Professional Documents
Culture Documents
BRM Chapter9
BRM Chapter9
BRM Chapter9
ANTITATIVE RESEARC H
QU
198
188 Time and cost 198
Introduction Non-response
ative sample 189 199
The importance of a represent Heterogeneity of the popula
tion
191 g
199
plin
Sampling error Types of non -probability sam 200
193 Convenien ce sam plin g
Types of probability sample 201
193 Quota sam pling
Simple random sample 203
.193
Systematic sample limits to gen eralization
194 20 4
Stratified random sampling fHor in survey research
194
Multi-stage cluster sampling ;eys 204
195 Sampling issues for onli11a iur
sample
The qualities of a probability 20 7
1< • ,r •t:ir.i;r-
197
Sample size 207
size 197 C.uesti ons fer re•;iew
Absolute and relative sample
This chapter is concerned with some important aspects Decide on topic/area to be researched
of conducting a survey, but it presents only a partial pic-
ture, because there are many other steps. In this chapter Review literature/theories relating to to
+
we are concerned with the issues involved in selecting
respondents for survey research, although the principles
Formulate res+rch question ~
+
involved apply equally to other approaches to
quantitative research, such as content analysis. Chapters 9,
10, and 11 deal with the data collection aspects of con- Consider whether a social survey is appropr t
ducting a survey, while Chapters 15 and 16 deal with the (if not, cons,'der an a Iternat ·rve research design) ra e
analysis of data.
Figure 9 .1 is intended to outline the main steps involved
+
Consider what kind of population will be appropriate
in planning and conducting survey research. The plan-
ning will begin with an examination of general research +
Consider what kind of sample design will be employed
issues that are to be investigated. These are gradually nar-
rowed down so that they become research questions,
which may take the form of hypotheses, though this is not
+
Explore whether there is a sampling fra me that can be
necessarily the case. The movement from research issues employed
to research questions is likely to be the result of reading
the literature relating to the issues. Decide on mode of adm inistration
+
Once the research questions have been formulated, (face-to-face, telephone, postal, email, online)
the planning of the fieldwork can begin. In practice,
decisions relating to sampling and to the research instru-
Develop questions (and devise answer
+
ment will overlap, but they are presented in Figure 9 .1 as alternatives fo r closed questions)
'--- - - -- -
part of a sequence. The survey researcher needs to
decide what kind of population is suited to the investiga- +
Review question::; and assess face validity
tion of the topic and also needs to formulate a research ' - - - - --
instrument and decide how it should be administered.
The 'research instrument' may be a structured Pilot questions
+
interview schedule (discussed in Chapter 10) or a
+
self-completion questionnaire (Chapter 11). There are
several different ways of administering such instru-
Revise questions J
ments. Figure 9.2 outlines the main types that are likely
Finalize quest ion naire/schedule
+ J
to be encountered.
Quantitative research almost invariably involves sam-
pling. In this chapter, we will be mostly concerned with L___ s_a_m.. .'p.__l.e_
____
+
.::.. fr_o:__m___::__ _J
po.. .'p.__u_la_t_io_n_ _ _ _
sampling for social survey research involving data col-
lection by structured interview or questionnaire. In
Administer quest ionnaire/schedule to sample
social survey research, sampling constitutes a key step in
the research process, a s illustrated in Figure 9.1. How-
ever, other methods of quantitative research also involve Transform completed questionnaires/schedules
sampling considerations, as will be seen in Chapter 13, into computer-readable data (coding)
where we will examine structured observation and con-
tent analysis. The principles of sampling involved are
r-----___:_
♦--~ Analyse data ~
more or less identical in connection w ith these other
methods, though frequently oth er considerations come r
r -_ __ ♦
...:..___-~
L Interpret fi ndings ______-,
to the fore as well.
Survey
Structured Self-completion
interview questionnaire
t t i ♦
Face-to-face Telephone
Supervised
5
I Po~tal I / internet I
I
t t
Paper + pencil
1
I C~PI I Paper ; pencil
B I
+I
Embedded
7
I
Atta~hed
your findings from your sample to the population from sample? There are various reasons, ofwhich t
which it was selected, the sample must be representative. stand out: he fo)]O\v.
lllg
This is a fundamentally important point, because the aim
of analysing data from a sample is to make inferences • The first two approaches depend heaVi
availability of students during the ti ly Upon th
about the larger population from which the sample was me or ti e
drawn, without having to try to collect data from the
you search them out. Not all students mes that
are lik 1
equally available at that time so the e Yto be
whole population. See Key concept 9 .1 for an explanation ' sample .
of terms concerning sampling. reflect those students who are not available. Will llot
Why might the strategies for sampling students previ- • The first two approaches also depend O h
n t e stud
ously outlined be unlikely to produce a representative going to the locations. Not all students Will ents
necessaruy
Sampling error
be equally split in
• . . . sam pli ng we wo uld ex pect ou r sample of 50 to
In order to appreciate the s1gmf1cance of there is a
r ter ms of thi s va ria ble, as in Figure 9.4. If
error for ach.ieving. conside tha t we ha ve one
F' a representative sample ' am ou nt of sampling err or, so
. ion of sm all
agme we have a po pu lat
7
igures 9 3-9
2oo plo·yee · · Im no t ap praised and one too
d Im ag ine em ployee too many who is
em we want a sample of 50 . . In Figure 9.6
as Well h s, an th . us is few wh o is, it will look like Figure 9.5
t at one of iables of concern to us de gr ee of ov er-
wheth rnotem I e var . rm an ce we se e a ra th er mo re se rio
ero p oyees receive regula r pe rfo no t receive
appraisal fr 0 the i · • d th at rep res en tat ion of employees who do
s m. r 1mmed1ate supervisor, an ee too many who are
the popuJa . 15
do and th tlon equally
h ose wh o do
div ide
.
