Researchers’ mobility and its impact on scientific ...
Researchers’ mobility and its impact on scientific performance
Ana Fernández-Zubieta §
Joint Research Centre – Institute for Prospective Technological Studies (IPTS)
Institute for Advanced Social Studies - Spanish Council for Scientific Research
Aldo Geuna*
Department of Economics and Statistics Cognetti De Martiis, University of Turin
BRICK, Collegio Carlo Alberto
Cornelia Lawson #
Department of Economics and Statistics Cognetti De Martiis, University of Turin
BRICK, Collegio Carlo Alberto
Acknowledgments
The authors are grateful to Cristiano Antonelli, Marco Guerzoni, Jacques Mairesse, Fabio Montobbio, Lia Pacelli, Chiara Pronzato and Paula Stephan for comments and suggestions. Thanks are due also to Daniel Lopez Gonzales and Manuel Toselli for their contribution to the creation of the database. Financial support from the European Commission (FP7) Project ’An Observatorium for Science in Society based in Social Models – SISOB’ Contract no.: FP7 266588 and the Collegio Carlo Alberto Project ‘Researcher Mobility and Scientific Performance’ are gratefully acknowledged.
§: Institute for Advanced Social Studies – Spanish Council for Scientific Research, Campo Santo de los Mártires, 7, 14004 Córdoba , Spain, Tel: +34 957760534, Fax: +34 957760153, email: afernandez-zubieta@iesa.csic.es.
*: Corresponding author - Department of Economics and Statistics Cognetti De Martiis, University of Turin, Lungo Dora Siena 100 A - 10153 Turin, Italy, Tel: +39 0116703891, Fax: +39 011 6703895; email: aldo.geuna@unito.it
#: Department of Economics and Statistica Cognetti De Martiis, University of Turin, Lungo Dora Siena 100 A – 10153 Turin, Italy; email: cornelia.meissner@unito.it
Abstract
This article analyses the impact of mobility on researchers’ performance. We develop a theoretical framework based on the job-matching approach and the idea that research productivity is driven by the availability of capital equipment (and human capital) for research, and peer effects. The empirical analysis studies the careers of a sample of 171 UK academic researchers, spanning 1957 to 2005. On the basis of a unique ranking of UK institutions that we were able to construct for the period 1982 to 2005, we develop an econometric analysis of the impact of job changes on post mobility performance over three-year and six-year periods, and the overall effect of mobility. Contrary to the assumptions underpinning most policy action in this area, we find no evidence that mobility per se increases academic performance. Mobility to ‘better’ departments has a positive, but not significant impact, while downward mobility reduces researchers’ productivity. In most cases, mobility is associated with short-term decrease in performance arguably or most likely due to associated adjustment costs.
Keywords: Academic labour market, Research productivity, Researcher mobility
JEL codes: O31, I23, J24
1. Introduction
The mobility of researchers and the establishment of research networks across different countries, fields and sectors have become major policy goals in recent years. For example, in the EU, the commitment to develop a European Research Area (ERA) implies the promotion of ‘greater mobility of researchers’ (EC, 2001: 1; EC, 2010: 11, 17). National reports also point to the need for greater intra-national researcher mobility and flexibility to increase knowledge diffusion to different institutions and sectors (e.g. CST, 2010). These policy papers assume that scientists’ mobility facilitates knowledge and technology transfer and the creation of networks, and increases research performance. In relation to the last, the dominant policy view in Europe (at both EU and national levels) is that the lower scientific performance of European researchers compared to their peers in the USA is due to the low levels of mobility in most national university systems. Policies are being developed to support the mobility of researchers aimed at increasing individual productivity and the overall performance of the system via the creation of positive externalities. However, due also to the difficulty involved in obtaining complete career information for researchers (including performance data), this policy view is not supported by statistical evidence.[1]
Whether and how mobility affects researchers’ performance, that is, the focus of this paper, has not been explored systematically. Several academic papers analyse spill-over and peer effects resulting from academic mobility (Cooper, 2001; Møen, 2005; Pakes and Nitzan 1983; Zucker et al., 1998, 2002), but very little attention has been paid to the consequences for the researchers involved. A few papers in sociology of science (see e.g. Allison and Long, 1990, and much earlier Hargens and Farr, 1973) study this topic and find some weak evidence of a negative impact of immobility and some suggestion that mobility is a characteristic of productive researchers (van Heeringen and Dijkwel, 1987; Allison and Long, 1987). Dietz and Bozeman (2005) worked on intersectoral mobility, finding weak evidence of some positive effect on productivity. However, due to data availability and modelling difficulties, these studies offer only very preliminary insights into the relationship between mobility and productivity and do not provide either a comprehensive theoretical framework or a full econometric modelling strategy.
In order to analyse this relationship, first, we develop a theoretical framework to predict the impact of mobility on research performance based on a job-matching approach to academic labour mobility that emphasizes research and reputation factors. Science is a social system in which opportunities for research and the symbolic and material rewards for its enquiry tend to be accumulated in a few individuals and institutions (Merton, 1968). This process leads to a structured system of production, and access to resources and recognition. As in all structured systems, mobility across different levels of the scientific social structure is more limited, making it possible to use this lower level social mobility to check the quality and impact of transitions. Job changes to a higher quality/reputation institution could lead to better academic performance. The idea of productivity being driven by the availability of capital equipment (and human capital) for research and peer effects lead us to expect medium-term positive effects of mobility on productivity only for job changes that imply a move to a higher quality/reputation institution. In our framework, a job change is associated always to a short term reduction in productivity due to mobility and adjustment costs.
Second, we perform an empirical analysis to address some of the shortcomings in the previous literature by focusing on the entire careers of a sample of mobile and immobile researchers. In a dynamic set up, we estimate a series of econometric specifications of our model to assess the impact of job changes on post mobility output at three and six years after a job move. We expect an initial decrease in performance associated with mobility costs, and a subsequent increase in performance only for those who move to higher reputation/quality institutions.
The empirical analysis is based on a unique database that includes detailed information on the employment patterns and publishing activities of a sample of UK academic researchers in science and engineering from the year of their first professional appointment, for the period 1957 to 2005. Reliable institutional level information on publications and citations needed to build an original time varying research ranking indicator, limited the econometric analysis to the 23 year period 1982-2005. In our sampling strategy we focus only on research active academics occupying ‘tenured type’ positions, that is, we do not include mobility due to non-renewal of contract. Thus, a change of job is the result of the researchers’ decision.
We found no evidence that mobility per se boosts the productivity of researchers. Mobility to lower ranked universities is accompanied by a decrease in both the number and impact of publications, while upward mobility is neither associated with a positive increase in productivity nor with a quality effect. Contrary to the assumption underpinning most policy actions in Europe, it seems that all types of mobility are associated in the short to mid-term (3-6 years post mobility) to an insignificant change in impact, or to a lower impact.
2. What do we know of researchers’ performance and mobility?
Labour market analyses based on job matching and the search theory model (Jovanovic, 1979; Mortensen, 1986) examine job changes in general; Zucker et al. (2002) examine the case of scientists, emphasizing the role of productivity for explaining mobility. However, only a few systematic studies try to assess the other side of the relationship - whether mobility has a positive or negative impact on short term scientific performance (Allison and Long, 1990). There is no systematic evidence of a causal effect between mobility and medium to long term researcher productivity.
