A key survey design question related to the use of measurements for variables. As can be seen from Table 5.1, the variables, which are necessary to answer the research objectives, were coded according to data type (e.g. ordinal, Likert). The decision about the appropriate data type for each variable is based on two questions: what data types can represent the variable (epistemology), and what data types are needed to reveal the variation in behaviour and to construct a typology. In this case the response to every variable is not comprised of the same data type, as for some variables the data would be difficult to accurately obtain (or may not be possible to represent in this way). The full survey can be found in Appendix C.
There are nine different data types in table 5.123. Some of these data type descriptions relate to intuitive categories (e.g. date [month and year] of an event), some are less intuitive (e.g. Likert). The following section describes the survey design considerations relating to the data types: Likert; text; binary; and categorical.
The use of Likert items, based on semantic differentials, was determined for the survey in order to provide a flexible scale, but with regular intervals that would be comprehensible for respondents. The use of Likert items is very common in social science research, particularly for the measurement of attitudes (Brill, 2008) across a range of declarative statements (Peterson, 2000) and has been used in housing market studies in identifying buyer preferences and behaviours (Bible and Brown, 1981; Greaves, 1985; Munro and Lamont, 1985). The use of Likert items also includes the possibility of extensive use of quantitative methods in the analytical stages of the research, as the items are readily quantifiable through the use of representative consecutive integers. Five point scales were used; although adding points to this scale allows additional granularity there is a trade off in terms of the precision with which people can accurately identify their attitudes with extensive scales24. Even number point scales (forced choice scales) were rejected, as it was considered important that households could indicate that they neither agreed nor disagreed. Two potential biases occur with Likert items that may be of issue here; acquiescence bias and central tendency bias (Welkenhuysen-Gybels et al., 2003; Bertram, 2003). These biases are normally overcome by asking multiple questions on the same topic, but framed differently, for example some positive and some negatively (Brill, 2008). However, given the extensive nature of the research topic this was not possible as the survey would have been overly lengthy.
Fig. 5.1: An example of the Likert item questions in the survey
A horizontal scale for the Likert items was selected (over a vertical scale). A horizontal scale allows respondents to move quickly through the range of questions that are grouped together (e.g. factors influencing the decision to search) and there is less ambiguity about where the tick should be placed (Bryman, 2008). Horizontal scale can be clearly related to numeric values (e.g. Strongly Agree (1), Agree (2) etc.) in order to save space in the survey, and hence avoid length based non-response or missing data (e.g. Fig 5.1).
Text based data
The survey included some free text boxes for respondents to reply in an open setting (see Fig 5.2 for an example). Whilst free text boxes allow greater detail into respondents’ perceptions of their behaviour their use may increase non-response rates (Bryman, 2008), and hence were used for only some of the questions, which were priorities for open text (see Table 5.1 for variables containing free text boxes). The text from these boxes could be transformed to quantitative data (for example counts of the numbers of specific words), but in this research project the quantitative variables required for analysing the variation and creation of the typology are all included in the survey in other forms (e.g. Likert and binary). The inclusion of text based questions enables qualitative checks on the survey responses and also provides examples to illustrate the narrative of housing search variation and classification.
Fig. 5.2: An example of a text based question in the survey
Some variables required only a binary response, these variables relate to the presence or absence of the variable. For example, for variables relating to the characteristics of the dwelling purchased this might relate to the presence or absence of a garden (a binary distinction). For these forms of question a tick box for ‘Yes’ and a tick box for ‘No’ were selected (see Fig 5.3 for an example). Other forms of binary response are possible, for example writing ‘Yes’ or ‘No’, but tick boxes were selected to require as little work as possible from the respondent (and therefore minimise fatigue) and to facilitate the coding of surveys upon completion. Horizontal options were selected in order to limit the space taken up by the questions and to follow a similar format to the more frequently used, Likert item, questions.
Fig. 5.3: An example of a binary question in the survey
Categorical data is required for some variables relating to household and housing characteristics. Variables that use categorical data must be capable of being represented by discrete (non overlapping) options, for example property types are normally known by discrete categories (e.g. semi-detached). For the same reasons as those for binary data, categorical data options were horizontally embedded in the survey.
A note on sensitive questions
Questions on household income typically receive the highest non-response rate in surveys (Ross and Reynolds, 1996). This may be because of concerns about the personal nature of the topic and confidentiality (Peterson, 2000). In this instance three measures were undertaken to mitigate non-response for this question. Respondents may also be concerned about their income being low on the list of options; the inclusion of low-income brackets may therefore encourage responses from low-income households (Peterson, 2000). First, the introductory letter highlighted the confidentiality and anonymity of respondents. This was repeated in the instructions for the income question, including the phrase “This information will be treated in the strictest confidence.” Second, sixteen categories of income ranges were selected for the survey so that households will not be required to reveal their precise income.25 Third, the question categories began with low-income levels (Below £5,000) in order to encourage owner-occupiers to respond.
The survey included an introductory letter setting out the research agenda, funding for the survey and that the research was conducted by the University of Sheffield, as there is some evidence to suggest that response rates are higher for university based research than other research bodies (Edwards et al., 2002). This letter was designed to fulfil these three aims in a clear and brief manner and perspicuity and brevity have been shown to increase response rates (Couper, 1997).
Download 13.27 Mb.
Share with your friends:
The database is protected by copyright ©sckool.org 2020