 1660
TER, and Ry denote the number of runs of ny objects of one type and n2
objects of another type in a sample of size m, + np then find the probability that
R, +R, =, for r even and r odd, and also‘the mean and variance of Ky + Kz
‘when all these m, + my observations arise from the same distibution.
Dethi Unie. [Link]. (Stat), 1991)
16. (Explain how the run test can be used to test randomness.
(i In the median test with samples of size 9 and 7 respectively, from two
populations find the probability density function of the random variable
representing the number of values of the samples from the first population in
{he tower halt of tne combinea sample. (Madras Unto, BrSe, 1988)
17. (a) The win-lose record of a certain basketball team for its last 50
‘consecutive games was as follows -—
WWWWWWLWWWWWWLWLWWWLLWWWW
LWWWLLWWWWWWLLWWLLLWWLWWW
‘Apply run test to test that sequence of wins and losses is random.
(®) Use an appropriate non-parametric test procedure to test for randomness
the following set ot 30 two-digit numbers :
15, 17, O1, 65, 69. 69, 58 41, 81, 16,
16, 20, 00, 84, 22, 28 26, 46, 66, 36,
86, 66, 17, 34, 49, 85, 45, 51, 40, 10.
18. At the beginning of the year a fist grade class was randomly divided
into two groups. One group was taught to read using a uniform method, where
all students progressed from one stage to the next at the same time, following
te teacher's direction. The second group was taught wo read using an individual
method, where, each student progressed at his own rate according 10 a
programmed work book, under supervision of the teacher. At the end of the year
each student was given a reading ability test with the following results:
First Grow Second Group
227 SS 184 202° 27163
116234447 14 151 284
25219488 16523553
149 247,81 in a7 228
16 92 22 99271
Use the Wald-Wolfowitz run-test to test for the equality of the distribution
functions ofthe two groups.
19. Using the number of runs above and below the median, test for
randomness te following set ofa table of 2-digit numbers:
15, 77, O1, 65, 69, 69, 58, 40, 81, 16, 16 20, 00, 84, 22,
28, 26, 46, 66, 36, 86, 66, 17, 43, 49, 85, 40, 51, 40, 10.
169. Sequential Analysis - Introduction We have seen that in
'Neyman-Pearson theory of testing hypothesis, n, the sample size is regarded as a
fixed constant and keeping a: fixed, we minimise B. But in the. sequential
analysis theory propounded by A. Wald n, the sample size is not fixed but is
regarded asa random variable whereas both cand B are fixed constants.
16-9:1. Sequential Probability Ratio Test (SPRT). The best
known procedure in sequential testing is the Sequential Probability Ratio Test
(GPRT) developed by A. Wald discussed below.1670 ‘Fundamentals of Mathematical Statistics
Suppove we want 10 test the hypothesis, Ho :0 = Oo against the
‘ternative hypothesis, Hj : 8 = ®,, fora distribution with pt ftx, ®) For any
Positive integer m, the likelihood function of a sample x3, 13, from the
Population with p.a.. f(x, 6) is given by
Lin = Tha 04) when H; is true,
andty Low = Lf 99 when iti tne,
and the likelihood ratio 2, is given by
Theo .
(m= 1,2.) (16-115)
‘The BPRT fur eating Mo agaist ff i Ueftwed ay flow»
‘At each stage of the experiment (at the mth tral for any integral value m),
the Hkelihood ratio Ay, (mt = 1,2, ..) is computed.
© 182,24, we semminate the process with the
rejection of Ho
(i IA, $B, we terminate the process with the
acceptance of Ho, aid elle
Gil) If B<2,
09
e
a-
= Ene
(i Accept Heit
S58 [pa -tieeg 80 2 ] she (#2)
(Oo + 0
= Ea sates eB gt P1200
and (iif) Continue taking additional observations as long as
mts) as on(t)
0.C. Function, First of all we shall determine h = h(0) #0, from
(16-122) iz,, from
» [te » a=
cee fix.) de= 1‘Statistical Inference -II (Soquantial Analysis) 1673
el labo
x G2e + Oot oo]f'a 1,[0n using (*))
= LJ” oid b2-2(@,-e9h+0)
al re 4 + 0:2 02)h}] dew
Ifwe take
fee)
A= (01-0) A+ }
AF = (0,2 - O92)h + 0?
