Qualitative_data_analysis
Qualitative_data_analysis
Qualitative_data_analysis
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
ASSIGNING CATEGORIES 125<br />
we ought to follow it as consistently as we can right through our <strong>analysis</strong> of the<br />
<strong>data</strong>.<br />
This does not mean, though, that we have to categorize every bit of <strong>data</strong>. Even if<br />
we have previously summarized our <strong>data</strong> and eliminated some of ‘the dross’, there may<br />
still be parts of the <strong>data</strong> which turn out to be less relevant than expected as the aims<br />
and direction of our <strong>analysis</strong> become more focused. There is no point in<br />
categorizing <strong>data</strong> which is not clearly relevant to the <strong>analysis</strong> as it develops.<br />
So far we have only considered how to identify bits of <strong>data</strong> through judgements by<br />
the analyst which take account of irreducible ‘units of meaning’ in the <strong>data</strong>. This<br />
approach respects the integrity of the <strong>data</strong> and also ensures that bits of <strong>data</strong> are<br />
meaningful both internally and with respect to the <strong>analysis</strong>. Another approach is to<br />
allow the computer to create bits of <strong>data</strong> for us. For example, we could simply<br />
divide up the text on an entirely arbitrary basis, asking the computer to demarcate<br />
bits of <strong>data</strong> by a specified number of characters or lines. This would certainly ensure<br />
consistency in the size of <strong>data</strong>bits, but at the expense of intelligibility. However, we<br />
can use this facility more selectively, by focusing on target keywords or phrases in<br />
the text, and asking the computer to extract all the bits of <strong>data</strong> which contain the<br />
specified target. The size of these extracts may be specified in different ways,<br />
according to the number of characters or lines before or after the target text, or (less<br />
arbitrarily) to include the sentence or paragraph in which it occurs. For example, the<br />
computer could extract for us all the sentences which contain the word ‘Cézanne’.<br />
This method of generating bits of <strong>data</strong> is quick, but it also has limitations. The<br />
boundaries which demarcate bits of <strong>data</strong> are arbitrary, even where the computer<br />
extracts the contextual sentences or paragraphs. The analyst no longer defines the<br />
appropriate ‘unit of meaning’, and has to make do with whatever results the<br />
computer produces. This disadvantage can be offset somewhat, if the computer<br />
allows us to retrieve the original context and modify the boundaries of the bits of<br />
<strong>data</strong> it has extracted as appropriate. The time gain may depend on how selective we<br />
can be in checking contexts and adjusting the boundaries of bits of <strong>data</strong><br />
retrospectively.<br />
Another drawback of this approach is that it identifies ‘units of meaning’ entirely<br />
in terms of key words or phrases in the text. These are unlikely to be the only bits of<br />
<strong>data</strong> relevant to the <strong>analysis</strong> as a whole, or even to the particular aspect of the<br />
<strong>analysis</strong> in which we are interested. Thus the computer cannot pick up any<br />
discussion of Cézanne where he is not explicitly mentioned in the text. Generating<br />
bits of <strong>data</strong> automatically is unlikely to exhaust all possibilities, and may therefore<br />
be better regarded as a way of complementing rather than substituting for judgements<br />
by the analyst.<br />
Let us now turn to some of the problems of assigning categories. This is likely to<br />
prove the most challenging but also the most rewarding aspect of analysing the <strong>data</strong>.