UNIT 18 COMMON TESTS AND

CONFIDENCE INTERVALS
Structure
18.1 Introduction
Objectives
18.2 Some Common Tests of Hypothesis for Normal Populations
18.3 Confidence Intervals
18.4 Chi-Square Test for Goodness of Fit
18.5 Summary
18.6 Solutions and Answers

18.1 INTRODUCTION
In Unit 17, you have been introduced to the problem of testing of hypothesis and also
to some basic concepts of the theory of testing of hypothesis. There you have studied
two important procedures fortesting statistical hypotheses,viz. using Neyman-Pearson
Lemma and the likelihood ratio test. In this unit, you will be exposed to the problem
of testing statistical hypotheses involving the parameters of some important
distributions through some selected examples. In this unit, you will also be exposed to
the problem of constructing confidence intervals for parameters of some important
distributions through some selected examples. You will also learn the use of chi-square
test for goodness of fit.

Objectives
After reading this unit, you should be able to:

derive test statistic for various testing of hypotheses problems as well as to

derive power functions,
construct confidence intervals for parameters of various distributions,
conduct large sample tests.

18.2 SOME COMMON TESTS OF HYPOTHESIS

FOR NORMAL POPULATIONS
In Unit 17, we have already described with examples two procedures for testing
statistical hypotheses. (See Example 1and Example 2). In this section we will .
employ Neyman-Pearson Lamma and likelihood ratio test for testing of hypothesis
related to a normal population.

Example 1: Let XI, . . .,X, and Y1, . . . ,Y, be independent random samples from
2 2
N (pl, u ) and N (%, u ), respectively. It is desired to obtain a test statistic for
2
testing Ho: p1 = against p1 : p1 * CL;! when u ( > 0 ) is unknown.

In order to obtain the test statistic, we use the likelihood ratio test. We have

We shall write 8 - (pl, k,u 2 )

,
.I --
We have Common Tub d
C d e o c e Interval8

Under Ho, p1 = k = p and the maximum likelihood estimate of p is

= U' (say)

Thus Sup L ( B ( X , Y )
BEQo
- 1
(ZSCU')~ 1
Under HI, the maximum likelihbod estimates of pl, k and a 'are respectively

and

The likelihood ratio test is thus

sup L @ ( x , y )

L(x7Y) - 8EQo
Sup L @ I X , Y )
-B E 9
Elements of Statlsticsl
Inference

X -Y
Now under null hypothesis, pl = & = p, and t = follows a

2 u(m+n)
Student's t distribution with m + n - 2 degrees of freedom, where S =
m+n-2

Thus

and

The likelihood ratio critical region is given by

where c is to be determined so that

Sup P, [A (X, Y) < c] = a

BE Q,

+ n - 2) we reject H,
- Since A (X,Y) is n decreasing function of t2/(m

if
 where q is so chosen that Common Testa and
Conlldence Inkmala

Let q - t,,, -2, in accordance with the diskbution oft under Ho. Thus, the
+

two sided &st obtained is

'

Let X I , . . . ,X,,be a random sample from N (p, u ), p is known and

2
Example 2:
o YO, is unknown. We wish to obtain a test statistic for testing H,: u2 - u
2

against an alternative H1 : a 2 - a 2 ( > u 2 ).

We have

*+* !'

- ., I Using Neyman-Pearson Lemma, the test statistic is

n
-> (u - u i) 2 (Xi - l?) 2 k', taking logarithms
1

i
-> 2 (xi-d)akl,
1
since o1
2 > a 2m under H1

11

Here kl is so determined that

Elements of Statistical
Inference
Under the null hypothesis, since o2 - 1 o
n

1
(X -p ? / o 3 has a )2 distribution

(chi-square distribution with n degrees of freedom). Let 2, be the upper -a

probability point of 2.The test statistic is thus
xn
(Xi - p )2> - k1 and hence

andkl/uo
2
- h2 a => kl - u 2~ 2& , ~ .

On the other hand, if the alternative hypothesis is H1 : u

test statistic is
2
- a,2 (al2 < uo),
2
then the

and hence

where -,is the lower a -probability point of the 2 distributionwith n degrees

of freedom.

Example 3 : Let XI, . . .,X, and Y,, .. .,Yn be independent random samples
2 2
from N (pl, u u 2). We wish to obtain a test statistic for testing
and N (CLL,
Ho:ul
2
- 2
u2againstH1:ul
2
t
2
a2.

