Bit-Wise Coding for Exploratory Analysis

Abstract

Getting data to yield their insights can be hard work. This is especially true with survey data, which tends to be oriented toward the presence or absence of characteristics, or attitude relative to some arbitrary midpoint. A good example of the first comes from surveying the web by looking at websites: Ho (1997) analysed the value-added features used by 1800 commercial websites to form a profile of commercial use of the Web within various industries; the U.S. Federal Trade Commission (2000) examined the characteristics of the privacy policy on 426 websites; West (2004) looked at 1,935 national government websites for 198 nations around the world, evaluating the presence of various features dealing with information availability, service delivery, and public access. A good example of the second comes from any number of studies that use the ‘Likert scale’. These studies are characterised by (a) large sample sizes and (b) an analysis that must incorporate a large number of ’indicator’ and ’categorical’ variables. Coding and analysis should be considered at design time: not only ’what information to collect’ but also ’how to store it’ and ’how to use it’. This is particularly true with web-based surveys using HTML forms, since the data can be stored automatically without the intermediate step of transcribing the answers from paper questionnaires into the computer. Getting a relevant graphical view of the data is often essential, since the human eye is a powerful analytical tool. The help files that come with statistical analysis applications explain particular techniques, but the importance of coding is often obscured by the description.