were used for documenting ideas, comments, and decisions throughout the analysis
stage, as such representing a running record of data analysis and interpretation that
allows for traceability of ideas and decisions. Memos functioned to capture and
document new ideas concerning the research materials. In addition, the memos also
evolved throughout the research process whenever the researcher obtained additional
information or new insights concerning the topic described in a particular memo.
4.6.3 Data reduction
The first activity in qualitative data analysis is data reduction (Miles and Huberman,
1994). In order to reduce the data we collected in the case studies we coded the interview
data collected in the field as well as the observation notes. In order to improve the
validity and reliability of the study, we used the qualitative analysis software Atlas.ti
5 (Atlas.ti, 2004) to manage the coded data. Using software leads to more systematic
analysis procedures and guards against information-processing biases (Miles and
Huberman, 1994; Eisenhardt, 1989).
We have followed the coding scheme as suggested by Strauss and Corbin (1998). Firstly,
open coding was used to assign tags or labels to the data collected as such identifying
the concepts under research as well as their properties and dimensions. Open coding
in its purest form is intended to start the data analysis openly and identify concepts
and dimensions from within the data only (Strauss and Corbin, 1998). We decided not
to start our data reduction phase completely from scratch
but from a brief collection
of generic codes based on literature concerning modularity aspects and practices. We
listed these generic codes in an initial codes list. This list facilitated us in analyzing
our data from a modularity perspective. Yet, because of its generic and open character
we ensured an open mind with a minimum amount of bias.
Each interview was coded independently by two raters who then compared and discussed
their codes to reach consensus on each of them. In total, three raters were involved in open
coding to improve the validity and objectivity of the findings. The principal researcher
involved in this study coded all interviews; the other two raters each coded half of the
interviews. During the process of open coding, the initial codes list was expanded to
encompass emerging themes and cover the richness and nuances of the collected data.
The next step is axial coding. The objective of this step is to regroup and link categories
with each other in a relational manner (Miles and Huberman, 1994, Voss, 2009). We
therefore bundled text fragments with similar codes and systematically analyzed
their contents to reveal the core concepts related to various aspects and practices of
modularity and long-term care. We were thus able to gain insight into the different
(sub) categories involved and revealed by the data.
80
Modular Care Provision
The final step is selective coding, i.e. selecting a core category and relating it to other
categories. For this step, for each research question we identified the (sub) categories
of codes that were most relevant for the question under research and selected the
main category from which to start. As such, we made different cross-sections of the
data that we considered the most appropriate for structured analysis in the light of
a particular research question. In this way we were able to develop insights into the
mutual relations among code categories from different perspectives.
4.6.4 Data display
The second activity in qualitative data analysis is data display (Miles and Huberman,
1994). Displays are organized, compressed assemblies of information that permit
drawing conclusions and action. Displays summarize the case evidence and indicate
how focal constructs have been measured, as such creating a strong link between
qualitative evidence and the research findings (Eisenhardt and Graebner, 2007). Data
displays
can be extended text, matrices, or other schematic overviews.
In the chapters that follow we will use both detailed narratives and summarizing tables
and figures to demonstrate our analysis process and show how we built our insights
and findings concerning modularity in long-term care from the rich interview data
and observation notes. The use of the different means for data display differs slightly
among the various research questions that we address in this study. In general, as
the total care process is our unit of analysis, most displays are organized around this
process.
In addition, given each research question, the displays
relate the modularity
aspect or practice of interest to stages of or activities within this process.
For each research question, we first analyzed the pattern of data within cases in order
to become familiar with each case as a stand-alone entity and to allow the unique
patterns of each case to emerge (Eisenhardt, 1989). In this respect, we created, among
others things, graphical flowcharts and causal networks (Chapter 5), taxonomies and
time ordered matrices (Chapter 6), and event listings (Chapter 7) for each of the four
cases. These displays formed the basis for the next and final step in qualitative analysis,
being conclusion drawing and verification.
4.6.5 Conclusion drawing and verification
As explained, for each of the research questions we made a different cross-section of
the data for structured comparison. In order to draw valid conclusions on each of the
cross-sections, we looked for cross-case patterns. This is essential for enhancing the
generalizability of conclusions drawn from the cases (Eisenhardt, 1989). Our overall
strategy was to look for patterns based on the descriptions and displays of each case,
81
Resear
ch desig
n
and methods
Chapt
er 4