d betw een
not. This split is repres
tho
en
se
ted
wh o appraisals. This time there are thr
by no t ap pra ise d an d three too few who are
. In Figure 9. 7
entation of employ-
I
t e vertical li
tha t div id h ion int o tw o we ha ve a ve ry ser ious over-rep res
haJVes (see p· ne es t e po . pu lat ce appraisals,
igure 9.3) · If the sample 1s resen tat ive , ee s wh o do no t receive pe rfo rman
rep
19 2 9 Sampling in quantitative research
P,ilj' , H r u Y :Si,!,■ -·
;±1 "dfM dU15W'ISll:Nt111Pi&'l'bf'Ulk jiH!i,N. 1,,11;;.114;,ji
··---·-- ,..,__,.,.__, ,_
Have performance Do not have
Do not have performance
Have performance appraisal Perform
appraisal appraisal a~Ce
appraisal O O O O O O O O O O
O o O O O O O O O 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 O 0 0 O0
0 0 0 0 0 0 0 0 0 0 Oo oo o o
0 O O O O O O O O 0
0 0 0 0 0 0 0 0 0 O 0 Oo
O o O o O O O O O 0 Ooo o o o
0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 O 0 00000 0
o o o O O O O O O 0
0 0 0 0 0 0 0 0 0 0 0 0
o o O o O O O O O 0 0 0 00000000 o o o - 00
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 000 0000 0 000 ° 00
o o O O O O O O O 0 0 0 0 O
0 0 0 0 0 0 0 0 0 0 O
o o O O O O O O O 0 o o o o o o ._
o _ o_ o_ o-+_o__:o:.....::o:._:::
o_'-o O o
0 O O O O O O O O 0 O0
00 0 0 0 0 0 0 0 0 0 0 0 0-0 0 0 00
O O O O O O O O O 0
0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 ° 0
O O O O O O O O O 0
0 0 0 0 0 0 0 0 0 0 0 0 0
O O O O O O O O O 0 0 0 00000000 0 00 00 0 00 00
0 0 0 0 0 0 0 0 0 0
I
f.ijHffii1A.,w,,;1;;;;.;1 1¥!liiiMLifi,iiiiiiMI
Do not have performance Have performance Do not have performance
Have performance
appraisal appraisal appraisal
appraisal
O O O O O O O O O O O O O O O O O O O 0 0 0 0 0 0 0 0 0 0 0 0 0 00 0 00 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 n n 0 0 0 000000000 0
0 0 0 0 IJ ,.. 0fo O 0 000000000 0
O O O O O O O O O O O O O O O O O O O 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 •0 0 0 0 0 0000000 0
J
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 .:, ,. ' 0 0 0 0 0 0 0 0 0 0 O OO
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 . 0 0 0 0 0 0 0 0 0 0 0 00
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00 , 0 0 0 0 0 0 0 0 0 0 0 00 --
1
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00 0 0 00 0 0000000 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 O O 0000 000000 0 0 0 0 0 0 0 0 00 ~
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 O O 0 000 000000 0 0 0 0 0 0 0 0 000
random selection process. It is importan t to ensure, how- and by whether or not employees are b
certain salary level or occupati onal a ove or
ever, that there is no inherent ordering of the sampling
frame, since this may bias the resulting sample. If there is to identify employe es in terms of th&racte. 1rit is ~I~~
. . . "bl ese s.... e~-tluJt
na, 1t 1s poss1 e to use pairs of cr·t . "atjh.•
some ordering to the list, the best solution is to rearrang e it. 1 ena O •nng .
(such as departm ental member shi r several C?iCrit,e_
occupati onal grade). p Plus &end . ~
. er 1>l11.
Stratified random sampling S tra tifie d sampIing 1s really feasible ...
• • • Only Wh
evant m1orma tton 1s avatlable . In other Word en the tel
• C
•
21 or
two were selected if the organization had six to 20 workplaces; and three were selected if the organization had
more workplaces. In the third stage, employees were sampled from within the sample of workplaces. In workplaces
with fewer than 200 employees, all employees were surveyed. In larger workplaces a random sample of employees
them,
was surveyed. This design produced a representative sample of Australian workplaces and employees within
which allowed workplace and employee data to be linked for the purposes of analysis.
Thus, we might randomly sample 10 companies from the • group 100 largest UK companie s by market
entire population of the 100 largest companies in the capitalization;
UK, thus yielding ten clusters, and we would then inter- • sample one company from each of the major
view 500 randomly selected employee, at er1ch of the 10 industries;
companies. Research in focus 9.2 give:; an e.xample of a • sample 400 employees from each of the companies .
study that used cluster sampling.