This paper tries to fill this gap. Starting from the traditional model of the analysis of scientific productivity (Cole, 1979; Levin and Stephan, 1991), we study scientific performance (sp) as a function of individual characteristics, environmental specificities and mobility events:
[pic] (1)
where M is mobility events, p is individual personal and academic characteristic and h is institution, field, country and time-specific environmental characteristics effecting scientific productivity.
Mobility might assert a positive impact on research performance only if the researcher finds better conditions for pursuing her research endeavour; for example, if she moves to a new job in order to increase her research performance. However, there are other reasons for mobility that are unrelated to research performance including salary, family demands, etc. To fully understand the impact of mobility on research productivity we need first to understand what drives researchers’ mobility and then to model the impact of mobility on performance controlling for those factors that might have a confounding effect. Below, we briefly review the main tenets in the literature on the drivers of mobility and discuss the distinctive characteristics of the academic labour market (Section 2.1); secondly, we propose a framework to model the relationship between mobility and performance (Section 2.2).
2.1. The academic labour market: Distinctive characteristics
Depending on the particular institutional setup, such as the public servant role of academics in some European countries, not discussed in this paper, the academic labour market is driven by traditional labour market factors such as wage and search costs, contextualized to the academic market, and a set of academia-specific factors related to research and reputation. Among labour market factors, the most important are: (1) wage related – the difference between current compensation and the new wage offer (particularly relevant for a move to a business job, usually associated with a much higher salary); (2) career related –promotion to associate or full professor usually associated with access to more resources for research and the possibility of hiring and directing doctoral and post doctoral fellows, in addition to a higher salary;[2] (3) employment opportunity related – non-permanent academic jobs are becoming more common in all countries and are associated with termination and non-renewal resulting in involuntary mobility; (4) market related –the fluidity of the job market differs across countries and disciplinary fields and the density of the market varies depending on the time period;[3] (5) mobility cost related – the relevance of the costs associated with mobility is not fixed and depends on previous mobility experience;[4] (6) family related reasons – partners moving, ageing parents and children’s education are all common reasons for involuntary mobility, and may reduce the propensity to move which introduces a gender and age bias.
Academic distinctive factors
The academic labour market is also characterized by some distinctly academic factors, which are the focus of this paper. Setting aside redundancy, generally the wage received is the single most important determinant of the choice to accept/leave a business job. However, this does not always apply to the academic labour market where research and “reputational” factors are also crucial (Levin and Stephan 1991). For academics, research (time and support) is the most important aspect of their job and provides the greatest job satisfaction (positive utility) while also being a work activity that produces outputs. The time spent doing research is perceived by academics partially as consumption time, resulting in their willingness to forego the higher wages available in business jobs which do not include independent research. Hence, all else being equal, academics are willing to earn less in order to be able to work on their chosen research (Stern, 2004; Sauermann and Roach, 2013). Another important argument in the utility function of a researcher is reputation, which is affected in part by institutional reputation (to simplify we do not distinguish between department and university). A researcher values employment in a highly prestigious institution because of its direct benefits, such as fewer teaching obligations, more research time, higher financial endowments, etc. but also because of the positive externalities attached to these positions which can reflect on her individual reputation. These aspects are important in the market for scientists where individual quality assessment is not straightforward, especially in the early stages of a research career, and publications are not perfect carriers of information. All else being equal, an academic will move to a better-ranked institution (expecting the benefits to outweigh the mobility costs), since research and reputation enter positively in her utility function. She can expect to increase performance in a higher ranked institution because there will be more capital available for research, crucial in the natural and biomedical sciences where laboratory costs are extremely high in terms of both equipment and human capital (Stephan, 2012). The researcher will benefit also from direct peer effects related to her new colleagues and indirect effects through access to their social networks. Moreover, institutional reputation may provide a higher probability of receiving future funding for research; in the context of funding agency selection, there are more excellent proposals than available budget, and institutional reputation can matter for the final selection.
In addition, especially in new and fast changing disciplines, mobility is driven by the prospect of accessing tacit knowledge and new equipment. In an early phase of development of a new discipline, knowledge is located in a small number of laboratories responsible for the original discoveries. Publications allow this knowledge to percolate through the university system, but due especially to the invention of new equipment (see e.g. the case of the production of the onco-mouse, Murray, 2011), some knowledge is ‘sticky’to a particular laboratory and can only be passed on via training and use of equipment. Researchers are willing to bear the costs of a move to these centres in order to acquire the tacit knowledge held there. Acquisition of tacit knowledge can be achieved through short stays (such as sabbatical leave) or job changes.
Finally, academic mobility is strongly affected by relative opportunity advantage. In a market with clear reputation/quality ranking, researchers working in high-ranked institutions have much lower probabilities of moving, everything else being equal.
2.2 The relationship between mobility and researcher’s scientific performance
The relationship between mobility and researcher’s scientific performance is bidirectional. To model it we need to understand the reasons of academic mobility to predict the impact of mobility on research performance. The probability of a job change (academic mobility) depends on the probability of receiving a job offer f(.) and the probability of accepting that job offer g(.). Let us define:
[pic] (2)
[pic] (3)
In the typical search theory model, the probability of receiving an offer f(.) is likely to depend on factors such as search effort (s), and environmental (e) and individual (p) labour characteristics. The probability of accepting an offer g(.) is likely to depend on the level of the wage offer (w) relative to the individual’s current compensation (b), and other mobility costs (c). We modify the basic model to include the academic labour market distinctive factor (r) that takes into account the research and reputation related effects discussed in the previous section.
The probability of receiving a job offer f(.) depends decreasingly on search effort (s). The academic profession being an intrinsically networked job, the more connected the researcher is to a densely populated network of public and private organizations the lower will be her search costs since she will be well informed about available positions. The extent of the individual’s social network, therefore, increases her probability of receiving an offer f(.). The probability of receiving a job offer f(.) also depends on environmental academic labour market characteristics (e) such as the existence of a strong potential demand. Potential demand in terms of flexibility and density of the academic market is scientific field, country and time dependent. The researcher’s personal characteristics (p) (such as PhD awarding institution, tenure, scientific productivity), which could be interpreted as signalling high individual productivity, positively affect the probability of receiving a job offer f(.).
In traditional job change models, the probability of accepting an offer g(.) depends on the salary offered (w) and the retention strategy of the sending company that could offer an increase in the salary (b); these factors can be affected by personal characteristics (p). In academia, the higher the academic’s position and longer the academic experience in that position, the higher will be the salary in the current job. However, academic salaries tend to vary within a well-defined national range, based on experience, with some limited flexibility at the top depending on the country considered. In the US, and less so in the UK (with the exception of business schools), professorial salaries can vary significantly. However, in most other countries, public employee contracts or tradition give little room for individual salary increases. In the academic labour market, this leads to a reduced effect of salary on the probability of moving. In Europe, the wage offer (w) relative to the individual’s current compensation (b) plays a very small role in explaining mobility. Thus, we can rewrite equation 3 as follows:
[pic] (4)
where the probability of accepting a new academic position depends on personal characteristics (p), mobility costs (c) and the research and reputation effect (r).
