The analytic framework of qualitative content analysis according
to Graneheim and Lundman (2004) was used in the analysis.
The analysis was built on codes, which were sorted into categories
and subcategories. Firstly, the interview transcripts were read
through by the interviewer while listening to the recordings, to
make corrections in the text. Each interview was read at least once
without any simultaneous coding. The initial analysis and the coding
of data were conducted by the first author, while the coauthors
took parts of the material and worked with the analysis
during meetings throughout the analytical process. Codes were
written as short sentences that mirrored the content of text segments
in order not to lose information contained in the text and
to maintain fidelity to the material. In the conduct of the analysis,
safety culture was considered as consisting of normative beliefs
and fundamental values (assumptions, expectations, philosophies,
norms, and rules), with regard to safety at a workplace (Mearns
and Flin, 1999). Text segments in which it was judged that participants
gave descriptions of safety that were normative or based on
values were identified as meaning units and coded. Reasons for
risk-taking were coded during the analysis, in which text segments
with explicit reasons for risk-taking were identified as meaning
units and coded. The coding of the data was inductive in the sense
that codes and categories were derived from the data and not identified
in advance. The codes and categories were shaped and
named (abstracted) successively during the analysis and were
not based on the questions in the interview guide. The qualitative
data analysis software Atlas.ti 6.2 (ATLAS.ti GmbH, Berlin) was
partly used in the analysis to code and sort the data.