Apotential problem, however, is rhar £;-ierc is no guar- In a sense, cluster sampling is always a multi-stag e
antee that these 10 companies reflect the diverse range of approach, because clusters are always sampled first and
industrial activities that are engaged in by the population then something else-eithe r further clusters or popula-
as a whole. One solution to this problem would be to tion units-is sampled.
group the 100 largest UK companies by industry. This Many examples of multi-stage cluster sampling entail
could be done using the codes from the Standard Indus- stratification. We might, for example, want further to
trial Classification system or SIC, which is used widely to stratify the companies according to whether their head-
classify companies for the purposes of collecting data. quarters are located in the UK or abroad. To do this we
The next step would be to randomly sample companies would group companies according to whether their head-
from each of the major groups. One company might then quarters were based in the UK or elsewhere and then
be sampled from each of the industries, and then approx- select one or two companies per industry from each of the
imately 400 employees from each of the companies would two strata (see Research in Focus 9.2 for an example of
be interviewed. Thus, there are three separate stages: this kind of approach) .
era li .
sam e. Let us say we wis h
to is the sim ple ave rag e.. Thi s gen zat ion from th
and t~e sam ple dat a as the me an to the pop u1at1. on me
an Will h
dev elo pm ent in our sam ple
of ave kn 0 ,,_ esa
exa m.m e the lev el of ski ll d ll on this see T· wn "1p1.
at ski ll dev elo pm ent as the
erro r. For mo re eta ' 1ps and 5 . ·••argj ,
4SO em plo yee s. We wil l tre 12
.
izm g fro m a ran dom sam ple
to the lti1Js •c; l'l.!or
com ple ted in the pre vio us enera1 Population•
num ber of tra inin g day s
day s
. .. . . . .. .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . .
me an num ber of tra inin g Access the onl ine resourc~~
·;~· ·•... .. ... ... ..'...
mo nth s. We can use the Watch a ·• ...
und ena ken by the sam ple
me an Cm). The me an, ormo
(X) to est ima te the pop ula
tion
I TI PS AN D SKILLS
Generaliz in g from a random sa m pl e to th e population
rb~ ~ompany, let us say that the
ry stud y of trai ning and ski/I development in a single nea .
Usi ng our ima gina
am oun t of training received in the pre v,ou s 12
day s of trai ning per emp loye e (the average . . . .
sam ple mean is 6 ·7 .
n it we used pro bab ility sampling, how confide
cial consideration here Is: eve
mo nth s in the sample) . A cru the population? If we take an
in fin~: can
num ber of 6. 7 train ing days is likely to be found in
we be tha t the mea n
estimates of the mean of the
variable under consideration Will
a pop ulat ion, the sam ple
num ber of samples from d curve known as a
ion mea n. This vari ation will take the form of a bell-shape
vary in relation to the populat lies tha t there is a clustering
of sample
ribu tion (see Figu re 9.8 ). The shape of the distribution imp ion mean; the
nor ma l dist or below the populat
the pop ulat ion mea n. Hal f the sample means will be at
means at or around population mean) or
the pop ulat ion mea n. As we move to the left (at or below the
oth er halt will be at or above that there would be fewer and
fewer
ve the pop ulat ion mea n), the curve tails off, implying ns
the righ t (at or abo n. The variation of sam mea ple
mea ns tha t dep art con side rably from the population mea
sam ples generating sured using a statistic known
as the
is the sam plin g erro r and is mea
aro und the population mean t a sample mea n is like ly to differ from the
erro r of the mea n. This is an estimate of the amount tha
sta nda rd
population mean.
•,?h~m4 C·iiii,,i·iii,,i¥i,t. -
en I
GI
Q. i
E
-~
0
.8E
::::s
z
-1.9 6 Population +1. 96
SE mean SE
Value of the mean
Note: 95 per cent of sample
means will lie within the sha ded
mean.
area. SE = standard error of the
6} + (1,96 X l.3)
and
6.7 - (1.96 X 1.3)
If the standard error were smaller, the range of possible values of the population mean would be narrower; if the
standard error were larger, the range of possible values of the population mean would be wider.
If a stratified sample is selected, the standard error of the mean will be smaller; this is because the variation
between strata is essentially eliminated, because the population will be accurately represented in the sample in
terms of the stratification criterion or criteria employed . This consideration demonstrates the way in which
stratification injects an extra increment of precision into the probability sampling process, since a possible source
of sampling error is eliminated.
By contrast, a cluster sample without stratification exhibits a larger standard error of the mean than a comparable
simple random sample. This occurs because a possible source of variability between employees (that is,
membership of one department rather than another, which may affect levels of training undertaken) is
disregarded. If, for example, some departments have a culture of learning in which a large number of employees
were involved, and if these departments were not selected because of the procedure for selecting clusters, an
important source of variability would hzve been omitted. It also implies that the sample mean would be on the low
side, but that is another matter.
Sample size
One question about research methods that we are asked Absolute and relative sample size
by students almost more than any other relates to the
One of the most basic considerations, and one that is pos-
size of the sample: 'How large should my sample be?' or
sibly the most surprising, is that, contrary to what you
'Is my sample large enough?' The decision about sample
might have expected, it is the absolute size of a sample
size is not a straightforward one: it depends on a number
that is important, not its relative size. This means that a
of considerations and there is no one definitive answer.
' national probability sample of 1,000 individuals in the UK
This is frequently a source of great disappointment to
is as likely to be as representative as a national probability
those who pose such questions. Moreover, most of the
sample of 1,000 individuals in the USA, even though the
time decisions about sample size are affected by consid-
latter has a much larger population. It also means that
erations of time and cost (see Tips and skills 'Sample size
increasing the size of a sample increases the precision of a
and probability sampling'). Therefore, invariably deci-
sample. This means that the 95 per cent confidence inter-
. about sample size represent a comprotn1se
sions · be tween
val referred to in Tips and skills 'Generalizing from a ran-
the constraints of time and cost, the need for precision,
dom sample to the population' narrows. However, a large
and a variety of further considerations that will now be sample cannot guarantee precision, so that it is probably
addressed.
