A non-agentic view on career success
The impact of career shocks
Research Master Business in Society
Thesis written by: Noud Schartman Studentnumber: UvA – 11429852 VU – 2521951 VUnet ID – nsn810 Thesis supervisor: dr. Stefan Mol Submission date: 18th of August 2018
2
Statement of Originality
This document is written by Noud Schartman who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 Table of contents Abstract………...4 Introduction………...5 Theory………...8 Career success………...………...8 Career shocks………...11 Methods………13 Sample………...13 Measures………...14 Analyses……….16 Results………...20 Topic labels………...20 Regression analyses………..24 Discussion………29 References………...33 Appendix A………..38 Appendix B………..41 Appendix C………..44
4
Abstract
This study examines the relationship between career shocks and career success. Prior research has focused on a positivistic and deterministic view of careers in general and career success specifically. By taking an agentic perspective and focusing mainly on the individual, researchers have neglected other factors that could have an impact. We argue that more attention to events in the context, operationalized as career shocks in this study, can supplement the current understanding of career success. We first employ Structural Topic Modeling to induce ten topics that represent career shocks from our interview data. This data, consisting of a sample of 188 Dutch employees, was collected specifically to gain information on career shocks. We then labeled the ten topics that we induced from the data to represent different forms of career shocks and integrate these topics into our dataset. With the aim to investigate the relationship between career shocks and career success we ran several regression analyses with the topics as predictors. We did not find any evidence of a relationship between career shocks and measures of objective and subjective career success. This study contributes to the literature by taking an inductive approach to career shocks and to explore the non-agentic and unpredictable side of career success.
5
Introduction
In recent years the topic of career success has been gaining more and more attention of organizational scholars (Akkermans & Kubasch, 2017). Much effort has been put into trying to identify predictors of career success. A meta-analysis by Ng, Eby, Sorensen & Feldman (2005) identified more than twenty-five antecedents which were classified in the overarching domains of human capital, organizational sponsorship, stable individual differences and socio-demographics. Career success is defined as ‘‘the accomplishment of desirable
work-related outcomes at any point in a person’s work experiences over time” (Arthur, Khapova &
Wilderom, 2005, p.179) and scholars often distinguish between subjective career success and objective career success (Hogan, Chamorro-Premuzic & Kaiser, 2013; Ng & Feldman, 2010). On the one hand, subjective career success refers to feelings of accomplishment and satisfaction regarding one’s career (Seibert, Crant & Kraimer, 1999) and is often operationalized as career satisfaction (Heslin, 2005). On the other hand, objective career success is often measured by looking at salary, promotions, or prestige (Hogan et al., 2013; Ng et al., 2005).
Prior research has, for the most part, focused on the career as something that is shaped by both the individual and the organization. However, since the beginning of this millennium, researchers have looked at new career paradigms (Akkermans & Kubasch, 2017), in order to address the growing complexity and dynamic nature of work. As a result of such developments, individuals are increasingly responsible for developing and managing their own career, and organizations are less likely to offer stable, identifiable career paths for employees (Converse, Pathak, DePaul-Haddock, Gotlib & Merbedone, 2011). The emergence of new career models such as the protean career (Hall, 2004) and the boundaryless career (Arthur et al., 2005) reflects that the literature on careers too, has shifted towards individual decision making and independence from pre-defined career paths, as opposed to the earlier ‘traditional’ career, which implied gradual advancements, starting at the lower ranks and advancing to more senior positions, often in large, stable firms (Arnold & Cohen, 2008). Perhaps in part due to the fact that careers in practice did not conform to this model, this model is increasingly being discredited as impeding individual initiative and promoting an unhealthy dependence on organizations (Arthur & Rousseau, 1996; Briscoe & Hall, 2006). The shift to newer career models has reduced the relevance of this model of vertical progression through promotion and led to a focus on career self-management and individual agency instead.
6
Research on career success has also started to focus more on agentic topics. For example, De Vos, De Hauw and Van der Heijden (2011) established a mediating relationship of employability between competency development and career satisfaction. Likewise, a study by Ng and Feldman (2010) examined the mediating relationship of task and contextual performance in the relationship of cognitive ability and conscientiousness with the objective career success indicators salary and promotions. Additionally, the relationship between a protean career attitude and career satisfaction has been found to be mediated by career insight (De Vos & Soens, 2008). Drawing on social capital theory, Seibert, Kraimer and Liden (2001) found that access to information was significantly related to the number of promotions and career satisfaction. They also found that access to resources affected salary and career satisfaction and career sponsorship was significantly related to all three measures of career success. These studies are examples to illustrate that the recurring theme is that they are developed from an agentic perspective to look at career success, in which ‘appropriate’ actions on the part of the focal individual ‘automatically’ bring about objective and subjective career success.
Contemporary careers research seems to take a deterministic perspective to careers in general and career success specifically, suggesting that it can be fully predicted. Although we may not agree with this premise, we do not argue against this specific premise. We do however, argue against how the contemporary careers research aims to support this premise. The current research suggests that the possibility and responsibility of achieving career success lies purely with the individual. We argue that this does not offer a complete understanding of career success.
One way that research fails to offer a complete view of career success is that most studies fail to account for context or contextual factors. Research that only focuses on individuals, their behavior, and their resources will miss events that happen outside an individual’s control yet can have a major impact on career development and therewith career success. For example, the passing away of a partner, or suddenly inheriting a lot of money, can both lead to major changes in one’s career. It seems unrealistic to expect that individuals have full control over how their careers develop (Akkermans, Seibert & Mol, 2018). Therefore, researchers have called for work on how chance events may hinder or facilitate career development (Akkermans & Kubasch, 2017; Bright, Pryor & Harpham, 2005; Holtom, Mitchell, Lee & Inderrieden, 2005). Chance events are events that happen to an individual, and that “cannot be anticipated and proactively acted upon or, even where anticipated, the
7
effects of the event are not anticipated.” (Akkermans et al., 2018, p. 2). Research on this topic
has been around for quite some time, but with the emergence of agency-based career models, the notion of chance events and the impact they can have on careers has been largely forgotten. To redress this agentic focus, scholars have thus started to revive research into chance events. In recent years, especially the notion of career shocks has surfaced (Akkermans et al., 2018, Seibert, Kraimer, Holtom & Pierotti, 2013). Career shocks are unexpected events in one’s career that trigger a deliberate thought process about the career (Seibert et al., 2013). By looking at these experiences that cannot be controlled, yet may be postulated to have a direct impact on both objective and subjective career success, researchers aim to explain how context may impact (individual decision making vis-à-vis) the career (Akkermans et al., 2018). These deliberations lead to the following research question: How do
career shocks impact career success?
