Anthropology, History, International Relations, Media, Philosophy, Politics, Quantitative Methods, Religion, Social Policy, Sociology, Theology, Urban Studies
Having recently returned from teaching English as a Foreign Language (EFL) in Brazil, I am working as a freelance private tutor. I have had similar experiences working as a teacher in Russia and as a summer camp teacher in Italy. In fact, most of my professional career has been devoted to education and helping others to learn. My first experience of teaching, which was in a Soviet-built ‘commuter town’ outside Moscow, was followed by my completing a PGCE in Post-Compulsory Education and Training. For my placement, I taught Sociology and Politics at a Sixth Form College in South London. After this, I lectured Sociology at a Further Education (FE) College and, later, at a Sixth Form College both of which were in Cambridge. I also worked as a Sociology and Politics lecturer at an FE College in London. In addition to my PGCE, I have a BA (2:1) in History and Sociology and, more recently, I obtained an MA in Sociology both from the University of Essex. In the future I would like to read for a PhD and pursue an academic career. My speciality is in Sociology and History as well as other related areas of Social Science and the Humanities.
What are the main forms of sampling and when would they be used? How would you sample a ‘hidden population’ (of your choice)?
This essay will examine the main forms of both probability and non-probability sampling. Although there are several sampling methods, for the purposes of this essay three specific examples will be discussed. These are: simple random sampling; cluster sampling; and snowball sampling. I will then explain the sampling method that I would use to sample a hidden population of my own choice: far-right political organisations.
Probability sampling methods rely on the principle that if you select certain members of the research population at random or with a degree of randomness, they will at least provide an unbiased snapshot of the research population as a whole. Daniel(2012) notes that probability sampling can be differentiated from non-probability sampling in the sense that although not every member of the research population has an equal chance of being selected, they do at least have some chance of being selected.
Daniel argues that probability sampling has a number of strengths. It minimises any selection bias on the part of the researcher and can be considered to be suitable for conclusive research which is likely to require a great deal of description, prediction, explanation and evaluation.
However, within probability sampling there are several different modes of data collection. Blakie states that simple random sampling is ‘the standard against which all other methods are judged.’
He adds that for simple random sampling to work, each member of the research population needs to be identified and numbered. Daniel notes that when the desired sampling frame has been determined, all the names of the research population are numbered using a computer programme. Then certain numbers are selected using a lottery method known as the ‘blind draw method’ or the ‘hat model’ until the desired sample size has been obtained.
Simple random sampling has a number of strengths and weaknesses. One of its strengths is that in-depth information about individuals within the research population is not required; the research population is not divided into sub-groups and each member has an equal chance of being selected, thereby increasing the representativeness of the method.
Dattalo (2010), however, argues that although simple random sampling minimises sampling errors, bias can still occur. For simple random sampling to be possible it needs to be carried out among a relatively small research population, which may affect the representativeness of the study. Despite respondents being selected at random, within such a small group they are likely to be homogenous.
Gilbert (2005) notes that the work required for simple random sampling can be cumbersome and time-consuming. He argues that it is an impractical method if the researcher wants to study large or geographically dispersed populations. Large scale simple random sampling studies are rarely undertaken due to the potential expense and practical inconvenience.
An example of simple random sampling comes from a study by Jenness, Maxson, Summer and Matsuda (2010) about sexual assaults by inmates on each other in Californian prisons. The researchers obtained rosters from the California Department of Corrections and Rehabilitation which identified all the prisoners within the appropriate research population. Once they had obtained this data, the researchers were able to exclude potentially inappropriate respondents such as those deemed to have high levels of mental incapacity and those with restricted status, housed in secure units. Once this process was completed, the researchers used statistical software to randomly select 100 inmates from each prison to be potential study participants.
The second form of sampling is that of clustering. This is a form of probability sampling whereby elements of the research population are randomly selected from naturally occurring units. For example, if a researcher was studying the experience of trainee teachers, rather than a using a simple random selection from all the trainee teachers in an educational establishment, he or she would focus on a small group of trainee teachers on a particular course, such as those training to be geography teachers.
Blakie argues that cluster sampling is usually used when it is very difficult to list all the population elements within the target population. He provides examples of when cluster sampling could be used, such as a classroom of students, a month of applications for citizenship, a year’s issue of a newspaper, or a street of residences. Groves et al (2004) have pointed out that cluster sampling is much cheaper than simple random sampling and can be used when constructing a large scale sampling frame would be impractical. Schutt (2009) notes that cluster sampling is often multiple-staged and that clusters can develop within clusters. A social researcher may initially decide to sample a particular geographical area such as Essex. He or she may then decide to sample particular towns in Essex such as Colchester and Chelmsford. He or she may then perhaps focus on particular roads in these towns before randomly selecting people who live in these roads.
Dorofev and Grant (2006) argue that cluster sampling can make research feasible that otherwise would be very difficult if not impossible. Conducting a survey of passengers departing from a major airport randomly would be very difficult to administer but cluster sampling could be employed to sample a small number of flights. The possibility of contacting passengers going on the same or similar flights would be high, thereby increasing the representativeness of the study.
However cluster sampling also has its weaknesses. Daniel  argues that sampled clusters may not be as representative as simple random sampling of comparable sizes. In addition, analysing data is more complex particularly in relation to the interpretation of the data by statistical software packages. Sapford (2007) notes that as cluster sampling is not a random sampling method there is a risk that certain people with particular attitudes or characteristics will be over-represented in cluster sampling, due to its limited geographical spread. He argues that there is a tendency for people who live in the same area, go to the same schools and have the same places of work to have similar attitudes and experiences.
