Professional Documents
Culture Documents
Statistics 2
Statistics 2
TEST OF
14 SIGNIFICANCE
(14.1 INTRODUCTION
The object of sampling is to study the features of the population on the basis of
sample observations. Acarefully selection sample is expected to reveal thesefeatres,
and hence we shall infer about the population from a statisticalI analysis ofi the
This process is known as Statistical lInference. sample.
There are two types of problems. Firstly, we may have no iinformation at allabou
some characteristics of the population,especially the values of the parameters involve
in the distribution, and it is required to obtain estimates of these parameters. This i
the problem of Estimation. Secondly, some information or hypothetical values of he
parametersmay be available, and it is required to test how far the hypothesis is tenable
inthe light of the information provided by the sample. This is the problem of Test of
Hypothesis or Test of Significance.
Sample variance
(S)= J-)? (14.3.3)
(14.3.6)
(n - 1)
becausNotee the distinction betweenE(s²)S and= o2s in which only he denominators are different.
(14.3.7)
Sis the variance of the sample observations, but s is the 'unbiased estimator' of the
varvance (o in the population.
Example
with 14.1 Show that the sample mean based on a simple random sample
Cement (srswr) is an unbiased estinator of the population mean.
[C.U., B.A(Econ)'78|
Statistical Methods
546
ns
X995
30.8 30.8
Or,
i.e.. 1.52 S o's 31.1
20.3 0.99
If two independent random samples of sizes n, and n, are drawn from two
Normal populations with unknown means M Hy and variances o, o
respectively, then 2 follows F distribution with degrees of freedom
(n, -1.n,- ). If F9ps and Fo2s denote the lower and the upper 2.5% points of
Fdistribution, we have with probability 95% the following inequalities
.025 (14.5.12)
The 95% confidence interval for ol¡; can be obtained from this as
< (14.5.13)
Fo25 F97s s5
where s and s; denote unbiased estimators of o, o respectively from the
two samples. (see Example 14.51)
14.6 THEORY OF TEST OF SIGNIFICANCE
statistical Hypothesis
In many practical problems, statisticians are called upon to make decisions about a
statistical population onthe basis of sample observations. For example, given a randon
sample, it may be required to decide whether the population, from which the sample
has been obtained, is a normal distribution with mean =40and s.d, =3. In
to reach such decisions, it is necessary to make attempi
certain
the characteristics of population, particularly about assumptions guesses
or
the
values of its parameters. Such an assumption or statementprobability
about distribution
the populationis
called a Statistical Ilypothesis. The validity of ahypothesis will beetested by analysing
the sample. The procedure which enables us to decide whether a certain hypothesisis
true or not, is called Test of
Significance or Test of Hypothesis.
Null Hypothesis and Alternative Hypothesis
In tests of significance, we start with a certain hypothesis about the Population
characteristics.This iscalled Null H, For
example, the null hypotheSIS may beHypothesis,
and denoted by the symbol
that the population mean is 40. We wrie
H,(u = 40)
Any hypothesis which differes from the null hypothesis is called Alternarive
Hypothesis, and is denoted by the symbol H,. The null hypothesis is tested against an
Estimation and Test of Sianifi cance 547
allernative hypothesis which in the above case, may be either that the population
not 40, or that it is greater than 40, or tat it is less than 40: i.e. any one of
meanis
H(u 40), H(u> 40), H,(u< 40)
Thesampleisthen analysed to decide whether to 'reject' or not to reject the null
hypothesis H. For this purpose, we choose asuitable statistic, called "Test Statistic"
andfind its
sampling distribution, assuming that H, is really true. The 'observed
the statistic in the sanmple will in general be different from the expected
value' of
because of sampling fluctuations. If the difference betwecn them is large, the
value the
hypothesis H, is rejected, and we doubt the validity of our assumption. If
null
difference is not large, HÍ is not rejected, and the difference may be considered to
havearisen solely due to
fluctuations of sampling. It is therefore necessary to decide
ndmuch of difference is tolerable before we are able to conclude that the null
bypothesis is acceptable.
