• No results found

Chapter 4 Case Study Method Selection of farmers, data collection and transcript

4.4 Analysis of Data [76]

D

ata analysis in qualitative research is labour intensive and may last many months (Miles Huberman, 1 984). This is because, unlike quantitative analysis where there are formal &

decision rules and procedures for identifying significant or non-significant relationships, there are no agreed formalised procedures for analysing qualitative data (Miles & Huberman, 1 994(a)). Also, the analysis of qualitative data focuses on fmding 'how' systems operate, or 'how' people behave, rather than cause and effect relationships within systems or people's situations (Strauss, 1 987; Dey, 1 993).

Qualitative data analysis has been likened to an investigative process similar to detective work (Miles

& Huberman 1984), or doing a jigsaw puzzle (Dey, 1993). In these instances the investigator or puzzle maker must make sense of the facts of the case, or pieces of the puzzle by comparing, contrasting, classifying, and cataloguing the data that they have before arriving at a conclusion to the case or completion of the puzzle. The fact finding, piece location, (sampling) and comparing, contrasting, classifying and cataloguing (analysing) activities are conducted progressively and

iteratively by the investigator or puzzler as part of the process of solving their problem (Miles &

Huberman, 1 984; Dey, 1 993). In a multiple-cilse example such as this study, the data collection and analysing activities are carried out in order to find differences and similarities between cases, to help

decide when to pursue ideas further or abandon unfruitful leads, and to arrive at an understanding of

what is being studied. The analysis of data helps the researcher find new categories of information

and to draw connections between these and existing theory (Richards & Richards, 1 994). Therefore,

data analysis forms part of the data collection process in qualitative research because it allows data collection techniques to be modified during the research period in order to ensure that accurate and complete records are collected (Hedrick et al., 1 993).

Dey ( 1 993) presented the process of qualitative data analysis as three steps to enable the researcher to proceed from raw data collection to presentation of analysed data. These three steps, description,

Case-study - data collection and analysis

classification and connection, are carried out iteratively and can be represented as a spiral (Figure

4.2)

'fhe first step, description, results in longhand field notes, interview transcripts, and detailed case

descriptions (with-in case analysis (Miles & Huberman, 1 994 (b))). In addition, matrices of case

information setting out the data about individual cases will also describe the case(s) under investigation.

Coses 3 - 7

Figure 4.2 The iterative analysis spiral

for multiple-case qualitative research

(adapted from Dey,

1993).

The second step, classification, involves sorting the pieces of the puzzle into different groups. The classification of the data obtained during semi-structured interviews is needed to 'order' the data so that sense can be made of the situation under investigation. Classification involves reducing the data (words, sentences, paragraphs) contained in verbatim transcripts and field notes into data-bits (text blocks) so these may be later allocated to defined categories.

Dey (1993

p

94

to

151)

also suggested that the classification process consisted of three steps. Glaser

(1978

&

1 992)

and Strauss & Corbin

(1994)

described this process as coding rather than classification. First categories are developed under which the data-bits can be classified (sorted).

Second the data-bits are classified under each category, and third the data-bits allocated to each category are re-classified if required by splitting categories into sub-categories if they are too broad, or joining categories if they overlap. Super categories can be created if the categories relate to underlying concepts (Figure 4.3). In order to classify the data-bits into categories, decision rules need to be developed which enable similar data-bits from different cases to be consistently allocated each time a transcript is analysed, yet be flexible enough to enable new categories or sub-categories to be created during the classification process. Computer program such as NUD.IST (QSR, 1 993) make it possible to combine steps two and three.

The advantages of using clearly delineated decision rules are that the classification process can be repeated for each case in a multiple-case case-study, and that the allocation of data-bits to particular categories is able to be clearly described and explained. The rules also allow the process to be repeated by different researchers using the same data or when investigating the same system. The use of decision rules also results in a structured hierarchy of categories and sub-categories that are useful when establishing links between the data. The criteria to decide if a data-bit is part of a category I

sub-category or not can also be specified, and logical relationships (links) between sub- and super­ categories can be established.

Maykut & Morehouse ( 1 994) called this coding process unitising where blocks of text are allocated to categories based on 'units of meaning', i.e: a block of text that must be able to be fully understood without additional explanation. Despite differences in terminology, Glaser (1 990), Strauss & Corbin ( 1 990), Dey ( 1 993) and Maykut & Morehouse ( 1 994) all recommend sorting data into appropriate categories to describe the situation being studied.

If

r

gnificant

Annotate

If not a p attern treat as a single item

If too m any

data-bits

j

Recategorise

Case-study - data collection and analysis

D E C I S ION RULES F O R A L L O C A TING T E X T B L O C K S D ata If not

f

gnificant Ignore If contextually relevant to a catego'y

!

Treat as a data-bit I f a pattern create a c ategory If a relationship create a link

If data-bits too If categories

b road overlap

j j

S ubcategorise Inter grate

(Split) (Splice)

Seek connections and create links to connect categories

If categories realte to underlying co

r

pts Create super- category

Figure 4.3 Decision rules for allocating data-bits to categories (adapted from Dey, 1993).

[79]

Although the transcripts of interviews used to classify categories are usually analysed by the researcher, the use of the interviewee in this role has distinct advantages (Ewing-Jarvie, 1 994; Kemp, 1 994 pers. comm.). First, it assists the researcher in pin-pointing the important concepts from each interview. Second, promising to provide the interviewee with a copy of the transcript may lower their resistance to the interview being recorded (Firlej & Hellens, 1 99 1 ).

The third step, connection, can be undertaken once classification has been completed. This process involves identifying links between the categories (Dey, 1 993; Maykut & Morehouse, 1 994). The links between data-bits extracted from verbatim transcripts, are identified by words. However, the words by themselves are often insufficient for determining links, and the context of the data-bit must also be considered when links are identified (Belkin et al., 1 987). Decision rules are needed to link data­ bits consistently. These rules, established qualitative data analysis, are determined by the context of the data as well as inferences drawn from the words surrounding the data-bit. Link-words, such as

because and therefore, create clear messages of a link between two data-bits, whereas links inferred

by context should be labelled in the same way as categories, i.e. X causes Y, X explains Y, doing Y leads to X, and so on (Dey, 1 993).

Having analysed the data from each transcript, the resulting categories are then compared across-case, i.e. categories existing in the individual cases are compared for similarities and differences. Matrices showing similarities and differences also show where there are gaps in the data. This is particularly useful where follow-up interviews are to be carried, as and these 'gaps' can be filled. The between­ case data comparisons can then be summarised to account for the data in a more general way (Eisenhardt, 1 989), and can be reported as a summary of the system being studied or presented graphically or pictorially, in order to aid understanding (Miles & Huberman, 1 994). As the researcher is able to focus on similarities and explainable differences, conclusions can then be drawn about the system.