Early career training and development of academic independence: a case of life sciences in Japan

ABSTRACT

Academic training is the initial step for junior scientists to learn to develop into independent scientists. This study investigates how supervisors decide to employ different approaches of early-career research training, and how these approaches influence the degree of trainees’ independence in their later careers. Drawing on survey and bibliometric data of life scientists in Japanese universities, this study presents the following findings. First, if scientists are allowed higher autonomy in upstream research functions in early-career training, they later tend to attain greater organizational independence with higher organizational ranks. Second, if scientists are encouraged to deviate from conventional research topics during early-career training, they later tend to achieve greater cognitive independence by producing original research output. Third, the differences in the training approaches chosen by individual supervisors are influenced by the training that they had received in their early-career training. Overall, the study suggests that training approaches and independence of scientists are socialized in the local training context and passed down from one generation to the next.

KEYWORDS:

• Academic training
• graduate education
• independence
• autonomy
• PhD
• postdoctoral researcher

Introduction

The modern knowledge society is based on frontier scientific and technological knowledge, and thus, it is crucial that scientists – as a core source of frontier knowledge development – be sustainably developed (Bozeman and Corley 2004; Laudel and Glaser 2008; Stephan 2012). In this regard, academic institutions play a central role in providing academic training (i.e. postgraduate research education and postdoctoral training), where substantial investment has been made to produce a growing number of scientists (Cyranoski et al. 2011; Gould 2015; NRC 1998).

Academic training is the initial step for junior scientists (i.e., graduate students, postdocs) to acquire knowledge and skills and learn to become independent scientists (Bozeman and Corley 2004; Gardner 2008; Laudel and Glaser 2008; Lee 2008). Junior scientists engage in scientific research projects under the supervision of established scientists (Blau 1994; Delamont and Atkinson 2001; Latour and Woolgar 1979), and this process of learning-by-doing transforms junior scientists into independent scientists (Long, Allison, and Robert 1979; Miller, Glick, and Cardinal 2005).

In an attempt to identify effective approaches of academic training, the higher education literature has investigated how academic training is delivered – in particular, how supervisors interact with junior scientists, and found substantial variations in the approaches of training between supervisors even in the same scientific field (Bastalich 2017; Hockey 1991; Mainhard et al. 2009; Marsh, Rowe, and Martin 2002; Shibayama, Baba, and Walsh 2015). This variation in training approaches influences junior scientists’ understanding of the scientist’s job as well as their skill development, thus shaping their future careers (Marsh, Rowe, and Martin 2002; Morrison et al. 2011; Sambunjak, Straus, and Marusic 2006; Shibayama and Kobayashi 2017). Despite the crucial role of academic training, the previous literature has two limitations. First, prior studies evaluated mostly the short-term impact of training, for example, on the basis of students’ satisfaction, whereas longer-term impact has been insufficiently explored. Second, the literature has been largely silent on how the observed variation in the training approaches is shaped in the first place.

This study aims to address these limitations with two objectives. First, we investigate the long-term impact of academic training, particularly in terms of the degree of trainees’ independence in later careers. In this inquiry, we assess trainees’ later independence in two dimensions: (1) organizational independence – the extent to which scientists are independent from organizational control by attaining higher organizational positions in a career ladder – and (2) cognitive independence – the extent to which scientists are thematically independent by establishing their own research agendas (Gläser and Laudel 2015; Laudel 2017; Laudel and Glaser 2008). In investigating academic training, we highlight two aspects: (1) functional autonomy – the extent to which junior scientists are encouraged to take responsibility in different research functions so that they learn to fulfill them, and (2) cognitive exploration – the extent to which junior scientists are encouraged to deviate from scientific conventions so that they learn to identify their original research agenda (Kam 1997; Shibayama, Baba, and Walsh 2015). We argue that a variation in these training approaches makes long-lasting impact on trainees’ later independence. The second objective of this study is to investigate the source of the variation in training approaches. In particular, we argue that scientists who are trained under a certain training approach will apply a similar approach when they become established – that is, training approaches are socialized in the local training context and passed down from a generation to the next generation.

To test our argument, we draw on a sample of mid-career life scientists and trace their career from their postgraduate education period up to 2018. Using a questionnaire survey, we inquired into the training experience at their PhD and early postdoc stages as well as into their current training activities as supervisors. We also surveyed their career status and bibliometrically analyzed their publications to measure their organizational and cognitive independence. Finally, we analyze the impact of training approaches they experienced in early-career stages on (1) their independence in later careers and (2) the training approaches they employ for their trainees.

The remainder of this paper is structured as follows. The next section discusses the previous literature. The subsequent section outlines the empirical setup and data. This is followed by the results of empirical analyses. Finally, the results are interpreted and implications are discussed.

Literature review

Academic independence

The academic career starts from the apprenticeship status, in which junior scientists are trained and guided by senior scientists, and gradually develops into the independent one, in which the trained scientists are responsible for setting their own research agendas and running their own research teams (Laudel and Glaser 2008).

Independence in academia can be understood from two aspects – organizational and cognitive. First, as scientists climb the career ladder, they generally attain higher organizational power and greater autonomy in mobilizing resources (e.g. time, budget) (Hackett 1990; Heinze et al. 2009; Hollingsworth 2004; Lee, Walsh, and Wang 2015), although the link between career progression and the level of organizational independence may not be perfectly linear. Second, scientists are expected to achieve cognitive independence by identifying and developing their own research agendas as an originator rather than a follower of other scholars (Hagstrom 1974; Merton 1973; Stephan 1996), though the expectation for originality might differ in its degree and nature between scientific communities.

The process of developing the two dimensions of independence is interrelated. On the one hand, because organizational independence helps invest in developing original research agendas (Andrews 1976; Heinze et al. 2009; Hollingsworth 2004), cognitive independence should follow organizational independence. On the other hand, if the science community expects scientists to demonstrate originality and this expectation is incorporated in the career system (NRC 2005; Macfarlane and Saitoh 2008), organizational independence should follow cognitive independence. The interrelation may be also ambiguous. Originality incurs substantial risk of failure, and original discoveries can be unfairly recognized or deliberately undermined by mainstream scientists (Azoulay, Graff Zivin, and Manso 2011; Nicholson and Ioannidis 2012; Wang, Veugelers, and Stephan 2017). Thus, cognitive independence can actually deter career progression. Nonetheless, scientists by and large aim to gain independence in both senses gradually throughout the careers.