With this guiding research question we formulate three research aims. The first aim is to use text mining techniques (topic modeling, see Methods for details) to induce career shocks from qualitative interview data. The second aim is to use the career shocks that we found in this process of topic modeling, to investigate the relationship between career shocks and career success. The third aim is to explore whether we can identify antecedents of career shocks.
Although there has been some research on career shocks, to the best of our knowledge, no study has systematically catalogued the types of shock that people experience in practice (Akkermans et al., 2018). A possible reason for this is that in the current, nascent stage of investigation on career shocks, much of the extant research has been conducted from a positivistic standpoint. To work towards a taxonomy of career shocks, inductive, qualitative research might be more suitable, in that it is difficult to imagine how a taxonomy of career shocks could be derived from contemporary careers theory. This study aims to contribute to the literature on careers by using qualitative data to generate insight in what career shocks are experienced in practice. Not many studies have used qualitative data to elucidate career shocks (for exceptions, see Dries, Pepermans & Carlier 2008; Sturges, 1999). One of the caveats of using qualitative data is that sample size is usually small (resulting in concomitant generalizability issues) and that analysis is very labor intensive. By deploying text mining as a novel means of analyzing a relatively large qualitative dataset, we expand the literature on career shocks by exploring what career shocks people experience in practice. To achieve this, we collected qualitative interview data with the goal of enhancing our understanding of the
8
career shocks that people encounter and how these affect their career. In order to address our research question, this study will report on the analysis of qualitative interview data in an effort to examine i) which career shocks people experience; and ii) how these shocks are related to both objective and subjective career success. As mentioned previously, the literature on career shocks is still in a nascent phase. Although the term itself was coined some time ago (Lee & Mitchell, 1994), the notion of career shocks has not yet been grounded in the careers literature. In sum, multiple researchers (Inkson, Gunz, Ganesh & Roper, 2012; King, Burke & Pemberton, 2005; Zeitz, Blau & Fertig, 2009) have called for research on the contextual factors that affect the development of careers. The second contribution we aim to make is to explore this non-agentic side of career success by investigating the relationship between career shocks and objective and subjective career success. By only focusing on personality, demographics and/or individual agency and control, our understanding of careers will not be complete. This study addresses events and forces that are beyond an individual’s control and sets out to broaden our understanding of careers in general and career success specifically. A third contribution is made in the field of career shocks. We aim to explore antecedents of career shocks. Career shocks always have an element of unpredictability (Akkermans et al., 2018), whether it is the occurrence of the shock that is unpredictable or the outcomes of the shock. Shedding more light on this will provide valuable insights into the (un)predictability of shocks.
This study is organized as follows: in the next section we will review the literature on career success and career shocks. We will summarize the existing knowledge and highlight that which is as of yet unknown. After that, we will explain how the research was conducted. Here we will report on the sampling procedure, data collection, operationalizations, and analyses. Next, the results will be presented and interpreted. Finally, we will discuss our findings, present the implications for theory and practice, and suggest directions for future research.
Theory
Career success
Career success is an important topic to both individuals and organizations. For obvious reasons, individuals will try to achieve success in their career, because it is related to the standard of living of people, happiness (Pan & Zhou, 2013) and the satisfaction they derive
9
from working (Ng & Feldman, 2010). Also, many individuals derive a sense of identity from their career. The notion of career identity represents the way individuals define themselves in the career context (McArdle, Waters, Briscoe & Hall, 2007). In this sense, a successful career can lead to an individual identifying as a successful person. For organizations, career success of their employees can contribute to organizational success (Judge, Higgins, Thorensen & Barrick, 1999). Examples can be retention of desired talent, satisfied employees, and better job performance (Ng, Sorensen, Eby & Feldman, 2007; Stumpf 2014).
Because of this importance it is not strange that researchers have continually tried to identify individual and organizational factors that affect career success. Ng et al. (2005) carried out a meta-analysis, summarizing the literature on predictors of career success. They classified more than 25 predictors into four categories: human capital, organizational sponsorship, socio-demographics and stable individual differences. From these categories, human capital is the most agentic one, organizational sponsorship focuses on the effect an organization can have on an individual’s career success and socio-demographics and stable individual differences represent individual characteristics and personality traits. Although these last three categories cannot be considered very agentic in the sense that they would be difficult to change for individuals, they differ from the non-agentic view we try to capture in this study. While the predictors of demographics and stable individual difference are dispositional characteristics and organizational sponsorship looks at the support from an organization, this study looks at unexpected, unplanned events, that are mostly outside the locus of control of the individual or at least have unknown consequences.
Prior research has identified two related but empirically distinct constructs of career success (Ng et al., 2005). Objective career success, also called extrinsic success (Judge et al., 1999), consists of observable career achievements which can be measured (Wayne, Liden, Kraimer & Graf, 1999). Objective career success indicates structural, social views about what is a successful career. Examples of indicators of objective career success are earnings or salary, occupational status, and promotions (Hogan et al., 2013; Seibert, Kraimer & Liden, 2001).
In the previous century, measures of objective career success dominated the career success literature (Heslin, 2005). However, in recent years due to the rise of the protean and boundaryless career (Arthur et al., 2005) models of career success more and more attention has been focused on the subjective side of career success. By forming one’s career to meet one’s needs, one may acquire a high level of subjective career success, even though this might
10
not show in terms of objective career success. An example can be a guidance counselor who feels he has a positive impact on the lives of other people, but is not being paid very much. Although objective indicators of career success are easily understood and often readily available, scoring high on these indicators does not necessarily make people feel happy or successful (Heslin, 2005). Individuals have different goals and place different values on factors as salary, work-family tensions, occupational status, development and so forth. Thus, each individual has a different understanding of what career success exactly is. In the career success literature, subjective career success is used to capture subjective judgements about individual career attainments (Ng et al., 2005). Measures that are often used to operationalize subjective career success are career satisfaction, job satisfaction, or constructs such as perceived employability or marketability (De Vos & Soens, 2008; Wayne et al., 1999).
Although a considerable amount of research has been conducted in the field of career success, a close inspection of the Ng et al. (2005) meta-analysis, reveals that most of the reported effect sizes are small to medium. The average correlations of human capital, organizational sponsorship, socio-demographics and stable individual differences with the objective success indicator salary, are respectively: .21, .13, .20 and .11. Squaring these correlations yields a rather disenchanting view on the percentage of explained variance in salary. For promotions the scores were even lower: .06, .10, .05 and .08. And for career satisfaction the respective scores are: .10, .31, .02 and .24. These empirical findings show that there still is a lot of variance left to explain in objective and subjective career success.
