Accident and Emergency Nursing (2007) 15, 234–238

Accident and

www.elsevierhealth.com/journals/aaen

EmergencyNursing

Clinical research 3: Sample selection q

Ruth Endacott RN, DipN(Lond), MA, PhD (Professor of ClinicalNursing) a,b,*, Mari Botti RN, RM, BN, PhD, MRCNA (Professor,School of Nursing) a,c

a University of Plymouth, Faculty of Health and Social Work, Earl Richards Road North,Exeter EX2 6AS, United Kingdomb La Trobe University, PO Box 199, Bendigo, Victoria 3552, Australiac Deakin University, Burwood, Victoria 3125, Australia

Abstract This research series is aimed at clinicians who wish to develop researchskills, or who have a particular clinical problem that they think could be addressedthrough research. The series aims to provide insight into the decisions that research-ers make in the course of their work and to also provide a foundation for decisionsthat nurses may make in applying the findings of a study to practice in their own Unitor Department. The series emphasises the practical issues encountered when under-taking research in critical care settings: readers are encouraged to source researchmethodology textbooks for more detailed guidance on specific aspects of theresearch process.

�c 2007 Elsevier Ltd. All rights reserved.

KEYWORDSMethodology;Sample size;Sample selection;Qualitative design;Quantitative design

0d

Cp

0

c

Introduction

This paper addresses two key principles of sam-pling: adequacy (the number of participants) andappropriateness (the characteristics of the partici-pants). The research aims and design will deter-mine whether you need a statistically based or

965-2302/$ - see front matter �c 2007 Elsevier Ltd. All rights reseroi:10.1016/j.aaen.2006.12.006

q This article was originally published in Intensive and Criticalare Nursing 2005 21(1) 51–55. The article is republished withermission from Intensive and Critical Care Nursing.* Corresponding author. Tel.: +44 01392 475155; fax: +441392 475151.E-mail addresses: ruth.endacott@plymouth.ac.uk (R. Enda-

ott), mari.botti@deakin.edu.au (M. Botti).

saturation based sample size (adequacy). Theappropriateness of the sample is addressed throughinclusion and exclusion criteria. These two aspectsare discussed below in relation to quantitative andqualitative designs.

Quantitative designs

By definition, quantitative designs rely on the con-trol or exclusion of all factors that might bias theoutcome of a study. In the reality of clinical re-search, this is rarely achievable. However, a crucialearly stage is the identification of confounding vari-ables and, for each one, to identify:

ved.

Clinical research 3: Sample selection 235

(a) whether it can be controlled;(b) if not, can it be measured or;(c) should it be used as an exclusion criterion?

Confounding variables that are measurable butnot controllable provide useful data for sub-setanalysis. These are usually referred to as secondaryoutcomes and phrased as secondary hypotheses.For example, when studying the intensity of post-operative pain in a group of patients, a usefulsecondary analysis would be to explore whether pa-tients with pre-existing chronic pain conditions (aconfounding variable that can be measured) expe-rience more post-operative pain. A confoundingvariable that could be used as an exclusion crite-rion for such a study might be different culturalbackground, unless the focus of the study was tocompare groups of patients from different culturalbackgrounds. If decisions made by the researcherare not made explicit, the rigour of the studymay be called into question.

Sampling is necessary because it is not usuallypossible to collect data from every individual,event or unit in the population we are interestedin. Data are collected from a smaller sub-set ofthe population – a sample – which ideally corre-sponds to the larger population of interest in termsof particular key characteristics. The intent andthe requirements of sampling differ depending onthe design of a study.