.oHel o£ Significance and Critical Region
The decision aboutrejection or otherwise of the null hypothesis isbased on probability
considerations.Assuming the nullhypothesis to be true, we calculate the probability
of obtaining adiffernce equal lo or greater than the observed difference. If this
nrobability is found to be small, say less than .05, the conclusion is that the observed
valhe of the statistic is rather unusual and has arisen because the underlying assumption
ie. null hypothesis) is not true. We say that the observed difference is significant at
5per cent level, and hence the 'null hypothesis is rejected' at 5 per cent level of
significance. If, however, this probability is not very small, say more than .05, the
observed difference cannot be considered to be unusual and is attributed to sampling
fluctuations only. The difference is, now, said to be not significant at 5 per cent level.
and we conclude that there is no reason to reject the null hypothesis' at 5 per cent
levelof significance. It has become customary to use 5% and 1% levels of significance,
although othe levels, such as 2% or 0.5% may also be used.
Without actually goingto calculate this probability, the test of significance may be
simplified as follows. From the sampling distribution of the statistic, we find the
maximum difference which is exceeded in (say (5) per cent of cases. If the observed
difference is larger than this value, the null hypothesis is rejected. If it is less, there is
no reason to reject the null hypothesis.
distribution, Since
Suppose, the smapling distribution of the statistic is a normal
5%, the
Ihe area under normal curve outside the ordinates at mean t 1.96 (s.d.) is only
probability that the observed value of the statistic differs from the expected value of
be
I96 times the S.E. or more is .05; and the probability of a larger difference will
stillo smaller. If, therefore.
(Observed value) - (Expected value) (14.6.1)
Z=
S.E.
than l.96, the
S either greater than 1.96 or less than -1.96 (i.e. numerically greater
values z>1.96 or
ulnypothesis H, is rejected at 5% level of significance. The set of
S-1.96, i.e. lzl196
OISitutes what is called the Critical Region for the test. Similarly since the area
sde mean 2.58 (s.d.) is only 1%. H, is rejected at 1% level of significance, if
?numerically exceeds 2.58, i.e. the critical region is Izl22.58 at 1% level.
Statistical Methods
548
and is called
rejection value
of observed
Critical Value. true,
ofthe test statistic falling in the critical region is oftenacceptancelity
the called the
probabi
"Size of Critical Region".
Size of Critical Region S Level of Significance
However, for a continuous population, the critical region is so determined that its
Size equals the Level of Significance (a).
Two-tailed and 0ne-tailed Tests
Our discussions above were centred round testing the significance of 'differenoet
between the observed and expected values, i.e. whether the observed value ie
significantly different from (i.e. either larger or smaller than) the expected value as
could arise due to fluctuations of random sampling. In the illustration, the null
hypothesis is tested against "both-sided alternatives" (u> 40 or u<40), iLe.
H, (u =40) against H, (u 40)
Thus assuming H, to be true, we would be looking for large differences on both
sides of the expected value, i.e. in "both tails" of the distribution. Such tests are.
therefore, called wo-tailed tests".
Sometimes we are interested in tests for large differences on one side only i.e. in
one 'one tail' of the distribution. For example, whether a new process of
manufacture
produces bricks with a 'higher' breaking strength, or whether a change in the
production technique yields ´lower' percentage of defectives. These are known as
"one-tailed tests".
For testing the null hypothesis against "one-sided alte
(u> 40), i.e., rnatives (right side)
H, (u =40)
the calculated value of the statistic z is against H, (u> 40)
under the standardnormal curve lies to compared with L645. since 5% of the lca
the right of 1.645. If the observed val
zexceeds 1.645, the null
hypothesis Ho is rejected at 5%
level were used, we would replace 1.645 by 2.33. Thus the significance. Ifa1%
levelcritical
of regions for test at
5% and 1% levels are z
For testing the null 1.645 and z 2.33 respectively.
i.e. hypothesis against one-sided alternatives (left side)"(u<40)
the value of zis H, (u= 40) against H, (u
compared with < 40) -2.33
for significance at 1% -1.645 for
significance at 5% level, and with for
5% and 1% levels level. The critical regions are now 2zs-1.645 andlzs-2.33
respectively.
In fact, the sampling distributions of many of the commonly--used statistics canbe
approximated by normal distributions as the sample size increases, Soo thatthese rules
areapplicable in most cases when the
sample size is 'large', say. morethan 30.
Estimation and Test of Signifi cance 549
It is evident that the same null hypothesis may be tested against alternative
hypothesisof different types depending on the nature of the problem (see Table 14.1).