A possible reason for these small effect sizes is the approach that has been taken to study career success so far. Most of the research on career success has looked at how certain aspects of the individual would impact career success. Looking at proactive personality (Seibert et al., 1999), employability (Hogan et al., 2013), personality traits (Judge et al., 1999), human capital and socio-demographics (Ng et al., 2005), only tells you something about how the individual affects career success. With research on social capital (Seibert et al., 2001), supervisor support (Wayne et al., 1999) and organizational sponsorship (Ng et al., 2005), other people are acknowledged to have an impact on career success as well. To increase and broaden this point of view on how the context and non-agentic predictors can impact career success, we believe that there should be more consideration for the context and events and forces that cannot be controlled by the individual. We argue that these kinds of events and forces can have an impact on the career success of individuals. Even though there
11
has been research on career success for many decades now, the small effect sizes can indicate that most of what brings about career success has not yet been identified.
Apart from the previously discussed empirical findings, there is also theoretical ground to recognize and address less agentic predictors. Careers are constructed over a longer period of time and develop under ambiguous and uncertain conditions (Seibert et al., 1999). Only a few authors in the careers literature take such conditions into account in researching careers (for an example see Converse et al., 2011). Indeed, even when they discuss the environment or context, the focus is still on how the individual can actively control or influence his or her environment. With the contemporary career paradigms of protean (De Vos & Soens, 2008) and boundaryless careers (Arthur et al., 2005), careers researchers have focused on how individuals can shape the career entirely to the desired form of the individual. As pointed out before, this premise seems quite unrealistic. Lewin (1936) constructed a theory where human behavior is the function of the dynamic relation between person and environment. Accepting this view, would lead to considering external forces in a model of careers. To try and account for these forces, this study aims to broaden our understanding of careers and career success by trying to predict career success with career shocks. In the next section we will discuss these career shocks.
Career shocks
As mentioned previously, one of the goals of this study is to shed light on contextual influences on careers and career success. To this end, this study examines career shocks. The term was supposedly used first by Lee and Mitchell (1994) when they discussed voluntary turnover. In their study they discussed “a very distinguishable event that jars employees
toward deliberate judgments about their jobs” (p. 60). This stream of research on shocks and
turnover is one of the origins of the career shock literature. The other stream of literature that is important is the work on chance events. This literature is fairly established, going back to studies by Roe and Baruch (1967) and Hart, Rayner and Christensen (1971) about the impact of chance events on careers. Later, theories in this field such as the Chaos Theory of Careers (Bright & Pryor, 2005) and the Happenstance Learning Theory (Krumboltz, 2009) also underscore the impact that chance events can have on careers.
To understand how career shocks influence career success, it is important to consider how career shocks differ from other contextual influences on career success. In the literature
12
on career success, organizational sponsorship (Ng et al., 2005), social capital (Seibert et al., 2001) and supervisor or mentor relationships (Wayne et al., 1999) have been used as predictors. These are predictors that can be seen as contextual since they represent factors outside the individual and the individual has no control over these factors. In this sense, organizational sponsorship overlaps with career shocks. Indeed, we could argue that events such as gaining the opportunity to enroll in a development program at work or being paired with a mentor who turns out to be very supportive and influential could be career shocks. Important to note is that these events would have to be largely outside the control of the individual and the event would have to be unexpected in some way. A difference between factors as organizational sponsorship and career shocks is that organizational sponsorship can happen over time and can be a process. Career shocks consist of the event that occurs and the thought process that comes after this event. As a result, it is more difficult to predict and control the occurrence and the outcomes of career shocks.
A career shock is defined as: “a disruptive and extraordinary event that is, at least to
some degree, caused by factors outside the focal individual’s control and that triggers a deliberate thought process concerning one’s career.” (Akkermans et al., 2018, p. 4).
Following this definition and the literature on career shocks, we can distinguish several characteristics of a career shock. First of all, a shock is an event that is disruptive and extraordinary. This refers to the unpredictable and unexpected nature of shocks. An event that can be predicted and is expected can hardly be called a shock, however, there are different degrees to the expectedness of a shock (Holtom et al., 2005; Morell, Loan-Clarke & Wilkinson, 2004). For example, losing a mentor at work due to a car crash is much more unexpected than having a baby. Having a baby is still considered a shock however, because although it is something that can to some degree be planned, the outcomes of the event can still be unpredictable (e.g. complications with giving birth, postnatal issues). This unpredictability of career shocks is particularly interesting, since it is in conflict with the deterministic perspective that contemporary careers research takes. We argue that with respect to career shocks there is always some unpredictability. This can manifest itself in two ways. Either the occurrence of the career shocks cannot be predicted, or the consequences of the career shocks cannot be predicted.
Another characteristic is the fact that shocks are caused, at least to some extent, by factors outside the individual’s control. This component of career shocks indicates that shocks occur not due to individual agency, but due to contextual factors (Akkermans et al., 2018).
13
This aspect of career shocks has much to do with the previous one, the unexpectedness. In some situations, the shock can be expected, but this does not mean that the individual has full control over the shock and/or its outcomes. For example, if you work at a company which is not doing well, you might expect cuts in salary or employees or even bankruptcy. However, this does not give the individual much control over the situation. The individual therefore cannot anticipate how this shock plays out and what the effects of the shock might be on his/her career.
The last characteristic that follows from the above definition is the notion that the shock triggers a deliberate thought process about one’s career. This aspect of the shock can lead to changes in one’s career development, decision making, or even career satisfaction, due to the initiation of psychological analyses about one’s career (Lee & Mitchell, 1994). It is supposed that during or after the shock, the individual will analyze what happened and that this analysis will lead to thoughts or behavior that impact one’s career (Seibert et al., 2013) in a positive or negative manner. Although triggering a thought process is a characteristic of a shock, this does not mean that the consequences of the shock only arise from this process. The shock itself can also yield direct impacts, for example becoming disabled because of an accident or disease.
Up until now, there has been little empirical work on the outcomes of career shocks. Hirschi (2010) questioned students about how chance events influenced the transition from middle school to vocational education and high school and found that encouragement from others and effects gained from personal and professionals connections (e.g., information about jobs, informal recommendations, job offers) affected this transition. Seibert et al. (2013), among other things, distinguish between negative and positive shocks. Negative shocks, such as a mentor leaving the organization, would negatively impact one’s career and positive shocks, such as receiving an unexpected promotion, would have a positive impact on one’s career. They found that positive career shocks positively affected the intention to pursue graduate education, suggesting that positive career shocks can impact the intention to pursue challenging career decisions. With this research Seibert et al. (2013) showed that one can distinguish between positive and negative shocks and that they can have different outcomes. Additionally, Holtom et al. (2005) investigated how shocks were related to voluntary turnover. They distinguished between different kinds of shocks. Expected or unexpected, positive or negative and work-related or personal shocks. The limited amount of work on career shocks indicates how little we know about this topic. This study aims to increase our
14
knowledge about career shocks by examining what shocks people actually experience in practice and whether such shocks have an impact on career success.