The intent of sampling in quantitative designs isto estimate or predict outcomes about the largerpopulation based on data from a sample of thatpopulation (Schofield, 2004). In other words, gen-

Table 1 Sampling methods in quantitative research

Random samplingSimple random All elements in a sampling frame have an

use of random numbers or similar methodSystematic Samples are drawn by beginning at a rand

each nth element, e.g. Every 20th event oStratified Participants are grouped according to stra

diagnosis. Equal numbers of participants aCluster The population is divided into smaller sub

and all or a random selection of individuain large national surveys to enable random

Non-random samplingConvenience The use of the most readily accessible ind

research to obtain an estimate of a particConsecutive A version of convenience sampling where

accessible population is chosen. The bestSnowball Individuals recruited for a study refer oth

maybe known within networks rather thanrare

Quota This method ensures that specific individusample, for example according to age gro

eralisations are made about individuals or unitsfrom whom data has not been collected. In orderto be able to make these generalisations it is nec-essary to ensure that the sample is in fact repre-sentative of the population. To do this, attentionis paid to the method used to sample the popula-tion and the size of the sample.

Sampling methods

Ensuring that the sample accurately represents thelarger population is more important than the sizeof the sample in quantitative research. The repre-sentativeness of the sample in terms of the largerpopulation has a direct impact on the externalvalidity of the conclusions of the research.

The first step in sampling then is to identify thespecific characteristics of the target population,i.e. the population to which you wish to generalisethe results. For example, a target population mayinclude all adult, post-operative day surgery pa-tients in Australia. Because it may not be possibleto study every post-operative day surgery patientit is then necessary to identify an accessible popu-lation. This is a sub-set of the target populationthat reflects specific characteristics such as age,sex and diagnosis, and is accessible (LoBiondo-Wood and Haber, 2004). For example, an accessi-ble population may include all post-operative daysurgery patients in a specified geographical area.

The next major consideration in the samplingprocess is to define the inclusion and exclusioncriteria of the accessible population. Inclusion

equal chance of selection. Selection is through the

om point in the sampling frame and then selectingr individual. Its advantage is its simplicityta that are important in a study such as age, sex orre randomly selected from each group-sets or clusters. These clusters are randomly selectedls within these clusters are selected. Commonly usedsampling in diverse populations

ividuals or units in a study. Used in exploratoryular element of interestevery available individual or event within anchoice of non-random samplinger potential participants. Useful when participantsgenerally or when a desired study characteristic is

als or units are included equally in a convenienceup or sex

236 R. Endacott, M. Botti

criteria are generally based on the research ques-tion and the research plan; they are applied toenable selection of homogenous samples and im-prove the feasibility of conducting a study. Forexample, it may be considered that pain manage-ment outcomes for day surgery patients may bebetter studied in patients undergoing abdominaland pelvic procedures. Exclusion criteria are gen-erally applied to exclude unique characteristicsthat may confound the results or to deal with eth-ical considerations relating to the research. Forexample, it may be necessary to exclude patientswho have additional medical problems that mayaffect their outcomes, or to exclude patientswho have cognitive impairment that impairs theirability to provide informed consent to participatein the study.

Once the target and accessible population char-acteristics have been defined, a sample is selectedfrom the accessible population. There are twomain types of samples: random and non-randomsamples, also referred to as probability and non-probability samples (see Table 1 for definitions ofcommon techniques used to sample within thesecategories).

Random or probability sampling refers to sam-pling processes that guarantee that each of thepotential participants, events or units under inves-tigation have an equal chance of being selected.Random selection is important for two reasons.First it reduces the risk of selection bias, and sec-ond, a randomly selected sample is a requirementfor inferential statistical analyses.

In non-random or non-probability sampling, thesampling technique is not random; therefore, allmembers of the population do not have an equalchance of being selected for recruitment into thestudy. Non-random samples cannot be assumed tofully represent the target population and conse-quently, conclusions about the generalisability ofresults to the target population should be qualified.

Although random sampling is the preferred sam-pling method for quantitative research, it can bevery difficult to achieve due to time, cost and eth-ical considerations. It is often necessary to useother sampling techniques. The use of appropriatesampling methods is one part of the process ofensuring a representative sample, the size of thesample is also important.