Correspondingly, the type of test and the critical region associated with each test will
alsobe different (Table
14.2).
mable 14.1 Formulation of Null Hypothesis and
ALternative Hypothesis
Problem Null Hypothesis Alternative
Hypothesis
Test whether u = 40, or
() uisdifferent from 40 H, (u = 40) H,(u 40)
(2) is more than 40 H, (u = 40) H{(u> 40)
(3) u is less than 40 H,(u= 40) H,(u< 40)
Table 14.2 Type of Test and Critical Region
RegionCr.
Critical Region
z2a2 or zS-za2
Za?
(2) One-tailed Test (3) One-tailed Test
uotd)
RegionCr. |
0 Za -Za
Critical Region Critical Region
zS Za(Left Tail)
z2Za (Right Tail)
Fig. 14.1 Critical Regions for TwO-tailed and One
tailed Tests
Statistical Methods
550
Type I error
Correct decision
0= Correct decision
Type llerror
the
we can measure in advance
Using the sampling distribution of the test statistic, rejeicd
probabilities of committingthe two types of errors. Since the null hypothesis is
only when the test statistic falls in the critical region,
Probability of Type I error itis truc
=Probability of rejecting H,(0= ,). whencriticalregion.
= Probability that the test statistic liesinthe
assuming 0= (a) ofthe
test.
The probability of Type lerror must not exceed the level of significance
Probabilityof Type I error s Level of Significance values ot 8
The probability of Type Il error assunies different Ivalues for ditterentcceptedonly
covered by the alternative hypothesis H,. Since the null hypothesis is
when the observed value of the test statistic lies outside the eritical region.
Probability of Type II error (when = ). false
= Probability of accepting H, (0= ,). when itis regionof
= Probability that the test stalistie lies in the
acceplance, assuming =0,.
Estimation and Test of Siqnificance 551
Itis Butthis is not possible, because for agivensample size an attempt to reduce
error.
of error is generally accompanied by an increase in the
other type of error.
of
one
lype
ofsignificance are designed so as to limit
the probability of Type I error to
tests
The (usually5% or 1%) and at the same time to minimise the probabil1ty
|value
specified
distribution.
a error. Note that when the population has a continuous
ofType error
Probability of Type I
= Level of significance
= Size of critical region
Test
Power of a
hypothesis H, (8= ) is accepted when the observed value of test statistic
The null
test procedure. Suppose that the
lies the critical region, as determined by the a specified alternative hypothesis
i.e.
Inie value of 0 is not ,. but another value 0,,
H, is not rejected, i.e. the test statistic
H«8=9,) is true. Type Ilerror is committed if
probability of Type II error is a function of
lies outside the critical region. Hence the true.
assumed to be
9..because now =, is
II error, when = , is true, the
If B(e,) denotes the probability of Type against the specified
the test
complementary probability 1-ße,) is called Power of
alternative H, (@= 0,).
Power = |- Probability of Type IIerror
= Probability of rejecting H, when His true
as possible for all critical regions
Obviously, we would like a test to be as 'powerful'
same size. Treated as a function of 0, the expression P(0) =l- B() is called
ofthe P(e)
Function of the test for , against 6. The curve obtained by plotting
Power
as Power Curve.
against all possible values of 0, is known
in a large lot is P. To test the null
Example 14.19 The fraction of defective items
numberfof defectives in a sample of8
hypothesis H,: P= 0.2, one considers the the hypothesis otherwise. What
hypothesis iff s6, and rejects
ems and accepts the test? Whau is the probability of type Il
error
probability of type I error of this
S Ihe [W.B.H.S., "79]
Corresponding to P = 0.1? not.
test whether the fraction (P) of defectives in the lot is 0.2 or
uClOn We are going to hypothesis H,(P 0.2). The test procedure is
as
ulhypothesis H,(P =0.2).
Alternative
follows: lot.
(1) Take a random sample of 8 items fromthe
in the sample.
14) Count the number of defectives found in the sample is 6or less (fs 6),
the number of defectives
9) Accept H, if of defectives in the lot may
be ),.2.
and conclude that the fraction
the number of defectives actually obtained in the sampie is 7 or
Reject H if 0.2: i.e.
Conclusion: The fraction of defectivesin the lot is not
8.
H, is not tenable.
552 Statistical Methods
8)Critical Region. The set of values of the test statistic which lead to rejection
ofthenull hypothesis is called Critical Region of the test. The probability
withwhich.a true nuill hypothesis IS rejected by the test is often referred to as
"Size"ofthe Critical Region. Geometrically, asample X, X: nis n',of size
looked upon as just a point x, called Sample point, within the region of all
possiblesamples, called the Sample Space (W). The critical region is then
defined as a subset (w) of those sample points which lead to the rejection of
the null hypothesis.