Methods
Sample
Our sample comprised 272 individuals, who were interviewed about career shocks. The interviews were conducted by graduate students, as a part of a course on (qualitative) research methods, who each interviewed two employees. The interviewers introduced the topic of career shocks and asked respondents about the positive and negative career shocks that they had experienced in their career so far, and how they reacted to these shocks. The interview protocol can be found in Appendix A. After the interview was completed, the interviewer completed a brief survey together with the participant, which consisted of questions about demographics and the career, the current occupation of the individual and questions about objective and subjective career success. The survey can be found in Appendix B. Of the 272 interviews that were conducted, 75 interviews were dropped because they were conducted in English. The reason for this is that with text mining different languages cannot easily be analyzed at the same time. Another nine interviews were dropped from further analysis because either the interview or the survey was not complete. Our final sample consisted of 188 Dutch individuals, of which 52.6% were female. Age ranged from 22 to 72, with an average age of 46.9 years old (SD = 13.0). 67 of the 188 individuals (35.6%) received a scientific education. The average organizational tenure was 9.8 years (SD = 10.1) .
Measures
Career success
Career success was measured in three different ways. The indicators of objective career success were promotions and salary and the measure for subjective career success was career satisfaction. Promotions were measured with the question: “Up until now how many times
have you been promoted in your career?” The answers ranged from 0 to 5, with 5 indicating
that the respondent had been promoted 5 or more times in his or her career. Salary was measured as gross annual salary and had five answer categories, ranging from 1 = less than
15
euros a year, 4 = between 28,500 - 33,999 euros a year to 5 = over 56,000 euros a year.
Subjective career success was measured using a career satisfaction scale adopted from Greenhaus, Parasuraman and Wormley (1990) to measure an individual’s overall career satisfaction. An example of an item included in the scale is: “I am satisfied with the progress
I have made towards meeting my goals for advancement”. Of these items the average was
taken to form one measure of career satisfaction. The response scale ranged from 1 = Strongly
disagree to 5 = Strongly agree. The Cronbach’s alpha for this scale was .83.
Control variables
Based on prior research, we include several control variables in our study. The positive effect of age on objective career success is one of the most consistent findings in career success research (Ng et al., 2005), older people will have more experience, which often results in a higher salary and older people have had more time and options to be promoted. These objective outcomes often accumulate over time (Judge, Cable, Boudreau & Bretz, 1995). Age is a continuous variable and measured in years. Gender is supposed to affect career success as well. Research has shown that women receive less salary and promotions (Judge et al., 1995; Ng et al., 2005). Furthermore, gender is also said to play a role in subjective career success. Men value objective outcomes more and derive satisfaction from those indicators, while women evaluate their career in broader and more subjective ways, leading to women and men perceiving success differently (Heslin, 2005; Powell & Eddleston, 2008). Gender was coded female = 0 and male = 1. Many studies include education as a form of human capital in their analyses. Research has shown that investing in education leads to significant returns in salary and promotions (Judge et al., 1995; Wayne et al., 1999). Acquiring a solid education promotes greater career success by increasing individuals’ knowledge and skills, which in turn are valued and rewarded handsomely in the labor market (Ng & Feldman, 2010). For education we distinguished between people with and without scientific education (0 = no scientific education, 1 = scientific education). Marital status is also used as a control variable in this study. Research has found that being married affects objective and subjective career success (Ng et al., 2005). Pfeffer and Ross (1982) mention that married individuals often are perceived as more stable, responsible and mature. A significant other can also act as a resource for individuals, offering emotional support and assisting with household responsibilities. The operationalization of marital status was altered somewhat. Since living together with a significant other provides the same benefits as being married, we code being married and living together as 1 and being single, divorced or widowed as 0. Another variable
16
that has been shown to be related to career success is organizational tenure (Ng et al., 2005). Individuals who have longer organizational tenure often have a lot of experience in their position and have gained more extensive knowledge about their specific environment (Wayne et al., 1999). This type of implicit knowledge is seen as an important determinant of career success (Ng & Feldman, 2010). Organizational tenure was measured by asking: “Since when
(what year) have you been working with your current organization?” The answer was then
subtracted from 2017 to derive respondents’ organizational tenure in years. The results of research on organization size have been ambiguous. A positive relationship between organization size and objective career success seems reasonable, since larger organizations have more resources to pay higher salaries and more available spots for promotions. However, there are also more employees, so more contenders for those resources (Judge et al., 1995). Organizational size was a self-reported answer to the question “What is, approximately, the
size of your organization in terms of number of employees?” The answers ranged from 0 to
65.000. Because of this large range, we used a natural logarithmic transformation of this variable for all analyses.
Analyses
Qualitative data
To answer our research question and address our three research aims, we need to take several steps. The first step deals with inducing career shocks from the topics we find in our topic models. We start this process by analyzing the 188 interviews on career shocks we have collected. We employ Structural Topic Modeling (STM) (Roberts et al., 2014b) to automatically extract topics from the data. We base our analysis on recent developments in machine learning and text mining to inductively search for topics in the text corpus. This type of model is often called an ‘unsupervised’ method, because the model does not assume the content of the topics, but instead enables the researcher to make inferences about the content based on the data. This is as opposed to ‘supervised’ methods, where the topics would be established in advance (Roberts et al., 2014b) and fed into the algorithm.
With STM, which builds on the earlier LDA model (Blei, Ng & Jordan, 2003), each document in the corpus is seen as pertaining to a mixture of topics. For example, document 1 can represent topic 1 for 10%, topic 2 for 25% topic 3 for 5% and topic 4 for 60%. Each word in a given document is assigned to exactly one topic. STM organizes the words from a set of
17
documents, in our case the 188 interviews, into (unlabeled) topics based on the cooccurrence of the words in the documents. Thus, STM gives the researcher an idea of which words appear together across respondents/documents (Rothschild, Howat, Shafranek & Busby, 2018). The software to employ STM is freely available in the stm package (Roberts, Stewart & Tingley, 2014a) on R, a statistical software program. We use R for all data (pre)processing and analyses.