198 9 Sampling in quantitative research
•
nts. But, above all, do no
make claims about your sample that are not sustainable. Do not claim that it is represent
ative or that you have a t
random sample when it is clearly not the case that either of these is true. In other words,
be frank about what you
have done. People will be much more inclined to accept an awareness of the limits of your
sample design than
claims about a sample that are patently false. Also, it may be that there are lots of good
features about your sample-
the range of people included, the good response rate, the high level of cooperation you
received from the firm. Make
sure you play up these positive features at the same time as being honest about your sample's
limitations.
better to say that increasin g the size of a sample increases cision (because the amount of sampling error will be
the likely precisio n of a sample. This means that as sample less). By and large, up to a sample size of around 1,000,
size increase s, samplin g error decrease s. Therefor e, an the gains in precision are noticeab le as the sample size
importa nt compon ent of any decision about sample size climbs from low fi gure:- of SO, 100, 150, and so on
could be how much samplin g error one is prepared to tol- upwards . After a certair> point, however , often in the
erate. The less samplin g error one is prepared to tolerate, region of 1,000, the sh arp in creases in precision become
the larger a sample will need to be. Fowler (1993) how- less pronoun ced, and, although it does not plateau, there
ever warns against a simple acceptan ce of this criterion . is a slow-dow n in the extent to which precision increases
He argues that in practice researche rs do not base their (and hence the extent to w hich the sample error of the
decision s about sample size on a single estimate of a vari- mean declines) . Consider ations of sampling size are likely
able. Most survey research is intended to generate a host to be profound ly affected by matters of time and cost at
of estimat es-that is, of the variable s that make up the such a juncture , since striving for smaller and smaller
researc h instrum ent that is adminis tered. He also increme nts of precision becomes an increasingly uneco-
observe s that it is not normal for survey research ers to be nomic proposit ion.
in a position to specify in advance 'a desired level of preci-
sion' (1993: 34). Moreover, since sampling error will be
only one compon ent of any error entailed in an estimate, Non-response
the notion of using a desired level of precision as a factor
Conside rations about samplin g error do not eild her~-
in a decision about sample size is not realistic. Instead, to
The problem of non-resp onse should also be borne •~
the extent that this notion does enter into decision s about . arnounto
mind. Most sample surveys include a certain of our
sample size, it often does so in a general rather than a cal-
non-resp onse. Thus, it is likely that only some If it is
cula ted way. sample will agree to pa rticipate in the research. are
our aim to ensure as far as possible that 4 So ernployeesceJlC
interview ed and ifwe think that there may be a zorers40--
Time and cost
rate of non-resp onse, it may be advisable to saJllP will
Time and cost consider ations become very relevant in this 90
SOindividu als, on the grounds that approxunatelY y queS·
context. In the previous paragrap h it is clearly being sug- be non-resp ondents. For example , of the 143 s~ e( 995)
gest ed that the larger the sample size, the greater the pre- .
nonna1r.
es sent to compani.es m . T c powell 1
• ·
Types of non-probability sampling 199
study of total quality managem ent, only 40 were companie s. Fulton (2018) claims that this poses a par-
returned and of these only 36 were usable, making a ticular threat to studies of organizati ons, where response
response rate of 25 per cent. This raises the question of rates tend to be lower than in other fields of research.
whether or not this sample is big enough to represent However, analyses of this issue by Baruch and Holtom
companies in the geographi cal area of the n orth -eastern (2008) and Anseel et al. (2010) suggested that there is
USA that the study claims to represent (see Chapter 11 little evidence of a generalize d decline in response rates
for a further discussion of acceptabl e respome rates) . in surveys in organizati onal research. Strategies that can
The issue of non-respo nse, and in particular of ,efusal to improve responses to survey instrumen ts such as struc-
participate, is of particular significan ce, because it has tured interviews and questionn aires will be examined in
been suggested by some researche rs that respon se rates Chapters 10 and 11.
(see Key concept 9.3) to surveys are declin ing in many
countries. This implies that there is a growing tendency Heterogeneity of the population
towards people refusing to participa te in survey
research. As long ago as 1973, an article in the American Yet another considerat ion is the homogene ity and hetero-
magazine Business Week carried an article ominousl y geneity of the population from which the sample is to be
entitled 'The Public Clams up on Survey Takers'. The taken. When a sample is very heterogen eous, such as a
magazine asked survey companie s about their experi- sample of a whole country or city, the populatio n which it
ences and found considera ble concern about declining represents is likely to be highly varied. When it is rela-
response rates (Business Week 1973). Similarly, in Brit- tively homogene ous, such as members of a company or of
ain, a report from a working party on the Market an occupatio n, the amount of variation is less. The impli-
Research Society's Research and Developm ent Commit- cation of this is that, the greater the heterogen eity of a
· 1975 pointed to concerns among market research
tee In populatio n, the larger a sample will need to be.
