ERRORS I N RESEARCH To a lar ge ext ent t he m aj or er ror sam ple and t his sam ple is not

com ponent s ar e r elat ed t o each r epr esent at ive of t he popula-

D ESI GN
ot her in an inverse m anner. I n- t ion concerned. For exam ple,
creasing t he sam ple size t o reduce a random sam ple of 500 peo-
A useful way of looking at t he
t he sam pling er ror can lead t o an ple com posed only of people
m arket ing research process is t hat
increase in non- sam pling er ror . I n bet ween 35- 50 years of age
involves t he m anagem ent of er-
m ost st at ist ical proj ect s – t he ef- m ay not be repr esent at ive of
rors. At all st ages, from problem
for t is t o m inim ise t he t ot al error adult populat ion. Sam pling
form ulat ion t o report present at ion
by m inim ising t he sam pling error er ror is affect ed by t he hom o-
– errors can arise. I t is rare that a
and assum ing t hat good training geneit y of t he populat ion un-
research proj ect will be error free.
of t he invest igat ors will reduce t he der st udy. I n gener al – m ore
Consequent ly t he research de-
non- sam pling error. However in hom ogenous t he populat ion;
signer m ust adopt a st rat egy for
m arket research t hat m ay not be t he sm aller t he sam pling er-
m anaging and m aint aining t he
t he possible as t he t im e and cost ror. Sam pling error falls to
error. We shall first look at t he
const raint s m ay pr event choosing zero in case of a census.
com ponent of errors and t hen t he
a sam ple size which will allow us
t ype of errors.
t o be confident t hat t he sam pling 3. Se le ct ion Er r or : I t is defined
error ( and hence t ot al err or ) has as t he “ Non correspondence
COM PON EN TS OF ERRORS
been m inim ised. I deally a m ar ket of sam ple select ed by non
researcher should strive to m ini- probabilit y m eans and t he
The obj ect ive underlying any re-
m ise bot h t he t ypes of error. Con- r epr esent at ive sam ple sought
sear ch pr oj ect is t o provide infor-
sidering t he t im e and cost lim it a- by t he resear cher .” I n ot her
m at ion t hat is t o be accurat e and
t ions t his can rarely be done. The words – it is a sam pling error
error free as possible. Maxim ising
m arket r esearcher m ust m ake a for t he sam ple select ed by
accuracy requires t hat “ t ot al er-
decision t hat involves a t r ade off non- probabilit y m et hods.
ror” be m inim ised. Tot al error has
bet ween sam pling and non- There is a nat ural t endency
t wo dist inct com ponent s:
sam pling errors. Unfort unat ely for t he invest igator t o select
very lit t le is known about t he rela- t hose respondent s who are
Tot al Error = Sam pling Error +
t ive size of t wo error com ponent s m ost accessible and agree-
Non Sam pling Error
– alt hough som e em pirical evi- able. Such sam ples are oft en
dence suggest s t hat non- sam pling com prised of friends and as-
Tot al error is usually m easured as
errors t end t o be lar ger of t he t wo sociat es who m or e or less
t ot al error variance – also know as
com ponent s. Sam pling er rors can represent t he desired popula-
t he m ean squared error:
t o a large ext ent be reduced t ion. Sam ples of t his t ype
and/ or cont rolled by following m ay not be repr esent at ive of