Among personal characteristics (p), a key determinant of the probability of accepting a job offer is the academic position of the researcher (pt). Non-tenured researchers are more likely to accept an offer than tenured university staff since they have a non-zero probability of non-renewal of contract (all non-tenured positions are based on ‘soft’ money that is time limited).
Scientific performance (sp) is one of the personal characteristics that directly affects the probability of receiving f(.) and indirectly affects acceptance of a job offer g(.). Researchers with a good publications track record will have better career and retention package prospects affecting g(.). However, academic researchers who are more productive have higher chances of receiving a job offer from another university since research performance usually is considered the most important criterion for selection (a conditio sine qua non). Scientific productivity can be seen as signalling a high quality researcher, increasing the probability of receiving an offer f(.) and decreasing the probability of accepting an offer g(.).
Finally, individual personal characteristics (pf), such as age and sex, can affect the probability of accepting an offer due to family related consideration that can both increase or decrease mobility costs. The probability of accepting an offer g(.) depends negatively on mobility costs (c). Mobility costs include the direct personal costs of moving to another city or country, and the skills adjustment costs - particularly important in high skilled jobs. If the researcher’s skills are university specific (i.e. not all the routines of the academic teaching and research work are transferable to the work in the new university and especially if the move is to/from a firm), she must learn new practices, protocols and routines and adjust to different management and administration procedures. Thus, a period of adjustment with lower expected efficiency may transpire. Even if these skill adjustments are minor, they can be considered sunk costs and may deter some researchers from moving.[5] This applies especially to mature academic researchers who have invested a lot of time in accumulating the skills and reputation needed to succeed in a specific university environment. Both the direct and skills adjustment mobility costs are decreasing in the number of times a researcher has moved, due to learning effects. Individual personal characteristics (pf) affect the assessment of mobility and adjustment costs.
At the same time, according to the discussion in the previous section, the probability of accepting an offer g(.) depends also on the researcher’s expectation of the higher research performance (r) achievable in the new job at a higher ranked institution. We can therefore rewrite the equations as follows:
[pic] (5)
[pic] (6)
[pic] (7)
We now turn to the impact of mobility on scientific performance (sp). A change of job can have an impact on the scientist’s research performance after the move. The short to medium term (say a window of 3-6 years after the change) post-mobility productivity of the researcher is affected by her reasons to move. For example, a researcher moves to a new job if the value Vt+1 of her utility function is higher than the value Vt before the move at time t. This may be due to the traditional job search related factors discussed above and/or because of an expected better research and reputation environment (r). Only if the job change is driven by research and reputation related motives can we expect a positive impact on the researcher’s performance. Hence, not all types of mobility are associated with increased research productivity.
In the basic job search model, the difference Vt+1 - Vt should be higher than the mobility costs (c) for a job change to happen. Mobility costs are assumed to be instantaneous. However, mobility can be associated with significant deferred adjustment costs that can have a negative impact on post-mobility productivity because the researcher will have less time to spend on research activities due to the need to devote time to learning tasks which would have been accomplished more efficiently in the previous job because of the scientist’s familiarity with practices, protocols and routines (van Heeringen and Dijkwel, 1987, Shaw, 1987; Groysberg, 2008). Following a job change in laboratory-based work, the researcher can show decreased performance associated with the setting up of a new laboratory. The extent and depth of the reduced performance depends on the relevance of the adjustment costs, which, in turn, depend on the learning required to adjust to the new job. Therefore, job changes may be associated with no change in short to medium term scientific performance if the reasons for moving are related exclusively to traditional job search factors, or to a positive increase if the mobility is driven by research and reputation reasons (r). In both cases we can expect a short run decrease in performance due to adjustment costs.
H1: Academic job mobility is associated with a short term decrease in research performance due to adjustment costs
Mobility is expected to be associated with an increase in productivity due to its effects on matching and networking. In terms of matching, the model predicts that researchers with high potential productivity unexploited in a lower quality department, will move to a higher quality department, where they can find better endowed laboratories (better equipment, more junior research staff) and will, therefore, increase their performance. In terms of networking, interpreted as better human (more diverse learning opportunities) and social (better network connections) capital, the model predicts that a move to a better department means a move to a better research group with positive peer and network effects that increase the researcher’s performance. Research group composition and local peer effects have been identified as important predictors of individual performance (Weinberg, 2007), and researchers are more productive if they collocate with productive scientists. However, Kim et al. (2009) find that peer-effects have diminished since the 1990s, perhaps due to better communication technology (see also Ding et al., 2010). Working in a department with high quality peers enhances performance not only through direct interactions but also through privileged access to their social networks. Moreover, mobile researchers benefit from their existing networks, which they bring to the new environment (Azoulay et al., 2010; Waldinger, 2012) thereby creating new extended networks with the potential for new combinations. It is very difficult to disentangle the matching effect from the social/human capital model since, in a departments with high reputation, both are present (funding for good labs, and high ranked peers which enable access to better quality social networks and more learning). Moreover, well-reputed researchers tend to concentrate in high ranked departments (Oyer, 2007) because they are the source of the ranking and, due to competitive allocation of resources, they are also the departments that receive the most funding.
Within this framework, we hypothesize that only a move to an institution of higher quality/reputation will be associated with a medium term increase in performance; after an initial period in which adjustment costs may constrain researchers’ productivity we should expect increased research performance. On the basis that scientific production is strongly affected by cumulativeness and self-reinforcement phenomena (Dasgupta and David, 1994), we would expect that improved medium term productivity will be persistent and, thus, will affect the long term performance of researchers.
H2: Academic job mobility to a higher ranked institution is associated with an increase in research performance.
Conversely, mobility to an institution of the same or lower quality/reputation level should be associated with short term lower productivity due to adjustment costs. These can be only slightly mitigated and at best stabilized at pre-mobility levels (for same rank changes) or lower performance levels, in the medium to long term, due to research resource constraints (such as financial and human support resources) and reputation, assuming the move involves a similar work profile (e.g. similar teaching and administration load).[6]
H3: Academic job mobility to a lower ranked institution is associated with a decrease in scientific performance.
We recognize the need for a dynamic perspective of researchers’ mobility. This implies seeing researchers’ mobility not as a one-step process, but one that takes account of short and long term return opportunities. Successive changes in job positions, associated or not with career advancement, in/from higher rank institutions, should be considered in order to assess possible opportunities for returns. However, the data required to estimate such a model are not currently available. This paper is a first step in that direction. We therefore estimate the following function:
[pic] (8)
where M is the mobility events, pt is individual academic characteristics such as career rank, pf is individual personal characteristics such as gender, and h is institution, field, country and time specific environmental characteristics affecting scientific productivity (e.g. there is a greater tendency to publish and cite more in medicine than in economics). In the next section, we present the data and the estimation model chosen in light of the data available.