Calculating sample size

Sample size calculation in quantitative researchdepends on a number of factors. These include:

1. research design;2. sampling method;3. the degree of precision required;4. the variability of the factors being investigated;5. the incidence of a particular variable in the

population.

As a general statement, the larger the samplethe higher the likelihood that the findings will accu-rately reflect the population because larger sam-ples have lower sampling error. Sampling errorrefers to the notion of estimating population char-acteristics from data collected from a sample. If 10samples were drawn from an accessible popula-tion, there would be 10 different patterns of re-sponses to the same question. This is known assampling error, as the sample size increases thelower the sampling error.

In descriptive studies, where the intent is toidentify the proportion of a particular phenomenonin a sample, sample size calculation is based on thelevel of precision and confidence required of theresults (Altman, 1980). This is determined by confi-dence intervals. Confidence intervals are a range ofvalues that surround a value obtained from a sam-ple that have a known probability (confidence) thatthe true population value is contained within thatrange. Normally the specified confidence is set at95% and the interval width is determined by theclinical significance of the precision of knowingthe occurrence of a particular phenomenon. Forexample if we were interested in determining thepercentage of patients who experience moderateto severe pain in the first day after surgery, wemay specify a level of confidence of 0.95 and aninterval width of 10%. These criteria form the basisfor calculating the sample size for the study. If thefindings show that 50% of the sample experiencemoderate to severe pain we can have 95% confi-dence that 50% ± 5% (between 45% and 55%) ofthe population of patients experience moderateto severe pain.

In explanatory studies where hypothesisedcause–effect relationships are tested, researchersbase sample size calculations on three desiredfactors:

1. The significance level – normally set at 0.05 or0.01 (p value). Statistical significance refers toreducing the probability of finding an effectwhen the effect does not exist, referred to asType I error. When the level of significance is0.05, the probability of a Type I error is 5%.

2. Power – normally set at greater than 0.8 (80%).Power refers to the probability of finding aneffect when the effect does exist. In situations

Clinical research 3: Sample selection 237

of low power (inadequate sample size) impor-tant effects may not be detected. This isreferred to as Type II error.

3. Effect size refers to the minimum size of the dif-ference to be detected and is determined by theclinical importance of the difference. The effectsize may be derived from a similar study, pilotwork or estimates of the likely effect.

It can be seen that sample size determination inexplanatory studies is a crucial part of the overallvalidity of the study (for an example see Bottiet al., 1998). In studies where the effectivenessof a particular intervention is tested, the size ofthe sample will determine the likelihood of detect-ing an effect of the intervention if that effectexists. Inadequate sample sizes may result in fail-ure to detect clinically important small to mediumeffects of interventions.

Controlling for bias

The best way to evaluate the suitability of a samplein terms of its representativeness of the targetpopulation is to scrutinise the sampling methodand evaluate the size of the sample. Potentialbiases in sampling can be reduced by rigorousrecruitment strategies that increase response rateand reduce the likelihood of systematically exclud-ing some people from the sample or over repre-senting others.

Qualitative designs

Qualitative designs rely on saturation based sampleselection to satisfy the sampling principle ofadequacy. Appropriateness of sampling is usuallyachieved through purposive or theoretical sampling.

The concept of saturation

Sampling to the point of saturation requires the re-searcher to continue to recruit participants until nonew data emerge. However, ethics committees re-quire an indication of likely recruitment; five toeight participants are usually sufficient for ahomogenous sample and 12–20 for a heterogenoussample, where it is important to maximise varia-tion across the sample (Kuzel, 1999, p. 42). Whenexploring the information needs of relatives inICU, you may wish to include relatives of both longand short stay patients, requiring a sample of fiveto eight in each category. If you were to explore

the same topic with nurses of varying experienceand ICU qualification, a single focus group of eightis unlikely to be adequate. A key strategy to ensuresaturation is the seeking out of confirming and dis-confirming cases.

