Data analysis

Table 10: Summary of the purpose of each data-collection method

Source of data Method for collection Purpose in relation to the research question

Interviews

x Semi-structured interviews with open-ended questions with agents and middle managers, conducted in 2 sets

x Non/semi-structured interviews/meetings with other company managers, conducted in 2 sets

- Set 1: Exploring a wide scope of views of potential elements influencing performance in call centers - Set 2: Testing and refining initial analysis of the drivers of performance in a call center setting based on further findings from 4 subcases (which generated 4 types of elements that explain performance).

Interviews provided a thorough understanding of elements and their character, the essence of the link between elements and performance in a call center setting.

Observations

Direct and participant observations of:

x Daily activities and behaviors at work, on- and off-schedule x Work-group meetings x Managerial meetings

Provided additional insights regarding:

- Impact of the work setting

- Dynamics and interactions between organizational actors

- Workplace behaviors (in real time)

Observations entailed an in-depth understanding of the character of elements influencing performance and as a complement to interviews regarding the link between elements and performance in a call center setting.

Archival data

x Collecting reports, documents and records

x Collecting data from company performance-measurement system

For a thorough understanding of performance in a call center setting.

- Provided proxies for establish areas and patterns of performance, which generated 3 performance categories

- Provided insights of challenges between categories and areas of performance over time

- Provided a selection and in-depth study of 4 subcases with different performance levels over time for covering a wide scope of individual- and group-based performance.

90

framework. This analytical process provided opportunities for elaborating on discrepancies between theory and empirical data, and enabled a detailed comparison between performance metrics utilized in the case company and those in prior studies. The combination of pattern-matching and iterations was important for staying as true to the data as possible during the research process (Lakatos, 1970) and revealing novel findings.

4.3.1 Coding the empirical data

All empirical data was analytically coded. This coding process was represented by taking apart segments of data to allow the complexity of the entire scope of data “to speak” (Bryman & Bell, 2007). Two separate analytical coding processes were carried out. One process (A) was based on coding the qualitative data from interviews and observations, whereas the other process (B) coded archival data. However, both processes were pragmatic methodological schemes for comparatively analyzing data between cases at the micro-level. Both schemes are beneficial for understanding a phenomenon at the aggregated organizational level (Lieberman, 2005). Both analytical coding processes also leaned on my analytical judgment to establish when saturation was met, both between empirical data and between empirical data and theory (Eisenhardt, 1989;

Glaser & Strauss, 1967). The following paragraphs describe how I carried out the two analytical coding processes.

Analytical coding process (A): Coding the qualitative data

Table 11 illustrates that the qualitative data was coded in three main stages.

91

Table 11: Illustration of the analytical coding process of qualitative data Coding stagesAimContent Output ExamplesCarried out Stage 1 Open coding process (raising raw data to a conceptual level) Iteratively categorized all quotes from interviews, group meetings and notes from interview set 1 and initial observations into themes by: a) merging data sources into an Excel spreadsheet b) initial coding of data material Generated 365 codes“We have a lot of statistics. It’s a good thing because then I can follow how I perform” (Agent, Case Beta, April 2012). Æ codes: absorbing control, work tools, statistics, attitude toward work “Sometimes I just want to talk calories instead of kW, but I feel a bit stuck since I live in the north and there are no other job opportunities” (Agent, Case Gamma, April 2012). Æ codes: location of the call center, attitude toward work, development, job satisfaction

Autumn 2012 Stage 2

Creating second-order coding families for a broad knowledge of the detailed data.

Interpretatively establishing key concepts of the study generated from the large number of codes. Established after follow-up interviews.

Generated 40 second-order coding families

“One’s perceptions of the level of surveillance at work has a lot to do with the manager you have” (Agent, Case Epsilon, March 2012). Æ Interpretatively coded as the coding family of managerial control “What would make me feel better is a quiet room…a quiet work environment is a utopia within customer service” (Agent, Case Gamma, April 2012). ÆInterpretatively coded as the coding family of perceptions of physical environment and well-being

Autumn 2012– mid 2013 Stage 3

Categorizing recurrent themes, categories and patterns Identifying main elements influencing performance and their relation to the three performance categories, or theorizing the drivers of performance. Empirical presentation following the performance categories.

Generated 10 coding families “I think that if I would be given better pay, my performance level would rise. It feels pointless to constantly outperform when it doesn’t give any dividends, despite I’m keen to do it” (Agent, Case Beta, 2013). Æ Categorized as the element of Appropriate levels of material rewards “You don’t see the same effort among those who have worked her for a long time and [that] don’t find it as fun as these younger ones coming from Manpower” (Middle manager, Case Gamma, Nov 2013). Æ Categorized as the element of Tenure

Autumn 2013 Identifying core elements influencing performance (individual and interpersonal elements) and their relation to the 3 performance categories again after interview set 2 and accompanying observations. Empirical presentation now instead following the elements.