3. Empirical Analysis
3.1 The Sample
The empirical study is based on a sample of 171 research active academics working at 53 different UK universities in 2005, in four scientific fields: chemistry, physics, computer science, and mechanical, aeronautical and manufacturing engineering.[7] For the purposes of this study, career information taken from CVs was coded in order to construct comprehensive profiles of the researchers, spanning their career from PhD award to 2005, which resulted in a panel for the period 1957 to 2005. Our econometric analysis is limited to job changes that occurred between 1982 and 2005 to allow an adequate number of researchers in each observation period, and because we need to create an original institutional ranking variable based on publication and citation data which are reliable only after the year 1982. There is also some additional information that could not be collected for all the researchers in the sample, leaving us with a set of 150 researchers and 2,367 observations for the econometric analysis.
CV data are very useful for analysis of academic careers since they provide information on job changes and also a reliable publications record (Cañibano and Bozeman, 2009). Using data collected from CVs combined with the ISI Web of Science (WoS) improved the accuracy of our data since it avoids mismatches arising from similar names and changes in researchers’ institutional affiliations. Researchers’ CVs include information on career paths and the timing and nature of job transitions.
In our analysis we focus on inter-institutional ‘real’ labour mobility (Crespi et al., 2007), which implies a change in job position from one institution to another. Changes in job position within the same institution are not considered (e.g. a move to a different department in the same university). We also only consider changes that occur after the first ‘tenure-track’ position in academia, or first full time position in industry, after the award of the PhD degree.[8] Thus, our analysis is limited to the influence of job mobility on productivity for researchers in tenured or tenure-track equivalent positions. Accordingly, postdoctoral research positions are not considered real labour mobility in our analysis.[9] In the UK, the minimum tenure-track positions in academia are lecturer or research fellow, followed by ‘senior lecturer’, ‘reader’ and ‘professor’. Research fellow positions are considered only if they are of 5 or more years indicating a long term relationship with the university. Academics in the UK are usually hired on permanent contracts, which, in the case of lecturer appointments or research fellowships, are subject to a three-year probation period. Thus, mobility in our sample is likely to be voluntary, that is, where researchers leave a permanent position for reasons other than termination of contract.
The sample consists of researchers aged 29 to 77, who were active in 2005. The mean age of the sample is 49 in 2005 (Figure 1). The first researcher joins our sample in 1957 and the last in 2003 (Figure 2). Accordingly, the career years recorded in our sample range from 3 to 49, with an average observation period of 20 years. In our sample of 171 UK academics, 145 (85%) started their career as lecturer or research fellow; 22 researchers (13%) took up their first position in industry, and 2 researchers started in senior academic positions (Figure 3). For two researchers first positions were not evident from their CVs. The mean starting age is 28.6 with a minimum of 22 years and a maximum of 38 years (Figure 4). [10] The mean PhD age is slightly lower at 27.2 years. Among the researchers, 45.2% took up their first position immediately after PhD award and 48.8% embarked on postdoctoral research; 6% of the researchers in our sample started their work careers during or before studying for their PhD.
CV information allowed us also to assign WoS publications to each researcher. Academics in our sample produced an average of 4.45 publications per year between 1982 and 2005. A total of 88 researchers (59%) published their first article before taking up their first tenured employment, either during their PhD study or postdoctoral appointment. The average number of publications per researcher per year increases from an average of 4.08 in 1982 to 5.05 in 2005 (Figure 5). There is a similar increase for publication quality. Quality is measured as the number of WoS citations received by publications in the first five years after publication. The number of quality adjusted publications increased from 46 in 1982 to 74 in 2005. This increase could be due in part to life-cycle, year or mobility effects which this paper attempts to measure.
3.2 Mobility and reputation
The academic market in the UK differs from that in the rest of Europe. It is characterized by its internationality - it attracts academics from across the world, and by the competition amongst its universities for the most promising scholars (BIS, 2011; Ziman, 1991). Further, the three-step promotion system and race for positions at the most prestigious institutions (Hoare, 1994) make the UK system more competitive than other academic systems in Europe. There is no obligation to move after PhD completion; however, mobility barriers are very low and mobility is usually rewarded, making the UK academic labour market very fluid.
In our base sample of 171 researchers, 109 (64%) changed jobs at least once during their career. In total, we have 159 job changes, with 31 academics changing positions twice during their career, 8 academics changing three times and one person moving four times. The mean number of years in one job is 10 years. Our econometric analysis is limited to job changes that occurred between 1982 and 2005 because some explanatory variables are not available for earlier years. Also, 21 researchers had to be dropped from the analysis because of missing personal characteristics. This left a sample of 150 researchers. In 1982, the number of observed academics is 51, increasing to 150 in 2005. We recorded 110 job changes for 84 researchers between 1982 and 2005. Of these, 59 involve mobility between UK higher education institutions, and 25 moves from industry to academia. The remainder involve international moves or mobility to or between firms.[11]
In the theoretical part of this paper, we stressed the importance of research and reputational factors for explaining the academic labour market. Access to resources and a better research environment are incentives to move and are fundamental when analysing the impact of mobility on scientific productivity. Wages play a less important role in the UK academic labour markets and are less central to explaining job changes, especially considering the high level of standardization in academic salary scales. We assume that mobility is driven by reputation factors and, therefore, identify job changes to either higher or lower quality/reputation institutions.
To measure universities’ prestige we built an original indicator of the research ranking of a university in a discipline, based on its publication productivity and quality. We used WoS publication data compiled by Thomson Evidence for UK Higher Education Institutions (HEI) in two main subject categories - natural sciences and engineering sciences - for the years 1982 to 2005.[12] Our data include information on researchers in chemistry, physics, computer science and mechanical engineering. The first two belong to the natural sciences and second two the engineering discipline. We had access to two indicators:, 1) raw number of publications (for each HEI, each year and each of the two categories); and 2) relative impact of a university within a discipline, measured as the ratio of its mean citation rate to the world average. We use the product of these measures which considers both department quality and research size within a specific subject field. We calculated our research ranking indicator as the share of the HEI’s quality adjusted output during a three year period over total output for UK HEIs in the same period. Thus, we measure the contribution of the particular HEI to the production of the UK sector. We consider a three-year period to adjust for possible annual fluctuations, bursts or sudden decreases. We then calculate percentile rankings based on the HEIs underlying distribution, reflecting their relative size differences. Thus, we normalize them linearly, dividing each indicator value by the maximum value in the year and field. Given the skewed distribution of the indicator, percentile ranking is preferred to an ordinal scale that takes no account of ranking differences.
Our measure of research reputation for a 23 year panel can be constructed only for UK universities; international institutions and firms are not included in the second part of the econometric analysis. Researchers in this reduced sample worked at 52 different UK institutions between 1982 and 2005, and 43 researchers moved between UK universities 53 times. Of the 52 UK universities in the sample, 47 are in the top 50% and 17 are in the top 10% in the fields of engineering and science. Upward mobility is defined as a move to a department ranked at least 5 percentile points above the previous department, in the year preceding the move (before the focal academic joined the new department), and downward mobility is defined as a move to a department ranking at least 5% lower than the previous department.[13] In our sample, between 1982 and 2005, 18 academics moved 19 times to more prestigious institutions and 19 researchers moved to less prestigious institutions.[14]
Figure 6 shows the mean number of publications for the 12 years surrounding a move. We plot the graph for the 53 moves between UK universities, and plot separate graphs for the 19 cases of upward and 19 cases of downward mobility. We assume a one year lag between the research and its publication. Thus, articles published in the year of the move (year zero) would refer to research undertaken at the previous institutions. The disruption caused by the mobility event will cause the publication pipeline to dry up and result in a decrease in publication numbers in year one. Figure 6 confirms the one year lag between move and publication output. Job mobility generally is followed by a decrease in publications in the year following the move. This may reflect the costs of mobility and adjustment, which likely result in a decrease in research efficiency in year t. However, the number of publications increases from year two onwards. For downward mobility, the rate of publication does not improve, but returns to pre-mobility levels. A downward move, on average, means that the mobile researcher performs worse than the immobile researcher. An upward move, demonstrates a higher number of publications for the mobile researchers than in the case of other types of mobility, even in the year before the move. The mean number of publications increases further, from year two after the move. Hence, academics moving to higher quality institutes are already performing above the average before the move, while academics moving to less prestigious universities are those with below average performance who are unable to benefit from a job move. The difference between the two increases following the move. The graph in Figure 6 is consistent with results in Allison and Long (1990) on positive department effects on productivity, but in contrast to their results, in our case the upward mobile group starts out with higher productivity than the downward mobile group.