( Tot al Error) 2 = ( Sam pling Error) 2
st at ist ical procedures, but such a t he populat ion but are cer-
+ ( Non Sam pling Error) 2
check is gener ally not possible on t ainly not select ed r andom ly –
non- sam pling errors leading t o st at ist ical problem s
Sa m plin g Er r or refers t o t he
in inferencing.
variable error result ing from
TYPES OF ERRORS:
chance select ion of elem ent s from
4. Fr a m e Er r or : I t is defined as
populat ion as per t he sam pling
1. Popu la t ion Spe cifica t ion t he non- correspondence of
plan. Since it int roduces random
Er r or : I t is defined as t he t he sought sam ple wit h t he
var iabilit y int o t he precision wit h
“ non correspondence of t he required sam ple. A sam pling
which sam ple st at ist ics are calcu-
required populat ion t o t he fram e is a m eans of account -
lat ed it is also called random sam -
populat ion select ed by t he in- ing for all t he elem ent s of t he
pling err or . N on Sa m pling Er r or
vest igat or ” . I t occurs when a populat ion. I t is usually a list -
consist s of all ot her error associ-
researcher select s an inap- ing of all t he elem ent s t hat
at ed wit h t he research proj ect .
propriat e populat ion from ident ify a populat ion. A per-
Such errors are diverse in nat ure.
which t o obt ain dat a. For ex- fect fr am e ident ifies each
They are oft en t hought of as bias.
am ple m any t im es package populat ion only once and does
However bias is a t ype of syst em -
goods m anufact urers conduct not include elem ent s of ot her
at ic er r or which ent ers int o t he
surveys am ongst housewives populat ion, which m ay lead t o
process because of un- calibrat ed
– because t hey are easy to overlap. For exam ple a sam -
inst rum ent s or prej udices of t he
cont act and because it is as- pling fram e t hat specifies t he
researcher. There can be com -
sum ed t hat as t hey are t he populat ion as all adult s using
plet ely random com ponents of
end users t hey m ake t he pur- oral care product s – m ay well
non- sam pling error. For exam ple a
chase decisions. This assum p- leave out people who prefer
m isrecording of a response during
t ion m ay not be always valid using neem and babool st icks,
dat a collect ion represent s a ran-
since husbands and children hom em ade t oot hpast e and
dom non- sam pling error. However
m ay significant ly influence t he powders – leading t o a fram e
using deliberat e “ loaded” quest ion
buying decisions. er ror in defining t he popula-
t o get desir ed response – which
t ion.
m aint aining t he st at us quo of ran-
2. Sa m plin g Er r or : I t is defined
dom ness – is called bias. Thus
as t he “ non correspondence 5. N on Re spon se Er r or : A non-
non- sam pling errors can arise
of sam ple select ed by prob- response error occurs, when
from var ious sources – but in
abilit y m eans and t he repre- t he obt ained sam ple differs
m arket ing research proj ect t hey
sent at ive sam ple select ed by from t he original select ed
arise m ainly from response and
t he r esearcher” . I t occurs sam ple. Non response can oc-
non- response.
when a probabilit y sam pling cur in t wo ways – ( a) Non-
m et hod is used t o select a Cont act i.e. t he inabilit y t o