Level of Significance. The maximum probability with which a true null
hypothesis is rejected is known as Level of Significance of the test, and is
denotedby a. In framing decision rules,the level of significance is arbitrarily
chosenin advance depending on the consequences of a statistical decision.
Customarily, 5% or 1% level of significance is taken, although other levels
ach as 2% or 2% IS also used. The level of significance ais used to indicate
the upper limit of the probability of committing Type I eror, i.e. the size of
the critical region.
(10) TypeI Error(or Error of First Kind), This is the error committed in rejecting
a nullhypothesis by the test when it is really true. The critical region is so
determined that the probability of Type I eror does not exceed the level of
significance of the test.
0) Type II Error (or Error of Second Kind). This is the error committed in
accepting anull hypothesis by the test when it is really false. The probability
of Type II error depends on the specified value of the alternative hypothesis,
and is used in evaluating the efficiency of a test.
12) Power. The probability of rejecting a false null hypothesis is called Power of
the test. Therefore, Power is the probability of drawing a correct conclusion
by the lest, when the nul hypothesis is false. For a specified value of the
parameter Consistent with the alternative hypothesis,
Power = |- Probability of Type Il error.
I the power is graphically plotted against all specified alternatives, the curve
tMained is known as Power Curve. The power of a test determines the extent to
Mch atest can discriminate between true and false null hypothesis.
teps in Test of Signi£icance
(|) Set up the "Null Hypothesis" H, and the "Alternative Hypothesis" H, on the
values of
vasts of the given problem. The null hypothesis usually specifies thealternative
Some parameters involved in the population: H, (0 = %). The
hypothesis may be any one of the following types: H,(6 6}) H,(6> e).
H(0 <6). The type of alternative hypothesis determines whether to use a
(right orTleft tail).
two-tailed or one-tailed
2) State the ppropriate teststatistic"
test and alsoits sampling distribution, when
the null statistic z =(T- ,)/S.E.
hypothesis is true. In large sample tests thedistribution, is often used.
(T), whichsample
In small tests, the follows
approximately population Standard Normal
is assumedto be Normal and various test
Chi-square, t or Fdistribution
Staistics are used which follow Standard Nomal,
exacly.
Statistical Methods
554
7,= (14.7.1)
(S.E.of p)
approximatelyfollows standard normal distribution.
hichlestingsignificance at 5% level, the rules are as follows:
For alternative Pis different'
hypothesis is that the population proportion
) Ifthe
from Po. reject M,When the value of zlies outside the range-1.96 to 1.96.
H, (P Po); Critical Region Iz12 1.96
(i)Ifthe
alternative hypothesis is that the population proportion Pis 'greater
when the value of zis greater than 1.645.
than Po reject H,
H,P>Po): Critical Region z 1.645
A f the alternative hypothesis is that the population proportion P is 'less' than
Pa, reject H, when the value of z is less than - 1.645.,
H,(P< P); Critical Region z s -1.645
Otherwise, do not reject the nullhypothesis H:
For testing at 1% level, the values 1.96, 1.645 and -1.645, given above, should
be replaced by 2.58, 2.33 and -2.33 respectively.
Table 14.4 Rejection Rules for H,(P = P)
Alternative Critical Region
Hypothesis
5% level 1% level
H,
P#Po lzl21.96 lzl22.58
z21.645 z2.33
P>Po
zSl.645 z S-2.33
P<Po
Confidence limits for the population proportion P are (see 14.5.2)
95% confidence limits =p+ 1.96 (S.E. of p)
9% confidence limits =p+ 2.58(S.E. of p) (14.7.2)
where S.E. of p= n
if H, is rejected.
ple 14.20 Adice was thrown 400 times and 'six'resulted 80 times. Do the
data justifythe hypothesis af an unbiased dice? [I.C.W.A. June "77]
uOn Let us assume that the dice is unbiased, i.e. the null hypothesis is that the probability
1
Obtaining a 'six' with the dice is 6'
H
Altermatively the dice is not ubiased, i.e. the alternative hypothesis
Statistical Methods
556
Onthe assumption that H, is true (i.e. thhe dice is ubiased), the expected value oof the
(P) of 'six' is proportion
Po= = 0.167
6
6 6
(S.E. of p)= 400 120
=.0186
Example 14.21 Ina big city 325 men out of 600were found to be Smokers. Does
this information support the conclusion that the najority of men in this city are
Smokers? (State the hypothesis clearly. [I.C.W.A., June '82]
Solution Null Hypothesis is that the proportion of smokers in the whole city is 50%,
50
1.e. = 0.5.