STUDENT EXPERIENCE
Using convenience sampling in a dissertation
Jordan conducted research on self-expression in the workplace. Her approach required a fairly large sample 01
employees. The one-year duration of her honours programme, and the need to fit her research around a
. . . h
demanding coursework load, meant that she had to find an efficient way to generate a sample wh1c would alloW
her to conduct her analysis. She chose to survey graduate employees in a private-sector organization, who were
·
undergoing a tra,rnng
· ·
programme run bya non-governmental organization (NGO) she had prev1ou · sly worked tor.
All 515 graduate employees who underwent the training programme were given a paper survey to comp_lete nd
10
during their training programme. She explained: '[the NGO] provides training to this particular organization:
all of the graduate employees ... 515 of them. They . .. provide this training in two places in Melbourne an, use
Sydney .. . the reason why I got 515 surveys was because time was set aside during the workshop. , Jordan sher
of convenience sampling provided her with an effective means to collect data which allowed her to conciuclloyees-
analysis in a situation In which she did not have the time or resources to generate a random sample of emP
Types of non-probability sampling 201
refuse to Publish papers based on student sam- have been decided upon, it is then the job of interview-
. rnaist0 pie is merely a convenient way to sam- ers to select people who fit these categories. The quotas
JO0 sthesam
ies, unJes ·n the case of part-time MBAs discussed will typically be interrelated. In a manner similar to
P gers, as 1 . . .
pie mana . The study described m Research m focus stratified sampling, the population may be divided into
~• earher. •ence sample taken from the ' rea l worId'
8 (icue strata in terms of, for example, gender, social class, age,
edaconveru
9.4 us from among students. and ethnicity. Census data might be used to identify the
ratberthan number of people who should be in each subgroup. The
numbers to be interviewed in each subgroup will reflect
Quota sampling the population. Each interviewer will probably seek out
m iing is used intensively in commercial individuals who fit several subgroup quotas. Accord-
Quota sa pch as market researc h and po1·1t1ca
. l op1mon
. .
ingly, an interviewer may know that among the various
search, su
re Uin&· The aim of quota sampling is to produ~e a sample subgroups of people, they must find and interview five
Po fl ctS a population in terms of the relative propor- Asian, 25-to-34-year-old, lower-middle-class females in
tbat re e .
. f eople in different categories, such as gender, the area in which the interviewer has been asked to
oons o P . . .
. •cy age groups, socio-economic groups, and region work. The interviewer usually asks people who are
etbntCI , . . .
'dence, and in combmanons of these categories. available to them about their characteristics in order to
0f rest
However, unlike a stratified sample, the sampling of indi- determine their suitability for a particular subgroup.
viduals is not carried out randomly, since the final selec- Once a subgroup quota (or a combination of subgroup
tion of people is left up to the interviewer. quotas) has been achieved, the interviewer will
Once the categories and the number of people to be no longer be concerned to locate individuals for that
interviewed within each category (known as quotas) subgroup.
RESEARCH IN FOCUS 9 .4
Convenience sampling in a study of
discrimination in hiring
Derous et al. (2017) conducted a !>i. 1.1,jy ;.;s a ;-neans of studying how ethnic cues influenced the outcomes of
curriculum vitae (CV) screening during rec;uitment processes. Most studies of CV screening have used samples
of university students (Derous et a!. 2017: 862), but this study utilized a sample of HR professionals as a means
to increase the likelihood that the conclusions drawn were representative of processes and outcomes involving
respondents who undertake recruitment in their daily working lives.
The study was conducted in Belgium. The authors needed a large sample for their study, but there was no
sampling frame available from which to draw a random sample of respondents, so the sample was drawn from
membership lists of HR professional associations, business publications, and the researchers' own networks.
This generated a sample of 1,463 respondents, of whom 424 agreed to participate. Participants were all
Caucasian. Clearly this strategy did not generate a random sample and the sample cannot be regarded as
representative in a statistical sense. Nonetheless, the sampling strategy represented a viable way to draw a large
sample of actual HR professionals which could not otherwise have been generated.
The researchers provided each respondent with job advertisements for two different jobs in two different kinds of
business. They were each also provided with four fictional CVs, each including a photo of a fictional applicant.
The CVs showed a variety of combinations of skin tone (light versus dark) and name (Flemish versus Arab/
Maghreb). Respondents were asked to rate each applicant using a three-item measure with Likert scale
responses. For example, 'Given all the information you read about this applicant, how likely is it that you would
invite this applicant for a job interview?' (1 = not likely at all; 7 =very likely). The findings showed the equally
qualified candidates with a dark skin tone were systematically rated less suitable for jobs than those with light skin
tone, with the effect varying depending on the nature of the job. Name did not appear to matter. The findings
suggested that there was systematic ethnically based discrimination in CV screening and that there were subtle
effects Which arose from the kinds of jobs involved.
202 9 Sampling in quantitative research
The choice of respondents is left to the interviewer, sub- • It is not permissible to calculate a st
andard
ject to the requirement of all quotas being filled, usually mean from a quo_ta sample, because the error 0rth
within a certain time period. Those of you who have ever method of selectton makes it impo 'b non.rand e
ss1 let o111
been approached on the street by a person toting a clip- the range of possible values of a pop 1 . 0 caiC\JI
u ation. <tte
board and interview schedule and have been asked about
All of this makes the quota sample look
your age, occupation, and so on, before being asked a · not favoured
· is
· no d oub t t h at 1t
there is b bet, and
a Poor
series of questions about a product or whatever, have . Y acade i
researchers. It does have some argume
almost certainly encountered an interviewer with a quota nts 10 its 111 c
however: favour,
sample to fill. Sometimes, they will decide not to inter-
view you because you do not meet the criteria required to • It is undoubtedly cheaper and quicker than .
fill a quota. This may be due to a quota already having view survey on a comparable probability sarnan lllter.
been filled or to the criteria for exclusion meaning that a example, interviewers do not have to s Pie. For
nd
person with a certain characteristic that you possess is not time travelling between interviews. pe a lot of
required. • Interviewers do not have to keep calling b k
A number of criticisms are frequently levelled at quota ac on
pie who were not available at the time they I>eo.
samples: Were fir
approached. st
• Because the choice of respondent is left to the inter- • Because calling back is not required, a quota sam .