Academic training and independence

In the path to academic independence, academic training is the initial and crucial step for junior apprentices to learn knowledge and skills from senior scientists (Bourdieu 1975; Bozeman and Corley 2004; Kuhn 1977; Laudel and Glaser 2008). Junior scientists usually begin to receive academic training during postgraduate degree programs, and after their degrees, undertake extra training as postdoctoral researchers (Kahn and Ginther 2017; Stephan 2012: Ch.7). The training occurs at multiple levels of academic institutions (Campbell 2003; Gardner 2007; Golde 2005), but a closer inspection reveals a laboratory or a small team of scientists as the main arena of academic training.¹ A lab consists of a senior scientist as the supervisor and junior scientists (e.g. graduate students) as members. The core part of academic training employs learning-by-doing through the apprenticeship model, where junior scientists are engaged in research projects and tasked with solving research questions under the instruction of supervisors (Blau 1994; Campbell 2003; Carayol and Matt 2004; Delamont and Atkinson 2001; Latour and Woolgar 1979; Lee 2008; Muller 2014).

To identify effective approaches of academic training, the higher education literature has investigated various aspects of supervising practices and styles (Bastalich 2017; Brown and Atkins 1988; Hockey 1991; Marsh, Rowe, and Martin 2002). Many studies refer to the psycho-social traits of supervisors, such as being empathetic, showing respects to students, and sharing anxiety with students, as well as the instrumental aspects of training, such as extending network and supporting career development (Green and Bauer 1995; Paglis, Green, and Bauer 2006; Tenenbaum, Crosby, and Gliner 2001). Several studies referred to the contrasting approaches in developing junior scientists’ independence. For example, Mainhard et al. (2009) highlighted dominance and submission as conflicting dimensions of student-supervisor relationships in social sciences in the Netherlands. Murphy, Bain, and Conrad (2007) contrasted controlling and guiding as supervisors’ roles in the context of engineering PhD programs in Singapore. Similarly, Wichmann-Hansen and Herrmann (2017) distinguished directive and non-directive supervision styles in PhD degree programs in Denmark.

In studying academic training and its contribution to academic independence, we focus on two aspects of training approaches – functional autonomy and cognitive exploration.

Functional autonomy

The first aspect is functional autonomy – the extent to which junior scientists are encouraged to take responsibility in different research functions so that they become capable of carrying out those functions on their own (Kam 1997; Shibayama, Baba, and Walsh 2015). We argue that having a high degree of functional autonomy in early-career training helps develop academic independence in later-career stage.

Scientific research usually starts from setting a research question and developing a research plan; then, the question is tested by experiments, simulations, and other methods; and finally, the test results are interpreted and used to advance extant knowledge; and this last stage often raises new questions for future research, and the whole process is repeated (Nightingale 1998). Each step in this process requires unique functional knowledge and skills, and labs often employ a division of labor to efficiently allocate research functions among lab members (Hackett 1990; Latour and Woolgar 1979; Laudel 2001; Shibayama, Baba, and Walsh 2015; Walsh and Lee 2015). Prior literature generally assumes that lab heads play a main role in upstream functions that require cognitive skills, such as problem identification, and members are in downstream functions that require physical labor, such as carrying out experiments (Delamont and Atkinson 2001; Laudel 2001; NRC 1998; Shibayama, Baba, and Walsh 2015). For example, Delamont, Parry, and Atkinson (1997), drawing on lab ethnographies in natural sciences in the UK, found that supervisors were responsible for identifying research projects and assigning them to students, while students typically considered their lab experience an opportunity to acquire technical skills for downstream tasks.

Nonetheless, for nurturing independence, a clear division of labor, in which junior scientists specialize in certain functions and remain incapable in others, would not be ideal, because they have to learn how to design and coordinate broader research projects as independent scientists, not just fulfill isolated technical functions (Bozeman and Corley 2004). A complete lack of experience in upstream functions might signal their incapability of running research independently. This would undermine junior scientists’ postgraduate career prospects, particularly when the modern academic system requires successful records of leadership and independence as conditions for employment and funding (NRC 2005).

Indeed, a few studies suggested variations in terms of functional autonomy. For example, Kam (1997) investigated the allocation of research functions between students and supervisors and identified students’ varying degrees of functional dependence on their supervisors. Shibayama, Baba, and Walsh (2015) also studied the allocation of functional tasks between supervisors, postdocs, and PhD students in life science labs, finding that junior scientists usually take on downstream functions but their responsibility in upstream functions varies between labs. Shibayama (2019) further found that engagement in upstream functions during the PhD period is associated with higher performance in later career stages.

Cognitive exploration

The second aspect of academic training is cognitive exploration – the extent to which junior scientists are encouraged to deviate from scientific conventions and learn to identify their original research agendas (Shibayama, Baba, and Walsh 2015). We argue that scientists who are given a high degree of cognitive exploration in early-career training are likely to gain greater independence, mainly in cognitive dimension, in later-career stage.

In general, junior scientists begin their careers by following their supervisors’ established line of research (Campbell 2003; Delamont and Atkinson 2001; Kam 1997), in which their cognitive independence is limited. This is because supervisors have a strong authority in deciding the subject areas and research questions of their member scientists (Delamont and Atkinson 2001; Laudel 2001; NRC 1998; Shibayama, Baba, and Walsh 2015). This is justified in that junior scientists can draw on resources and capabilities accumulated in the supervisors’ lab and avoid risks entailed in identifying their own research lines.

The impact of junior training can be long-lasting because of path dependency in research topics (Austin 2002; Campbell 2003; Laudel and Glaser 2008). Research topics selected in the training period can influence those at later career stages, potentially limiting the level of cognitive independence in the long term.

Despite such a tendency, junior scientists are also encouraged to achieve cognitive independence by identifying their own original research agendas and deviating from conventions (Clarke and Lunt 2014; Lienard et al. 2018; NRC 2005; Shibayama 2019). Junior scientists often move across multiple fields during the training periods by working in multiple labs to explore their own agendas. It is also plausible that junior scientists under supervision are allowed to explore beyond supervisors’ expertise but still lies within their interest (Kam 1997; Lee 2008; Lee, Dennis, and Campbell 2007; Mainhard et al. 2009; Shibayama 2019).

To develop cognitive independence, scientists need not only to learn domain-specific knowledge concerning an original scientific agenda, but also more importantly, to learn to explore and identify an original agenda in the first place. For the latter, early-career training offers learning-by-doing experiences.

Original research agendas can be explored through various routes. For example, scientists can deviate from traditional mono-disciplinary research and aim at recombination of multiple disciplines (i.e. interdisciplinarity) (Boudreau et al. 2016; Foster, Rzhetsky, and Evans 2015; Mednick 1962; Nelson and Winter 1982; Simonton 2003). Alternatively, scientists can tackle neglected topics, while most scientists are interested in well-recognized problems (Bourdieu 1975). It is also plausible to avoid the over-crowded mainstream and to search for niche, unique spaces (Horlings and Gurney 2013). Scientists can aim at up-to-date, topical areas, rather than traditional long-studied areas (Ion and Ceacero 2017). Scientists can also use a heterodox approach, rather than an orthodox one, even if the focal research question is not original. All these approaches can lead scientists to originality, thus helping them attain cognitive independence.