3.3 Econometric Specification
We estimate count data models as the number of publications and citations are by nature positive values. The data is characterised by overdispersion and we thus employ pooled negative binomial models of the form:
[pic] (9)
where sp’it is the count variable representing scientific performance (sp) as either the publication count (Pubit) or the number of citations per publication per year (Citit) of researcher i in year t. Mit is the mobility event, Xit represents a set of explanatory variables including personal and academic characteristics (pt, pf) and institutional effects (h). ci is the individual time-invariant unobserved effect, including ability and attitude, τt is the time fixed effect and υit other time-variant unobserved effects.
To analyse the difference in research performance between mobile and immobile researchers we look at the effect of the mobility indicator Mobileit. This is done to see if mobile researchers have a performance premium over immobile researchers. Secondly, we estimate the model again but drop post-mobility observations of mobile academics. In doing so, we can estimate the potential performance difference before mobility occurs and thus investigate if mobile researchers have a performance premium even before being mobile. This estimator corresponds to a pre-mobility indicator and shows whether researchers were more productive before the move.
To measure the performance difference between the pre- and the post-mobility period we reduce the sample to mobile academics. First, we assumes a lasting career effect of mobility on publication outcomes, thus recording mobility as a one-time shift by defining PostMobit=1 for all the years following the first move (or the first upward / downward move). As the effect of mobility may vary and different short- and long-term effects could be envisaged, we introduce an indicator variable Mobit, which takes the value one in the year of the move, and include its lags in the regression. We consider lags of three and six years after job transition to investigate the evolution of post-mobility research performance. Different regressions for different time windows are chosen as deeper lags always come with a loss in observation numbers.
We estimate pooled models which have the advantage that they relax the strict exogeneity assumption of a fixed effects model. However, they do not control for unobserved individual heterogeneity (ci). In our case such unobserved effects could be specific skills of each researcher that are positively correlated with the right hand side variables such as mobility and a potential endogeneity problem arises. For example, the literature suggests that more able researchers have many more opportunities to change their jobs as universities screen researchers for their ability and hire the most productive. If unobserved individual heterogeneity (ci) is present, the estimated coefficient of the mobility variables would be upwards biased. We can cope with this challenge if pre-sample information of the dependent variable is available. Specifically, Blundell et al. (1995) suggest a solution which controls for individual heterogeneity (ci) by specifying the average productivity of the academic before she enters the sample. The pre-sample mean of the dependent variable is a consistent estimator of the unobserved individual effect (Blundell et al., 1995) if it mainly corresponds to the intrinsic ability of an academic and her motivation, both factors that are not directly observable but may affect scientific productivity. Following Blundell et al. (1995) we can therefore account for unobserved individual heterogeneity (ci) by using pre-sample information of publications and citations. Blundell et al. (2002) show in Monte Carlo simulations that the estimator is consistent in the presence of unobserved heterogeneity and pre-determined regressors, as is the case in our estimation. They also show that the efficiency of the estimator increases with longer pre-sample observation periods. We measure the average number of publications (or citations) published since the start of the PhD and before the academic enters the sample (before she arrived in her first position or before 1982), resulting in pre-sample observation periods of at least three and up to 21 years with a mean of 4.6 years (median of 4 years).
Theory further suggests that research activity is subject to dynamic feedback (Dasgupta and David, 1994), i.e. heterogeneous dynamic effects, as each researcher’s performance is driven by cumulative unobserved factors (υit), like learning, family and health that are not controlled for through fixed effects. Blundell et al. (1995) therefore argue that it is important to consider continuous, sample-period dynamics when modelling research outcomes. This knowledge stock changes over time and while it increases with experience as a by-product of research, it decreases at a rate of δ as the quality of this knowledge decreases over time. Thus, to proxy for dynamic feedback within the sample period we calculate the depreciated stock of publications (or citations) published during the observation period. We assume that knowledge depreciates at a constant rate of 10%[15] and the sample period feedback measure is hence defined as:
[pic] (10)
The pre-sample value and the stock variable are included in all estimations. This dual approach helps to address the problem of endogeneity that arises from correlated individual effects and through feedback from the dependent variable.
However, the problem of reverse causality of our mobility variables could persist especially of the PostMobit and Mobit dummy variables as predicted research performance could be related to both, the decision to be mobile and to past levels of productivity. In Appendix A we address endogeneity through an instrumental variable approach. The test for exogeneity is not rejected and our model, thus, does not seem to suffer from endogeneity bias, suggesting that the ‘feedback model’ is able to address the main endogeneity concerns.
3.4 Variables
Our primary objective is to measure the effect of job mobility on research productivity in terms of publication and citation numbers controlling for career patterns and past mobility. The main dependent variable in our specifications is the number of publications in year t (PUBit). Additionally, we introduce a proxy for the quality of the research output: the total number of citations received by the researcher’s publications in the five years following the publication (CIT5YRit). As some publications may receive citations at a later date we further look at the total number of citations received before the April 2013, the date of data download, (CIT2013it) that may act as a proxy of the long term quality of publications in robustness checks. Looking at the number of all citations received to date results in different citation windows for different years with a minimum of eight years. We further consider the average number of citations received in the five years following the publication (AVGCIT5YRit) which represents the average quality of a researcher’s publications in a given year.
The main explanatory variables in the regression refer to the mobility event. Firstly, we analyse if mobile academics perform better than immobile academics before they become mobile. We thus define a dummy variable Mobilei that takes the value one for researchers that were mobile between 1982 and 2005. In other words, this variable measures the performance difference between mobile and immobile academics. Secondly, to measure the potential performance difference between pre- and post-mobility periods, we introduce two dummies that measure the mobility event, (1) PostMobit,, that switches from zero to one in the year of mobility, thus clearly indicating the pre- and post-mobility periods; and (2) Mobit that takes the value one only in the year of the move, indicating a one-time shock. As our main focus is on mobility between universities, we run additional models focussing on moves between UK higher education institutions while excluding all researchers that have had other types of mobility experiences (UNIMobilei, PostUNIMobit, UNIMobit). We argued above that mobility is affected by reputation of the sending and receiving institution and we therefore use additional measures for mobility that consider the nature of transition: (1) Upward Mobility (UPMobilei, PostUPit, UPit) defining a move to a university of higher esteem, and (2) Downward mobility (DOWMMobilei, PostDOWNit, DOWNit) defining a move to a university of lower esteem.