which being the total area under normal probability curve with mean 2 and
variance ois always unity, as desired. Thus h = h(8) is the solution of (***)
and is given by
2-02) h+ & = (,- 89+ OP
PBA = (6,092 296, ~60A
‘Since h = h(8) #0 and 0, # Gp, on dividing throughout by (8, ~ 8)h, we
Pa
+8) =@-09 4428
= He) gage
Substituting for h(@) in (16-121) we get the required expression for the
0. tunton
‘ASIN function, We have
Za tog FEE Site [- 54] (From *)]
8; - 6
Be) =“bar* (28) - O~ 04]
0, ~ 94
= "og (20-0 -0,)
Substituting in (16-123), we get the required A.S.N. function.
Example 16-12. Let X have the distribution :
fe, (1 -0)'-452=20,1;0<0<1
For testing Ho : © + Q against Hy : 6 = @,, construct S.P.R.T. and obtain
its ASN. and O.C. functions.
Welhi Univ. [Link]. (Stat. Hons.), 1993, 1985]0% Fundamentals of Mathematical Stati
Solution. We have
LG Xpy ney Sq LH)
ial (CE ET)
fn ada
Ad a-o) | |
(2 :
%
Jog hy = E5105 0/8) + (m— som(tet)
$ @ 80) ~ 9
“ae v4 @a(t=6) a vor Tet
Hence SPRT for testing Ho: @ =O» against H, > @ =A. is given by
(ef. (6-119) :
sanuitirnsne( 185 ae
oe b—m log [(1- 0/1
tet Ens Seon opm sayfa
(i) Reject #7 (Accept Hy) it tog Ae 2 og 1=B. = a, (say)
it Bx, ote mon ld = 0/0 - 09)
Oe fog [6,(1 ~ 09)/09(1 -.8,)) = Fam (529).
Gi) Continue sampling if
DK On 0, then
1@.-») a (
=> ‘L(@,—h) =B*. (0, h)
(1 = 8,
cs &m) “id AI
son (2) ttytore Fundamentals of Mathematical Statisticg
Formulae (vl) and (vi) are very convenient .o use for obtaining the points
00 OC. curve for arbitrary negative values of h.
EXERCISE 16(4)
1.(@) Describe Wald's Sequential Probability Ra:io Test.
(®) Explain how the sequential test procedure viffers from the Neyman-
Pearson test procedure,
2, Define the OC function and ASN function in sequential analysis.
Derive their approximate exproesions for tho sequential probability ratio tact of
simple hypothesis against a simple alternative.
3. Describe Wald's SPR.T, Let X be a Bemoulli variate with pd.f.
‘fe 0) = OF(1 -0)!-#32 =, 1505051.
Etnploy [Link]. for testing Hy : @ = O against H, : 0 = 6,, and obtain its
ASN. and O.C. functions.
Dethi Univ. [Link]. (Stat. Hons), 1998, 85]
4. (a) Explain how the sequential test procedire differs frdm the Neyman-
Pearson test procedure,
Develop the S.P.R.T. for testing Ho: x= Ro against Hy : X= m;, based on
a random sample from a binomial population with parameters (n, x), m being
known, Obiain its O.C. and A.S.N. functions.
(©) Oban the sequential probability ratio test of the hypothesis Ha: © =
against Hy : © =} for the distributic
{ (1 0)!-4, forx = 0,1
“1 0 otherwise
- (Madras Univ. [Link], 1988]
5, Develop S.P.R. test for testing Ho: 0 = Oo against H,:0 = 0,
(©, > 0,), where 0 is the parameter of a Poisson distribution. Find approximate
‘expressions for OC function and ASN’ function of the test.
Wethi Univ. [Link]. (Stat. Hons.) 1988)
6 Describe SPR.T., its OC and ASN functions.
Construct S.P.R.T. for testing Ho: 0 6p against Hy :0 = 0, (0< 0) <0),
nthe basis of random sample dawn fom te Pareto distribution wit density
function :
Fe ;8)
W285, x2 as
‘Also obtain its 0.C. function and A.S.N. function,
Wethi Unio. [Link]. (Stat. Hons), 1989)
7. Explain how she sequential test procedure differs from the Neyman-
Pearson test procedure,
Develop the [Link].T, for testing Hy += 6p against Hy:
based on a random ssinpe of size n from a population with pdf.
= 68),