Variation in academic training approaches

The literature has suggested that training approaches are not uniform but can substantially differ between supervisors (Bastalich 2017; Brown and Atkins 1988; Hockey 1991; Marsh, Rowe, and Martin 2002). For example, Mainhard et al. (2009) presented eight types of supervisor-student interpersonal relationships. Hockey (1996) identified various supervisors’ motives for training, such as the moral obligation for education and the need for research labor. Importantly, as above discussed, considerable differences are suggested in the approach to nurture independence.

The variation can be attributed to many sources. It may be a strategic choice by supervisors. For example, allowing higher functional autonomy to junior scientists can compromise supervisors’ research performance while limiting it can better exploit physical labor of junior scientists (Shibayama 2019). Some senior scientists may choose to nurture the independence of junior scientists at the cost of their own performance. Training approaches may also vary because the best practice of academic training has not been fully established (Lee, Dennis, and Campbell 2007), and consequently, supervisors employ a certain training approach only because it is a practice in the local context. That is, we argue that one important source of variation in training approaches is socialization (Austin 2002; Golde 2005). Functional autonomy and cognitive exploration form part of local norms, and scientists who experienced a certain approach in the junior stages employ the same approach for their trainees when they are established (Halse 2011).

Data and methods

Empirical setup

As an empirical ground, we draw on a sample of mid-career life scientists affiliated to Japanese universities. We chose Japanese academia as an empirical arena for two reasons. First, the postgraduate education system in Japan relies primarily on lab-based training. As the role of school-wide training is minor and the practice of academic training differs substantially between labs, we expect the role of labs and supervisors in junior training to be highlighted. Second, the socioeconomic context is rather homogeneous, for example, with the majority of postgraduate students and supervisors being Japanese.² This allows us to observe the phenomenon of our interest with a limited noise.

In Japan, approximately 700 universities offer four-year undergraduate programs, with approximately 400 of those offering PhD-level postgraduate programs. Universities are categorized into three groups based on governing bodies: 86 national, 92 regional (of prefectures or cities), and 603 private. Among the three groups, national universities are the main providers of both academic training and research, while most private universities are oriented to undergraduate education. For example, national, regional, and private universities accounted for 72%, 5%, and 23%, respectively, of the approximately 12,000 PhD degrees awarded in 2019,³ and for 69%, 7%, and 24%, respectively, of the national research funds allocated in 2019.⁴ Most postgraduate programs in Japan consist of two parts: a two-year master’s program and a three-year PhD program. Students decide whether to pursue a PhD degree during their master’s program. When enrolling in PhD programs, the majority of students remain under the same professor’s supervision in the same lab (Kato and Chayama 2010).

As in many other countries, a PhD degree is usually required for academic employment in Japan. Typically, PhD graduates have postdoc experience before finding a faculty position. For example, 44% of PhD graduates in science from 2002 to 2006 became postdocs, while 6.2% obtained faculty positions immediately after graduation (Misu, Horoiwa, and Chayama 2010).

Data

Our sample consists of 188 scientists who earned their PhD degrees in life sciences⁵ in 1995–2011 and is currently acting as supervisors. The primary data were collected with a questionnaire survey conducted in 2018. The survey asked a wide range of questions, of which this study mainly uses the parts concerning respondents’ training experience during the PhD period, during the first postdoc period, and in the current supervising activities.

The survey respondents were selected based on the following criteria. First, we chose 20 research-intensive universities that publish online dissertation databases. Second, we identified 1697 PhD graduates who earned a degree in 1995–2011 in the field of life sciences in these universities. Third, we excluded 151 non-Japanese graduates for practical reasons.⁶ Fourth, we investigated their postgraduate careers and identified 587 graduates who were in academic employment as of 2017. We mailed our survey to these scientists and collected 270 responses after two waves of requests (response rate = 46%).⁷ We evaluated the non-response bias by comparing 270 respondents and 303 non-respondents in terms of publication counts during the PhD period, genders, and university rankings, finding no significant difference.

Of the 270 respondents we find that 188 (70%) have attained faculty position and supervised at least one graduate student, based on which the following analyses are carried out. The mean age is 42, and 13% are female.

Measures

Functional autonomy

To measure the degree of functional autonomy, we surveyed the magnitude of responsibility in different research functions. On the basis of our interviews of scientists and prior literature (Kam 1997; Shibayama and Kobayashi 2017; Shibayama, Baba, and Walsh 2015), we defined five functions: (a) setting a research subject, (b) formulating a hypothesis, (c) planning an experiment, (d) performing an experiment, and (e) writing a paper. For each function, the respondents answered the degree of their own responsibility as a trainee during the PhD period as well as during the first postdoc period. The response takes a three-point scale: (0) no responsibility, (1) minor responsibility, and (2) major responsibility. The same questions are asked for the current supervising activity. We asked the respondents to choose one graduate student who they supervised most recently, and to answer his/her responsibility (but not of the respondents themselves) in the five functions. When the respondents did not supervise a graduate student at the PhD level, they refer to a student at the master level.

Cognitive exploration

To measure the training for strategic independence, we surveyed the features of research carried out in the lab to which the respondents were affiliated during the PhD period and during the first postdoc. Drawing on the interviews and previous literature (Azoulay, Graff Zivin, and Manso 2011; Dirk 1999; Guetzkow, Lamont, and Mallard 2004; Shibayama and Wang 2019; Yegros-Yegros, Rafols, and D’este 2015), we identified five research directions with which scientists differentiate themselves from others, and developed questionnaire items on (a) topicality, (b) community recognition, (c) niche, (d) methodological heterodoxy, and (e) interdisciplinarity (Table 1). Each item consists of a pair (A and B) of contrasting characteristics. The respondence chose: (1) mainly A, (2) rather A than B, (3) A and B to a similar extent, (4) rather B than A, or (5) mainly B. Similarly, we asked the same questions as to the research directions in the respondents’ current lab where they play a role of supervisors.

Table 1. Research directions for cognitive exploration.

	A	B
(1) Topicality	Traditional	Up-to-date
(2) Community recognition	Widely recognized as a relevant topic in the scientific community	Not recognized as a relevant topic in the scientific community
(3) Niche	In the mainstream	In a niche
(4) Methodological heterodoxy	Draws on an orthodox approach	Draws on a heterodox approach
(5) Interdisciplinarity	Aims to deepen a specific research area	Aims to bridge multiple research areas

Note: The respondents answer whether their research topic is/was considered A or B, as of the time when the research was carried out, in a five-point scale: (1) mainly A, (2) rather A than B, (3) A and B to a similar extent, (4) rather B than A, and (5) mainly B.

Organizational independence

As a measure of organizational independence, we draw on the organizational rank. An ordinal variable is coded 0 if a respondent is currently an assistant professor, 1 if an associate professor, and 2 if a full professor.