. I . ple15
viewer, the proponents of probability sampling argue easier to manage. tis not necessary to keep track of
that a quota sample cannot be representative. It may people who need to be recontacted or to keep track of
accurately reflect the population in terms ofsuperficial refusals. Refusals occur, of course, but it is not neces.
characteristics, as defined by the quotas. However, in sary (and indeed it is not possible) to keep a recordof
their choice of people to approach, interviewers may which respondents declined to participate.
be unduly influenced by their perceptions of how • When speed is of the essence, a quota sample is invalu-
friendly people are or by whether the people make eye able when compared to the more cumbersome proba-
contact with the interviewer (unlike most of us, who bility sample. Newspapers frequently need to know,
look at the ground and shuffle past as quickly as pos- quickly, how a national sample of voters feel about a
sible because we do not want to be bothered in our certain topic or how they intend to vote at that time.
leisure time). Similarly, if there is a sudden major news event, such
• People who are in an interviewer's vicinity at the times as a major terrorist attack or a natural disaster, the
they conduct interviews, and are therefore available to news media may seek a more or less instant picture of
be approached, may not be typical; in that case, those the nation's views or responses. Again, a quota sample
who are included in the sample will not be typical. will be much faster.
There is a risk, for example, that people in full-time • As with convenience sampling, quota sampling is use·
paid work may be under-represented. ful for conducting development work on new meas·
• The interviewer is likely to make judgements about ures or on research instruments. It can also be usefully
hich
certain characteristics in deciding whether or not to employed in relation to exploratory work from w
approach a person: in particular, judgements about new theoretical ideas might be generated.
age. Those judgements will sometimes be incorrect- hould not
• Although the standard error of the mean 5 .15 AS
for example, when someone who is eligible to be inter- · frequentlYwriters
·
be computed for a quota sample, it
viewed, because a quota that they fit is unfilled, is not Moser and Kalton (1971) observe, somde . quota
approached because the interviewer makes an incor- etho in
argue that the use of a non-random m oJJIPu·
rect judgement (perhaps that the person is older than . tosuch ac
sampling should not act as a barrier f error is
they look). In such a case, a possible element of bias is . b . . "fi
tatlon ecause its sigm cance as a source o aY artse
.
being introduced. that rn
small when compared to other errors go on to
• It has also been argued that the widespread use of . . H ever theY pe
m surveys (see Figure 9 .9). ow ' plillS 1
social class as a quota control can introduce difficul- argue that at least with ran
dom sam
f saJ11P 1
rnS
mount O t j1S
ties, because of the problem of ensuring that inter- researcher can calculate t h e a d aboll
viewees are properly assigned to class groupings error and does not have to be concerne
(Moser and Kalton 1971) . potential impact.
Limits to generalization 203
fjijfP-il·ill:iH@h'lddMIGM
Error
'. , . 1. , .
Sampling Sampling-related Data collection Data processing
error error error error
L1m1•ts to genera11
I /
•z,rnvn S
One point that is often not fully appreciated is that even Similarly, we should be cautious of overgeneralizing in
when a sample has been selected using probability sam- terms oflocality. A frequent criticism made of research on
pling, any findings can be generalized only to the popu- employee motivation relates to the limited extent to
lation from which that sample was taken. This is an which it can be assumed to be generalizable beyond the
obvious point, but it is easy to think that findings from confines of the national or regional culture on which
a study have some kind of broader applicability. If we the study is based. For example, in their famous study
take our imaginary study of training and skill develop- of motivation Herzberg et al. (1959) conducted
ment among employees of a company, any findings semi-structured interviews with 203 accountants and engi-
could be generalized only to that company. In other neers in the Pittsburgh area in the USA. Most of the com-
Words, you should be very cautious about generalizing panies that constituted sites for the study were involved in
to employees at other companies. There are many fac- heavy industry, such as steelmaking or shipbuilding. The
tors that may imply that the level of training and skill population from which the sample was selected consisted
development is higher (or lower) in that company than of all accountants and engineers who worked for these
a~ong company employees as a whole. There may be a companies. Respondents were chosen randomly accord-
higher (or lower) level of skill required in order to do ing to certain criteria for stratification, including age, job
th
e jobs that the company requires its employees to do; title, level in the company, and length of service. It is inter-
there maY be more (or less) money m . the company' s esting that there is no mention of gender in the study,
training budget; there may be more (or Jess) of a cu J- although we can fairly safely assume that, given that this
. at this company· or the company may
ture of Jearning was a study of accountants and engineers in the late
recruit h" '
a igher (or lower) proportion of employees 1950s, there is likely to be a male bias to the study. The
h
w oar I maximum number of individuals selected for interview in
t e a ready skilled. There may be many other fac-
ors too. each company was approximately SO. As the authors
204 9 Sampling In quantitative research
acknowledge, 'the fact that this work was done within a area, is there a point at which we have
thirty-mile radius around Pittsburgh will inevitably raise findings applied to the Pittsburgh to say, 'We11
area the , th~
questions about the degree to which the findings are have changed, and we can no long n, but t~,"'St
er assllrn "lltt
applicable in other areas of the country' (1959: 31). The apply to that or any other locality'? ,.. e that &i
vve are aft th
findings may also reflect the values of high individualism, to assuming that changed circumst ' er all ey
ances . , llsed
self-interest, and high masculinity that have been identi- changed behaviour and attitudes. To ta.k WtU Pro
ea sirn lllpt
fied as characteristic of American culture (Hofstede pie: no one would be prepared to assum h Pie ex:.....