100
H(P = 0.5)
We are interested to see if the proportion of smokers is more than 50%; i.e. Alternative
Hypothesis is
H,(P> 0.5)
The proportion of smokers in the sample is 325 out of n = 600. Using sample proportion,
325
Observed Value (p) = = 0.542
600
If the null hypothesis H, is true,
Expected Value (P) = 0.5
and 0.5(1-05)
S.E. of p= 600
=0.0204
The test statistic is
p= (14.7.3)
The expected value of the difference (p, -P) is then 0, and
558
test statistic
n,and n, are large the
sizes
Ifthesarmple PIP2
approximately follows
standard normal
S.E.
distribution. The critical
significance regions are (1474)
lzl2 1.96 at 5% level of significance
lzl 2.58 at 1% level of
P, are
(see 14.5.5)
limits for P, -
Confidence
limits = (p,-p)t 1.96 (S.E.)
95% confidence (S.E.)
= (p,-P) t2.58
99% confidence limits
PI9L P242if H, is rejected.
(14.15)
S.E. = n
where
Example 14.23 In alarge city A, 20per cent of arandom sample of 900 school
children had defective eye-sight. In another large city B, 5 per cent of a random
sample of 1600 children had the same defect. Is this difference between the two
proportions significant? Obtain 95% confidence limits for the differnece in the
populationproportions.
Solution Nullhypothesis is that the population proportions are equal.
H, (P, = P,)
Alternative hypothesis is that they are not equal, H,(P, # P)
n =900 p, =20% = 0.20
n, = 1600, p, = 15% = 0.15
Under Ho, the estimate of the equal but unknown proportion Pis
900 x0.20+ 1600 >x0.15
p= =0.168
900+ 1600
n
P292 approximately
Estimation and Test of Signi ficance 559
Hence,the
95% confidence limits for P, - P, are
(0.20- 0.156)#+1.96 (.016) =.05 +.03l=.019 and .081
Example
14.24 A machine produced 20 defective articles in a batch of 400.
overhauling it produced 10 defectives in a batch of 300. Has the machine
Afer [C.A., Nov. 81]
imporved?
Solution Null Hypothesis is that the proportions of defectives before and after overhauling
after
decreased
equal. Alternative Hypothesis is that the proportion of defectives has
Ie
overhauling.
H, (P,= P,) against H, (P >P)
We have
Sample size Sample proportion
20 1
Sample 1: n, = 400 Pi 400 20
10 1
Sample 2: n, = 300 P2= 300 30
be
sample of 400 were found to
Exampls. e
Smoker
14.25 In a certain
In another city, the
city
number of
l00 men
smokers
in a
was 300in a random sample of
800.
smokers in the second citv
Does this indicate that there is a greater proportion of
than in the
first?
Statistica1 Methods
560
300
n, = 800. = 0.375
800
0.25-0,375
2= =-4.31
.029
Here the alternative hypothesis H, (P, < P,) is
critical regions are one sided, and for this
one-tailed test the
zs-1.645 at 5% level of significance
zs- 2.33 at 1% level of
Since the observed value of z is less
than
significance
null hypothesis and conclude that -2.33, it is significant at 1%
level, We reject the
in the first. the proportions of smokers is greater in the second city than
[Note: Which Level of
(1) The critical region atSignificance to use-5% or 1%?
1%
the former is a subset of level is always included in the critical region at
the latter). 5% level (ie.
is found to be Therefore, if an
observed
the other hand,'sifignificcant' at 1% level, it is value of the test statistic
the observed value is 'not no doubt significant at 5% level also. On
significant at 1% level.
(2) The significant' at 5% level, it is obviously not
following general rules may be
atest:
the testConsider the critical regions both followed in the choice of
statistic is at 5o and 1%o levels. Iflevel of significance o
(a) not the observed value o
significant at 5%
at 1% level.level, mention that only. It is
(b) significance See Examples unnecessary to consider
significant at 5% level, but 14.20, 24, 26,
not significant of 1% 40, etc.
significant
at 1% level.
at 5%
level. Do not level, mention only that itis
(c) See
Examples mention
14.21 and
that the observed value is not significant
significant
5%
at
level. See 1% level, mention that only. It is
42.