Cognitive independence

To measure cognitive independence, we use a bibliometric indicator to assess whether or not the respondents are currently capable of producing original research output. To this end, we downloaded all publications of the respondents from the Web of Science and computed the originality of each paper based on the recombinant novelty indicator (Wang, Veugelers, and Stephan 2017).⁸ Then, we code a dummy variable 1 if a respondent published at least one original publication in the last five years and 0 otherwise.

Control variables

In the later section, we carry out regression analyses to test our argument, in which we include several control variables. First, we control for a gender as it may affect a supervising style (e.g. Pezzoni et al. 2016). A dummy variable is coded 1 for female respondents and 0 for male (female). Second, we include the number of years after graduation, as scientists can learn more through a longer career, which can affect their training approaches (#years since PhD). Third, we control for the training stage of the respondents’ students. A dummy variable is coded 1 if the student under supervision is at the master level, and 0 if the student is at the PhD level (supervising MS student). Fourth, we control for the respondent’s publication performance before they started PhD thesis projects – i.e. during the two-year master program (#pub before PhD). The descriptive statistics of all variables are presented in Table 2.⁹

Table 2. Descriptive statistics.

Category	Variables	Mean	SD	Min	Max
Independence	Organizational independence	1.610	.702	1.000	3.000
	Cognitive independence	.399	.491	.000	1.000
Functional autonomy (as supervisor)	Subject setting (PhD)	1.039	.767	.000	2.000
	Hypothesis formulation (PhD)	1.140	.755	.000	2.000
	Experiment planning (PhD)	1.455	.689	.000	2.000
	Experiment execution (PhD)	1.804	.520	.000	2.000
	Writing (PhD)	1.251	.748	.000	2.000
Functional autonomy (PD)	Subject setting (PD)	1.473	.720	.000	2.000
	Hypothesis formulation (PD)	1.718	.576	.000	2.000
	Experiment planning (PD)	1.883	.369	.000	2.000
	Experiment execution (PD)	1.952	.259	.000	2.000
	Writing (PD)	1.717	.508	.000	2.000
Functional autonomy (PhD)	Subject setting (PhD)	1.346	.722	.000	2.000
	Hypothesis formulation (PhD)	1.670	.566	.000	2.000
	Experiment planning (PhD)	1.897	.354	.000	2.000
	Experiment execution (PhD)	1.973	.219	.000	2.000
	Writing (PhD)	1.632	.603	.000	2.000
Cognitive exploration (as supervisor)	Topical (PhD)	3.182	1.030	1.000	5.000
	Unrecognized (PhD)	2.611	1.225	1.000	5.000
	Niche (PhD)	3.244	1.142	1.000	5.000
	Heterodox (PhD)	3.066	1.200	1.000	5.000
	Interdisciplinary (PhD)	3.111	1.494	1.000	5.000
Cognitive exploration (PD)	Topical (PD)	3.271	1.182	1.000	5.000
	Unrecognized (PD)	2.691	1.202	1.000	5.000
	Niche (PD)	3.170	1.086	1.000	5.000
	Heterodox (PD)	2.755	1.158	1.000	5.000
	Interdisciplinary (PD)	2.420	1.410	1.000	5.000
Cognitive exploration (PhD)	Topical (PhD)	2.930	1.084	1.000	5.000
	Unrecognized (PhD)	2.973	1.219	1.000	5.000
	Niche (PhD)	3.459	1.053	1.000	5.000
	Heterodox (PhD)	2.688	1.243	1.000	5.000
	Interdisciplinary (PhD)	2.075	1.284	1.000	5.000
Control variables	#Years since PhD	13.170	3.978	7.000	23.000
	Female	.128	.335	.000	1.000
	#Pub before PhD	1.584	1.943	.000	14.000
	Supervising MS student	.292	.456	.000	1.000

Results

Training at PhD and postdoc stages

Functional autonomy

First, we illustrate the functional autonomy that our respondents experienced in their early careers (Figure 1(A,B)). In the PhD stage, most of our respondents were given a major responsibility in planning and performing experiments but a lower responsibility in other functions (subject setting, hypothesis formulation, and writing). This is consistent with the stereotypical understanding on the functional task allocation, where junior apprentices have low autonomy in functions that require cognitive skills (Delamont and Atkinson 2001). Nonetheless, 46% of the respondents were given major responsibility in subject setting, 68% in hypothesis formulation, and 71% in writing. As engagement in the upstream tasks is considered a route for independence (Shibayama 2019), this result suggests that not a few supervisors are willing to nurture independence in the PhD stage.

Figure 1. Functional autonomy in early-career training.

Note: (A) In the respondents’ PhD stage. (B) In the respondents’ first postdoc stage. (C) In the respondents’ current training activity as supervisors. The respondents reported their own responsibility in (A) and (B), they referred to the responsibility of their students in (C).

Comparing the PhD stage and the postdoc stage, we find that the autonomy given to junior trainees is higher in the later stage. Respondents with major responsibility in subject setting increased from 46% in the PhD stage to 58% in the postdoc stage; from 68% to 76% in hypothesis formulation; from 71% to 76% in writing. Thus, as expected, scientists attain greater functional autonomy as the career progresses.

Overall, junior trainees’ responsibility in the two stages is significantly correlated.¹⁰ This can be because junior trainees simply stayed in the same lab after graduation and continued the same working styles. In our sample, 32% of the respondents stayed in the same lab while 68% changed their labs after graduation. Excluding the former respondents – i.e. focusing on mobile scientists – we find that the correlations remain significant. Hence, junior trainees who had a major responsibility in the PhD period are likely to continue to do so in the postdoc period even after changing labs and supervisors.

Cognitive exploration. Second, we investigate the opportunity of cognitive exploration that the respondents experienced in their early careers (Figure 2(A,B)). We measured the five directions of the research topics that our respondents engaged in as junior trainees. First, 34% of our respondents engaged in a traditional topic and 30% in an up-to-date topic.¹¹ Second, 37% engaged in a well-recognized topic and 31% in a yet-to-be-recognized topic. Third, 22% considered that their research was in the mainstream whereas 49% in a niche area. Fourth, 52% considered their approach was orthodox while 22% believed they employed a heterodox approach. Fifth, 73% believed their research was based on a single discipline while 18% believed that they aimed at bridging multiple disciplines. These results suggest that not a few supervisors give their trainees research topics that deviate from scientific conventions.

Figure 2. Cognitive exploration.

Note: (A) In the respondents’ PhD stage. (B) In the respondents’ first postdoc stage. (C) In the respondents’ current training activity as supervisors.

When the PhD stage is compared with the postdoc stage, a few changes are observed. The topicality increases from 30% to 44%, and interdisciplinarity increases from 18% to 28%. These imply that postdoc researchers are geared towards original directions. However, mainstream topics, as opposed to niche topics, increased from 22% to 26%. The methodological heterodoxy is unchanged across the two stages. Thus, the temporal change in cognitive exploration is less clear at this early stage.