As controls we include the academic’s age (AGEit) to account for potential life-cycle effects (Levin and Stephan, 1991) and the gender of the researcher (FEMALEi). Further, we control for the academic rank of a researcher. The UK university system requires researchers to fulfil minimum requirements to be considered for promotion. Thus, academics in lower ranks should have more incentives to publish. Professors on the other hand have access to research assistance and funding that may allow higher rates of publications. We hence consider three “ranks” in our analysis: Lecturer or Research Fellow before first promotion (RANK1it-1), senior position or rank after first promotion (RANK2it-1), and professorship (RANK3it-1). In the first set of regressions that also include mobility between firms and universities we introduce a dummy of firm employment in t-1 (RANK0it-1), which represents the omitted category. We also consider an indicator for postdoctoral research experience (POSTDOCi). To account for commercial orientation of the researcher (Crespi et al., 2010) we further include a dummy (PATENTit-1) that measures if a patent was filed in the previous year. We include subject dummies to control for discipline effects. To account for any potential department effects in terms of access to resources and networks we include the university’s rank in t-1 (UniRankingit-1) in the set of regressions that consider only UK institutions. We further may expect a London effect due to proximity to funding bodies and networks that may positively affect research output and therefore include a London dummy (Londonit-1). We also include year fixed effects and the individual effects proxy in all regressions.
A summary of all variables used in the regressions and their descriptive statistics can be found in Table 1.
4. Results
We estimate pooled negative binomial regressions. Standard errors are clustered at the individual level and robust to heteroschedasticity and serial correlation. Tables 2, 3, 4 and 5 show the results for the mobility dummy and for the mobility shift variable, and three-year and six-year lags respectively of these mobility variables.
4.1 Mobility effect - Full sample regression
Table 2 shows the results for all mobility regardless of its type and including international and business mobility. The number of observations in column 1 is 2,367 which reduces to 1,125 in column 5 due to deeper lags that require a minimum of seven observation years and consider only academics whose careers began before 2000. The mobility dummy in column 1 is positive but insignificant indicating that mobile academics do not perform better relative to the group of immobile researchers. In column 2 we exclude post-mobility observations of mobile academics and find that mobile academics outperform their immobile peers in the pre-mobility period but the effect is only 85% significant. Column 3 shows publication performance changes for mobile academics after the mobility event. The mobility variable is insignificant and negative, indicating that mobile academics do not perform better after mobility. Also, in column 4, which looks at the yearly effects of the mobility shock the results are significant only in year 1 after the move, indicating an initial drop in performance. However, column 5 shows there is a tendency from an initial negative effect towards a positive coefficient in later years, albeit insignificant. These results are robust when we consider the full sample not just the sub-sample of mobile academics.
In relation to the quality adjusted variable (columns 6-10), mobility effects are insignificant, indicating that there is no significant difference between mobile and immobile academics or between pre- and post-mobility performance of mobile academics in relation to quality adjusted publication numbers. The coefficients are similar to the ones found for publication numbers.
Thus, overall our first hypothesis of an initial negative effect on research performance is weakly supported. We find no support for the policy assumption that mobility has a positive impact on scientific performance.
4.2 Mobility effect - UK university sample regressions
To introduce our ranking measure which takes account of the quality of the university department we consider only mobility between UK universities (Table 3). We exclude all researchers that moved internationally or to industry since their inclusion could introduce potential bias (i.e. they might appear to be immobile).[16] The number of observations reduces to 1,662 in column 1 and to 474 in column 5.
Model 1 shows that the mobility dummy is insignificant. Thus, mobile academics do not perform better relative to the group of immobile academics if we consider moves only between UK institutions. Model 2 excludes post-mobility observations of mobile academics. Again, we find no difference in performance between mobile and immobile researchers for the pre-mobility period, although there is a positive sign. Column 3 reports how publication performance changes for mobile academics after the mobility event. The mobility variable is significant and positive, indicating that mobile academics perform better after mobility. In column 4, which looks at the effects of the mobility shock, the signs are negative, but insignificant. In column 5, again we observe a tendency to change from an initial negative effect to a positive coefficient in later years. These results are robust when we consider the full sample and not just the sub-sample of mobile academics. For the quality weighted publication count (column 6-10), the effects are insignificant, but the signs are equivalent to those on publication counts. Overall these results show that mobile academics do not outperform immobile academics and give weak support to our hypothesis of an initial negative effect following mobility that turns positive in later years.
4.3 Control variables
The coefficients for non-mobility variables are consistent across the different mobility measures and lags. We report their results in Table 3, which includes our university ranking measure. Several individual factors are associated with productivity. Age is positively correlated with number of publications, but this effect is insignificant for the mobile group. We also included a quadratic term in our regression, which suggests a decrease in productivity over the life-cycle; however, again it is only statistically significant in columns 2, 6 and 7. We find a significant gender effect in the publication count regressions in column 2. The sign is positive indicating that women produce more research which is in line with Crespi et al. (2011) which uses the same sample of researchers. The indicator is insignificant and negative in estimations that only consider the mobile group. The discrepancy may be explained by the few women researchers in the sample.
We find no evidence of an increase in publication numbers along academic ranks. Senior academic staff are not expected to publish more than researchers in the category RANK 1.Only in column 2 do we find a positive effect for professors. In Table 2, which also considers researchers working outside academia we find a positive effect of university employment compared to the base category RANK 0. We find further that a postdoctoral research stay does not improve future publication numbers or citation counts. Instead, we observe a negative effect, which may be due in part to job insecurity and fragmented career path associated with postdoctoral appointments (Stephan, 2012). The patent dummy is positive and significant in all estimations confirming prior research that found a positive link between patent and publication output.
Looking at academic disciplines, we see that researchers in chemistry publish significantly more, and are more frequently cited, than colleagues in other fields, with computer sciences researchers producing the least publications and receiving the smallest number of citations.[17]
The university ranking has no significant effect on publication numbers. However, we find a strong positive sign for the quality adjusted measure. Thus, researchers at the most prestigious institutions may not produce more, but may produce publications that are of better quality and achieve more visibility than those produced by their peers. In the models that consider only mobile academics, we see a positive effect of university ranking that is significant only in column 5.
The location dummy for London is negative and only significant in column 2 indicating that researchers at a university in the capital do not have access to better networks resulting in more and better publications but instead may produce less. The coefficient is positive in the regressions that only consider the mobile group but remains insignificant.
The pre-sample means and the feedback measure are significant and positive in regressions using the full sample. This underpins the importance of controlling for unobserved heterogeneity and dynamic feedback. In estimations that do not control for dynamic feedback the coefficients of our mobility measures increase and turn positive in estimations in column 1 and 2 of Table 2, suggesting that we the feedback measure is able to capture the reverse causality. The pre-sample means turn insignificant in some of our estimations that are limited to mobile academics probably due to the small number of researchers left in these estimations.