. . et atth ...,1.
1984). This is pan of the reason that there have been so of a study m 1980 of UK uruversity Stud e fincli11
ents' b &s
many attempts to replicate the study on other occupa- and personal finance habits would apply to Stud lldgeting
tional groups and in other localities, including different early twenty-first century. Quite aside fr ents in th
om.ch e
cultures and nationalities. might have occurred naturally, the erosi·o anges that
. nand Vi
However, there could even be a further limit to gener- mantlmg of the student grant system in the Ttl!a] dis.
alization that is implied by the Herzberg et al. sample. The introduction of tuition fees have changed thlJI< and the
main study was conducted in the late 1950s. One issue dents finance their education, including pe h e Ways stu.
r apsa
that is rarely discussed in this context, and that is almost reliance on part-time work (Lucas 1997) greater
, a greate
impossible to assess, is whether or not there is a time limit ance on parents, and use of loans. But even h r reli-
, w en the .
on the findings that are generated. Quite aside from the no definable or recognizable source of releva h re 1s
ntc an
fact that we need to appreciate that the findings cannot (or this kind, there is nonetheless the possibility ( ge of
. . . oreven Uk
at least should not) be generalized beyond the Pittsburgh lihood) that findmgs are temporally specific. e-
• Ahousehold may have one computer but several users. Hewson and Laurent (2008) suggest that when there is no
, Internet Users are a biased sample of the population,
in that they tend to be better educated, wealthier,
younger, and more often male than female (Blasius and
Brandt 2010) •
, Few sampling frames exist of the general online popu-
sampling frame, which is normally the case with samples
to be drawn from the general population, the main
approach taken to generating an appropriate sample is to
post an invitation to answer a questionnaire on relevant
newsgroup message boards, to suitable mailing lists, or
•
lation, and most of these are likely to be expensive to on web pages and social media. The result will be a sam-
acquire, since they are controlled by ISPs or may be ple of entirely unknown representativeness, and it is
impossible to know what the response rate to the ques-
confidential.
tionnaire is, since the size of the population is also
Such issues make the conduct of online surveys using
unknown. On the other hand, given that we have so little
probability sampling principles challenging.
knowledge and understanding of online behaviour and
This is not to say that online surveys should nut be con-
attitudes relating to online issues, it could reasonably be
sidered. Indeed, for researchers in the field of bu~1ness,
argued that some information about these areas is a lot
where respondents are often contacted via the ,_iip n.iza.-
better than none at all, provided that the limitations of the
tions they work in, there may be more opport1Jr.iues than
findings in terms of their generalizability are appreciated.
for researchers in other areas. For example, in many
A further issue in relation to online sampling and sam-
organizations, most if not all non-manual workers are
pling-related error is the matter of non-response (see Key
likely to be online and familiar with the details of using concept 9.3). There is growing evidence that online sur-
email and the internet. It is relatively easy to obtain a list veys typically generate lower response rates than postal
of potential respondents from an organization and to dis-
questionnaire surveys (Tse 1998; Sheehan 2001;
tribute questionnaires to them (Saunders et al. 2017). In Pedersen and Nielsen 2016) . A meta-analysis that com-
such circumstances, surveys can be conducted using
pared web surveys with other modes of survey adminis-
essentially the same probability sampling procedures as
tration found that, on average, web surveys produce a
those outlined earlier in the chapter and email-based sur-
response rate that is 11 per cent lower (Lozar Manfreda
veys may present sampling problems that differ little
et al. 2008). However, response rates can be boosted by
from offiine modes of administration (Fricker 2017).
Suni)'
• following three simple strategies:
arly, surveys of members of commercially relevant
0
~ine groups can be conducted using probability sam- 1. Contact prospective respondents before send-
Phng Principles. If we are interested in collecting data ing them a questionnaire. This is regarded as basic
from members of on1·
me commum.nes . for example t h
ose 'netiquette'. Bosnjak et al. (2008) found that response
Using so . l
cia networking technology' for business pur- rates to a web-based panel survey could be enhanced
poses the .
' n vutually by definition they would need to be by pre-notifying prospective participants. They found
contacted l' .
(199 on tne to generate a sample. C. B. Smith that pre-notifications sent by text (SMS) message
(p
7) cond d
ucte a survey of web presence providers· were more effective than when sent by email but that
eople or org . . . .
and . an1zanons that are involved m creatmg a combination of both was more effective than text
illatntaj ·
rung web content). She acquired her sample messages alone.
206 9 Sampling in quantitative research
Aguinis et al. (2020) conducted a review of the use of MTurk in management research and report that ·t .