Overall, the five research directions are significantly correlated between the two career stages. This may be because junior trainees simply continued the PhD thesis topics even after graduation. In our sample, 14% of the respondents continued their thesis projects in the postdoc stage, whereas the rest started new topics. Excluding the former group – i.e. focusing on those who changed topics – we still find that the research directions are positively correlated between the two stages. Hence, junior trainees who sought new research directions in the PhD period are likely to continue doing so in the postdoc period.

Training approaches and independence

Organizational independence

We examine how the training approaches in early careers influence scientists’ independence in later careers. First, we examine the impact of functional autonomy (Table 3(A)) and cognitive exploration (Table 3(B)) in the early careers on organizational independence measured by the current organizational rank. Comparing Table 3(A) and (B), we find that organizational independence is mainly explained by functional autonomy but not by cognitive exploration. Focusing on functional autonomy, Table 3(A) indicates that greater responsibility in upstream functions (Model 1: subject setting, Model 2: hypothesis formulation, and Model 5: writing) is associated with faster promotion. Model 6 further collapses the five measures of functional autonomy into (1) the upstream function – the mean of subject setting, hypothesis formulation, and writing – and (2) the downstream function – the mean of experiment planning and experiment execution – and tests their effect on organizational independence.¹² The result indicates that organizational independence is significantly positively associated with autonomy in upstream functions (b = 1.039, p < .01) but not with that in downstream functions. Computing the marginal effect, junior scientists who had major responsibility in upstream functions in the PhD stage, compared to those who had no responsibility, is 37% more likely to be promoted to associate professors and 8.2% more likely to be promoted to full professors.¹³ These results are consistent with our expectation that skills in upstream tasks are essential in running their own research teams (Shibayama 2019). Model 7 also incorporates functional autonomy in the postdoc stage, finding that organizational independence is attributed to functional autonomy in PhD training rather than to that in postdoc training.

Table 3. Prediction of organizational independence by early-career training.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
(A) Functional autonomy
#Years since PhD	0.363***	0.363***	0.345***	0.343***	0.359***	0.374***	0.376***
	(0.0517)	(0.0513)	(0.0500)	(0.0497)	(0.0510)	(0.0523)	(0.0531)
Female	0.0379	−0.0528	−0.0536	−0.0441	0.0650	0.122	0.0727
	(0.546)	(0.546)	(0.536)	(0.535)	(0.534)	(0.548)	(0.560)
#Pub before PhD	0.183*	0.193*	0.194*	0.199*	0.197*	0.185*	0.182*
	(0.0918)	(0.0942)	(0.0934)	(0.0931)	(0.0903)	(0.0906)	(0.0918)
Subject setting (PhD)	0.424^†
	(0.245)
Hypothesis formulation (PhD)		1.013**
		(0.370)
Experiment planning (PhD)			0.304
			(0.512)
Experiment execution (PhD)				−0.183
				(0.723)
Writing (PhD)					0.546^†
					(0.307)
Upstream (PhD)						1.058**	0.846*
						(0.406)	(0.415)
Downstream (PhD)						−0.339	−0.340
						(0.725)	(0.723)
Upstream (PD)							0.529
							(0.439)
Downstream (PD)							1.355
							(1.217)
Field dummies	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	178	178	178	178	178	178	178
Log likelihood	−134.0	−131.4	−135.4	−135.5	−133.9	−131.9	−129.1
Chi^2 test	76.35***	81.47***	73.63***	73.33***	76.56***	80.64***	86.13***

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
(B) Cognitive exploration
#Years since PhD	0.365***	0.344***	0.344***	0.368***	0.343***	0.352***	0.355***
	(0.0512)	(0.0497)	(0.0499)	(0.0521)	(0.0497)	(0.0508)	(0.0510)
Female	−0.174	−0.145	−0.124	−0.00621	−0.123	−0.160	−0.151
	(0.541)	(0.529)	(0.526)	(0.533)	(0.530)	(0.526)	(0.529)
#Pub before PhD	0.211*	0.209*	0.206*	0.184^†	0.212*	0.198^†	0.201*
	(0.102)	(0.101)	(0.101)	(0.103)	(0.101)	(0.102)	(0.102)
Topical (PhD)	0.0106
	(0.157)
Unrecognized (PhD)		−0.0585
		(0.132)
Niche (PhD)			0.0832
			(0.162)
Heterodox (PhD)				0.300*
				(0.138)
Interdisciplinary (PhD)					−0.0361
					(0.128)
Unconventional (PhD)						0.212	0.144
						(0.201)	(0.231)
Unnoticed (PhD)						0.0325	0.0259
						(0.169)	(0.179)
Unconventional (PD)							0.138
							(0.219)
Unnoticed (PD)							0.00827
							(0.178)
Field dummies	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	177	178	177	178	178	178	178
Log likelihood	−132.1	−135.2	−134.6	−132.9	−135.3	−134.8	−134.6
Chi^2 test	78.11***	73.16***	72.35***	77.79***	73.05***	74.08***	74.47***

Note: Ordered logistic regressions. Unstandardized coefficients (standard errors in parentheses). Two-tailed test. ^†p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.

We control for three variables in the regression models. First, the result suggests a positive coefficient of the number of years since graduation because organizational promotion takes time. Second, the result does not show that females are less likely to attain higher organizational positions, unlike previous studies (Lawson and Shibayama 2014). Third, we control for the respondents’ publication performance during the master level (the two years before starting PhD). This is because we are concerned about a selection bias, in that junior scientists who had higher ability and greater potential to be promoted were given higher autonomy. Though we do find a positive impact of the prior publication performance, the effect of functional autonomy on organizational independence is not affected regardless of controlling for the variable or not.

Cognitive independence

Similarly, we examine the impact of functional autonomy (Table 4(A)) and cognitive exploration (Table 4(B)) in early-career training on cognitive independence measured by the originality of current publications. Comparing Table 4(A) and (B), we find that cognitive independence is mainly explained by cognitive exploration but not by functional autonomy. Focusing on cognitive exploration, Table 4(B) indicates that topicality (Model 1), heterodoxy (Model 4), and interdisciplinarity (Model 5) are associated with later cognitive independence. As a factor analysis finds that the five measures can be collapsed into two factors,¹⁴ we prepare two measures: (1) the unconventional direction – taking the mean of topicality, heterodoxy, and interdisciplinarity – and (2) the unnoticed direction – taking the mean of niche and community recognition. Of the two directions of cognitive exploration, Model 6 finds that the unconventional direction is significantly positively associated with cognitive independence (b = .806, p < .001). Computing the marginal effect, junior scientists who engaged in unconventional topics in the PhD stage, compared to those who engaged in conventional topics, are 67% more likely to produce original publications after established.¹⁵ Model 7 further incorporates cognitive exploration in the postdoc stage, finding that cognitive independence is explained by cognitive exploration in PhD training but not by that in postdoc training. As to the control variables, the result shows that females are slightly less likely to attain cognitive independence.