4.4 Mobility and quality/reputation of the department
In Tables 4 and 5, the mobility effect is conditioned by the nature of the job transition. Table 4 reports the results for upward mobility (UP). They show that upward mobile researchers publish significantly more than their peers even when we exclude post-mobility observations (significant at 13%). It might be that this effect is driven by the expected mobility event. However, if we exclude the three years before the upward move we still find a strong positive effect. Curiously though, the effect on the quality adjusted measure is insignificant although still positive. Thus, while upward mobile researchers are more productive than their immobile peers, they do not produce research of greater impact.
Column 3 compares pre- and post-mobility publication numbers of upward mobile researchers. The effect is insignificant and close to zero. A detailed look at the short-term effects shows that scientific output decreases initially and turns positive in later years, but the effects are insignificant. The estimations for citation numbers confirm the short-term negative effect of upward-mobility, but the coefficients are again insignificant. The university ranking control variable is positive in models 6 and 7 which consider the quality adjusted publication outputs of all researchers. This indicates that while not all researchers that are mobile produce better quality research (as indicated by the insignificant coefficient UPMobile), researchers who move to more prestigious departments produce more visible research. Mobile academics that move to higher ranked departments may not produce better quality research than their peers in the new department (the mobility effect is insignificant), but potentially outperform their peers in their old department (belonging to a higher ranked department is associated with more citations).
Table 5 reports the results for downward mobility (DOWN). They show that downward mobile researchers do not perform worse than their immobile peers or colleagues who move to higher ranked institutions (the effects are insignificant). Column 3 shows that post-mobility productivity tends to be lower for downward mobile researchers, though again the result is insignificant. However, when we look at short-term productivity, we find a significant negative effect for most years, which does not diminish over the longer term (columns 4 and 5). Also, after six years, we find that the expected number of publications for a researcher that moved to a less prestigious university is lower than the expected number of publications before the move. Thus, downward mobility is generally associated with a reduction in productivity - possibly due to a decrease in resources. However, for the majority (all but 4) of researchers moving downward, the job change involves a promotion and, thus, potentially more resources. The negative effect, therefore, indicates that lower ranked institutions do not offer better packages that compensate for loss of institutional prestige and departmental peers. These negative signs are confirmed for the quality adjusted publications measure, however, all the coefficients remain insignificant.
For the department quality measure we find an additional negative effect for mobile academics (columns 3 to 5, and 8 to 10). This indicates that downward mobile researchers in higher quality departments perform worse than downward mobile researchers in lower quality departments. However, the effect in columns 6 and 7 is positive, suggesting that while downward mobile researchers may not perform worse than their immobile peers (the mobility variable is insignificant), researchers in lower ranked institutions produce lower quality publications.
Thus, the positive effect of upward mobility previously observed in the descriptive statistics is not confirmed by the regression analysis and Hypothesis 2 is rejected; but we find some evidence to confirm the negative effect of downward mobility giving support to Hypothesis 3. See Appendix B for robustness checks analysing other quality adjusted publication measures.
5. Discussion and Conclusions
This article analysed the impact of mobility on researchers’ productivity. We addressed the relationship by developing a theoretical framework based on the job-matching approach for academics, and the idea of productivity as driven by capital availability and peer effects. We studied job changes characterizing them in terms of upward and downward research and reputation ranking mobility.
The econometric analysis is based on the careers of a sample of 171 UK academic researchers in the period 1982 to 2005. Based on this sample, which should not be biased towards mobility, we find a high level of job mobility: two-thirds of researchers changed jobs at least once, and one–third was involved in two job moves. In this respect, the UK academic labour market resembles the US system rather than other European systems.
First, we analysed the difference in performance between mobile and immobile researchers. We found a positive albeit insignificant overall effect of mobility, but this result may be biased by the variety of different mobility events, including mobility to/from industry. Second, based on a unique robust research ranking system for UK university institutions over the 23 year period of our panel, we studied performance pre- and post-mobility to a better or a worse department than the original one. Contrary to our expectations, we found that mobility to a more prestigious university has no significant impact on performance, while downward mobility seems to decrease performance levels. We found some evidence of decreased productivity in the first year after a job change - probably or most likely due to adjustment costs - for the whole sample. While mobility to a lower ranked institution is associated with a persistent decrease in performance up to six years after the move, we found no evidence of increased performance for mobility to a higher quality department after the initial period of adaptation. Although upward mobile (though not downward mobile) researchers are more productive than their peers, their scientific performance does not improve in the short to mid-term after the mobility event. These results contradict the view that mobility is associated with higher productivity and challenges the predictions of the theory and the assumptions underlying most policy initiatives that mobility to a higher research quality and more prestigious institution has a positive impact on individual research performance. Our results seems to be consistent with the view that scientists’ mobility is driven by other factors, such as family related reasons, rather than strategies to enhance research performance (Franzoni et al., 2012).
These results should be taken with some caveats due to the small number of observations used. Although mobility is more frequent in the UK science system it is difficult to build a complete career dataset for a large sample of researchers. Thus, we are not able to develop more complex econometric models that would take account of the interactions among various determinants. We also acknowledge that, although we use a dynamic model lagging possibly endogenous variables and we include a feedback measure, our results may be still biased due to reverse causality problems. We tried to address the endogeneity problem by controlling for unobserved heterogeneity and dynamic feedback and through instrumenting the endogenous variable. The test of exogeneity following our instrumental variable approach in appendix A indicates that either we do not have a significant endogeneity problem or that our instrument is weak. We tried to collect new data to build other instruments, but the requirement related to our 23 year panel hindered this attempt.
Due to the complexity involved in collecting full career information, and quantity and quality of the research output, our sample is small in size and may not be representative. However, we do not suspect the presence of bias linked to either research performance or mobility except for the fact that the faculty included in the sample had to be research active (they had to have been awarded at least one EPSRC grant).
This study provides preliminary results for a small sample of UK researchers, that provides interesting, original evidence on mobility and challenges the commonly accepted policy view that mobility is beneficial and should be encouraged. Our results point to a complex interaction between mobility and productivity, which only in certain circumstances might results in a positive impact of the former on the latter. Mobility is far from been always beneficial for individual researchers, instead, mobility is associated with a short-term decrease in performance due to adjustment costs and mobility to lower ranked department seems to result in decreased performance in the mid term as well. Further research on the specificities of mobility, for example mobility associated with career progress, mobility to and from business, mobility to a foreign country, and the career period in which the mobility occurs is needed to properly assess the impact of mobility and inform policy, especially in Europe, to support possible alternative forms of mobility.[18] If our results are confirmed by future work, this would call for a rethinking of policies related to researcher mobility.
Mobility may not always have positive results for individual researchers, but might contribute to improved research from peers at the new institution, thus making it positive for the science system as a whole by increasing the diffusion of ideas. This paper analysed only the individual returns to mobility since analysis of the social returns would require a more comprehensive framework (with higher data requirements), which should be a goal for future analyses.
References
Allison, P.D. and Long, J.S. (1987). Interuniversity mobility of academic scientists. American Sociological Review, 52, 643-652.
Allison, P.D. and Long, J.S. (1990) Departmental Effects on scientific productivity. American Sociological Review, 55 (4), 469-478.
Azoulay, P, Zivin, J.S.G. and Wang, J. (2010). Superstar extinction, The Quarterly Journal of Economics, 125 (2), 549-589.
