structured codingin QUESTIONNAIRE design, analysis is assisted by the questions being 'structured’, so constraining response. Structured questions can be either open-ended or closed. Closed questions are those where respondents are presented with a list of options and are asked to indicate their answers by ticking a box or circling a number. Open-ended questions are those where respondents are asked to write down their replies which are then coded according to a pre-coded schema.
When using structured questions care must be taken that the questions asked refer to only one dimension. For example, religion has at least two distinct dimensions – the strength of each person's religious conviction and their nominal religion (into which they have been initiated). Care should also be taken to indicate that the answers are exclusive of one another.
Another problem is the purely practical one of including every possible reply in the coding schedule. The use of catch-all (‘other) categories should be avoided as far as possible at the coding stage. since such a category may contain many different replies.
To assign numerical codes to the data the researcher should, where possible, make use of coding schemas which have been professionally designed by competent experts in the field of study. Not only does this procedure simplify the researcher's task but it also leads to the accumulation of data which is readily comparable. For example, in undertaking occupational research the researcher is generally advised to make use of the full Registrar General's classification devised for the analysis of CENSUS material.
Generally, however, the researcher is forced to develop his or her coding schema. This involves first taking account of the level of measurement (see CRITERIA AND LEVELS OF MEASUREMENT) which is being employed. Where interval variables are being employed the researcher is advised against precoding the data into a series of numerical bands since a precoded question might result in clustering within one band. Where a variable has ordinal properties it is generally best to code it as such as this will increase the number of statistical tests which are acceptable to use. When coding nominal data attention should be given to whether it is possible to group the data into a more convenient structure at a later stage of the research programme.