Sei sulla pagina 1di 6

Statistics

Section – I Various measures of dispersion, minimal property of mean deviation, Root mean square
deviation, variance and S.D., Moment about mean, origin and any point. Skewness and Kurtosis, Pearson’s
β and γ coefficients.

Dispersion: The measure of central tendency are inadequate to give in a complete idea of the distribution.
[E.g. mean of (i) 7,8,9,10,11 (ii) 3,6,9,12,15 (iii) 1,5,9,13,17 is 9 and n for all of them is 5] They must be
supported and supplemented by some other measures. One such measure is Dispersion.
Literal meaning of dispersion is scattered ness. It is studied to have an idea about
homogeneity or heterogeneity of the distribution. We see that distribution (iii) above is more heterogeneous
(more scattered) than series (i) or (ii)
Characteristics for an ideal measure of Dispersion (same as those for an ideal measure of central
tendency) viz:
(a) It should be rigidly defined
(b) It should be easy to calculate and easy to understand
(c) It should be based on all the observation
(d) It should be amendable to further mathematical treatment
(e) It should be affected as little as possible by fluctuations of sampling
Measure of Dispersion
(a) Range
(b) Quartile deviation or semi-interquartile range
(c) Mean deviation and
(d) Standard deviation
(a) Range: It is simplest but crude measure of dispersion. Since it is based on two extreme observations
which themselves are subject to chance fluctuations, it is not at all a reliable measure of dispersion.
If A = greatest observation in a distribution
B = smallest observation in the distribution
then Range = A − B
(b) Quartile Deviation: (Or semi- interquartile Range) − Q : It is given by
Q = ½ (Q3 − Q1)
Where Q1 = first quartile of distribution and
Q3 = Third quartile of distribution
It is better than range in the sense that it makes use of 50% of the data. But since it ignores the other 50% the
data, it can not be regarded as reliable measure.
(c) Mean Deviation: If xi/fi, i = 1,2,….., n is the frequency distribution, then mean deviation from average a
(usually mean, median or mode) is given by
Mean Deviation = 1/N Σ fi |xi − A|, Σ fi = N
Where |xi − A| represents the modulus of deviation.
It is better than range and quartile deviation as it is based on all the observations. But ignorance of sign of
(xi − A) create artificiality and renders it useless for further mathematical treatment.
Note: Mean deviation is least, when taken from median.
(d) Standard Deviation and Root Mean Square Deviation (σ ): for a frequency distribution xi/fi, i = 1,2,
….., n it is given σ = √{1/N Σ fi(xi - X)2}
where X = arithmetic mean of distribution and Σ fi = N
Here the drawback of ignoring sign of (xi – X) is overcome. It is also suitable for further mathematical
treatment. Moreover, it is least affected by fluctuations of sampling.
Note: S.D. given greater weight to extreme values and as such has not found favour with economists
or businessman, who are more interested in results of the modal class.
Variance = σ 2 = 1/N Σ fi(xi - X)2
Root mean square deviation S = √{1/N Σ fi(xi - A)2}
Where A = any arbitrary no
Relation between σ and S:
We have S2 = 1/N Σ fi (xi − A)2
= 1/N Σ fi (xi − X + X − A)2
= 1/N Σ fi [(xi − X)2 + (xi − A)2 + 2(xi − X)(xi − A)]
= 1/N Σ fi (xi − X)2 + (xi − A)2 1/N Σ fi + 2(xi − A) Σ fi (xi − X)
= σ 2 + (X - A)2 {1/N Σ fi = 1 & Σ fi(xi - X) = 0}
=σ +d2 2
where d = X − A
Note: S2 will be least when d = 0 i.e. X = A
⇒ S2 & S is least when the deviations are taken from A = X
⇒ Variance is the minimum value of mean square deviation OR standard deviation is the minimum
value of root mean square deviation.

Different formulae for calculating variance


By def. We have σ 2x = 1/N Σ fi (xi – X)2 { σ 2x = variance of x}
= 1/N Σ fi (xi2 – X2 − 2xiX)
= 1/N Σ fi xi2 + X2 . 1/N Σ fi – 2X . 1/NΣ fixi
= 1/N Σ fixi2 + X2 – 2 . X2 {As 1/NΣ fi =1 & 1/N Σ fixi =
X}
= 1/N Σ fixi2 – X2

σ 2x = 1/N Σ fixi2 – (1/NΣ fixi)2


If x & f are large, we have
σ 2x = 1/N Σ fi (xi – X)2 = 1/N Σ fi (xi − A + A – X)2
= 1/N Σ fi(di + A – X)2, where di = xi − A
= 1/N Σ 2 2
fi (di + (A - X) – 2(A - X)di)
= 1/N Σ fidi2 + (A - X)2 . 1/NΣ fi – 2(A -X) . 1/N Σ fidi
= 1/N Σ fidi2 + (− 1/N Σ fidi)2 + 2(− 1/NΣ fidi)(1/NΣ fidi)
= 1/N Σ fidi2 – (1/N Σ fidi)2 = σ 2d {As X = A+1/NΣ fidi }
{ ∴ A – X = −1/NΣ fidi}
⇒ Variance and consequently standard deviation in independent of change of origin.
If we take di = (xi - A)/h so that xi – A hdi, then
σ 2x = 1/N Σ fi (xi – X)2 = 1/N Σ fi(xi – A + A – X)2
= 1/N Σ fi(h di + A – X)2
= h2 1/N Σ fidi + (A - X)2 + 2(A - X) h . 1/N Σ fidi
using X = A = h. Σ fidi/N, we get
σ x = h2 [q/N Σ fidi2 − (1/NΣ fidi)2] = h2σ 2d
2

⇒ Variance is not independent of change of scale.

Some solved examples


Ex 1 Calculate mean and S.D. for the following table giving are distribution of 542 members
Age in years: 20−30 30-40 40-50 50-60 60-70 70-80 80-90
No. of students: 3 61 132 153 140 51 2

Solution:
We take d= (x - A)/h = (x – 55)/10

Age group Mid value Frequency d = x–55 fd fd2


(x) (f) 10
20-30 25 3 −3 −9 27
30-40 35 61 −2 −122 244
40-50 45 132 −1 −132 132
50-60 55 153 0 0 0
60-70 65 140 1 140 140
70-80 75 51 2 102 204
80-90 85 2 3 6 18
N = Σ F = 542 Σ fd = −15 Σ fd = 765
2

X = A + h (Σ fd/N) = 55 + 10(-15)/542 = 55 – 0.28 = 54.72 yrs. Ans

σ 2 = h2 [1/N Σ fd2 – (1/N Σ fd)2] = 100 [765/542 – (−15/542)2] = 100 [765/542 – (0.028)2]
= 100 × 1.4107 = 141.07
σ = √(141.07) = 11.9 yrs Ans

Example 2 Prove that for any discrete distribution standard deviation is not less than mean deviation
form mean .
Solution Let xi/fi, i = 1,2,3,..….. n is any discrete distribution, then we have to prove that S.D. <!
Mean deviation from mean.
⇒ (S.D.)2 <! (Mean deviation from mean)2 ⇒ (S.D.)2 ≥ (Mean deviation from mean)2
⇒ 1/N Σ fi (xi – X) ≥ [1/N Σ fi | xi − X|]2
i.e. 1/N Σ fi (zi)2 ≥ (1/N Σ fizi)2 where zi = xi – X
i.e. 1/N Σ fizi – (1/N Σ fi zi) ≥ 0
2 2

i.e. 1/N Σ fi (zi − Z)2 ≥ 0


i.e. σ 22 ≥ 0
Which is always true. Hence the result.
Note: 1/N Σ fi (zi − Z)2
1/N Σ fi (zi2 + Z2 – 2zi Z)
1/N Σ fizi2 + Z . 1/N Σ fi − 2Z 1/N Σ fizi
1/N Σ fizi2 + Z2 – 2 Z Z
1/N Σ fi zi – Z2
1/N Σ fi zi – (1/N Σ fi zi)2

Example 3 Find the mean deviation from mean and standard deviation of A.P. a, a + d, a + 2d, ….a +2nd
and verify that the latter is greater than the former.
Solution Mean of an A.P. = mean of its first and last term
= ½ [ a + a + 2nd] = a + nd

xi |xi − X| (xi − X)2


a nd n2d2
a+d (n -1)d (n -1)2d2
a + 2d (n - 2)d (n - 2)2d2

a + (n-2)d 2d 22d2
a + (n-1)d d d2
a + nd 0 0
a + (n+1)d d d2
a + (n+2)d 2d 22d2

a + (2n -2)d (n−2)d (n−2)2d2


a + (2n -1)d (n−1)d (n−1)2d2
a + 2nd nd n2d2

Mean deviation from mean


= 1/(2n+1) Σ | xi − X|
= 2d[1+2+3+….+n]
2n+1
= n(n+ 1)2d = n(n+ 1)d
2(2n+1) 2n + 1
σ = 1/(2n+1) Σ (xi - X)
2 2

= 1/(2n+1) . 2d2[12 + 22 + 32 + ….. + n2]


= 2d2 . (n)(n + 1)(2n + 1)
6(2n + 1)
= n(n+ 1)
3
Hence S.D. = √{[n(n+1)d2]/3}
Verification S.D. > (M.D from mean)2
If (S.D)2 > (M.D from mean)2
if {[n(n+1)d2]/3}> {[n(n+1)d]/2n+1}2
if (2n + 1)2 > 3n(n+1)
if 4n2 + 1 + 4n – 3n2 – 3n > 0
if n2 + n + 1 > 0
if (n + ½ )2 + ¾ > 0
which is always true