If the level of Examples 14.23, 25, 27 useless to mention significance at
Examples
(3) An 14.22, sig
32,nifi
33,cance
36,
is
mentioned in the
to 31, etc.
39, 41,43 to 45, problem, there is no choice, but to it. See use
observed value of the test
"significant"; significant at 1% level,statistic
if
etc.
that is
it is said significant
to be at 5% level is said to besimply
"highly significant ]
ot imation and
Test of Significance 561
Z= S.E.of X
(14.7.6)
ayproximatelyfollows standard normal distribution. The critical region of the test,
On
depends the nature of the alternative hypothesis, is given in the table below.
atich
Table 14.5 Rejection Rules for H (å = u,)
Alternative Critical Region
Hypothesis
5% level 1% level
H,
lzl1.96 lzl2.58
z2 1.645 z22.33
ZS-1.645 z S-2.33
(14.7.8)
S.E. =
n
populations. In orderto test the
here oy and O, are the standard deviationsin the two
hypothesis that the means are equal
Hu, = )
Use the test statistic
S.E. -n (14.7.9)
S.E.
Statistical Methods
562
distibution, For
follows standard normal testing
which approximately
critical regions are:
against H (u, # u) the lzl21.96 at 5% level of significance
lzl> 2.58 at 1% level of significance
equality t
mean
are (see 4.5.4)
Confidence limits for u, -u, - )±1.96 (S.E.)
confidence limits =(
95% (S.E.)
confidence limits = ( - ) +2.58
99%
where S.E. is given by(14.7.8).
Example 14.26 Asample of 400 male students is found to have a mean height of
(14740,
Can it be reasonably regarded as a sample from a large population wth
171.38cms.
mean height (17.17 cms and
Solution
s.d. 3.30 cms).
assume that the sample really
We
comes from alarge population [LC:Wwith.A., meanJune'8)
3.30 (cms). Null Hypothesis is 17141
and s.d.
H,(u =171.17, o =3.30)
Alternative Hypothesis is that the sample does not come Trom such a population
H,(u ÷ 171.17)
Since the sample size n = 400 is large, the sample mean (x) is approximately normaly
distributed.
Observed value () = 171.38
Expected value (u,) = 171.17
3.30
S.E. of x = = 0.165
n 400
When Ho is true, the z-statistic (14.7.6) follows N (0, 1).
Observed value - Expected value 171.38171.17
Z= = 1.27
S.E. 0.165
Since the alternative hypothesis is both sided (ie. u is either more than or less than 171.1).
trhe critical region of the test is also two tailed (i.e. given by two tails of Standard Nomd
curve). At 5% level
Critical Region is |zl1.96
The value of test statistic z(viz. 1.27) does not fall in the critical region and hence s
Significant".We have therefore no reason to reject the null hypothesis at 5% level of sIg
and conclude that the sample may be regarded as having arisen from the given Population.
Sincethesamplesize (n) is large, the sample mean (r) is approximately normally distributed
and SE. = g/Nn However, since the population s.d. o is not known, an
meanu
with
proximatevalue
of S.E. is
S 0.10
S.E. = VI00 0.01
Vn
1.94-2.0
Therefore, =-6
0.01
Sincelzl=6 exceeds 2.58., we reject the null hypothesis at 1% level of significance and
concludethat the machine is not funcioning properly.
Examples 14.26 and 27 are two-tailed tests for specified mean, the former with o
(Note:
known, and
the latter with Gunknown. Example 14.28 is a one-tailed test (o unknown).]
Example 14.28 Amachine part was designed to withstand an ave rage pressure
A random sample of size 100 from alarge batch was tested and it was
of120units.
dthat the average presSure which these parts can withstand is 105 units with a
standard deviation of 20 units. Test whether the batch meets the specifications.
[C.U., M. Com. "72, '74]
Colution Null hypothesis: Mean pressure in the population which a machine part can
120 units.
uithstand is 120 units. Alternnative hypothesis: Mean pressure is smaller than
H,(u=120) against H,(u< 120).
S
S.E. of x= approximately,(since o is not known)
20
= 2.