Table 4. Prediction of cognitive independence by early-career training.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
(A) Functional autonomy
#Years since PhD	0.00785	0.0199	0.0193	0.0169	0.0299	0.0224	0.0195
	(0.0412)	(0.0410)	(0.0408)	(0.0408)	(0.0417)	(0.0414)	(0.0417)
Female	−1.083^†	−0.947^†	−0.989^†	−0.985^†	−0.835	−0.945^†	−0.905
	(0.577)	(0.568)	(0.566)	(0.565)	(0.572)	(0.571)	(0.582)
#Pub before PhD	0.0165	−0.00348	0.00263	0.00931	0.00986	0.00162	−0.00513
	(0.0826)	(0.0836)	(0.0828)	(0.0824)	(0.0847)	(0.0830)	(0.0838)
Subject setting (PhD)	−0.284
	(0.228)
Hypothesis formulation (PhD)		0.450
		(0.307)
Experiment planning (PhD)			0.393
			(0.472)
Experiment execution (PhD)				0.610
				(0.864)
Writing (PhD)					0.594*
					(0.289)
Upstream (PhD)						0.190	0.0167
						(0.347)	(0.365)
Downstream (PhD)						0.500	0.471
						(0.718)	(0.696)
Upstream (PD)							0.427
							(0.389)
Downstream (PD)							0.525
							(0.849)
Field dummies	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	179	179	179	179	179	179	179
Log likelihood	−118.1	−117.7	−118.5	−118.6	−116.6	−118.3	−117.0
Chi^2 test	5.091	5.770	4.257	4.135	8.025	4.699	7.316

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
(B) Cognitive exploration
#Years since PhD	0.0211	0.0174	0.0171	0.0332	0.0276	0.0446	0.0487
	(0.0421)	(0.0405)	(0.0411)	(0.0418)	(0.0418)	(0.0435)	(0.0437)
Female	−1.269*	−1.075^†	−1.075^†	−0.962^†	−1.212*	−1.239*	−1.239*
	(0.571)	(0.561)	(0.563)	(0.567)	(0.572)	(0.581)	(0.583)
#Pub before PhD	0.0376	0.0395	0.0428	0.0115	0.0323	0.00516	0.0123
	(0.0913)	(0.0860)	(0.0862)	(0.0873)	(0.0875)	(0.0882)	(0.0887)
Topical (PhD)	0.507**
	(0.164)
Unrecognized (PhD)		0.0206
		(0.128)
Niche (PhD)			0.0593
			(0.154)
Heterodox (PhD)				0.308*
				(0.133)
Interdisciplinary (PhD)					0.339**
					(0.125)
Unconventional (PhD)						0.822***	0.720**
						(0.214)	(0.238)
Unnoticed (PhD)						0.205	0.208
						(0.171)	(0.181)
Unconventional (PD)							0.266
							(0.229)
Unnoticed (PD)							−0.0597
							(0.173)
Field dummies	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	178	179	178	179	179	179	179
Log likelihood	−112.5	−118.3	−117.4	−115.6	−114.5	−109.9	−109.1
Chi^2 test	15.16^†	4.593	4.647	10.12	12.27	21.41*	22.96*

Note: Logistic regressions. Unstandardized coefficients (standard errors in parentheses). Two-tailed test. ^†p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.

Determinants of training approaches

Functional autonomy

Finally, we examine the determinants of different training approaches. Analyzing the variation in functional autonomy in the PhD stage, we first find that the training approaches differ between scientific fields. One-way ANOVA indicates field differences in the level of responsibility in subject setting (p < .001), hypothesis formulation (p < .001), experimental planning (p < .01), and writing (p < .05), but not in experiment execution. Nevertheless, the contribution of field differences is only minor. Although the largest field difference is found in subject setting, it accounts for only 6% of the total variance. Furthermore, a careful analysis suggests that the field difference is caused mainly by the specificity in the field of medicine. Once medicine is excluded, field difference turns insignificant. Therefore, as prior studies suggested (Bastalich 2017; Brown and Atkins 1988; Hockey 1991; Marsh, Rowe, and Martin 2002), training approaches vary substantially between individual supervisors.

Our central argument is that training approaches are socialized through the training itself. Thus, we investigate if the training approaches that junior scientists experienced as trainees is associated with the training approaches that they later employ as supervisors. To this end, we inquired into functional autonomy that our respondents as supervisors currently give to their trainees (Figure 1(C)). Note that the respondents reported the autonomy of their trainees but not of themselves. The correlation analysis finds that the responsibility of the respondents as junior trainees (Figure 1(A,B)) and that of the respondents’ trainees (Figure 1(C)) are significantly positively correlated, supporting our argument on the socialization of training approaches.¹⁶

To test this argument more rigorously, we draw on regression analyses (Table 5). The dependent variables are the autonomy in different research functions that the respondents currently give to their trainees. The independent variables are the autonomy in the same set of functions that the respondents were given in their PhD and postdoc periods. As the dependent variables are ordinal, we draw on ordered logistic regressions.

Table 5. Prediction of functional autonomy by early-career training.

	Model 1 Subject setting	Model 2 Hypothesis formulation	Model 3 Experiment planning	Model 4 Experiment execution	Model 5 Writing
#Years since PhD	0.0989^†	0.0653	0.0460	−0.126^†	0.0319
	(0.0509)	(0.0498)	(0.0529)	(0.0722)	(0.0495)
Female	0.286	0.509	−0.289	0.645	0.0732
	(0.512)	(0.521)	(0.530)	(1.135)	(0.563)
Supervising MS student	−1.103**	−1.411**	−1.833***	−0.339	−1.488***
	(0.425)	(0.435)	(0.456)	(0.625)	(0.445)
Assistant (baseline)
Associate	−0.601^†	−1.037**	−0.651	0.00432	−0.260
	(0.361)	(0.385)	(0.398)	(0.568)	(0.384)
Full professor	−1.110^†	−0.233	0.0429	3.077^†	−1.100^†
	(0.609)	(0.608)	(0.684)	(1.696)	(0.592)
Subject setting (PhD)	0.755**
	(0.249)
Subject setting (PD)	0.227
	(0.239)
Hypothesis formulation (PhD)		0.623*
		(0.307)
Hypothesis formulation (PD)		0.646*
		(0.306)
Experiment planning (PhD)			−0.0202
			(0.546)
Experiment planning (PD)			−0.354
			(0.530)
Experiment execution (PhD)				1.780**
				(0.689)
Experiment execution (PD)				−15.08
				(2,484)
Writing (PhD)					0.0659
					(0.319)
Writing (PD)					0.293
					(0.360)
Field dummies	Yes	Yes	Yes	Yes	Yes
Observations	155	155	154	155	152
Log likelihood	−157.1	−150.0	−127.7	−64.84	−146.5
Chi^2 test	22.27*	32.69***	28.03**	21.21*	25.09**

Note: Ordered logistic regressions. Unstandardized coefficients (standard errors in parentheses). Two-tailed test. ^†p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.