We shall use 8 - 2
(PI, k, <J 1 , ~ 2 ) .
2

Also L (8 1 X, Y)

The maximum likelihood estimates of pl, 112, u 2 u 2 are respectively

 Common Test#and
C m M e l l Intmalr
~~

Further, if o - :-
o o 2, the maximum likelihood estimate of dl is

I Thus

and

exp {-(m + n)/2]

The likelihood ratio test is thus

I
I

Now
 Elements of Statlstld
Inference
I

m
(X -Zf/(m - 1)
1
Writing f = n

we have

The likelihood ratio test criterion rejects Ho if h (X, Y) < c

It is easy to see that h (X, Y) is a monotonic function of f and h (X, Y) < c is

equivalent to f < cl or f > $. Under q,

has Snedecar's F (m - 1,n - 1) distribution, so that cl, 9 can be selected, such that

sup Po [h(X, Y) < c]

eEa, -
- a

Thus $ = F (m - l,,n - 1, a/2) is the upper a12 probability point of

F (m - 1,n - 1) distribution and cl = F (m - 1,n - 1,l- a/2) is the lower a / 2
probability point of F (m -1, n - 1)

Let X,, . . . ,X, be a random sample from N (p, u2), both p and a 2 unknown.
Obtain a test statistic for testing Ho : a -a 2
against all its alternatives.
(X2,Y2), . . . ,(X,,,Y,,) be a random sample from a bivariate
E2) Let (XI, Y1),
2 2
normal distribution with parameters pl, k,u a 2, p. Obtain a test for testing
Ho: p = 0 against H1 : p rr 0, say.
 Common Testa and
18.3 CONFIDENCE INTERVALS Confidence Intervals

In Section 15.6 of Unit 15 you have been briefly exposed to some notions of
interval estimation of a parameter. In this section we discuss in detail the problem
of obtaining interval estimates of parameters and describe, through examples, some
methods of constructing interval extimates of parameters. We may remind you
again that an interval estimate is also called a confidence interval or a confidence
set. We first illustrate through small examples the need for constructing confidence
intervals. Suppose X denotes the tensile strength of a copper wire. A potential user
may desire to know the lower bound for the mean of X, so that he can use the wire
if the average tensile strength is not less than say go. Similarly, if the random

'
t
variable X measures the toxicity of a drug, a doctor may wish to have a knowledge
of the upper bound for the hean of X in order to prescribe this dmg. If the random
variable X measures the waiting times at the emergency room of a large city
hospital, one may be interested in the mean waiting time at this emergency room. In
this case we wish to obtain both the lower and upper bounds for the waiting time.

In this unit we are concerned with the problem of determining confidence intervals
for a parameter. A formal definition of a confidence interval has been given in
Section 15.6.However, for the sake of completeness we define some terms here.

Let XI, X2, . . . ,& be a random sample from a population with density (or, mass)
function f (x,0),0 E S2 G R'. The object is to find statistics rL ( XI . . . . . ,& ) and
ru (XI, . . . ,X,,) such that
-
\

X ~ X,,) s 0 s ru(Xl ,..., X,,)]r. 1 - a f o r a l l 0 E h d S th the

P , [ ( ~ ~ (,...,
interval rL ( !
( )
!), rU ( X ) is called a confidence interval and the quantity

will be referred to as the confidence co-efficient associated with the random interval.
We now give some examples of construction of confidence intervals.

Example 4: Let XI, X2, . . . ,X,, be a random sample from a normal population,
N (p, u 2). We wish to obtain a ( 1- a ) level confidence interval for p.

Let fl - '2
n- Xi. Consider the interval @ - a, fl + b). In order for this to be a

(1 - a)-level confidence interval, we must have

~ ~ ~ - a < ~ < ~ + b } ~ l - a

Thus

?7 -
Since, -&/N'~N (0,l) we can choose a and b to satisfy
<I
\
Elements of Statlatical
Inference -
provided that a is known. There are infinitely many such pairs of values (a, b). In
particular, an intuitively reasonable choice is a b = c ,say
In that case

-
c 6 = Zd where Zd is the a12 percent point of the standard normal
a
distribution, and the confidence interval is

( z - ( 0 1 6 ) Za/2, X + (a/&) Zd2 )

The length of the interval is ( 2 a / 6 ) ZaI2 Given a and a one can choose n to get a
*
confidence interval of desired length.

Rgulp 1 :Probability density curve of normal dlstrlbutlon with mean p and variaance &.Shows
area ah in each of two tails

If a is unknown, we have from

that

and

-
It is known that - '-
s /-f i ,t - We can choose pairs of values (a, b) using a
students t-distribution with (n - 1) degrees of freedom such that

In particular; an intuitively reasonable choice is a I b - c say. Then

 r

(
a d j7 - (sifi) t,,-1,, , + ( S / m t,,- ,, ,
)
is 1- a level mnfidence interval Common Tat3 md
for p. The length of the interval is (2316)&, l,, , which is no longer constant.
Therefore, in this case one cannot choose n to get a fied length confidence interval
of level 1- a.The expected length is, however,

which can be made as small as we want by making a proper choice of n for a given
a and a.

