Survey research is a way of collecting information from a large and dispersed group of people rather than from the very small number, which can be accommodated in a case study. The aim of all survey research is to obtain information about some specified population.
The two foundation concepts of this method of research are the concepts of the population and the sample. By population is meant the target group (i.e. the group of interest for the researcher). A sample is the group of individuals, who are selected from within a larger population by means of a sampling procedure. The purpose of the sample is to ‘stand in’ for the population from which it was taken, and therefore a researcher is not so much interested in the sample data itself, but in what it can tell in general about the population from which the sample is taken. The critical factor in determining the validity of these generalisations is the extent to which the sample can be regarded as truly representative of the population in question. This issue is related to the concept of ‘sampling error’.
Designing a survey
The process of designing a survey can be divided into the following three activities:
Suveys are undertaken in order to describe (descriptive surveys) or to explain (explanatory surveys), but in practice the distinctions between the two becomes somewhat blurred.
Descriptive survey: Description of a group in relation to some specific characteristics. An educational researcher may want to survey a representative group of school-leavers in order to find out to what extent they are motivated to succeed in their chosen fields of works. The aim is to provide a clear and complete description of the motive to achieve of those individuals studied, with the assumption that such information will also apply to the wider group of school-leavers to some degree.
Explanatory survey: This kind of surveys aim to find explanations of phenomena by asking questions about possible cause-effect relationships between different variables. These surveys are more complex than the descriptive type and therefore more complicated to carry out successfully. An example of this kind of survey could be one which looks into how far parent’s behaviour contributes to childen’s motivation to achieve highly at school. In order to carry out such a study, the researcher will need to design the research so that it is possible to draw conclusions about the population and also find parameters in parental conduct that should be isolated as possible factors on the motive to achieve.
The most common survey designs are
In the one-shot design, the data are collected from a single sample drawn from the population of interest. This design permits only descriptive research to be carried out, since it makes no provision for the collection of data with which a comparison could be made, and without that it is not possible to draw any inferences about the possible causes of the data. This must be emphasised, eveen though often there may appear to be an implied comparison with another group of people.
In the before-after design, data are collected from the members of a single sample on two distinct occasions, which are separated from each other by some kind of treatment or event, and where the presumption is that the second set of results will be influenced by the intervening process. By comparing the second set of data with the first, it is possible to see whether the second set of results has changed in response to the intervening event. The validity of this line of reasoning depends on being able to show that only the intervening event could have influenced a second set of results, and this may be extremely difficult to arrange in a survey. One example of this could be a researcher who is interested in whether the experience of getting married might influence attitudes towards those who choose to remain unmarried. That is, people who marry might undergo a shift in attitudes towards the unmarried, caused by the change in their own marital status. This question could be adressed simply by taking a sample of to-be-married couples and administering an appropriate attitude questionnare to them six months before they are due to get married and again six months after the marriage. The two set of results could then be compared, and if they differed significantly, the argument could be advanced that the difference was due to the fact that the couples were married in the interval between the first and the second data collection points. However, if we should accept such a comparison it must be possible to argue from theory for an intervening event to have an influence on the variables measured, i.e. there has to be some reason for supposing that there is a connection between the variables being measured and the intervening event so that the intervening event could provide an explanation of the results.
In the case of the newly-weds, Festinger’s theory of cognitive dissonance (1957) would predict some change in attitudes as a result of the decision to marry, and it is reasonable to suppose that this is likely to be accentuated by the occasion of the wedding ceremony itself. But there may also be other explanations such as the passage of time (the couple may have matured), uncontrolled variables (marital happiness is not always a fact), and the repeated measures effect. Consequently, it is not clear cut that the results of such a survey reflects reality.
The two-groups controlled comparison design
Data are collected from two different groups. Here a questionnaire could be given to unmarried people and to married people, and the data collected compared in terms of the research question and the theories under investigation. This method can also be used to compare cognitive dissonance in smokers and the groups to compare would be smokers and non-smokers.
Sampling
The ideal situation is a random sample (i.e. every individual has an equal chance of being selected) because this makes it more reasonable to generalise the results, but this is not always possible. The sample size is also important, and generally the greater the sample the more it is possible to generalise. Another possibility is the stratified sample, a quota sample, a systematic sample, a volunteer sample, a cluster sample, or an opportunity sample (also called a convenience sample).
There may be errors in the sampling procedure which makes it impossible to generalise so it is important to control for samling errors.
Data collection
In surveys data is collected through interviews (i.e. self-report), questionnaires and attitude scales.
