Qualitative data analysis

ions are powerful means of making comparisons, but we must also remember their origins and limitations. In making comparisons, it is helpful to distinguish between two forms of relation between objects or events (cf. Sayer (1992:88). On the one hand, we can identify ‘substantial’ relations of connection and interaction. When we laugh at a joke, for example, there is a substantial connection between the joke and our laughter. On the other hand, we can identify purely ‘formal’ relations of similarity and difference between things. Thus we can distinguish between jokes and laughter, as different types of phenomena. This type of comparison involves categorizing phenomena according to their similarities or differences. In this and the following two chapters, I focus on how we can categorize qualitative data, before considering issues raised by substantive connections. In this chapter, I shall look at the problems of generating categories for the analysis. In the following chapter I turn to the issues raised in assigning these categories to the data. Then in Chapter 10, the last dealing with categorizing data, I consider how we can ‘split and splice’ categories—in other words, subdivide or integrate categories as ways of refining or focusing our analysis. Naturally, we have to create categories before we can assign them, and we have to assign them before we can split or splice them. But despite this logical precedence, in practice we may find ourselves shifting backwards and forwards between these different aspects of categorizing data. The very quality of qualitative data—its richness and specificity—makes for problems when we try to make comparisons between observations. For what are we comparing? There are no standard categories in terms of which to compare observations. Indeed, there are no clear boundaries as to what constitutes an observation. To compare observations, we must be able to identify bits of data which can be related for the purposes of comparison. How can this be done? The answer is deceptively simple. In principle, we could organize the data by grouping like with like, so that any observation which seems similar to or related to others can be grouped with those observations. We can put all the bits of data which seem similar or related into separate piles, and then compare the bits within each pile. We may even want to divide up the items in a pile into separate ‘sub-piles’ if the data merits further differentiation. We can then compare observations within each pile or subpile, looking for interesting similarities or differences within the data. We can also make comparisons between the different piles or sub-piles, again looking for patterns or variations in the data. However, this procedure begs two important questions. First, what is an observation? We referred above to ‘bits’ of data, but how are these bits to be identified or distinguished from the rest of the data? There must be some criterion or criteria which allow us to distinguish one bit or observation from another. Second, how can an observation be judged similar to or related to some other observations? Why put a bit of data into one pile, but not into another? Because they are alike, or related? But things are not just alike or related—they are alike or related in some respect or another. Although we may say that observations are alike or related without explaining why this is so, nevertheless there must be some respect or other in terms of which this judgement is made. If we distinguish between an employer and an employee, for example, we implicitly refer to a variety of social and CREATING CATEGORIES 101