How to use the software to support your analysis
The information here describes in a generic fashion what kinds of things you should be doing when undertaking computer assisted qualitative data analysis. In particular these notes emphasise the logic that lies behind the buttons. In other words it tries to explain what the computer functions will achieve and why you should use them.
To be completed.
Work to be done - Explanation and examples
Work to be done -Explanation and examples
Pros and cons of searching.
Work to be done - a few basic thoughts, there will be more to come.
Searching for codes can also be used for checking hunches and ideas; in fact a form of hypothesis checking. For example, in the study of carers for those with dementia, you might have a hunch that men who are carers get more or different kinds of assistance from both state services and from voluntary and charitable organisations. A search of the variable gender against the codes for the various kinds of help given by the various services and organisations will retrieve text for men and for women that can be compared. You might find that some codes are entirely absent for women, indicating that they didn't mention this kind of help at all. But more important than the absence or presence of coded text is a comparison of the actual content of what men said about the services and what women said. By making such comparisons you will get a real feel for how the assistance women and men get is different (or similar).
On the other hand, code and variable searching is only as good as your coding or assignment of attributes. If your codes are poorly defined, inconsistently applied and even conceptually overlapping and confused then the results of code searches will be biased and unreliable. It is up to you to ensure that codes are clearly defined and that the text coded at them is relevant and consistent. You will need to keep a constant eye out for text that has been missed in your coding. But even with near perfect coding, code and variable searching is not a perfect tool for hypothesis testing and pattern searching. Finding relationships (or failing to find them) is only reliable if the text reflects the assumptions built into such searching. These assumptions include, the recorded text is complete (it records all the relevant things that actually happened, could have been said, etc.) and the text is well structured (all the discussion about an issue are near together in the documents).
The last point is a particular problem for proximity searches. These rely on the fact that when people talk about one thing they tend to talk about related things just before or just afterwards. This may not be the case. For instance:
- People may draw together particular issues at one moment because it serves their purpose at that point in the interview or discussion, and later they may bring together quite different sets of related issues.
- Not uncommonly, later in an interview respondents remember things they meant to say earlier.
For this reason, related issues may not appear together or even near each other in transcripts. It is therefore worth remembering the caution expressed by Coffey and Atkinson:
'Given the inherently unpredictable structure of qualitative data, co-occurrence or proximity does not necessarily imply an analytically significant relationship among categories. It is as shaky an assumption as one that assumes greater significance of commonly occurring codes. Analytic significance is not guaranteed by frequency, nor is a relationship guaranteed by proximity. Nevertheless, a general heuristic value may be found for such methods for checking out ideas and data, as part of the constant interplay between the two as the research process unfolds.' (Coffey and Atkinson 1996, p. 181)
Once you have coded your data you can can do searches to explore your research questions by making comparisons and identify themes and patterns.
As you begin to search your codes you may find you need to do further coding because
- A code is too general for what is going on in the data
- Your searches have identified a new theme
Attributes and variables
Variables, also known as attributes, are familiar in quantitative research, but can also be used in qualitative analysis. Typically each case in a study might be assigned a value for each variable. (They may have no value if the variable is not applicable.) Common examples are the gender of a respondent - male or female, age of respondent in years and respondent's place of residence. Often this information is recorded in the document summary sheet.
This is similar to the use of variables in quantitative research, but in qualitative analysis we can also apply variables and values to other units in a study, like settings or events. Thus for settings, like different firms in a study, we might record the number of staff, company name and manager, or for events, the date, time, and place.
The distinctive feature of such data is that, in contrast with most analytic work in qualitative work (including coding) the values are not usually, initially a matter of interpretation. They are usually matters of fact about the person, setting etc. However, later in your analysis you might develop classifications or even a taxonomy that can be represented as an attribute and applied, perhaps, to different cases. A classification of job search strategies is such an example. The assignment of such attributes to cases is a matter of interpretation for you in doing your analysis.
Coffey, A. and Atkinson, P. (1996) Making sense of qualitative data: complementary research strategies, Sage Publications, Thousand Oaks, California.