The important thing is that all questions are asked, so that the data represent your research question. The researcher has to generate a preliminary set of questions, refine the raw ideas into interview or questionnaire items, organise the items into an appropriate sequence and thest the questionnaire by means of a pilot study and evaluate the results.
Open-ended or closed questions?
The questions in the interviews or questionnaires can be closed (i.e. they provide a range of possible responses to a question completely determined by the researcher, and the respondent is simply asked to select from a range of possible answers).
Example:
Q1: Which sex are you? (tick to indicate) Male Female
Q2: How old are you? (tick to indicate) 20-30 30-40 40-50 60 or older
Asking people to tick is generally more reliable than asking them to indicate those answers which do not apply, and the advantages are that the answers are more precise as to the research question (but less precise in terms of the respondent personally) and relatively easy to analyse and calculate.
Another way is to make a checklist of possible answers that people can tick, or they may be invited to rank certain items in terms of preference, frequency or some other variable.
Questions can be open-ended (i.e. they do not limit the nature of the response in any way). A question is asked and the respondent is simply provided with space to write in (or in interviews the researcher write the answer).
Example:
Question 1: Describe what specific event made you consider the problem of health in cigarette smoking?
Open-ended question may reflect the respondent’s position to a larger degree than closed questions but the disadvantage is that it is more difficult to analyse, so many researchers prefer to make closed questions, i.e. question that only have a certain range of answers possible because this will give quantitative data that can be used for statistical comparisons.
Interviews may be structured or semi-structured. The structured interview do not differ much from the use of a questionnaire, except that the questionnaire is used while the researcher is in one-to-one contact with the respondent (the term given to a participant who answers a questionnaire). Surveys consists of administering a questionnaire to a large sample of respondents, usually using a fairly structured procedure.
Semi-structured interviews are probably the most frequently used type in contemporary psychological research. Here the idea is to ask a pre-arranged set of questions but to attempt to make the interview situation informal and relaxed.
Graded response questions
Often a questionnaire designer needs to be able to provide a way to enable those people who will be completing the questionnaire to express degrees of magnitude in their answers. For example, it may be desirable to measure degrees of agreement or disagreement, or frequency. There are three types of graded response question that enable data collection
1. Attitude statement questions: the respondent is provided with a statement, which reflects an attitude to a particular issue or topic and is the invited to select an answer form a continuum which most closely matches to their response to that statement. This is an example of scaled response type of questions).
|
Example: Here is a statement about cigarette smoking and health. Choose the answer that is most likely to match your own. ‘Cigarette smoking is so dangerous to health and so expensive for society that it should be forbidden by law’. Do you: (tick only one)
|
2. Likert-type questions (scale) are virtually the same as attitude statement questions, except that they include a numerical scale, which enables the respondent’s response to be expressed directly as a numerical value. This makes it possible to gather quantitative data and make statistical comparisons.
|
Example: How dangerous would you consider cigarette smoking to be? Indicate the degree of danger by placing an x on the scale below. 1 indicates ‘hardly dangerous’ and 5 indicates ‘extremely dangerous’ --/1/2/3/4/5/-- |
Scale items such as this can have either an odd (typically 3,5,7, or 9) or an even number of points. Scales with an odd number possess a neutal zone, and depending on the nature of the question, it may be better to use a scale with even numbers because this forces a choice.
One of the advantages of Likert scale questions are the fact that they generate numerical data, which can easily be summarised across a sample.
3. Semantic differential type questions use a technique developed by Osgood and his co-workers (1957) for providing a way of quantifying feelings, emotions, perceptions and similarly subjective variables, which are otherwise difficult to express numerically. All that is required is that a number of dimensions of the target variable are able to be represented by one or more pairs of bi-polar adjectives. Respondents are asked to identify the position on each of the continuums which best describes their feelings. Each of these can subsequently be represented numerically when the data are analysed.
|
Example Q8. Place an x on the scale to indicate how you would describe your relationship with your partner as it is at this moment. Satisfying/.........../........../.........../.........../......../......../........./Unsatisfying Undemanding/..../...../....../....../....../......../......./demanding Unstable/....../...../...../....../....../......./....../stable |
It is important to ensure that questions are formulated in such a way that there is no possibility for misunderstanding. In answers to tick avoid using words that are imprecise and may be interpreted differently by different people, e.g. never, seldom, sometimes, always etc. Rather say: once a week, more than once a week etc.
Questions to avoid:
Survey