• • I is now
the most common online method for participant recruitment and that .,ts use mcreased over 2,000 per cent
between 2012 and 2019. Based on their review, they identify four maJor benefits of using MTurk:
They note, however that there are a number of important challenges inherent in using MTurk, which can under .
' mine
the validity of results. Among these are self-selection bias, high attrition rates, vulnerability to malicious software
intended to fraudulently collect payment for participation, and social desirability bias (Aguinis et al. 2020: 826-l).
The authors suggest that the challenges need to be weighed up against the potential advantages and that care needs
to be taken in planning and implementing studies, as well as reporting results, as a means to mitigate their effects.
While concerns about data quality are of course extremely important, there are also potential ethical
considerations in using online recruitment methods such as MTurk. Pittman and Sheehan (2016) argue that
MTurk study participants are usually paid quite small amounts for their participation and thus can potentially
considered to be analogous to 'sweatshop' workers in manufacturing. Gleibs (2017) argues that there has been
relatively little attention paid to the ethical challenges of MTurk and other on!in2 ;:;!:1tforms and that scholars
should be more attuned to these issues. She focuses on the fact that t here is 2 ;J')Wer imbalance between
researchers and participants. This means that underpayment is likely, but a!so ,;-,ises issues around whether
participants are able to withdraw from a study without risking being denied work in fut ure and that they have little
or no recourse if they feel that they have been underpaid or unfairly excl uded from studies by researchers.
2. As with postal questionnaire surveys, follow up non- Students in the sample were initially sent an email invit·
respondents at least once. ing them to visit the website, which allowed access, via a
3 . There is evidence that incentives can increase response password, to the questionnaire. Some of those emailed
rates in online surveys (Pedersen and Nielsen 2016) . were led to expect that the questionnaire would cake ~ght
to 10 minutes to complete (in fact, it would take consider·
The case for the first of these two strategies in boosting ably longer); others were led to expect that 1t · would. cake.
response rates is not entirely clear (Sheehan 2001) but d beJieve1t
20 minutes. As might be expected, those le to . .
seems to be generally advisable. However, as previously h inVJtanon.
would take longer were less likely to accept t e ver
. Howe •
noted, with many online surveys it is impossible to calcu- resulting in a lower response rate for this group. wbO
late a response rate, since, when participants are recruited ndents
Crawford et al. also found that those respo -" onlY
through invitations and postings on discussion boards, · uld tll"-e
were led to believe that the questionnaire wo n [be
etc., the size of the population of which they are a sample . h
e1g t to 10 minutes were more e lik ly to give up o
- unus·
is almost impossible to determine. . . h resulting I0
quesnonnaire part of the way throug , t cases-
. in cnos
Crawford et al. (2001) report the results of a survey of able partially completed questionnaires ere 111os1
students at the University of Michigan that experimented dents W
Interestingly, they also found that respon f the waY
with a number of possible influences on the response rate. . . . part o
likely to abandon their questionnaires
Questions for review 207
·n the middle of completing a series of on a screen together than when they appeared singly.
h when l . .
chroug . The implications of this findmg reflect the Respondents also seemed less inclined to omit related
uest1ons. . .
open q will encounter in Chapter 11: that 1t 1s proba- questions when they appeared together on a screen rather
advice you k as few open questions in self-complet ion than singly.
blY best to as .
. nnaires as possible. However, it is important not be too sanguine about
quesoo vidence suggests that having a progress some of these findings. One difficulty with them is that
Further e
• an online survey can reduce the number of the samples derive from populations whose members are
•ndicator in
1 . .
h abandon the questJ.onnaire part of the way not as different from one another as would almost cer-
eople w o
p h mpletion (Couper et al. 2001) . A progress indi- tainly be found in samples deriving from general popula-
throug co
. sually a diagrammatic representatio n of how far tions. Another is that it must not be forgotten that, as
catons u
dent has progressed through the questionnair e previously noted, access to the internet is still not univer-
cherespOn
ticular point. Couper et al. also found that it sal, and there is evidence that those with internet access
at any Par
took less time for respondents to complete related items differ from those without, both in terms of personal char-
(for example, a series ofLikert items) when they appeared acteristics and attitudinally.
(i)
■
KEY POINTS
G)I
QUESTIONS FOR REVIEW
········· ·······•·············· ··· ······················ ······················ ·················································· ···································•·········•
• What does each of the following terms mean: population, probability sampling, non-probability
sampling, sampling frame, representative sample, and sampling and non-sampling error?
Sampling error
• What is the significance of sampling error for achieving a representative sample?
• How far does a stratified random sample offer greater precision th an a simple random
or 5Yste
sample? l'llatic
• If you were conducting an interview survey of around 5 oo people in a large city, What ty
? ~~
probability sample would you choose and why.
• A researcher positions herself on a street corner and asks one person in f ive who walks b
· until·
interviewed: she continues doing this she has a samp Ie Of 2 50 • How 1·I kely ·1s she toyach·
to be
representative sample? ieve
a
• What factors would you consider in deciding how large your sample should be when devising a
probability sample?
• What is non-response and why is it important to the question of whether or not you will end up with a
representative sample?
Limits to generalization
• What are the main challenges in selecting a sample for an on line survey?
··· ············· ···· ··· ····· ······· ····· ······ ··· ·· ···· ······· ·· ·· ····· ······ ····· ······ ······· ·· ····· ············· ······
ONLINE RESOURCES
0 www.oup.com/he/brm6e/
In addition to the video interviews already highlighted, the book's online resources include th e following
features for this chapter specifically: