'

STANDARD ERROR OF ESTIM ATE _

With the help of regress ion equatio ns perfect prediction is practically

impossible. For examp le, the revenu e for the year from gasoline sales ()1
based on automobile registr ation (X) as of· a certain date would· no doubt
be approximated fairly closely, but prediction may not be exact to. the
nearest rupee nor probab ly to the neares t '000 rupees. What is needed.
~en, is a measur e which would indicat e how precise the prediction of Y
18 • based on X or conver sely, how inaccu rate the prediction might be. The
me~sure is called the standa rd error of estimate. The standard error .of
t:Stirnate, symbolised by S is the same concep t as the stand ahrd ddevia-
• er-
ti . yx ' .
:on discusse d in· Chapte r 8 The standa rd deviati on measur es t . eestima ~sp t
e
s1on b • • d d· r1·or ot
a out an average such as the mean. The stan ar e • line
•
measure~ th . •. \ • . 11 d the regression
The . ..:, e dispers ion about an aver:ig e hne. ca ~ . .
formula for calcula tin the standa rd error of estimat e
15
~
r (Y - Yc) 2 ✓une~lained Vartatiof!_
S,,.x=
::, - -N- - , i·. e ., - - N
also ✓ 2
Where Sy-x = cry 1 - r . . f yvatue s from Ye- . w
lbis fo Sy•x_= t~e_.~t andard error of ~egre~s10n o utational point 1o~a~!
becaus . nnuia 1s not ·conve nient from v the. comp venient forn u
e 1t req -· . . - -) .Kttiore con
uires the compu tation· of (Y - 1 c • •· .·

S - ✓.L
2
.Y - a-r Y - b 'f, X y

Y·x- • . N
 470

va lue s from Xe is
The sta nd ard err or of regressio2n of X 2
-✓!.(X-Xc) al so S. =c l' ✓I -r
Sx .y- . N x-y

✓J:,X 2 - a! .X - bl .X Y
N
Also Sx, y=
n of X va lue s from Xe-
Sx-y = the sta nd ard err or of reg res sio
of est im ate me as ur es th e ac cu rac y of th e estfmateci
The sta nd ard err or err or of es tim ate , th e clo se r Will be
e sm all er the va lue of sta nd ard
figures. Th
reg res sio n lin e an d th e be tte r th e es tim ate s ba se d on th
the dots to the
thi s lin e. If sta nd ard err or of es tim ate ts ze ro, th en the re is n'0
equ3.tion for lat ion wi ll be pe rfe ct. Th us wi th the hel
variation ab ou t the lin e an d the co rre
es tim ate it is po ssi ble for us to as ce rta in ho w good an~
of standard error of as a de sc rip tio n of th e av era ge relation.
representative the reg res sio n lin e is
sh ip between two ser ies .
•- ,. .
• • . \ I'

data : 8
Illustration 14. Given the following 10 .. 1 4
X 6 2 7
11 5 8 ' .J

Y ·g
ression equations and calculate the standar~ error of the estimate (Sy-x and
Find the two reg . •. . , ~ "'
•
- . . • -
5x-y). ns are :
Refer to pag e E-1 1 .11 , Illustra tion 2, the two regression equatio
Solution. . - -.
- Y = 11.9 - 0..65 X and X = 16.4 - :1.3 Y out
sion equatio n of Y on X for various values . of X we ca n find fhe
From th_e regres n of X on y we can find ou t x e- Th_ese vaIues are
g Y values , and from the equ atio .
correspondin
as follows :
2 2
y Ye Xe (Y - Ye) (X -X c)
X
8.0 4.7 1.00 1.69
6 9
2.1 0.16 0.01 -_
2 11 10.6 ~ ,.

10 .. 5 5.4 9.9 0.16 ·· · ·. · : .· 0.01 ~

• .t - • •

4 8 9.3 6.0 ' •. · , 40 0

\ .
8 . 7· 6.7. 7.3 ~- : .. 0.49
I Ye= 40 I Xe= 30 • : •• 2 2
I (X - Xc) = 6.20
IX =3 0 I Y= 40 I ( Y - Ye) = 3.1 :=,
.
• s··~~ ✓ l.(Y-Yc)2 Ar;;-.·_
V~ =,/0.62 =0.787
•

y · N . =

Sx-y = ✓(IX - Xe)2 =~V~ •

5= ✓1.2
,
4=
•
1. 114
. N
494 STATISTICAL METHoos •

LIST OF FORMULAE
Regressio n Equation ofX on Y :
• X - X = bxy (Y - Y)
C1x '1:xy .
bxy or r ay = l: ,j .
(if deviations are taken from actual means of X and Y)
('l: dx) (l: dy)
L dxdy N
N l: dx dy - (1; dx) (l; dy)
b - - - - - - - - - - - - - or
~- (l: d ) 2 N "1: d~ - (l; dy) 2
"£d2 y
y N
(if deviations are taken from assumed means of X and Y)
Regressio n Equation of Yon X:
Y - Y = byx (X - X)
cry '1: xy
byx or r - = - - (Deviations taken from actual mean)
O"x l:~
(1: dx) (1: dy)
d
l: xdy- N • N:f.dxdy -(:f.dx)( I;d)
~=-- ----- -oc Y
l:: d~ - (l:: dx)2 N :f. cf; - ( :f. d,J2
N
Regressio n CoeJficients :
(Deviations taken from assumed mean)
. C1x
bxy or r - is the regression coefficient of X on Y.
C1y . .
(1
byx or r c/; is the regression coefficient of Yon x
r= ✓bxy x byx
Standard Error ofEstimate :·
_ ✓~ (X - XJ2 • • 1
S
. x.y-+ N ·also Sx•y=crxV1 ..-r 2 .

Sy.x = + ✓1; (Y; Yd2.' : also Sy,x = ay ✓1~-: r 2