To estimate the topic model, the corpus first has to be preprocessed. Preprocessing is breaking the data down into components that statistical programs can read. The purpose is to get rid of irrelevant words, symbols or other things that could be present in the data and are not pertinent to the analysis (Kobayashi, Mol, Berkers, Kismihók, & Den Hartog, 2017). The first step is removing numbers, converting upper case letters to lower case letters, removing punctuation and getting rid of superfluous whitespace. Next, we examined words that can be spelled separately, but are meant together. An example is fake news, if you leave those words like that they will be seen as two separate words indicating fake and news. In our case, the word burn-out could be very relevant. However, due to punctuation removal and spelling variations in the interview transcripts, we had to specify that burn and out should be replaced by the single term burnout. The next step is to remove common stopwords and particular words. Stopwords are words such as ‘the’, ‘a’, ‘okay’ and ‘well’. These words are almost always uninformative due to their lack of meaning and high frequency. Furthermore, the removal of particular words is an iterative process. When a model is estimated and the results show a number of words that are associated with certain topics, it is possible that some of these words are not informative or relevant at all. In that case, one ought to proceed by manually removing these words from the corpus, in order to ensure that topics only represent relevant information. Another common preprocessing technique is stemming (Willet, 2006). With stemming, the researcher tries to obtain the root, or stem, of words. The assumption behind this process is that words with the same stem also have the same meaning. For example, play, played and playing all become the stem play. Stemming, however, does not work perfect with every language and every kind of text and researchers should consider what works best in their specific context (Kobayashi et al., 2017). In this research, the analyses were done with and without stemming and we did not observe differences in the highest ranked words per topic or in topic proportions. Because of the ambiguous interpretation of some stemmed words in Dutch, we decided to use the cleaned corpus without stemming for our analyses.
18
The clean corpus is now a collection of documents, that are filled only with relevant words or terms (unique words in the corpus). To be able to analyze this corpus the next step is to transform it into a Document by Term or Term by Document matrix. In doing so, text is transformed into mathematical structures (Kobayashi, Mol, Berkers, Kismihók, & Den Hartog, 2018). One way to construct such a matrix is to assign each document to a row and each term to a column. This yields a term by document matrix (TDM), and if one assigns the documents to the columns and the terms to the rows one can obtain a document by term matrix (DTM). In these matrices, each cell denotes the ‘weight’ of the term in each document. In other words it denotes the term frequency, how often a term occurs in a document. However, term frequency is not the only way to ‘weight’ terms. A term that is present in every document or in only one document has little discriminatory power. To account for this, the inverse document frequency (IDF) can be used. By combining this with the term frequency weighting, one obtains the popular TF-IDF measure that is used to simultaneously take into account the importance and occurrence of a word and its specificity (Kobayashi et al., 2018). Although the input for the structural topic model needs to be a TDM with term frequency weighting, later on in the process, the model can also account for the exclusivity of terms (Bischof & Airoldi, 2012). Exclusivity measures whether a term is important in many topics or if it is relatively exclusive to one topic. The idea behind this FREX (frequency and exclusivity) score is that the model includes fewer words that are used frequently throughout the entire corpus. The last step before estimating the model was to remove sparse terms. This discards all words that are used in less then a chosen percentage of documents in the corpus. We decided on a sparsity of .95 which means that terms that occur in less than 5% of the documents in the corpus are removed. So a term had to occur in at least 10 documents to show up in one of our topics. The final TDM consisted of 188 documents with 1589 unique terms and 113.383 words.
With the TDM as input, the topic model can be estimated. To estimate this model, the researcher needs to indicate how many topics he or she wants to get out of the model. This choice is not a direct decision, but an iterative process. The first indication can be on the basis of diagnostic values such as the held-out likelihood (Wallach, Murray, Salakhutdinov, & Mimno, 2009), semantic coherence (Mimno, Wallach, Talley, Leenders, & McCallum, 2011) and residuals (Taddy, 2012). For this research, this first, rudimentary, analysis indicated a number of topics between 4 and 16.
19
When deciding on a number of topics we look at two things, interpretability and parsimony. Interpretability is about being able to make sensible interpretations of the topics. If there are few topics, there might be terms in the same topic that indicate different content. This way one topic can include multiple distinct ideas. If there are too much topics there might be terms that indicate the same content in different topics. A larger amount of topics might lead to redundancy among the topics. Thus, the right amount of topic is approached when topics have distinct and unambiguous meanings. The principle of parsimony has to do with simplicity, the researcher should not go looking for more complex explanations when a simple one is sufficient (Banks, Woznyj, Wesslen & Ross, 2018). A greater number of topics, thus a greater complexity of the results, should be justifiable only with the reasoning that the researcher is better able to interpret the topics when there are more topics. This is consistent with the literature on construct redundancy in the organizational science literature (Banks, McCauley, Gardner & Guler, 2016; Shaffer, DeGeest & Li, 2016).
To label the topics, researchers should consider and/or discuss what the emerging topics represent. Important questions to ask are “How are these topics similar to one another?”, “How are these topics different?” and if they are different “In what way are they different?” (Cowan & Fox, 2015). To aid the interpretation of the topics supportive examples from the data should be identified. The goal of the process described here and in the previous paragraph, is to verify that the labels that are assigned to the topics are robust and characteristic of the data (Banks et al., 2018). Although researchers should always try to adhere to certain standards, it is worth noting that the interpretation and human judgement that is needed to label topics is still an arbitrary task. In some cases it might be easy to identify the content of a topic based on the highest ranked words and in other cases the researcher will have more trouble with assigning an unambiguous and robust label to a topic.
For this study, we arrived at a 10 topic model. We started by looking at a model with 4 topics and intended to work our way up to 16 topics, each time labeling the different topics. When we arrived at topic 13 it was clear that the topics were not going to become more clear or specific. At this moment we stopped and decided on our 10 topic model. For the first part of the process we noticed that topics contained words that indicated different content. Topics were not specific enough. After we hit 10 topics, we noticed that topics began to overlap and topics were formed that provided less relevant content. On basis of this assessment we choose the 10 topic model. To validate the topics we had formed, we went through the two of the
20
most representative interviews for each topic to this helped us to ensure that the words we find in our topics have the same meaning in their original context as in the FREX score list.
Quantitative data
The second and third aims of this research are answered with a quantitative analysis. To investigate the effect of career shocks on career success we run three OLS regressions, with salary, promotions and career satisfaction as dependent variables. Career shocks are integrated in the model as the topic proportion of the document with regards to the respective topic. To make interpretation easier, this proportion is multiplied by 100 to represent a percentage. Since the career shocks are all proportions of a single document that accumulate to 1, they are bound to be correlated. We checked for correlations between the topics, but the correlations we found were weak. Our third aim was to look at antecedents of the various career shocks. To this goal, we run several regression analyses with the different career shocks as our dependent variables.