© Rohit Vishal Kum ar - 1 - MBA Evening - I I SWBM

 cont act all t he m em bers of m at ion because based on lead t o an inclusion of bias in
t he sam ple and ( b) Refusal t hese inform at ion we t ry to t he exper im ent s. For exam ple
i.e. t he non- response of som e predict t he fut ure behaviour a st udy – t rying t o find out
or all t he it em s of t he m eas- of t he respondent s. Secon- reasons for cancer m ay con-
urem ent inst rum ent . dary sources of dat a are an- clude t hat as all t he pat ient s
ot her source of surrogat e in- under observat ion used t o eat
Non- cont act errors arise due form at ion. Surrogat e inform a- bread t herefore eat ing bread
t o t he inabilit y t o reach t he t ion err or can be m inim ised is t he cause of cancer.
respondent . This m ay be be- by ensuring t hat t he inform a-
cause t he respondent is NAH t ion used is highly correlat ed CON CLUSI ON
( Not at Hom e) on t he init ial wit h t he act ual inform at ion
call ( or call backups) or m ay obt ained. For any r esearch proj ect ,
have m oved away from t he recognising t hat pot ent ial error
area eit her perm anent ly or for 7. M e a su r e m e nt Er r or : This exist s is one t hing – but doing
t he period of t he survey. Non m ay be defined as t he non- som et hing about it quit e anot her
Cont act errors can be reduced corr espondence of infor m a- m at t er.
by careful analysis of t he se- t ion obt ained by m easure- Ther e are t wo basic approaches
lect ed sam ple. m ent process and t he infor- for r educing er ror s.
m at ion sought by t he re-
Refusal arises when t he re- sear cher . I t is gener at ed by The first is t o m inim ise errors
spondent does not respond to t he m easur em ent pr ocess it - t hrough research design. I n t his
a par t icular it em or t o m ult i- self and represent s t he differ- process effect ive use of r esearch
ple it em s of t he quest ion- ence bet ween inform at ion m et hods and t echniques are ut il-
naire. Mont hly Household I n- generat ed and inform at ion ised t o lessen t he im pact of bot h
com e, Religion, Sex and Poli- want ed by t he r esear cher. sam pling and non- sam pling er-
t ics ar e t opics t hat m ay elicit Such errors can pot ent ially rors. However, cost const raint s,
it em r efusal – norm ally cat e- arise at any st age of t he and at t im es t he peculiar nat ur e of
gorised as refused. Don’t m easur em ent pr ocess – from error t hey prevent com plet e
Know / Can’t Say ( DKCS) r e- t he developm ent of t he in- m inim isat ion of error t hrough t his
fusals ar ise when t he respon- st rum ent t ill t he analysis of m et hod.
dent is aware of t he it em but findings.
is unable t o provide an ade- The second is t o est im at e and
quat e answer t o t he it em . The error can also occur at m easur e err or . I n spit e of all t he
Const raint s of t im e on t he t ransm it t al st age - when t he precaut ions undert aken, not all
part of t he respondent , a bad int erviewer is quest ioning t he er rors – especially t hose r elated t o
experience wit h som e previ- respondent . Fault y wording of fieldwork – would be elim inat ed.
ous int erviewer, ill healt h et c. quest ion, non- preparat ion of I n such a sit uat ion if we can have
m ay be ot her reason for re- non- verbal clues, behaviour of an est im at e of err or we can say
fusal. Refusal rat es could be t he int er viewer et c m ay all how accurat e t he research design
brought down t o m anageable cont ribut e to how t he respon- was. However only sam pling er-
levels by giving t raining t o t he dent int erpret s t he quest ion. rors are m easurable w it h som e
int erviewers and cont inuously degr ee of confidence.
m onit oring t he invest igat ion I n t he response phase – when
process. t he r espondent is r eplying – Eit her way, est im at ing is not an
error m ay occur because t he easy t ask due t o t he peculiar na-
6. Su r r oga t e I n form a t ion Er- respondent gave a wrong an- t ur e of t he er ror s. St at ist ics help
r or : This is defined as t he swer or t he corr ect answer us t o reduce t he sam pling error t o
non- correspondence of t he in- was wrongly int er pr et ed and a large degree but for non-
form at ion being sought by t he recorded. I n t he analysis sam pling error researchers st ill
researcher and t hat is re- phase, errors of incorrect have t o rely on t heir int uit ion.
quired t o solve t he problem . edit ing, coding and/ or
Or in ot her words, inform at ion descript ive sum m arisat ion
is obt ained from subst it ut es and infer ences can cont r ibute
rat her t han original sam ple. subst ant ially t o m easurem ent
The necessit y to accept sur- error .
r ogat e inform at ion arises 8. Ex pe r im en t a l Er r or : I t can
from eit her t he inabilit y or t he be defined as t he non-
unwillingness of t he respon- correspondence of t he “ t rue
dent t o provide t he needed im pact of” and t he “ im pact at -
infor m at ion. t ribut ed t o” t he independent
var iable. When an exper im ent
Decisional orient ed behav- is conduct ed, t he researcher
ioural research is always con- at t em pt s t o m easure t he im -
cent r at ed wit h predict ion of pact of one or m or e m anipu-
behaviour usually non- verbal. lat ed independent variable on
This lim it s m ost m arket ing re- som e dependent variable –
sear ch proj ect s t o use proxy while cont rolling t he im pact of
inform at ion – dat a from past exogenous var iable. Unfor t u-
behaviour. At t it udes, beliefs nat ely cont rol over all t he ex-
and SEC classificat ion are all ogenous variable is not al-
exam ples of surrogat e infor- ways possible – which m ay

© Rohit Vishal Kum ar - 2 - MBA Evening - I I SWBM