Figure 2 :t values such that there Is on a m cz/2 in the mtla11end a12 in the left tell of the
distribution.

Example 5 :Let XI, X2, .. .,X, bb a random sample, from N ((r, o 1' . It is desired
.to obtain a confidence interval for u when C( is unknown.

Consider the interval (as2, bS?), a, b > 0, S? - (n 1)-

n
' 1(Xi -%f. We have
1

so that

It is known that
(n-l)s2/dZ-

We can therefore choose pairs of intervals (a, b) from the tables of the chi-square
distribution. In particular we can choose a, b so that
 Elements of S tatistlcal
Inference
n-1
Then -
2 n-1 2
a m % - 1 . o m d ~= x,,- ,, - a and the 1- a level confidence
I
'
interval for a when p is unknown is

n
If however, p is known then (n - 1) S' is replaced by (Xi - ( I ) and
~ the degrees
1
n
of freedom of *' is n instead of n - 1, for
1
(Xi - Cy/dl d.

Probability density curve .

of chi-square distribution
with n - 1 Jf

-
Figure 3 :Chl-squarevalues such that area 1 an and a/2 are b thelr right.

.
Example 6: Let XI, .. . ,X, and Y1, . . ,Y, denote respectively independent
random samples from the two independent distributions having respectively the
probability density functions N ((I,, a ') and N (112, o 1
' . We wish to obtain a
confidence interval for p1 - k.

(z
Consider the interval ( - y)- a, (E-Y)+ b 1. In order that this is a (1 - a ) level
confidence interval, we m b t have

which is the same as

 Common Teiitr and
Confkieneo Intenah

--
Since -_ *
(X- Y) - (PI -. h)

we can choose a and b to satisfy

.
provided that a is known There are infinitely many such pairs of values (a, b). In
particular, an intuitively reasonable choice is a = b = c, say. In that case
r .

The length of the intaval is 2u Za. Given a and a one can choose n
and m to get a desired length confidence interval.

If a 2 is unknown, we have from

that

-%l2+ 1(yi-Tf
where s2 = 1 1
I (n - 1) S: + (m - 1) S;
(n +m-2) n+m-2

It is known that

--
w -Y) -(PI- k)
* , -%+m-? We can choose pairs of values (a, b) using Student's

t-distribution with n + m - 2 degrees of freedom such that

In particular, an intuitively reasonable choice is a - -
b c, say. Then

-
is a 1 a level confidence interval for v1 - k.

Example 7: Let XI, . . . ,X,, and Y,, . . . ,Ym,n, m > 2, denote respectively
independent random samples from the two distributions having respectively the
2
probability density fundions N (h,o :) and N (R, o 2). We wish to obtain a
/o:
confidence interval for the ratio o :lo when and k are unknown.

Consider the interval a S2/Sl, b S2/S1 a, b > 0, where

( 2 2 2 2 )

so that

2
-
It is also known that if X and Y are independent random variables with m and n
degrees of freedom respectively, the random variable F (X/m)/(Y/n) is said to
have an F-distribution with (m, n) degrees of freedom. It is also known that if X has
.a
Fa -a-
F (m, n) distribution then l/X has an F (n, m) distribution, and
1 4 ,a a. Therefore
&. / Common Tertr and
We can therefore choose paris of values (a, b) from the tables of F-distribution. In
1 Confidence Intarvalr
r particular, we can choose a and b so that

Then - ,
a
2
1

2
F - , and
1
- Fa ,1-(.aand the 1- a level confihnce interval
for a 2 / ~is

E3) Let XI, .. .

,X,, be independently and identically distributed b (1, p) random
variables. Obtain a confidence interval for p.

18.4 CHI SQUARE TEST FOR GOODNESS OF FIT

An important limiting distribution used in connection with categorical data is the
x2-distribution. By categorical data we mean data which are presented in the form
of frequencies falling in different categories or classes. The categorisation may be
with respect to a character which is either an attribute or a variable. The
categorisation may also be with respect to two or more characters.
In this section we introduce some tests of statistical hypothesis that are commonly
2
called -tests. These tests have been immensely useful in practical applications I
particularly in problems dealing with categorical data. A test of this sort was
originally proposed by Karl Pearson before the formal theory of testing of
hypothesis was developed.

Suppose XI, . . .,X,, is a random sample from a discrete distribution given by

- 0, elsewhere,

2 pi - I.' We wish to obtain a test statistic for testing the null

k
where pi > 0 and
1
hypothesis

where poj are known numbers, on the basis of the observations.