Results
Topic labels
The first step we take is to infer topic labels from the highest ranked words that are assigned to each topic in the estimated 10 topic model. Like previous research (Lucas et al., 2015; Roberts et al., 2014b; Rothschild et al., 2018; Tvinnereim, Fløttum, Gjerstad, Johannesson & Nordø, 2017) we use the FREX score (Bischof & Airoldi, 2012) to rank the words. The FREX score indicates words that are frequent and exclusive to each topic, as opposed to only using frequency as the measure to rank words. To apply a form of validation, we examine and present quotes from exemplar documents for a topic. The documents that we use are the ones with the highest topic proportion. This means that that specific topic draws the most on that particular document. In Table 1 we show the topics, the highest ranked words and their topical prevalence. The topical prevalence indicates how much of the corpus is associated with a certain topic.
21
Table 1. The 10-topic model with most frequent/exclusive words and topic prevalence
Label Topical
prevalence Most important words based on FREX score (for original Dutch words, see 1 New opportunities .126 internship, resignation, choice, position,
hired, chance, offered 2 Coworkers and
supervisors .117 team, manager, fun, noticed, office, conversation, learned
3 Burn-out .116 burn-out, nice, days, journey, busy, partner, pregnancy
4 Control over situation .109 strategy, control, growing, job, role, person 5 Large event .099 project, executive, bank, board, abroad,
suddenly
6 Personal circumstances .095 study, reorganization, children, education, school, hospital, part-time
7 Professional
development .090 organization, vocation, certain, development, get, responsibility, energy 8 Entrepreneurship .083 company, sold, selling, bankrupt, wage
labor, entrepreneur, self-employed 9 Family .082 Father, husband, mother, children, wife,
family, firm 10 Gaining permanent
occupation .079 bank, shock, career, permanent, started, worked, department
We will now discuss the topics and give quotes from exemplar topics associated with the topic. While labeling the topics, we found that it could be difficult to consider all topics directly as shocks. For example, the topic family in itself does not indicate any sign of a shock. However, the exclusive subject of the interview was career shocks, and with this knowledge we argue that all the topics represent career shocks in that the context in which the words are found is all about career shocks. The topics are ordered according to topical prevalence. The quotes that are included are taken from the two most representative documents for the topic.
Topic 1. New opportunities
Words associated with this topic indicate that the individual encounters or is forced to look for new opportunities. Individuals get fired or get a job offer and need to make a choice about
22
what they want to do. In the two interviews that are most representative for this topic both individuals discuss that they had to make a choice between two options.
“After the application process, (…), I got the news that I could start at X. However, now I faced a dilemma, because I also had the option to start working at Y.”
“(…) This has definitely helped me to make a well-advised choice and eventually make that switch (…).”
Topic 2. Coworkers and supervisors
Coworkers and supervisors can have a large impact on an individual’s career. Words as ‘team’, ‘fun’, ‘conversation’ and ‘learned’, can indicate how the social environment of an individual affects his or her career. In the interviews we see that teams can play a large role.
“My manager played a very large role. And I repeatedly asked him for help. (….) My team played a very large role.”
“When you work together in a team and you break up the team, you need to ensure that, somewhere else, you again end up in a [different] team. That is a pretty big deal.”
Topic 3. Burn-out
The words associated with burn-out were not very easily interpretable. Words like ‘nice’, ‘days’ and ‘journey’ do not directly come to mind when you think about burn-outs. However, when going through the interviews we noticed that the individuals that talked about burnout are were often indicating small positive points.
“ (…) I go to a psychologist, which is nice because you learn a lot about yourself.. I think that is really nice.”
This could be one explanation for the word ‘nice’ occurring frequently together with burn-out. Another explanation could be that individuals mentioned that things are actually ‘not nice’. Due to stopword removal these indications could have been removed from the data.
Topic 4. Control over situation
This topic does not come as a big surprise. In the interview protocol (Appendix A) some questions discuss the issue of whether the individual had control over the situations that occurred. What could the interviewee control, what couldn’t he or she control and what strategies did or did not work. Interviewees mentioned:
23
“In most cases you don’t have it under control and it just happens to you, so you have to move with the flow and anticipate (…) In most cases you don’t have control over it.”
“I actually only had control over what I would do, I wasn’t in the position, as a subordinate, to ask my superiors to change their attitude. What I could decide is that I started to look for a different job.”
Topic 5. Large event
This topic was the hardest to interpret. The highest ranked words do not immediately let you come up with a label for this topic. Reading the two most representative interviews again in full did not offer any immediate solace either. Both interviewees discuss an event that was a large incident in their career. One interviewee got involved in a accounting scandal abroad and the other interviewee faced large consequences due to the financial crisis. With this knowledge the words ‘executive’, ‘board’, ‘bank’ and ‘abroad’ indicate the level of the playing field in this topic.
“Well, the most important one [shock] is that I worked for a Dutch company that (…) got engaged in an accounting scandal.”
“Yes, the board of directors, because everyone…. even in the board of directors a lot of people had to go.”
Topic 6. Personal circumstances
This topic was not directly unequivocal. There seem to be multiple things going on. ‘Study’ will for the most part refer to the study or training that the individual followed, is following or was offered to follow. ‘Education’ most likely deals with the education of people’s children or indicates that the interviewees works in education/teaching. Both interviewees worked at schools and were teaching. Children also play a big role in this topic, because of these influences that come from the individual self, this topic is labeled as personal circumstances.
“No, at that point I said that I didn’t want to do it anymore, but I would in the future. Now, after having two kids, I have the opportunity pick it up where I left it.”
“But when I got kids, that became priority number one.”
Topic 7. Professional development
The words ‘vocation’, ‘organization’ and ‘responsibility’ go together well. Even without going back and forth with the interviews these terms indicate that this topic is related to the
24
professional career of the individual. Add ‘development’ to this notion and the topic label is quite clear.
“Because I wanted to make a certain progress and because I, yeah, wanted to continue my development in my career, I decided to take the step.”
“… and after that I got into accounting, and you notice that you develop yourself in that vocation. That is very positive, because you get more and more responsibility.”
Topic 8. Entrepreneurship
This topic is relatively specific, we notice that the words definitely indicate similar content. ‘Entrepreneur’ and ‘self-employed’ are practically opposites of ‘wage labor’ indicating that the topic revolves around this dividing line. ‘Company’, ‘sold’ and ‘bankrupt’ suggest that the individual had to make choices about the company. So this topic deals with all the successes and struggles of entrepreneurship.
“No we sold the old company. And at that time my dad already got out.” “And we started with the three of us and we had to do everything ourself.”