Purposive sampling

Purposive sampling is commonly used to allow theresearcher to include participants with key experi-ence of the issue to be studied. Decisions aboutwhom to sample are typically made before data col-lection commences. Again, a key decision iswhetheryou need a homogeneous or heterogeneous sample.Also bear in mind that the synergy you would wish tocreate in a Focus Group interviewmay require you tosample a multi-professional team.

Specific qualitative methodologies also assist inpurposive sampling decisions:

1. Grounded theory studies require a homogenoussample and are referred to as ‘theory-based’;

2. phenomenology studies require sampling thosewho have experienced the phenomenon – a ‘cri-terion based’ sample;

3. ethnography studies require sampling of the cul-tural group – a ‘representative based’ sample.

Random sampling is inappropriate for qualitativestudies as key informants may be missed throughthe randomisation process; randomisation also as-sumes that the issue to be explored is normally dis-tributed across the population sampled (Kuzel,1999).

Theoretical sampling

With theoretical sampling, the process of data col-lection is guided by the emerging theory and sam-pling decisions are taken during data collection.

Controlling for bias

The very nature of qualitative research, with itsfocus on deliberate targeting of the sample, lendsitself to criticisms of bias. Unfortunately, discus-sion related to the process of identifying saturationis frequently omitted from published research. Oneapproach frequently used to reduce bias is triangu-lation. The traditional definition of triangulationuses the analogy of navigation, where knowledge

238 R. Endacott, M. Botti

of two points is enables calculation of a third. Thisis often referred to as triangulation for conver-gence. Within qualitative studies the term is alsoused to indicate the use of multiple strategies/sources to provide a fuller picture of the situationbeing studied, referred to as triangulation for com-pleteness (Redfern and Norman, 1994). For a fullerdescription of triangulation, see the first paper inthis series (Endacott, 2004).

Summary

Regardless of the research approach taken, ade-quacy and appropriateness of the sample areessential components of research design. It isequally essential that these processes are de-scribed for consumers of research, both to de-mys-tify the steps involved and, more importantly, toallow the reader to judge the credibility of thefindings (Selby et al., 1990). All decisions madeby the researcher about sample size should bemade explicit and the steps taken to control biasshould be described.

References

Altman, D.G., 1980. Statistics and ethics in medical research III.How large a sample? Brit. Med. J 281, 1336–1338.

Botti, M., Williamson, B., Steen, K., McTaggart, J., Reid, E.,1998. The effect of pressure bandaging on complications andcomfort in patients undergoing coronary angiography: amulticenter randomized trial. Heart Lung: J. Acute Crit.Care 27, 360–373.

Endacott, R., 2004. Clinical research 1: research questions anddesign. Intens. Crit. Care Nurs. 20 (4), 232–235.

Kuzel, A.J., 1999. Sampling in qualitative inquiry. In: Crabtree,B.F., Miller, W.L. (Eds.), Doing Qualitative Research, seconded. Sage, Thousand Oaks.

LoBiondo-Wood, G., Haber, J., 2004. Nursing Research: Meth-ods, Critical Appraisal and Utilisation, third ed. Mosby, StLouis.

Redfern, S., Norman, I., 1994. Validity through triangulation.Nurse. Res. 2, 41–55.

Schofield, M., 2004. Sampling in quantitative research. In:Minichiello, V., Sullivan, G., Greenwood, K., Axford, R.(Eds.), Handbook Research Methods for Nursing and HealthScience, second ed. Pearson Education Australia, FrenchsForest NSW.

Selby, M.L., O’Pry Gentry, N., Riportella-Muller, R., Legault, C.,1990. Evaluation of sampling methods in research reported inselected clinical nursing journals: implications for nursingpractice. J. Prof. Nurs. 6 (2), 76–85.

Available online at www.sciencedirect.com