Let nj be the number of X's that equal Xi in a sample of size n. Then

njts,j - i, . ..,k are random variables, satisfying

k

1
nj - n. Further, the

distribution of (nl, .. . ,nk- l) is under H,,,multinominal with parameters .d

(pil, ...,Pw - We are not considering 4 @t) because 1nj and 1pj are fured.
1 1
Elements of Statistical It is known that nj has a marginal binomial distribution with parameter pj Hence

-
Inference
E (nj) npj, so that E (nj/n) = pi

and we can use nj/n, j P 1, . . . ,k respectively as unbiased estimates of p,, . . . ,p,.

Intuitively, it appears meaningful to compare the observed proportions nj/n with the
postulated proportions poj,j = 1, . . . ,k.. The various way which can be used for
such a comparisonare, e.g.,

Another measure most commonly used in practice is the weighted sum of squares

Since, under 4,npoj, is the expected value of nj, we would feel intuitively that the
observed value of Q should not be large if I-&, is true. Thus large values of Q would
lead to large discrepancy between the observed data and the postulated hypothesis
and thus lead to the rejection of I-&,.

What is the distribution of Q under the null hypothesis? This is answered below.
We however, omit the proof,

Theorem 2 :In,a random and large sample of size n,

follows approximately chi-square distribution with k-1degrees of freedom, where

nj is the observed frequency and npoj is the corresponding expected frequency of the
-k
j t h c a s s j = I,. . . ,k 1, Enj = n..
1

Since the distribution of the ran'dom variable Q, under is approximately with

k-1 degrees freedom, the test rejects the null hypothesis Hoif Q t c where c is so
determined that
pH, (Q z C) = a (-the desired level of significance).

Thus, we get c = &-

2
,, and reject the hypothesis 4 at level a if Q > xk.
a
2
.,

-
For k = 2, Q has approximately a chi-square distribution with 1degree of freedom.
Indeed, for k 2, the problem reduces to testing % : p
sampling from biilomial population.
-
po against H : p * po in
 Probability Common Tats and
density Coatldence Intervals
function
4

-* -- , -,
r' Rejection region
Do not reject H, I Reject H,, '
Figure 4 :Probablllty density curve for chl-square distribution with k - 1 df and the decision rule.

Also under Ho,

-X
1
1
-npq
npq (1 -PO) - k

1
(1 -pq)

We can use the test statistic Q to test that a random variable X has a specified

-
(- w, a)into k intervals 11, . . . ,Ik and let poj pf [X G Ij], j -
probability density function f. In order to do that, we divide the entire real line A
1, . . . ,k. Let pj be
the observed proportion of observations in the sample that belong to the interval Ij

Then the statistic

has approximately a chi-square distribution with k - 1degrees of freedom, under H.

But, then an important question which arises now is as to how to choose the
intervals 11,. . . ,Ik. The answer to this question comes from the fact that the
chi-square distribution is only an approximation to the true distribution of Q. So the
choice has to be made in such a way that this approximation is good. Secondly, the
un!erlying assumption to the chi-square approximation is that each random variable
(nPj has approximately a normal distribution with mean zero and variance
5
12 poi. This holds provided n is large .

( 2
But if npq is small, then the term n i j -jpoj] /npq in Q will have a dominating
effect on the other terms because of a small enominator. Although the
approximation is often good even for npoj as small as one the following rules of
thumb are generally used in deciding upon the intervals 11, . . . ,Ik.
I
Elements of Statistical
Inference
i) Choose I,, . . .,1, such that under H,,,pOj
n
- p E Ij] is approximately

equal to l/k and each npq = - r 5.

k
ii) If any npg's is less than 5, pool' the corresponding interval with one or
mdre intervals to make the cell frequency at least 5. The decision which
intervals (or cells) to pool is arbitrary but we restrict pooling to a
minimum. The degrees of freedom associated with the chi-square
approximation, after pooling, are reduced by the number of classes
pooled.
._
the underlying density is f (x)
follows :
-
Example 8: A random sample of 80 points is picked from the interval (0,l) so that
1,0 < x < 1,and zero elsewhere: The data are as

.03 .69 .30 .92 .10 .85 .53 .6

.35 .O1 .70 .24 .22 .93 .22

.9 1 .57 .28 .47 .ll .69 .14

We construct the frequency distribution with equal class intervals of equal width.
We wish to test the null hypothesis I+,, that the sample comes from the uniform
distribution on (0,l).
The following table gives the observed and the expected frequencies.