Topic 9. Family
One of the easiest topics to interpret. All the words speak for themselves, with maybe the exception of ‘firm’. In the most representative interview we see that the interviewee wanted to take over the architecture firm from his father. However, due to the untimely death of his father, this did not happen. The frequent occurrence of the word firm in combination with father will have lead to the inclusion of this term. As mentioned in the start of this section, it might be difficult to directly infer a shock on the basis of table 1. However, going to the interviews reveals that often the topic is discussed in the context of a shock.
“But the moment and how, that is something I could have never seen coming, because it had to do with the untimely death of my father. And the intention was that I would continue the firm of my father while he still lived.”
“The serious illness of my husband, causing us to have a family income of zero, that definitely had an effect.”
Topic 10. Gaining permanent occupation
The last topic, also the topic with the lowest topical prevalence. This topic was relatively hard to label. The presence of less relevant and distinct words as ‘started’, ‘work’ and ‘department’
25
make interpretation difficult. The inclusion of the word ‘permanent’ gave an idea about the content and in combination with going through the two most representative interviews lead to this topic label. The term ‘bank’ is presumably included because both interviewees have a career in the financial industry.
“So all factors that played an important role in the background. And, that made me continue with this job.”
“Uhm, the most positive event that influenced my career is that I got a new side job through a friend. And through this job I eventually got a very nice permanent occupation.”
Regression analyses
Table 2 shows the means, standard deviations, and correlations of the primary variables in this study along with the Cronbach’s alpha score for the variable with multiple items (career satisfaction). The dependent variables all correlated significantly with each other. This was expected based on previous research (Ng et al., 2005). Furthermore, age correlated positively with both objective measures of career success and gender correlated positively with all three measures of career success. It is remarkable that education only correlates with income as previous research has shown education to correlate with promotions and career satisfaction as well (Judge et al., 1995, Ng et al., 2005). Organizational tenure is associated to income and organizational size. Marital status correlated significantly with age. This seems reasonable since not many people marry in their early twenties.
26
Table 2.
Means, standard deviations, and correlations among study variables Mean Std. dev. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Promotions 1 2.69 1.69 Income 2 3.56 1.29 .374** Career satisfaction 3 3.78 .70 .229** .165** .83 Age 4 46.91 12.97 .343** .287** .056 Gender 5 .47 .50 .201** .226** .175* .086 Education 6 .36 0.48 .013 .186* .115 .053 .051 Marital status 7 .74 0.44 .137 .083 -.011 .309** .067 .028 Org. tenure 8 9.98 10.17 .110 .192** .034 0.454 -.023 -.034 .187
Org. size (log) 9 6.24 3.17 .032 .141 .024 -.042 -.125 .194** -.132 .153*
Personal Circumstances 10 9.85 18.80 -.088 .048 -.033 .034 -.059 -.104 .055 .126 .078 Permanent occupation 11 7.89 18.67 .041 -.048 .018 .059 -.032 -.045 .018 .030 -.051 -.118 Professional development 12 9.01 18.66 .138 .084 .172* .081 .124 .033 .014 .089 .020 -.091 -.142 Entrepreneurship 13 8.25 18.00 .045 .041 .126 .096 .046 -.134 .093 .059 -.094 -.097 -.066 -.104 Burn-out 14 11.59 20.69 .000 -.034 -.047 -.22** -.086 .053 -.011 -.031 .002 -.119 -.108 -.150* -.135 New opportunities 15 12.68 21.23 -.138 -.066 -.017 -.180* -.100 .024 -.220** -.123 .065 -.153* -.134 -.108 -.112 -.143 Large event 16 9.92 19.20 .077 -.158* .022 .078 .015 .042 .086 -.061 -.096 -.122 -.108 -.075 -.091 -.161* -.112 Control 17 10.94 19.53 -.028 .045 -.031 .075 .056 -.034 .021 -.021 .109 -.136 -.081 -.075 -.104 -.128 -.158* -.160* Family 18 8.21 16.78 -.020 .085 -.027 .168* .025 .171* .134 .126 .037 -.064 -.041 -.102 -.076 -.069 -.164* -.103 -.084 Coworkers 19 11.72 18.37 -.009 .029 -.177* -.131 .038 -.002 -.147* -.161* .084 -.075 -.175* -.131 -.151* -.076 -.048 -.069 -.093* -1.64
n = 188. Coefficient alphas appear on the diagonal for multiple-item measures. ⁎ p < .05 (two-tailed).
27
We see that career shocks related to professional development correlate significantly with career satisfaction. Burn-out shows a negative correlation with age, suggesting that older people are less prone to a burn-out. Although we have seen from a preliminary check that the career shocks generally show weak correlations, some of the shocks are still significantly related to each other. New opportunities shows significant, negative correlations with age and marital status. The career shock relate to large events is negatively associated with income. In addition, we observe that family related shocks correlate significantly with age and education. Finally, coworker and supervisor related shocks shows significant negative correlations with career satisfaction, marital status and organizational tenure.
Although we mentioned previously on the basis of preliminary checks that the career shocks generally show weak correlations amongst themselves, there are a few that correlate significantly with each other. We observe a significant negative correlation between personal circumstances and new opportunities. Gaining a permanent occupation is negatively associated with shocks related to coworkers. Professional development and burn-out are also negatively correlated. We see that entrepreneurship is negatively correlated with coworkers and supervision. Large events show a significant negative correlation with burn-out. Shocks
28
related to control are negatively correlated to new opportunities and large events. Family related shocks are also significantly negatively associated with new opportunities. Finally, control shows a negative correlation with coworkers and supervision. Even though we observe these significant correlations, the strength of the correlations is still quite low, ranging from r = .041 to r = .175.
Table 3 shows the results from our first regression analysis. In this regression we look at income as our dependent variable. We first ran a model with only the control variables, this allows us to examine whether adding the career shock variables indeed improves the model. We start by looking at our F-statistic (F = 6.44) and observe that it is significant indicating that the model is significant. The adjusted R2 is .149, indicating that the control variables
explain 14.9% of the variance in income. Age, gender and education are all significant predictors of income, while organization size is only marginally significant. As a second step, we add our ten career shock predictors. The F-statistic (F = 3.125) is still significant and our adjusted R2 has decreased by 0.3% to 14.6%. Only career shocks related to large events seem
to have an impact on income. With a coefficient of B = -.016, a one 1 percent increase of topic proportion on career shocks related to large events leads to a decrease of .016 on the 1 to 5 scale of income.
29
shocks on promotions. The model is significant (F = 5.63) and the adjusted R2 is .129. We see
that age and gender are both positive significant predictors for promotions. This suggests that men generally receive .6 promotions more than women and that older people receive more promotions. Our second model doesn’t change much, organization size becomes marginally significant and the adjusted R2 increases by .4%.