Model 1 shows that the autonomy of the respondents’ trainees in subject setting is significantly positively associated with the autonomy that the respondents had in their PhD period (b = .762, p < .01). Computing the marginal effect, supervisors who had major responsibility in the PhD stage, compared to supervisors who had no responsibility, is 27% more likely to give major responsibility to their trainees. Similarly, Model 2 shows that trainees’ autonomy in hypothesis formulation is positively associated with the respondents’ autonomy in their PhD period (b = .724, p < .05; marginal effect = +26%) and with that in the postdoc period (b = .498, p < .1; marginal effect = +22%). Model 4 also indicates that the trainees’ autonomy in executing experiment is positively associated with their supervisors’ autonomy in the PhD period (b = 1.680, p < .01; marginal effect = +56%). Models 3 and 5 do not find significant association as to the autonomy in experimental planning and writing. We also run the same set of regressions after excluding 22 respondents (14%) who have been staying in the same lab since their PhD period (i.e. inbred), obtaining qualitatively similar results.

These results support our argument that the training approaches for functional autonomy, importantly in upstream functions, are socialized in early-career training and transferred from one generation to the next.

Cognitive exploration

We analyze the determinants of different training approaches for cognitive exploration. We similarly analyze the variation in cognitive exploration in the PhD stage, finding that the training approaches slightly differ between scientific fields. Nonetheless, a statistical difference is found only in community recognition (p < .05), and the field difference accounts only for 4.9% of the total variance. Thus, training approaches for cognitive exploration differ largely at the individual level between supervisors even in the same field.

We then test if the training approaches that junior scientists experienced as trainees are associated with the approaches that they employ as supervisors. We inquired into cognitive exploration that our respondents currently employ (Figure 2(C)). The correlation analysis indicates that cognitive exploration at the established stage (Figure 2(C)) is overall positively correlated with that at their early-career stages (Figure 2(A,B)), supporting our argument on the socialization of training approaches.

We further test this argument with regression analyses (Table 6). The dependent variables are cognitive exploration in five research directions in the respondents’ current labs, and the independent variables are cognitive exploration in the respondents’ early-career training periods. Since the dependent variables are ordinal, we draw on ordered logistic regressions.

Table 6. Prediction of cognitive exploration by early-career training.

	Model 1 Topical	Model 2 Unrecognized	Model 3 Niche	Model 4 Heterodox	Model 5 Interdisciplinary
#Years since PhD	0.0173	−0.0224	−0.0543	0.000288	0.0149
	(0.0476)	(0.0448)	(0.0454)	(0.0461)	(0.0452)
Female	−1.022*	0.613	−0.105	−1.105*	−0.690
	(0.476)	(0.467)	(0.488)	(0.466)	(0.463)
Supervising MS student	−0.606^†	0.132	−0.205	−0.896*	−0.741*
	(0.367)	(0.353)	(0.359)	(0.354)	(0.360)
Assistant (baseline)
Associate	−0.0287	−0.662^†	0.485	0.191	−0.0222
	(0.355)	(0.348)	(0.350)	(0.357)	(0.343)
Full professor	−0.0496	0.793	0.886	0.535	0.565
	(0.582)	(0.571)	(0.563)	(0.563)	(0.590)
Topical (PhD)	0.373*
	(0.167)
Topical (PD)	0.237^†
	(0.143)
Unrecognized (PhD)		0.0793
		(0.134)
Unrecognized (PD)		0.356**
		(0.138)
Niche (PhD)			0.269^†
			(0.158)
Niche (PD)			0.866***
			(0.171)
Heterodox (PhD)				0.417**
				(0.154)
Heterodox (PD)				0.577***
				(0.161)
Interdisciplinary (PhD)					−0.0301
					(0.132)
Interdisciplinary (PD)					0.577***
					(0.132)
Field dummies	Yes	Yes	Yes	Yes	Yes
Observations	161	161	160	162	161
Log likelihood	−214.0	−228.6	−216.8	−217.6	−231.3
Chi^2 test	25.65**	28.35**	52.25***	58.59***	35.37***

Note: Ordered logistic regressions. Unstandardized coefficients (standard errors in parentheses). Two-tailed test. ^†p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.

Overall, the result shows positive associations between cognitive exploration employed as supervisors and that given as junior trainees, suggesting socialization in the PhD and postdoc stages. Model 1 shows that the topicality of the respondents’ current labs is significantly positively associated with that in their PhD period (b = .386, p < .05). Computing the marginal effect, the respondents who engaged in an up-to-date topic in their PhD stage, compared to those who engaged in a traditional topic, are 13% more likely to coordinate up-to-date topics after established. Models 2 and 5 indicate that cognitive exploration as supervisors is positively associated with that in their postdoc training, in terms of community recognition (b = .305, p < .05, marginal effect = +10%) and in terms of interdisciplinarity (b = .561, p < .001, marginal effect = +40%). Finally, Models 3 and 4 show that cognitive exploration as supervisors is positively associated both with that in their PhD training and with that in their postdoc training, in terms of niche (PhD: b = .292, p < .1, marginal effect = +10%; postdoc: b = .899, p < .001, marginal effect = +37%) and methodological heterodoxy (PhD: b = .434, p < .01, marginal effect = +20%; postdoc: b = .554, p < .001, marginal effect = +27%). We also run the same set of regressions after excluding the inbred respondents, finding qualitatively similar results.

These results support our argument that the training approaches for cognitive exploration in all dimensions are socialized in early-career training and transferred from one generation to the next.

Discussions and conclusions

Academic training is the initial step for junior scientists to learn to be independent scientists (Bozeman and Corley 2004; Gardner 2008; Laudel and Glaser 2008; Lee 2008). Drawing on a sample of life scientists in Japanese universities, this study investigates how different approaches in early-career training (i.e. postgraduate education and postdoc training) nurture organizational and cognitive independence, and how different training approaches are passed down from one generation to the next generation in a local training context. In particular, focusing on two aspects of training approaches, functional autonomy and cognitive exploration, this study presents a few key findings.

First, we confirm that training approaches substantially differ between individual supervisors. As to functional autonomy, we observe relatively low responsibility of junior scientists in upstream functions compared to downstream functions. Although the result suggests that functional autonomy slightly increases in postdoc training after PhD training, not a few postdocs are still dependent on their postdoc supervisors, implying the existence of ‘postdoc slave.’ While these results are consistent with a stylized view of junior scientists’ roles (Delamont and Atkinson 2001; Laudel 2001; NRC 1998; Shibayama, Baba, and Walsh 2015), some junior scientists believed that they played a major role in upstream functions. In terms of cognitive exploration, we find trainees tend to be engaged in conventional research topics. This is understandable as deviating from scientific conventions is risky for the career development of junior scientists (Bourdieu 1975). Nonetheless, not a few supervisors give their trainees research topics deviating from conventions, willing to nurture their cognitive independence.