Interval (.005, .105) (.105, .205) (.205, .305< (.305, .405) (-405,505)
nj 8 11 8 13 6
npQi 8 8 8 8 8
Interval (.505, .605) (.605, .705) (.705, ,805) (.805, .905) (.905,1.00)
nj 5 8 9 4 8
"Poj 8 8 8 8 8

We have Q -110

1
(nj - nPq)z

"Poj
-
8.0

The value of chi-square at 9 degrees of freedom and .05 level of significance is

16.919.
Hence the observed value of Q is not significant at 5 percent level and we cannot
reject Q.

We have just described the chi-square test of goodness of fit when the cell
probabilities are prespecified. But a more interesting situation is one when instead
of fitting the data reasonably with one specified distribution, the data can be fitted
by one of a family of distributions. The question to be answered in such a situation
is; could the observations have come from some Poission distribution ? Some
normal distribution ? Some binomial distribution ? Some exponential distribution ?
etc. In particular the null hypothesis Q is not a simple hypothesis now. Under Q,
the probability mass or density function is specified except for one or more
parameters.
 Let XI, . ..,X,, be a random sample from a discrete distribution given by Common Testa and
Confidence Intervals

P [X = xi] = pj (€9,j = 1, . . . ,k and zero ekwhere, where

pj (8)> 0,
k

1
pi (8) - 1 and 8 is a scalar or a vector parameter that is unknown. Let

2 nj - n and under
k
nj be the number of X's in the sample that equal xj.Then
1
& : (nl,. . . ,nt - has a multinomial distribution .Let

We wish to test the null hypothesis that for some value of 8,pj (8),j
a good fit to the data. To achieve this, minimise Q (8)for all possible values of 8 in
-
I, . . . ,k, is

the parameter space. In other words, find that value of 8 for which pj (8)best fits ,
the data. Let be the value of 8 for which Q (8)is minimum. 8 is called the
6
minimum chi-squara estimate of 8.Under certain conditions and for large n tht?
minimum value Q (8)of Q (8)given by

tirl
has approximately a chi-square distribution with k - 1- s degrees of freedom where
hi I
2
p and a unknown, then s -
-
s is the dimensionality of 8.If 8 is scalar, then s 1,if 9 = (p, o '), both
2, and so on. It can be shown that for sufficiently
large n the 8 that minimizes Q (f) is approximately the maximum likeGhood
estimate of 8 and moreover Q (8)has a chi-square distribution ,when 8 is the
maximum likelihood estimate of 8.

We reject & at level a if Q (6)r xk2 - l,a.

In the continuous case the procedure is the same as described earlier. let X be of
continuous type with density function f (x,8)where 8 is unknown. Partition the real
line into k intervals 11, . . . ,Ik. under Ho,

Pj (8) a Jf 9'( 8)k ,

'j

and the result stated above applies. We illustrate tbese ideas through an example.
Example 9: A random sample of 200 families, each with four children has the
following frequency distribution of the number of girls.
Number of girls 0 1 2 3 4
Number of families : 5 32 65 75 23
We wish to test if the data comes from a binomial distribution.

Let X denote the number of girls in any family of four children. We wish to test the
composite nu!l hypothesis
Ekmenta of Statistical
Inference
against all alternatives. In this example n
maximum likelihood estimate of p. We have
-
4. Since p is unknown, we compute the

= 0.5987
In order to apply the test, we need

We have

p3 ($) - - .34448, p4 ($) - .I2848

,
We now compare the observed and expected frequencies as follows :

j o 1 2 3 4
A Observed 5 32 65 75 23
Expected 5 31 69 69 26
Thus, we get

The number of degrees of freedom on the conservative side is 3.Since the value of
2
&,O.OS. is 7.81,there is not much evidence to reject Ho. We therefore infer that the
data could have come from a binomial distribution.

18.5 SUMMARY
-In this unit we have
'
1. demonstrated the use of Neyman-Pearson Lemma and likelihood ratio test for
testing hypothesis about parameters of normal populations,
obtained confidence intervals for parameters of some standard distributions,
3. described the use of chi-square test for testing the goodness of fit problems.

18.6 SOLUTIONS AND ANSWERS

E 1) We have
2

and
O - 2
{ ( p , ~ ~ ) , - m < ~ c , m>, oU}
Now

Sup L (g (
!!Eno
iJ -"Sup
0'-oo
1
' (2nu0)&
The maximum likelihood estimate 3f p is
exp
[ 1
-7
20. : I
(q- d2
 Common Tests and
Confldeacc Intervals

i Thus

Also

sup L (8 1 X)
gcn
- sup
140
1
(2~r.o~)~
L
The maximum likelihood estimates of p and o 'are respectively j -a and

b Thus

SUPL(8lX)
gen
- 2Jt
n

[%-x)2
exp ( - n/2 )

The likelihood ratio test is

Under the null hypothesis

Thus

It therefore follows that A ( ) < c is equivalent to u < c1 or u > % Since .

2,
under H,u has a x distribution, cl can c2can be selected. It is ysual to take

~ [ 4)
2u - d 2 and P { U s c,] - 1-d2

In other words % is the upper a / 2 probability point of 2- and c1is the lower
1- a / 2 probability point of d-,.
E2) We have
no - {(pl,p27u;,o;p):-m~14<~70,2 P O 7 i - 1727~ - 01
And
 2 2
Elements of Statlstlcal Also we shall write 8 = (pl, k,u u 2, p)
Inference
and

2
Under S2, p = 0, and the maximum likelihood estimates of pl, ~ ( 2u
, u 2 are
given respectively by

Thus 1

under 62, the maximum likelihood estimates of pl, ~ ( 2 , and ui are the same I
as under a,, whereas the maximum likelihood estimate of p is
m

I
A A

"1'( 2'(

Thus

Sup L ( 0 1 & 3
Beno
- 1
(2~;~;~)~
2 n/2 exp (- n)
(I -T )
The likelihood ratio test rejects H,,if

Sup ~ -.( I X,
8- Y).
h (X, Y) - -e e i j
SUP L @ I X , Y )

or equivalently, if
>~'
I'I>C,
where Cl is chosen that
 The distribution of r under H, can be determined. It is known that Common Tests and
Confldence Intervals

under null hypothesis has a students' t distribution with n - 2 degrees of

freedom. Using this fact thd value of c, can be computed for a given a.
E3) There is another possible procedure of obtaining confidence intervals. This
method has universal applicability and can be used for large or small samples.
But this proadure usually yields confidence intervals that are much too large.
The method uses the well known Chebychev's inequality
1
P ( J X - E ( X ) ( < E ~ - } > ~ - ~ , for r > O
E

It follows that
1
P { ~ X - ~ ~ < E ~ ~ } > ~ - ~
E

since p (1- p) s 1/4, we have

One can now choose E and n . Qtherwise if n is fixed one can choose E to get
the desired level confideme intkrval.
 APPENDIX
Table Vnlues of the Stnndal-cl Normal Distribution Function

B.W.Lindgren, Shtisticnl Tbcory. l l i e Mncrnillnn Company, 1960.

 Appendlx Statistical Tables

Table (Cuntd.)
 1'
l k b l e of X2 Dlslrlbutlon
If X has a X2 distributios with n Jegrces of freedom, this table gives the vnlue of x such that
P r ( X s x 1 =p. I
\

Adnpted with permission from Biornetrika Tubles for Siatisticians, Vol. I 3rd ed. Cambridge
University Press, 1966, edited by E. S. Pearson end H.O. Herrley: and from "A new table of
percentage point of the chi-square distributior~."lliometriku, Vol. 51(1964), pp. 231 - 239, by 1-I.L.
Harter, Aerospace Research Laboratories.
 e X2 Dlstrlbutloa (Continued)
~ n b i of
~ l e m e n dof Statistical
Inference

Tnble of the t dlslributlon

1IXhas a t distribution with n degrees of Irgedom, the table gives the value o f x such that P(X s x)
'p.

n p = 55 .60 .65 .70 .75 .80 .85 .9O .95 .975 .99 .995

1 .I58 ,325 .510 .727 1.000 1.37631.963 3.078 6.314 1x706 31.821 63.657
2 ,142 .289 .445 .617 .816 1.061 1.386 1.886 2.920 4.303 6.965 9.925
3 .I37 .277 .424 .584 .765 .978 1.250 1.638 2.353 3.182 4.541 5.841
4 ,134 ,271 .414 ,569 .741 .941 1.190 1.533 2.132 2.776 3.474 4.604
5 .132 .267 .408 .559 .727 .920 1.156 1.476 2.015 2.571 3.365 4.032
6 .I31 .265 .404 .553 .718. ,906 1.134 1.440 1.943 2.447 3.143 3.707
7 ,130 .263 .402 .54Y .711 .896 1.119 1.415 1.895 2.365 2.998 3.499
8 .130 .262 .309 ,546 .706 .889 1.108 1.397 1.860 2.306 2.896 3.355
9 .I29 .261 .3YS ,543 .703 .S83 1.100 1.383 1.833 2.262 2.821 3.250
10 .129 ,260 .397 .542 .700 ,879 1.093 1.372 1.812 2.228 2.764 3.169
11 .12Y .260 .396 .540 ,697 .876 1.088 1.363 1.796 2.201 2.718 3.106
12 .I28 .259 .395 539 .695 373 1.083 1.356 1.782 2.179 2.681, 3.055
13 .I28 .259 .394 .538 .694 .870 1.079 1.350 1.771 2.160 2.650 3.012
14 .I28 .258 .393 .537 .692 368 1.076 1.345 1.761 2.145 2.624 2.977
15 .I28 .255 .393 .536 .691 366 1.074 1.341 1.753 2.131 2.602 2.947
16 .I28 .258 .392 535 ,690 ,865 1.071 1.337 1.746 2.120 2.583 2.921
17 .I28 .257 .392 ,534 .689 .863 1.069 1.333 1.740 2.110 2.567 2.898
18 .I217 .257 .392 .534 .688 .862 1.067 1.330 1.734 2.101 2.552 2.878
19 .I27 i257 .391 533 .685 .861 1.066 1.328 1.729 2.093 2.539 2.861
20 .I27 .257 .391 .533 .687 ,860 1.064 1.325 1.725 2.086 2.528 2.845
21 .I27 .257 .391 ,532 .686 ,859 1.063 1.323 1.721 2.080 2.518 2.831
22 .I27 .256 .390 .532 .686 .858 1.061 1.321 1.717 2.074 2.508 2.819
23 .I27 .256 .390 .532 .685 .858 1.060 1.319 1.714 2.069 2.500 2.807
24 .I27 .256 .390 ,531 ,685 .857 1.059 1.318 1.711 2.064 2.492 2.797
25 .127 .256 .390 .531 .684 .856 1.058 1.316 1.708 2.060 2.485 2.787
26 .127 .256 .390 ,531 .684 .856 1.058 1.315 1.706 2.056 2.479 2.779
27 .I27 .256 .389 .531 ,684 .S55 1.057 1.314 1.703 2.052 2.473 2.771
25 .I27 .256 .389 .530 .683 ,855 1.056 1.313 1.701 2.048 2.467 2.763
29 .I27 .256 .389 .530 .683 .854 1055 1.311 1.699 2.045 2.462 2.756
30 .I27 .256 .389 530 .6S3 .S54 1.055 1.310 1.697 2.042 2.457 2.750
40 .I26 .255 .3SS .529 .681 351 1.050 1.303 1.684 2.021 2.423 2.704
60 .I26 .254 .387 527 ,679 .84S 1.046 1.296 1.671 2.000 2.390 2.660
120 .126 .254 .3S6 .526 .677 .S45 1.041 1.289 1.658 1.980 2.358 2.617
00 .I26 .253 ,.385 .524 .674 .842 1.036 1.282 1.645 1.960 2.326 2.576