Table 5 shows us the regression with career satisfaction as dependent variable. Unfortunately the F-statistic isn’t significant for either of the models, which means we cannot reject the hypothesis that all regression coefficients equal zero, thus the linear model cannot be supported.
To check whether adding the career shocks indicators to our model significantly predicted our dependent variables better, we ran three ANOVA analyses between the models with and without the career shock indicators. Unfortunately, adding the indicators did not significantly improve the fit of any of the models (p-values of .50, .37 and .15, respectively).
Furthermore, we ran additional regression analyses to test whether our control variables were perhaps predictors of the career shock indicators that we integrated in our
30
model. We fitted 10 models and checked the F-values. Only the models with burn-out and new opportunities had a significant F-value. The career shock indicator for burn-out was significantly and negatively predicted by age (B = -0.448, p < 0.01). This suggests that the older you get, the less likely it is you will get a burn-out yourself. In the model with new opportunities, marital status is a negative predictor (B = -8.35, p < 0.05) suggesting that married individuals will less likely encounter new opportunities.
Discussion
In this study, we employed Structural Topic Modeling to infer indicators of career shocks on the basis of topics that were derived directly from interview data. Subsequently, we incorporated these indicators into our dataset to investigate the relationship between career shocks and career success. In this discussion, we will try to answer our research question and explain how our findings can be interpreted in light of prior literature. Furthermore, we will discuss the theoretical and practical implications. Lastly, we will discuss the limitations of this work and suggest directions for future research.
The first analysis of this study was the inference of topic labels on the basis of words that were ranked on frequency and exclusivity (Bischof & Airoldi, 2012). In the analysis we noticed that the characteristics that we discussed in this study were present. For the topic new opportunities, we saw that in accordance with the literature on career shocks, being presented with career opportunities indeed triggered a deliberate thought process about what the individual wanted to do with his or her career (Lee & Mitchell, 1994; Seibert et al., 2013). Both individuals also indicate, later on in the interview that they made the right choice and very content with their career. Based on these findings, the relationship between career shocks and, in this case subjective, career success, seems reasonable. For the topic coworkers and supervisors individuals indicated that their environment played a large role in their career experience. The literature agrees, mentoring, supervisor sponsorship and social capital have all been found to have an effect on career success (Seibert et al., 2001; Wayne et al., 1999). We were content to find that the terms that our model yielded, in combination with the interviews, related well to the literature on career shocks. Other examples are less work-related and more person-work-related. We found, not too surprising, that having children plays a big role in the career of individuals and that illnesses and deaths affect how careers develop. Interesting is how individuals perceive career shocks. In the interviews a fair amount of individuals was dealing or had dealt with a burn-out. Even though our first instinct was to see
31
this as a very negative shock, these individuals often suggested that the burn-out eventually turned into a positive experience for them. Akkermans et al. (2018) also mention that having twins can have a detrimental effect on one’s career, but it will very likely be identified as a positive shock. This first step in our analysis fulfills the first of our three research aims. We induced career shocks from qualitative interview data, by using text mining techniques.
The second aim we posited, was to use these career shocks to investigate the relationship between career shocks and career success. This step in our analyses was not as fruitful as we had hoped. The indicators of career shocks that we incorporated from our topic model did not predict much variance in our career success indicators. The only significant relationship between a career shock indicator and career success can be found in table 3. Here we see that the topic labeled large events negatively predicts income. It is difficult to interpret this finding. We could suggest that large shocks as seen in this topic are not beneficial to one’s career. An accounting scandal and a bank that needs to be saved by the government are likely not advantageous to one’s career. However, these kinds of connections can be easily made, and should not be made lightly. The absence of any other significant relationship and the virtually absent increase in variance explained leads me to consider that this significant relationship is more of a coincidence then an actual relationship between two variables. Apart from the absence of effects on the part of the career shock indicators, the control variables didn’t display many significant relations either. Gender, age and education did demonstrate significant relationships, as was expected (Judge et al., 1995, Ng et al., 2005). Marital status, organizational tenure and organizational size were less present in the regression analyses. For organization size these findings can been seen to correspond with previous findings, the findings on organization size were already ambiguous and this study suggests that organization size does not affect career success (Judge et al., 1995). We could argue that in the view of contemporary careers, organization size is less relevant for career success. Where working at General Motors, or ING or KPN used to be a big deal, nowadays organization size seems less relevant. This argumentation is similar for marital status. When an individual has a spouse who stays at home, this leads to extra resources (Pfeffer & Ross, 1982). When an individual has a spouse who is also pursuing his or her career, this can lead to an increase in costs (time, effort, emotional support). Nowadays the latter is maybe more usual than the former.
Our second and third aim are not fully achieved. We did not find a relationship between career shocks and career success. Trying to find predictors for the career shocks also
32
proved futile. This is unfortunate, as it makes answering our research question very difficult, but this study can still prove to be useful.
As for theoretical implications, this research used a more inductive approach to study career shocks. By letting individuals talk about their experiences, we explored qualitative data on career shocks to inductively find career shocks. We encourage other scholars to do this as well, not necessarily with the use of text mining techniques. There are very few qualitative studies on career shocks and the field could use more. This study contributes partly to this aim, by analyzing qualitative data. Another contribution to the literature is the goal of this research to draw more attention to the non-agentic and unpredictable perspective of careers. We hope that researchers will take note of our argument that the research on careers could be less deterministic. Not everything can be predicted or explained from the view of the individual. Of course, the individual plays by far the largest role in his or her career, but failing to acknowledge the presence of other influences will impede a complete understanding of careers. Practically, we had hoped to shed a little bit more light on the mechanisms of career shocks. How would career shocks impact career success? If we had gained some information on this relationship we could discuss whether the reaction to an unpredictable shock is more important than anticipating unpredictable consequences of a somewhat predictable event and more of these kinds of deliberations.
In addition to the contributions we made, our study had several limitations. Presumably the biggest one has to do with the process of labeling the topics. Even with the help of well-developed and automated statistical methods and measures, the human judgment is still needed. This also means that the labeling of the topics is, at least for now, arbitrary. Based on the data presented to the researcher he or she makes an informed decision. However, this decision, or interpretation, can be flawed. In this study it is very well possible that the interpretation of the topics does not cover the content completely. Another limitation regarding the topics, was the decision to include all topics in the regression analyses. We explained why we regarded all topics as actual indicators for shocks, but it is not said that this was the optimal decision and further research should consider this issue carefully. As a sort of robustness check we did run the regression analyses with input from a 5 topic model and a 16 topic model. Both models explained even less variance than our original 10 topic model, thus it seems that the model selection procedure went well. Furthermore, we have a possible explanation as to why we found no results in the regression analyses with the career shocks indicators. As an example, in the topic covering entrepreneurship, there are positive and