100
Here, sample mean is =105. Therefore,
105 - 120
=-7.5
2
(critical region at 1% level for the one
Since the valuje of z=-7.5 is smaller than -2.33
1% level of significance and conclude
lailed test is z S-2.33),we reject the null hypothesis at than 120units, i.e. that the batch of
in the population is less
diat lneaverage breaking strength specification.
acnine parts does not meet the (If the value is found to be significant at 1%
H, than at 5% level).
C, We have stronger grounds for rejection of
n, = 250 n, = 400
Sample sizes
Sample means I = 120 gms X, = 124 gms
Sample S.D.s S, = 12 gms S, = 14 gms
sanple S.D.s to calculate
Since the population S.D.s are not known, we use the
Error of the difference (I, I,) between the two sample means (see 14.7 8) the
S.E. of +- 7= |
Standad
/122 14
= L.03
250 400
The observed value of the test statistic is
- 120 - 124
=-3.9
S.E. 1.03
Critical region for both sided alternatives H, (u, #u,) is lzl>1.96 at 5% levelel andlzl2.58
at 1% level of significance. Since lzl= 3,9 exceeds 2.58, we reject H, at 1% 1level
signiticant; i.e. u, # Uo.lof
and conclude that the difference between two mneans is
99% confidence limits for u, - u, are(see 14.7.10). significance,
(7-) +2.58 (S.E.) = (120 - 124) + 2.58(1.03)
= -4+ 2.66 =- 1.34 and -6.66
This means that confidence limits for the
difference U, - u, are 1.34 and 6.66 øms
Note:Examples 14 29 to 31 are all two-tailed tests for equality of two means. The two
Ayulaions.d.s are(i) unknown in Example 14.29: (ii) known and equal, but sample s.d.s are
in Example14.330: and (iii) known and equal, but the sample s.d.s are also given in
fample
14.31,
]
pxample14.31 Themean yield of wheat from a district A was 210 lbs. with S.D.
peracrefrom a sample of 100 plots. In another district B, the mean yield
0lbs. with S.D. =}12ibs. from asample of 150 plots. Assuming that the standard
nationofyieldin the entire state was II lbs., test whether there is any significant
benween the
sfirencebe mean yield of crops in the two districts.[C.A., May '76]
Solution Let u and l, denote the mean yields of crops in districts Aand Brespectively.
H,(u, u,)
H =4)against n, = 100 n, = 150
Given
A = 210 lbs. I, = 220 lbs.
S, = 10lbs. S, = 12 lbs.
Population S.D. (o)= || lbs.
Me mav assume that the S.D. of yicld in the whole state is the S.D. of yields in the two
knc's That is, the two populations have the same S.D. (o) |1 lbs. Using (13.7.3a)
4
SE. of -, =
= 142
I50
The observed value of the test statistic is
A-) 210- 220
=-7.04
S.E. 142
Since lzl=7.04exceeds 2.58, we reiect H, at 1% level and conclude that there is a significant
ilference in the mean yields of crops in the two district.
Using Chi-soquare ( ) Distribution
small
C-square distribution (Page 519) is used in both large sample and
nple tests. It is mainly used in
) Test for goodness of fit.
2) Test for independence of attributes.
8) Test for a specified standard deviation (Small Sampletest).
y) The test, devised by
Test for goodness of fit (Pearsonian in good agreement with
Parson. is used to decide whether the observations are spposed to have arisen
thyparedpospecihetiwithcalfiedthedistribution.
i.e. whether the sample may
Population.
be
The observed
expected frequencies
different classes are
frequencies (fo)of
(f) by he test statistic
(14.7.11)
Statistical Methods
566
"goodness-of-fit chi-square".
"Pearsonian
Chi-square" or When
This is called hypothetical
population)
hypothesis, viz. the
the null
in agreement withchi-square distribution with(k:- ) degrees
are
H,(Dateapproximately follows observed value of the statistic
statistic classes. Ifthe
is true,the
where k is the number of level, the null hypothesissiis rejected.
of freedom, given
tabulated value ofx at a
exceedsthe pea-breeding. Mendel obtained the
on
14.32 n his experiments yellow-315; Wrinkled and yellow--l01;
Example seeds : Roundand
following frequenciesof
and green-32; Total-556. Theory predicts that
Wrinkled Examinethe correspondence
Round and green108; the proportions 9:3:3:1.
frequenciesshould bein value of for3df. is 7.815)
the observations. (Given that 5%
between theory and theory, and that the divergences hau
with
assumption that the data agree probabilities for the classes should bo
Solution On the
sampling fluctuations, the
arisen only due to below:
3 1 expected frequencies are calculated
9 3 respectively. The
16, 16 16' 16
Observed Frequency
Expected Frequency
Class
9
x 556 = 313
Round and vellow 315 16
3
X 556 = 104
Wrinkled and yellow 101 16
3
108 x 556 = 104
Round and green 16
3
32 x 556 = 35
Wrinkled ande green 16
556 556
Note: The totals for Observed and Expected frequencies must be equal.