h i s table is taken from Table 111 of Fisher LQ Yates : Statistical Tables forBiol&ical, Agricultural
and Medical Researcli, published by Longman Group Ltd. London (previously published by Oliver
and Boyd Ltd.. Edinburgh) and by permission o f the authors and publishers.
 Appendix: StstlPtiarlTables

Table of the 0.95 Quantlle ofthe F Dlstrlbutlon

If X has an F distribution with n l and n degrees of freedom the table gives the value of v such that P, (N s x ) = 0.975.

1 161.4 199.5 215.7 224.6 230.2 234.0 236.8 238.9 240.5 241.9 245.9 248.0 250.1 251.1 252.2 253.3 254.3
2 18.51 19.00 19.16 19.25 19.30 19.33 19.35 19.37 19.38 19.40 19.43 19.45 19.46 19.47 19.48 19.49 19.50
3 10.13 9.55 9.28.9.12 9.01 8.94 8.89 8.85 8.81 8.79 8.70 8.66 8.62 8.59 8.57 8.55 8.53
4 7.71 6.94 6.59 6.39 6.26 6.16 6.09 6.04 6.00 5.96 586 5.80 5.75 5.72 5.69 5.66 5.63
5 6.61 5.69 5.41 5.19 5.05 4.95 4.88 4.82 4.77 4.74 4.62 4.56 4.50 4.46 4.43 4.40 4.36
6 5.99 5.14 4.76 4.53 4.39 4.28 4.21 4.15 4.10 4.06 3.94 3.87 3.81 3.77 3.74 3.70 3.67
7 5.59 4.74 4.35 4.12 3.97 3.87 3.79 3.73 3.68 3.64 3.51 3.44 3.38 3.34 3.30 3.27 3.23
8 5.32 4.46 4.07 3.84 3.69 3.58 3.50 3.44 3.39 3.35 3.22 3.15 308 3.04 3.01 2.97 2.93
9 5.12 4.26 3.86 3.63 3.48 3.37 3.29 3.23 3.18 3.14 3.01 2.94 2.86 2.83 2.79 2.75 2.71
10 4.96 4.10 3.71 3.48 3.33 3.22 3.14 3.07 3.02 2.98 2.85 2.77 2.70 2.66 2.62'2.58 2.54
15 4.54 3.68 3.29 3.06 2.90 2.79 2.71 2.64 2.59 2.54 2.40 2.33 2.25 2.20 2.16 2.11 2.07
20 4.35 3.49 3.10 2.87 2.71 2.60 2.51 2.45 2.39 2.35 2.20 2.12 2.04 1.99 1.95 1.90 1.84
30 4.17 3.32 2.92 2.69 2.53 2.42 2.33 2.26 2.21 2.16 2.01 1.93 1.84 1.79 1.74 1.68 1.62
40 4.08 3.23 2.84 2.61 2.45 2.34 2.25 2.18 2.12 2.05 1.92 1.84 1.74 1!69 1.64 1.58 1.51
60 4.00 3.15 2.76 2.53 2.36 2.25 2.17 2.10 2.04 1.99 1.84 1.75 1.65 1.59 1.53 1.47 1.39
120 3.92 3.07 2.68 2.45 2.29 2.17 2.09 2.02 1.96 1.91 1.75 1.66 1.55 1.50 1.43 1.35 1.25
00 3.84 3.00 2.60 2.36 2.21 2.10 2.01 1.94 1.88 1.83 1.67 1.57 1.46 1.39 1.32 1.25 1.00