Second, we find that the independence of scientists in later careers is considerably influenced by how they are trained in the initial career stage. In particular, autonomy in upstream functions in the PhD training is associated with organizational independence, whereas cognitive exploration through unconventional research topics is associated with cognitive independence.

Third, we show that the supervisors’ choice of training approaches is substantially influenced by their own experience in their early-career training. The result indicates that functional autonomy in upstream tasks and cognitive exploration in the investigated five dimensions are socialized in the local training context and passed down from senior supervisors to their junior trainees. Junior scientists who experienced a certain training approach keep employing the same approach after established. Importantly, this is true even when scientists change supervisors or change research topics.

These findings clearly indicate the importance of facilitating junior scientists’ autonomy and exploration in early careers, which not only helps develop their independence but also continues affecting following generations of scientists through inter-generational transfer of training practices. Emphasis on early independence is in line with guidelines from global science communities. For example, Lee, Dennis, and Campbell (2007) suggested that supervisors should avoid micro-managing their students’ research projects. In Japan as well, policies have been reformed since the 2000s. For example, a larger amount of research budget has been allocated to junior scientists to facilitate their autonomy. Multiple senior scientists are often engaged in supervision and degree qualification (Cressey 2012; De Janasz and Sullivan 2004), which can contribute to cognitive exploration.

Limited autonomy and exploration at the junior stage may be culturally rooted in hierarchical social structure common in East Asia. Previous studies do find cultural factors in Japan shape research and teaching practices (Macfarlane and Saitoh 2008). Still in this context, not a few supervisors do offer a high level of autonomy and exploration, which suggests that cultural factors are not the only cause of the problem. In fact, limited early independence is universally observed, especially under increasing competition and emphasis on short-term outcomes (Enders 2005; Gould 2015). An international survey shows that supervising styles in Japan are fairly comparable to those in other developed economies (Cummings 2009).¹⁷

What may be more unique to the Japanese context is the steady succession of training and working styles over generations. Although we believe that inter-generational transfer is a general mechanism, it may be reinforced by a rather closed career system in Japan, exemplified by the practice of inbreeding and strong master-apprentice relationships (Morichika and Shibayama 2015). Supervisors’ training practices are imitated by their students, and further passed down to the following generations. Importantly, this helps maintain a high level of early independence at least among part of the community descended from pro-independence supervisors, while it also means a continuous lack of independence in other parts of the community. Thus, although the closed career system in Japan has been criticized, the overall impact of the system on independence is rather ambiguous. Nevertheless, the system has been reformed to increase openness, for example, by prohibiting inbreeding under some conditions and facilitating open recruitment, the effectiveness of which must be evaluated from various angles.

The theoretical contribution of these findings is twofold. First, this study demonstrates long-lasting impact of academic training. Though the literature suggested the critical role of academic training in socialization (Austin 2002; Campbell 2003; Laudel and Glaser 2008), previous studies on academic training mostly evaluated short-term impact. Addressing the limitation, this study presents the critical impact of different training approaches on junior trainees’ future independence. Second, this study shows that training approaches are socialized through academic training and succeeded across generations. Although previous literature suggested a considerable variation in training approaches between supervisors, the reason behind this variation was unclear (Bastalich 2017; Hockey 1991; Mainhard et al. 2009; Marsh, Rowe, and Martin 2002; Shibayama, Baba, and Walsh 2015). This study suggests that the academic training itself that scientists received in their early careers determines the training approaches they choose to employ in later careers.

The findings of this study need to be interpreted with reservation for a few limitations. First, the results may be subject to the specificity of our empirical context. Postgraduate education system and academic career system differ between countries and scientific disciplines. As discussed, hierarchical institutional system and culture may facilitate socialization in the local lab context. Second, our sampling strategy excludes junior scientists who exited from academic careers before reaching the supervisor status. In other words, our sample is biased towards scientists who attained relatively higher organizational independence. Third, although this study aims to present causal links between training approaches and independence, selection bias is a concern. Although we attempt to mitigate the concern by controlling for pre-PhD performance, there still is a risk that respondents’ unobserved ability biases the estimation.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by Japan Society for the Promotion of Science (JSPS Research) Grant [19K01830].

Notes

1 This is the case in many fields (e.g. natural sciences, engineering, medicine), but the roles of the lab and the senior scientist can be different in other fields (e.g. social sciences, ‘big science’ fields).

2 As of 2018, 4.6% of faculty members and 23% of PhD students are non-Japanese (source: the School Basic Survey).

3 Source: the School Basic Survey.

4 Source: https://www.jsps.go.jp/j-grantsinaid/27_kdata/data/3-3-1/3-3-1_r1.pdf.

5 The degree fields include Science (44%), Medicine (19%), Agriculture (11%), Bioscience (10%), and others (16%).

6 First, this allows us to administer the survey in a single language (Japanese), and the resulting homogeneous sample simplifies the later analyses. Second, non-Japanese students are less likely to say in academic careers in Japan – for example, only 50% of non-Japanese graduates stay in Japan while 95% of Japanese graduates stay in Japan (Nistep 2018).

7 Of the 587, we could not reach 14 scientists.

8 The indicator takes a value of 0 for non-novel publications and positive values for novel publications with higher indicator values corresponding to greater novelty. Since the indicator takes positive values only for 10% of all publications, we transform the indicator into a dichotomous measure coded 0 for non-novel and 1 for novel publications.

9 The correlation matrix of the variables is found in Online Supplement.

10 Spearman's rank correlation is calculated between the three-point scale (major – minor – no responsibility) measures for all functions in the PhD period and in the postdoc period. The analysis finds significantly positive correlations for subject setting (p < 0.001), planning (p < 0.001), and experiment (p < 0.001), but not for hypothesis formulation and writing.

11 The percentages of the respondents choosing either mainly A (or B) or rather A (or B) than B (or A) are summed and reported.

12 The distinction of upstream and downstream functions is supported by a factor analysis (see Online Supplement).

13 The marginal effect is computed with all variables set at their mean values except for the focal independent variable.

14 The result of the factor analysis is found in Online Supplement.

15 In computing the marginal effect, the comparison is made between a case where the focal independent variable takes the value of 5 and a case where the independent variable takes the value of 1.

16 The respondents reported significantly higher autonomy for their own training period (Figure 1(A,B)) than for the training of their students (Figure 1(C)). We believe that this is due to the respondents' cognitive bias.

17 For example, the study found that 60% of the Japanese in PhD training received intensive faculty guidance while 47% of respondents on average across 22 countries were given such guidance (e.g. 23% in the UK and 70% in the US). Similarly, 60% of the Japanese chose their PhD research topics, while 65% is the global average (64% in the UK and 85% in the US).