Generated 4 coding familiesRepresenting the 4 elements described in the empirical findings 1) The individual element of coping - The three interpersonal elements: 2) The contextual elements 3) The control-based elements 4) The cultural elements (All described in Chapter 7) Spring 2015

92

The first stage of this coding process was an open-coding process aimed at raising the “raw data to a conceptual level” (Charmaz, 2006; Corbin &

Strauss, 2008) and gaining an initial understanding of the wide scope of elements influencing performance in the case company. This process was carried out by categorizing quotes from all interviews (in set 1) and notes from group meetings and observations (during the same time period) into themes. This was an iterative process between the data and the preliminary theoretical framework. This process generated 365 codes on a cross-case basis, since the codes were based on data from actors within all work groups at Eon CS at that time.

The second stage of this coding process was aimed at keeping a clear link between the details of the rich data, the theoretical point of departure, and the initial empirical findings, while aggregating the 365 codes into second-order coding families. By interpretatively establishing key concepts of the study (from patterns of recurring answers and notes) and creating a broader picture of the data, this stage of coding eventually resulted in 40 coding families.

Finally, the third stage of coding aimed to create a comprehensive picture of recurrent themes and categories in the empirical data. Becoming intimately familiar with each case (Eisenhardt, 1989) and including archival data (triangulating the analysis) eventually resulted in 10 coding families. These families (representing the main elements influencing performance in this context) were first presented according to the performance category they influenced. This presentation primarily highlighted the outcomes rather than the actual elements (to avoid repeating descriptions of their impact). However, these findings were further refined after conducting the second set of interviews and making additional observations. This was done because the findings aimed to test the initial analysis, and the validity of the 10 coding families and their relation to performance. By iterating between additional collected data of the 10 coding families and additional organizational theory to gain a better theoretical understanding of the findings, this analytical process provided a more thorough understanding of my research question. The 10 initial coding families were refined into four coding families by testing, refining, and further analyzing the findings. These four coding families, which I refer to as elements in this study, are further presented in relation to their impact upon the three performance categories in Chapter 7.

This three-stage coding process provided a detailed comparison between raw qualitative data and aggregated categories. The analytical coding process of the qualitative data not only facilitated a novel interpretation of empirical data (Alvesson & Sköldberg, 2009), but also gave a broad overview of the context (Miles & Huberman, 1994) and revealed the complexities in how to manage performance in a call center context.

Analytical coding process (B): Coding the archival data

During the research process, collected archival data (Chapter 4.2.1) provided hints of how performance was conceptualized, defined, measured, and evaluated at Eon CS. These hints were crucial, given that I used actual performance data (derived from company databases) to establish individual- and group-based performance at Eon CS (subjective performance metrics were excluded from this study; see Table 22 in Appendix 3). To make sense of the rich, detailed, performance data while relating it to theory, an analytical coding process of archival data (such as performance metrics and actual performance data) was carried out.

The analytical coding process of the archival data was carried out in three steps. During the first step, I aggregated the individual performance data into the group level for each work group (14 in total) at Eon CS (both company sites) to establish performance levels of each work group while keeping a certain richness of the performance features at the group level.

The aggregated performance within each work group and metric was then arranged according to the performance outcome for three years (2011–

2013), which was also compared against the average performance in the company.²⁹

This second step of the coding process clearly shows that the work groups (as an average of all agents included in each group) generally performed equally within certain metrics (compared to other groups and the company average) and over time, although some differences were found within the groups at the individual level (Chapter 6.3.1).

These patterns at the group level over time enabled me to cluster various performance metrics into three different performance categories for the third step of this coding process. The performance categories generated from this step represented various types of performance: Routine-based

29 Agents’ performance data from 2011–2013 was the primary data for the performance categories. However, performance data from 2014 was later compared to this data set, which overall aligned with this data material.

94

efficiency (performance category A); social efficiency (performance category B); and problem-solving efficiency (performance category C).

These performance categories were further established while conducting this study (Chapter 4.2.1, interview set 2). This analytical categorization of performance metrics that are regularly and systematically measured at Eon CS at the individual and group levels, and the labels used for each category, reflect my analytical interpretation and theoretical understanding of performance.³⁰ The performance categories generated by this analytical coding process will be used when analyzing the empirical data of this study (Chapter 8).