Adapled with permission from Biomefrika Tables for Sfafisticia~is,Vol. 1, 3rd ed., Cambridge University Prcss, 1966, edited by
E.S. Pearson and H.O. Hartley
If X has an F distribution with m and n degrees of freedom, the table gives the vnlue of x such that P,(X s x) = 0.975

2 3 4 5 6 7 8 9 10 15 20 30 40 60 120 a0

1 647.8 799.5 864.2 899.6 921.8 937.1 948.2 956.7 963.3 968.6 984.9 993.1 1001 1006 1010 1014 1018
2 38.51 39.00 39.17 39.25 39.30 39.33 39.36 39.37 39.30 39.40 39.43 39.45 39.46 39.47 39.48 38.49 39.50
3 17.44 16.04 15.44 15.10 14.88 14.73 14.62 14.54 14.47 1442 .14.25 14.17 14.08 14.04 13.99 13.95 13.90
4 12.22 10.65 9.98 9.60 9.36 9.20 9.07 8.98 8.00 8.84 8.66 8.56 8.46 8.41 8.36 8.31 8.26
5 10.01 8.43 7.76 7.39 7.15 6.98 6.85 6.7 668 662 6.43 6.33 6.23 6.18 6.12 6.07 6.02
6 8.81 7.26 6.60 6.23 5.99 5.82 5.70 5.60 6.52 5.46 5.27 5.17 5.07 5.01 4.96 4.90 4.85
7 8.07 6.54 5.19 5.52 5.29 5.12 4.90 4.90 4.82 4.76 4.57 4.47 4.36 4.31 4.25 4.20 4.14
8 7.57 6.06 5.42 5.05 C82 4.65 4.53 4.43 4.36 4.30 4.10 4.00 3.89 3.84 3.78 3.73 3.61
9 7.21 5.71 5.08 4.72 4.48 4.32 4.20 4.10 4.03 3.96 3.77 3.68 3.56 3.51 3.45 3.39 3.33
10 6.94 5.46 4.83 4.47 4.24 4.07 3.95 3.85 3.78 3.72 3.52 3.42 3.31 3.26 3.20 3.14 3.08
15 6.20 4.77 4.15 3.80 3.58 3.41 3.29 3.20 3.12 .3.06 2.86 2.76 2.64 2.59 2.52 2.246 2.40
20 5.87 4.46 3.86 3.51 3.29 3.13 3.01 2.91 2.84 2.77 2.57 2.46 2.35 2.29 2.22 2.16 2.09
30 5.57 4.18 3.59 3.25 3.03 2.87 2.75 2.65 2.57 2 51 2.31 2.20 2.07 2.01 1.94 1.87 1.79
40 5.42 4.05 3.46 3.13 2.90 .2.74 2.62 2.53 2.45 2.39 2.13 i.07 1.94 1.88 1.80 1.72 1.64
60 5.29 3.93 3.34 3.01 2.79 2.63 2.51 2.41 2.33 2.27 2.06 1.94 152 1.74 1.67 1.58 1.48
120 5.15 3.80 3.23 2.89 2.67 2.52 2.39 2.30 2 22 2.16 1.94 1.82 169 1.61 1.53 1.43 1.31
m 5.02 3.69 3.12 2.79 2.57 2.41 2.29 2.19 2.11 2.05 1.83 1.71' 1.57 1.45 1.39 1.27 1.00

Adapted with permission from Biometr.ika Tables Jor Stafisficians, Vol. 1, 3rd ed., Ca;lbridge llniversity Press, 1966, edited by
E.S.Pearson and H.O. Hartley.
I

ERRATA
(MTE-11,
Block 1)

Page No. Line No. Should be

15 7 .... similarly the value 5.3

15 8 ....between 5.25 and 5.35. Thus, ....as 4.6-5.3 in

15 9 .... and ends at 5.35 ......

34 E 1) b) show that Fi = n - F;', ,(for i > k - 1)
. a .

51' last line s2 = ... = 133490.18

61 2 (from below) deleted

62 6 (from below)
-x = A + c v ; ,......
/-

14 (from below)

1 73 2 (from below) last column gives the ......

84 6 (from below) and Y, = f,,~,

J

90 Fig. 6 The top most should be Group 1, and the lowest

group should be Group 3