Example 4 Show that in a discrete series of deviations are small compared with mean M so that (x/m)3
and height powers of x/M are neglected, we have
(i) G = M (1 – ½ σ 2/M2)
(ii) M2 − G2 = σ 2 and
(iii) H = M(1 − σ 2/M2)
Solution Let xi/fi , i = 1,2,3,..….. n is given freq. Distribution. Then we are given that xi = Xi – M
i.e. xi = xi − M
Σ fi xi = Σ fi (xi − M) = 0 (1)
Also Σ fixi2 = Σ fi (xi – M)2 = σ 2 (2)
By def we have
G = (x1fi x2f2 …. xnfn)1/N where N = Σ fi
Log G = 1/N Σ fi log xi
= 1/N Σ fi log(xi + M)
= 1/N Σ fi log [M + (1+ xi/M)]
= 1/N Σ fi [log M + log (1 + xi/M)]
= log M + 1/N Σ fi log (1+xi/M)
= log M + 1/N Σ fi [xi/M – ½ xi2/M2 + 1/3(xi/M)3 …..]
log G = log M + i/NM Σ fixi – 1/2M2 . 1/N Σ fi xi2
= log M - σ 2/2M2 On using (1) & (2)
−σ 2/2M2
= log M e
−σ 2 2
G = M e /2M = M (1 − σ 2/2M2) (neglecting higher power)
Hence G = M (1 − ½ σ 2/M2)

Example 5 In a survey of 950 families in a village, the following distribution of number of children was
obtained
No of children 0-2 2-4 4-6 6-8 8-10 10-12
No of families 272 328 205 120 15 10
Find the standard deviation of this distribution
Let A = 5 here h = 2
Solution
C.I. Mid value Freq di = (xi-A)/h fidi fidi2
0-2 1 272 −2 -544 10088
2-4 3 328 −1 -328 328
4-6 5 205 0 0 0
6-8 7 120 1 120 120
8-10 9 15 2 30 60
10-12 11 10 3 30 90
950 −692 1686

Mean = A (1/N Σ fidi)h


= 5 + 1/950 (-692) × 2
= 5 – 1384/950
= 5 – 1.457
= 3. 543
S.D. = σ = h × √[Σ fidi2/N – (Σ fidi/N)2]
= 2 × √(1686/950 − (-692/950)2)
= 2 × √(1.7747 − (− 0.7284)2)
= 2 × √(1.7747 − 0.5306)
= 2 √(1.2441)
= 2.28

Example 6 Goals scored by two teams A and B in a football season were as follows:
No of goal scored in match No of Matches
A B
0 27 17
1 9 9
2 8 6
3 5 5
4 4 3
Find out which team is more consistent
Solution
C.V. for the team A C.V. for the team B
2
Xi fi di fidi fidi xi fi di fidi fidi2
0 27 −2 −54 108 0 17 −2 −34 68
1 9 −1 −9 9 1 9 −1 −9 9
2 8 0 0 0 2 6 0 0 0
3 5 1 5 1 3 5 1 5 5
4 4 2 8 56 4 3 2 6 12
53 −50 138 40 −32 94

Mean = Mean
A + (Σ fidi)/N = 2 – 50/53 = 2 − 0.94 = 1.06 A + (Σ fidi)/N = 2 – 32/40 = 2 − 0.8 =
1.2

Now σ = √[Σ fidi2/N – (Σ fidi/N)2] Now σ = √[Σ fidi2/N –


2
(Σ fidi/N) ]
= √(138/53 − (-50/53)2) = √(94/40 − (-32/40)2)
= 1.31 = 1.3
Coff. Of variation = σ /X × 100 Coff. Of variation = σ /X × 100
= (1.31 × 100) / 1.06 = (1.3 × 100) / 1.2
= 123.6 = 108.3
As C.V. for the team B, so team B is more consistent

Example 7 If the mean and s.d. of a variable x are m and σ respectively, obtain the mean and S.D. of
(ax+b)/c where a, b, and c are constants.
Solution Let u = (ax + b)/c
Let u and σ u be the mean and S.D of u respectively
Then U = q/N Σ fi (axi + b)/c = 1/c {a 1/N Σ fixi + b .1/N Σ fi}
= a/c X + b/c = (am + B)/c
Now u – U = a/c(x - X) {As σ is independent of change of origin}
∴ σ u = 1/NΣ fi(u - U) = a /c . 1/N Σ fi (xi – X)2 = a2/c2 σ 2
2 2 2 2

⇒ σ u = |a/c|σ and U = (am+b)/c

Potrebbero piacerti anche