Hester Paanakker

levels have different interpretations or understandings of values? To what extent do you feel different staff levels have differing views on how to enact values in practice? This generated patterns of mutual value perceptions that could be used to contrast the views of the different levels. The second set of questions started with an open question on how respondents perceive that the value divergence they witness impacts the way prison officers carry out and experience their work at street level. Our questions were directed at all respondents, including the ones from the policy and management arena, and asked them to reflect explicitly on the impact on implementation level. As participatory observation and document analysis revealed the occurrence of many practical implementation problems and potential moral dilemmas, we added follow-up questions on these specific effects. With respect to practical problems, we asked: Do you feel the divergence that you, or other levels, witness causes practical problems on the shop floor? With respect to moral dilemmas, we added: Do you feel the divergence you, or other levels, witness results in incompatibility with prison officers’ own ideas about delivering good work? Interviews were recorded and transcribed verbatim (411,954 words) and lasted approximately one hour. The interview items were also items in our observation and document analysis protocols. 6.3.2 Heuristics To find patterns in research data, Eisenhardt (1989, p. 541) suggests using cross-case techniques that “force investigators to go beyond initial impressions, especially through the use of structured and diverse lenses on the data” to establish “a theory with a close fit with the data.” In this study a software-supported (MAXQDA) systematic content analysis was conducted using the strategy of two-stage coding (Bazeley, 2007; Friese, 2012). This means the coding system was developed largely inductively, “going back and forth between data and codes” (Weiss, 1994, p. 156). During the first stage of open coding, subcategories were explored and created to provide “a good description of heterogeneity and variance in the data material” (Friese, 2012, p. 113) . In the second stage we established common denominators by renaming, modifying, and integrating subcategories into larger overarching value categories to build a methodological hierarchical coding system of mutually exclusive codes (Friese, 2012, pp. 130- 131). This validated version was used to code the data set at large and allowed us to compare 140 Chapter 6