(315-313)' (101-104) (108104) +
(3235)
313 104 104 35
4 9 16
+
313 104 104 3S
= 01 + .09 + .15 +.26 = 0.51
Since there are 4 classes. Degrees of freedom = (4 - 1)=3
We are given that the 5% value of y for3 d.f. is 7.815. Since the observed
than the tabulated value, the null hypothesis cannot be value 0.51 iS less
support the theory. rejected. We conclude that the observai
Example 14.33 Adice was thrown 60 times with the following results:
Face 2 3 4 Total
Frequency 10
13 11 12 60
Estimation and Test of Signifi cance 567
is 60x
and the expected frequency 6
= 10 for each.
Observed Frequency f) 10 13 12
16 0 4 4
16 4 9 4
+ + =3.4
10 10 10 10 10
Example 14.34 5 identicalcoins are tossed 320 times, and the number of heads
appearing each time is recorded. The results are:
Number of Heads 2 3 4 5 Total
14 45 80 12 61 8 320
Frequency
Would you conclude that the coins are biased ? (Given x as = l1.07 and yo =
I5.09 for 5degrees of freedom).
Solution This is a problem for testing "goodness of fit" of a binomial distribution. The nuli
hypothesis is that the coins are unbiased, i.e. the probability of obtaining head is 1/2 for each
coin.
H, (data support Binomial distribution with P= 1/2)
The test
ItH, is true, i.e. the coins are assumed to tbe unbiased, the probability of obtainingr heads
n One throwof the set of 5 coins is given by the binomial distribution
5-r
P(r) = 32
SO, the expected frequency () of r heads in 320 tosses is
f,=320. P(r) =10x C.
Inese are shown below for different values of r:
Number of Heads () 4 Total
4 5 20° 12
10 50 100 100 50 10
0.4 =10.36.
= 1.6+0.5+ 4.0 + 1.44+2.42+
There are 6 classes: so De grees of freedom=6-l= 5
tabulated value at 5 laya
Since the observed value of z (viz. 10.36) is less than the conclude that
We therefore
(Viz. 11.07), we cannot reject H, at 5% level of significance.
coins may not be biased.
Contingency
(2) Test for independence of attributes |Note: (1)Section 12)or
(see
Table-In Statistics,sometimes we have to deal with "attributes
qualitative characters of members which cannot be measured accurately, although
the members can be divided into twoor more categories with respect to the attributes.
Let us consider two attributes A and B, where A is shown in mcategories A,, A,, ...
form of a two
Am, and B in n categories B,, B,. .... B,. The data can be shown in the
way table with m rows and n columns, as in a biv ariate' frequencydistribution (Tables
9.2 and 9.7). This two-way frequency table for attributes is known as (m x)
Contingency Table (see Example 14.35). The frequency of members belonging to
both the categories A, and B, simultaneously is shown in the cell at the i-th row and /
th column, and denoted by (A,B.).
Similarly. (A) and(B,) denote the frequency of members belonging to categories
A, and B, respectively, and N the total frequency, as in the table below.
(3 x 4) Contingency Table
Atribute B
B, B, Total
B4
A, (A,B) (A, B,) (A,B) (A, B,) (A,)
A (A,B,) (A,B,) (A,B) (A,B,) (A,)
A (A,B,) (A,B,) (A,B,) (A,B}) (A,)
Total (B,) (B,) (B) N
(B,)
Oten, the members are divided into two categories only in respect of each attribute. according
to the presence or absence of attribute A (denoted by A and a) and presence or absence o
attributeB (denoted by Band B). See Example 14.36.
(2 x 2 ) Contingency Table
Attribute A
A Total
B (AB) = a (a B) =b (B) = a +b
B (AB) =c (a ß) = d (B) =c+d
Total (A)= a+c (a) =b + d