The third form of sampling to be discussed is one of the non-probability sampling techniques. Blakie  notes that although probability sampling may be necessary to answer research questions using large research populations, some studies do not need to generalise about a population or are unable to identify a suitable sample in a particular research population. Research on people suffering from AIDS would be very difficult using probability sampling as lists of sufferers would not be publically available.
Daniel indicates that non-probability sampling is the preferred method when researchers need a quick decision and there is no need to target specific elements of the population or a representative sample. This method is also suitable when no statistical data analysis is required or there is no requirement to minimise selection bias. It is also useful when studying a research population that is highly scattered and difficult to access; if resources are extremely limited; a sampling frame is not available; or if an extremely small sample size is targeted.
One of the main methods of non-probability sampling is snowballing. This is often used with research populations that are hard to identify or ‘hidden’. It is known as snowball sampling as it involves contacting one or more members of the target population who then suggest other potential respondents; for example, drug users who then supply names and addresses of others they know who could be contacted by the researcher.
One example of snowball sampling was the study by Fagan and Chin (2006) into violence and the distribution of crack cocaine in New York City. The researchers contacted people who had been arrested for drug offences and were awaiting trial in Manhattan. Respondents were then asked to nominate potential respondents who were also drug users but had so far avoided arrest.
Taking into account the sampling methods examined in this essay, I will now discuss how I would sample a hidden population of my own choice; far-right political organisations such as the British National Party or the English Defence League. As well as being a sensitive topic, there would be ethical considerations. Sociologists such as Fielding have used covert participant observation to study these groups. However, bearing in mind the ethical issues of deception and the use of covert research, this is not an option when considering different sampling techniques.
It can be argued that when targeting a specific hidden or closed group, simple random sampling is not appropriate because it relies on existing records to base the randomness of its selection procedure. Even the study of assaults among inmates in California prisons by Jenness et al relied upon existing records provided by the Californian prison authorities. It is unlikely that official records of far-right political party members would be accessible to social researchers. Those being targeted for research may feel that there are biased political agendas in opposition to their beliefs and policies. Since the leak of British National Party membership records in 2008, this has become a pertinent consideration.
Therefore it is argued argue that snowball sampling would be the best sampling method to use with this ‘hidden’ population. This view is supported by evidence that snowballing has been used by previous researchers to access other ‘hidden’ populations where conventional data and statistics are either non-existent or misleading. Whyte’s work on Street Corner Society (1943) and James Patrick’s study into gangs in Glasgow (1973) both employed elements of snowball sampling through the use of ‘gatekeepers’ in order to gain access to the groups being studied. It should be pointed out that both studies are usually classified more generally as participant observation.
Further uses of snowball sampling to reveal ‘hidden ‘ populations include work into drug use (Avico et al 1998; Griffiths 1993) prostitution (McNarma 1994), pickpockets (Incardi 1977) aids sufferers (Pollack 1989) and the seriously ill (Sudman and Freeman1988).
It is believed that, using snowballing sampling, one member of the ‘hidden’ group would be contacted and after some meetings, mutual trust would build up. After explaining why this particular research was being undertaken, the researcher would ask the member to suggest other members of the group who may be prepared to take part in the research. Every effort would be made to ensure that these individuals were as varied as possible in terms of age, social class and perspective regarding the movement. Consideration would be given to the point made by Gilbert that because potential respondents were likely to be very similar to the initial ‘gatekeeper’, a diverse range of respondents was unlikely which may limit the representativeness of the sample.
Criticisms of the use of snowball sampling amongst ‘deviant’ populations offered by Lee (1993) must not be ignored. Lee stated that although snowball sampling has advantages with hidden ‘deviant’ populations and is useful for finding respondents, there is a key problem with its method; intermediaries who arrange meetings with potential respondents may be unclear about the aims of the research. This may lead to an unrepresentative and unfocused study.
In conclusion, it is believed that the three main forms of sampling used in social research are simple random, cluster and snowballing. This essay has attempted to differentiate these sampling methods and explain when it might be suitable to use them. The strengths and weaknesses of these techniques have been evaluated well as those of probability and non-probability sampling generally. When considering the appropriate sampling technique for the study of a hidden population, attempts have been made to balance their individual benefits and drawbacks before deciding on the most suitable technique. Inevitably, whichever form of sampling is chosen, there will always be arguments regarding its accuracy.
Atkinson Rowland and Flint John (2001) Accessing Hidden and Hard-to-Reach Populations: Snowball Research Strategies, Social Research Update, Issue 33, Summer 2001
Bethlenham Jelke (2009) Applied Survey Methods A Statistical Perspective, John Wiley and Sons, New Jersey
Dattalo Patrick (2010) Strategies to Approximate Random Sampling and Assignment, Oxford University Press, Oxford
Daniel Johnnie (2012) Sampling Essentials Practical Guidelines for making Sampling Choices, Sage, London
Dorofeev Sergey and Grant Peter (2006) Statistics for Real –life Sample Surveys Non-Simple Random Samples and Weighted Data, Cambridge University Press, Cambridge
Gilbert Nigel (2008) Researching Social Life 3rd ed, Sage, London
Groves Robert, Fowler Floyd, Couper Mick, Lepkowski James, Singer Eleanor, Tourangean Roger (2004) Survey Methodology, Wiley –Interscience, New Jersey
Lee Raymond (1993) Doing Research on Sensitive Topics, Sage, London
Marsden Peter and Wright James (2010) Handbook of Survey Research 2nd ed, Emerald, Bingley
Schutt Russell (2009) Investigating the Social World, Pine Forge Press, California
Sapsford Roger (2007) Survey Research 2nd ed, Sage, London