Identify the purpose and objectives of the survey

When carrying out a survey, you will initially need to establish a certain understanding as to the nature and purpose of the study. Is it to be a poll of people's behaviour, attitudes or opinions? Or is it an analytical study, examining correlation (the relationship) between sets of data? What do you hope to learn from the process? What will happen to the results? Who will be asked, what will they be asked and how will they be asked? Being clear about the overall aims of the survey will help to establish the population to be assessed via the survey, which may comprise of individuals, specific groups or sectors of the general population.

After an overall aim for the study has been established, it may be useful to propose more precise goals. For example, if the overall aim of a survey is to establish service users' opinions of the care they receive, more precise goals of enquiry could include particular objectives of the service, such as quality of care.

Surveys are a good means of answering 'what?', 'how?', 'when?' and 'where?' questions, but are not really appropriate for 'why?' questions, that is, they should not really be used to draw conclusions about causal relationships.

Identify at whom the survey is aimed

Who is the intended audience for the survey's results? National or local politicians, service providers or purchasers, or recipients of care? Being clear about the target of the results and what you hope to show with them can help in deciding how to carry out the research.

Identify what survey instrument is to be used

Whatever choice is made, you should ensure that, as far as possible, their tool for collecting data is:

• Reliable "“ that it reproduces the same results when used on several occasions. It should not produce different answers because a different person is asking the questions, or, when questions are put to different individuals, they should be understood in the same way, with no room for personal interpretation.
• Valid "“ it should measure what it claims to. In research, the following types of validity have become common areas of concern:
• Face validity "“ at a glance, does the measure actually assess what it purports to? For example, asking care managers about money spent on a service does not have face validity as a measure of client satisfaction.
• Content validity "“ the relevancy of the measurement to the area being measured. For example, if we want to assess improvements in depression following an intervention, it is important that the tool used actually measures this factor. Every part of the measurement should relate to aspects of the area under investigation (e.g. depression).It is also important to assess whether different issues are treated fairly and equally within the survey, and whether all aspects of the issue being examined have been covered.
• Construct validity "“ how far does the tool measure what it claims to measure? For example, does a satisfaction survey of an entire service actually measure satisfaction with all or just part of that service?
• Concurrent validity "“ compares results from the survey under review with other independent measures of the same area, at around the same time. For example, you would expect there to be a fair amount of agreement between children's results on a 'mock' examination and the actual exam itself.

When deciding what survey instrument to use, you have three choices:

Use an existing survey instrument

It is cost and time effective to use a scale of measurement that has already been produced, preferably if information on its level of reliability and validity are available. This will give a degree of credibility to your research. Unless there is a clear description of an instrument's reliability and validity, and how these areas were assessed, they should not be assumed. Searching through articles and books on your area of interest may be a means of identifying appropriate, existing survey instruments.

Develop a new survey instrument

If well designed, such instruments can capture information that is of particular relevance to the area under investigation, which may not be addressed by existing tools.

Adapt an existing instrument to suit the particular area under investigation

Modifying, adding or cutting out parts of an existing instrument can help to ensure that it is more tailored to the area you are particularly interested in researching.

Identify a sampling plan

It may be decided that a census is possible, that is a survey involving every individual in the population of interest. However, this is rarely possible (due to time and financial considerations) and, therefore, a sample of the population of interest has to be targeted. When using a sample it is imperative that those chosen are representative, as far as possible, of the overall population, so that information from the survey can be generalised to the population as a whole. A representative sample means that findings can be said to relate to those not included in the sample. The following steps can help to ensure that a representative sample is chosen:

• Identify precisely the population of interest "“ e.g. all service users in a particular authority, or just those using a specific service or facility. A representative sample should provide a reduced version of the larger population in terms of specific, predefined, relevant characteristics (e.g. age, gender, ethnicity).
• Identify the 'sample frame' "“ before choosing a sample from the overall population, it is necessary to produce a sample frame, a list of all those within the population of interest from which a sample can be selected.
• Identify the form of sampling to be used "“ a number of ways of selecting people for inclusion in a sample exist, some of which are associated with a greater chance of achieving representativeness than others. Different forms of sampling include:
• Simple random sampling "“ whereby the names of those to be included in the sample are randomly selected from the population, so everyone has an equal chance of being included and excluded from the sample. 'Randomisation' can be achieved via computer programmes (e.g. Microsoft Excel), tables of random numbers, drawing numbers out of a hat, or the tossing of a coin. It ensures that factors that cannot be controlled for (e.g. motivation or apathy towards the area under investigation) will be evenly represented (for more information on constructing a random sample.
• Stratified sampling "“ the population is split into 'strata' thought to be important variables to the results produced, e.g. age, gender. The same principles of simple random sampling are then applied to each stratum. This process is used to ensure that no sector of the population is under or over represented.
• Systematic sampling "“ from a list of names it is decided that every nth person will be selected for inclusion in the sample, until the sample size required has been attained. Care should be taken with this method because, due to the way the sample frame is produced, an unrepresentative sample can be chosen.
• Cluster sampling "“ is applied when a population naturally falls into groups. For example, in a school, classes can be used as the unit to be randomised, rather than individual schoolchildren.
• Opportunity/convenience sampling "“ involves using those people that are to hand to form the sample, e.g. clients that you meet with on a regular basis. Such samples are rarely likely to be representative of the population of interest, making it difficult to state how far the results can be generalised to others.
• Snowball sampling "“ occurs when a small group of people are contacted and these individuals are then used to make contact with others from the population of interest. This may be employed if there is a lack of an accessible sampling frame from which to draw the sample or because the population is likely to be highly changeable. Again, such an approach is problematic due to its lack of representativeness.
• Quota sampling "“ a common method used by market researchers. The structure of a sample that would represent the population of interest is decided in advance. For example, a sample of service users at a mental health day centre may be structured so as to reflect the make up of all users in terms of specific characteristics (e.g. age, gender, mental health problem). It is then the researchers' task to interview people until this specification has been satisfied. Again, the drawback of such an approach is that it is unlikely that those chosen will represent all those fitting the specification.

Identify how large the sample needs to be "“ it should be borne in mind that a sample only constitutes a proportion of the real population. Hence, results from the sample that are generalised to the overall population are at risk of inaccuracy; the choice of an alternative sample could give different responses, generating different results. Such inaccuracies can be reduced through the use of random sampling and careful selection of the sample size. When deciding on a sample size, researchers need to consider various aspects of their research, including:

• The research questions, e.g. the variables being measured;
• The level of diversity within the population as a whole (e.g. degree of variability in terms of qualifications, job, gender);
• The degree of precision required from the sample in relation to the overall population;
• Whether to stratify sampling to avoid over or under representation of specific groups.

As a general rule, the greater the certainty required in terms of transferability of results to the population as a whole, the larger the sample size needs to be (which will make the study more expensive). However, a large size survey does not ensure representativeness, unless careful sampling has been carried out. For example, if you wanted to gain the opinions of staff working in an authority and managed to gain responses from 50% of staff, that could constitute a large sample; but it may not be representative if the majority of those sampled happened to be male administrative workers. Random sampling should reduce the likelihood of this situation arising. Statisticians have devised tables to help with deciding on how large a sample should be to account for different sized populations, based on the level of accuracy required from the survey. Yet many authors suggest that the need for precision should be weighed up against considerations of cost and time, and the kind of analysis to be carried out, e.g. the number of different variables to be considered in analysis. If many variables are to be compared and contrasted a greater number of people will need to be sampled. For example, more people should be sampled for a study investigating opinions of mental health service users, attending a variety of day centres, with age, gender, ethnicity and mental health problem all acting as variables, compared to a study simply examining attitudes of service users at a specific day centre. When assessing how many people to sample, the issue of non-response should also be a consideration. 'Oversampling' is often carried out to compensate, in advance, for non-respondents by estimating the likely non-response rate as a percentage of the initial sample size and adding that to the sample size.

Identify how the sample is to be chosen and contacted

Once a sample frame has been established, it is important to decide how the sample will be selected (e.g. computer generated, from a book of random numbers, tossing of a coin). It is then necessary to decide how those selected will be contacted; by letter, by phone or in person. Mailed or telephone surveys tend to be less expensive because they do not require the time and cost of an experienced, qualified interviewer. However, a better response rate is likely to emerge from face to face interviews, compared to telephone and mail surveys, and from telephone surveys compared to mailed one.

Identify a strategy to plan for optimum response rate

It is important to gain as large a response rate as possible from the survey. The lower the response rate, the less one can generalise conclusions to the population as a whole, since those individuals failing to respond may prove to be systematically different to those who do, biasing results as a consequence.

Whether surveys are face to face, telephoned or mailed, it can prove advantageous to personalise all correspondence with respondents, so that they can feel part of the overall process. A pre-notification letter to those in the sample is a good idea, stressing anonymity and confidentiality, explaining how people were selected, why the research is necessary, and how the results can benefit them. Second and third reminders are often required to increase the response rate, although financial limits may prohibit this.

Acting on the results

Plans need to be made for how to analyse, present and distribute results from the study in advance, with structures in place to ensure that respondents' anonymity is maintained. It is also advisable that those who have participated in the study receive a copy of the research, in its entirety, or in a summarised form.