Data analysis followed the Constant Comparative method described by Glaser & Strauss (1967). They recommend that the researcher begins by coding each incident into as many descriptive categories as possible. At the same time each data item in one category is compared with all other items in the same category. From this constant comparison it is possible to begin to generate theoretical properties of each category. Glaser & Strauss also recommend that thoughts which arise during the coding process are recorded separately as theoretical memos. These form a continuing record of the process of theory development, recording the blind alleys as well as the ideas which eventually prove to have explanatory power. As coding continues, the constant comparisons change from comparing one data item with another to comparing new data items with the properties of the categories which emerged from the initial comparisons. There will come a time when it will be possible to integrate categories and their properties into an overall schema, unified by a smaller set of higher-level concepts. Theoretical saturation will also gradually limit the number of new categories and properties. The process of analysis is deemed to be complete when the theory advanced describes the existing data, when it is grounded on data of sufficient breadth to answer the aims of the enquiry, and when new data adds little or nothing to the properties of established categories. Glaser & Strauss argue that theory generation using this method is well worth the time and effort:
"Theory based on data can usually not be completely refuted by more data or replaced by another theory. Since it is too intimately linked to data, it is destined to last despite its inevitable modification . . . " (Glaser & Strauss 1967: 4)
The key words here are "based on data". One problem with Grounded Theory methods is that coding and theory development take place in the mind of the researcher, while the sheer bulk of data in a qualitative study makes it impossible to present more than illustrative examples of original data to the critical reader. What guarantee can be offered to the reader that the researcher has in fact grounded the theory upon the body of data? This potential criticism applies particularly to a study using Critical Incident Technique. Schneider & Locke (1971) published a critique of the reliability of the categories developed from critical incidents by Herzberg et al. (1959) in an influential book, "The Motivation to Work". In that study, critical incidents were used to collect information about factors which motivated or de-motivated employees. Schneider & Locke demonstrated that the original classification system devised by Herzberg et al. was logically flawed. It was an artificial system externally imposed upon the data and it did not fit the data soundly. As a safeguard, Schneider & Locke recommended that the researcher should always try to enlist the help of the informant when coding material.
In a study using Grounded Theory methods this is not fully possible, since the coding scheme develops over the whole period of data collection and is not finalised until quite a late stage. However, it is possible to conduct inter-rater reliability studies to test the extent to which the final coding scheme will give the same results when external raters code the same data. Desmond Cormack (1983) in a study using critical incidents offered a measure of inter-rater reliability based on a formula developed by Cohen (1960). Cohen's formula has been adopted for the present study, since it allows one to produce a coefficient of agreement for nominal scales from which the amount of agreement expected by chance has been removed. Chance agreement is a function of the length of the coding scheme. Cohen's formula is more appropriate than a Chi-square test in this context because Chi-square only measures association. Thus with Chi-square, significant disagreement between raters would affect the reliability score as much as significant agreement. With Cohen's formula, this problem is eliminated.
The formula is quite straightforward to apply. It involves counting ratings and placing them on a grid, calculating chance agreement by finding the joint probabilities of the marginals, then applying the formula. This gives a measure of K, which is the measure of agreement after chance agreement has been removed. Further calculations can be performed to estimate the degree of significance. In the present study, a significance level of at least 0.05 was specified as a requirement at the outset.
THE CODING SCHEME
The coding scheme applied equally to the three groups of incidents: those from written questionnaires, from interviews with patients and from interviews with nurses. The scheme makes use of two levels - AREA and CATEGORY. There are five Areas describing the five main elements of the study:
A. How the nurse first became aware of the patient’s anxiety, worry or distress.
B. The source of the patient's concerns. C. The nurse's actions.
D. The nurse's intentions.
E. The outcome of the nurse's intervention.
Detailed descriptive categories were developed within each of the five Areas. For example, Area B dealing with the source of the patient's concerns comprised categories for worries about health, treatment, the environment of care, relatives and friends, being discharged, and a catch-all other concerns section (full details of all categories are shown in Appendix Three). The category headings were developed from the constant comparison of incidents, which facilitated the identification of similarities and differences. Where the number of categories in any one Area was large, these were subsequently grouped under aggregate
category headings, which were used to summarise the main themes of each group as they appear in the Results Chapters.