BIS (2011) International comparative performance of the UK research base – 2011. Elsevier: .
Blundell, R., Griffith, R. and Van Reenen, J. (1995) Dynamic count data models of technological innovation. Economic Journal, Royal Economic Society, 105 (429), 333-44.
Borjas and Doran, (2012) “The Collapse of the Soviet Union and the Productivity of American Mathematicians”, NBER Working Paper No. 17800
Cañibano, C. and Bozeman, B. (2009) Curriculum vitae method in science policy and research evaluation: the state-of-the-art. Research Evaluation, 18 (2), 86–94.
Chan, K.C., Chen, C.R., Steiner, T.L. (2002) Production in the Finance Literature, Institutional Reputation, and Labor Mobility in Academia: A Global Perspective. Financial Management, 31, 131-156.
Cole S. (1979) Age and Scientific Performance. American Journal of Sociology, 84, 958-977.
Cooper, D.P. (2001) Innovation and reciprocal externalities: information transmission via job mobility. Journal of Economic Behaviour and Organization, 45, 403-425.
Cotgrove, S. and Box, S. (1970) Science, industry and society. George Allen and Unwin. London.
Crespi, A.G., Geuna, A. and Nesta, J.J.L. (2007) The mobility of university inventors in Europe. Journal of Technology Transfer, 32, 195-215.
Crespi, G., D’Este, P., Fontana, R. and Geuna, A. (2011) The impact of academic patenting on university research and its Transfer. Research Policy, 40, 55-86.
CST, Council for Science and Technology (2010). A vision for UK research. London.
Dahl, M. S. and Sorenson O. (2010) The social attachment to place. Social Forces, Vol. 89, No. 2, pp. 633-658.
Dasgupta, P. and David, P.A. (1994) Toward a new economics of science. Research Policy, 23 (5), 487-521.
Dietz, J.S. and Bozeman, B. (2005) Academic careers, patents, and productivity: industry experience as scientific and technical human capital. Research Policy, 34, 349-367.
Dietz, J.S., Chompalov, I., Bozeman, B., Lane, E.O. and Park, J. (2000) Using the curriculum vita to study the career paths of scientists and engineers: an exploratory assessment. Scientometrics 49 (3), 419–442.
Ding, W.W., Levin, S.G., Stephan, P.E. and Winkler, A.E. (2010) The impact of information technology on scientists' productivity and collaboration patterns. Management Science, 56, 1439-1461.
Eurobarometer (2005) Mobility in Europe. Analysis of the 2005 Eurobarometer survey on geographical and labour market mobility. Dublin: European Foundation for the Improvement of Living and Working Conditions.
European Commission (2001) High-level Expert Group on Improving Mobility of Researchers. Final report: Brussels.
European Commission (2006) Mobility of researchers between academia and industry. Brussels.
European Commission (2010) Communication from the Commission “Europe 2020 Flagship Initiative Innovation Union” SEC(2010) 1161.
Elton, L. (2000) The UK research assessment exercise: unintended consequences. Higher Education Quarterly, 54 (3), 274 - 283.
Franzoni, C., Stephan, P., Scellato, G. (2012), Foreign Born Scientists: Mobility Patterns for Sixteen Countries, Nature Biotechnology, 30 (12), 1250-1253.
Groysberg, B. (2008) How star women build portable skills. Harvard Business Review, 86, 74-81.
Hargens, L.L. and Farr, G.M. (1973) An examination of recent hypotheses about institutional inbreeding. American Journal of Sociology, 78, 1381–1402.
Hoare, A.G. (1994) Transferred Skills and University Excellence?: An Exploratory Analysis of the Geography of Mobility of UK Academic Staff. Geografiska Annaler. Series B, Human Geography, 76 (3), 143-160.
Hoisl, K. (2007) Tracing mobile inventors. The causality between inventors mobility and inventor productivity. Research Policy, Vol. 36, pp. 619-636.
Imbens, G. and Lemieux, T. (2008) Regression Discontinuity Designs: A Guide to Practice, Journal of Econometrics, Vol. 142, pp. 615-635.
Jovanovic, B. (1979) Job matching and the theory of turnover. The Journal of Political Economy, 87 (5), 972–90.
Kim, E.H., Morse, A., Zingales, L. (2009) Are elite universities losing their competitive edge. Journal of Financial Economics, 93, 353–381.
Lawson, C. and Soos, S. (2013) Higher education ranking indicators: A sensitivity analysis. University of Turin, mimeo.
Levin, S.G. and Stephan, P.E. (1991) Research productivity over the life cycle: Evidence for academic scientists. American Economic Review, 81, 114–132.
Long, J.S. and McGinnis, R. (1981) Organizational context and scientific productivity. American Sociological Review, 46, 422-442.
Marcson, S. (1960) The scientist in American industry. Princenton, Princenton University Press.
Merton, R.K. (1968) The Matthew effect in science. Science, 159, 56-63.
Meyer, J.B. (2001) Network approaches versus brain drain lessons from the diaspora. International Migration, 39 (5) Special Issue 1, 99-110.
Møen, J. (2005) Is mobility of technical personnel a source of R&D Spillovers? Journal of Labor Economics, 23, 81-114.
Mortensen D. (1986) Job search and labor market analysis. In O. Ashenfelter and R. Layard (eds) Handbook of Labour Economics, Volume 2, Amsterdam: North Holland.
Moser, P., Voena, A. and Waldinger, F. (2013) German Jewish ÉMIGRÉS and U.S. Innovation (April 29, 2013). Available at SSRN:
Murray, F. (2011) The Oncomouse that roared: Hybrid exchange strategies as a source of productive tension at the boundary of overlapping institutions. American Journal of Sociology, forthcoming.
Oyer, P. (2007) Is there an insider advantage in getting tenure?. The American Economic Review, 97 (2) 501-505.
Pakes, A. and Nitzan, S. (1983) Optimum contracts for research personnel, research employment, and the establishment of 'rival' enterprises. Journal of Labor Economics, 1, 345-365.
Sauermann, H. and Roach, M. (2013) The price of silence: Science and engineering PhDs’ preferences for publishing in industrial R&D, (June 14, 2013). Available at SSRN: .
Shaw, K. (1987) Occupational change, employer change, and the transferability of skills’. Southern Economic Journal, 53 (3), 702-718.
Stephan, P. (2012) How Economics Shapes Science, Cambridge (MA): Harvard University Press.
Stern, S. (2004) Do scientists pay to be scientists? Management Science, 50 (6), 835-853.
Thistlewaite, D.L. and Campbell D.T. (1960) Regression-Discontinuity Analysis: An alternative to the ex post facto experiment, Journal of Educational Psychology, Vol. 51, pp. 309-317.
Toole, A.A., Czarnitzki, D. (2010) Commercializing Science: Is There a University “Brain Drain” from Academic Entrepreneurship? Management Science, Vol. 56, pp. 1599-1614.
UNESCO (2005) Tertiary students abroad. Learning without borders. Montral, Canada: UIS.
van Heeringen, A. and Dijkwel, P.A. (1987) Effect of mobility on productivity. Scientometrics, 11, 267-280.
Waldinger, F. (2012) Peer effects in science - Evidence from the dismissal of scientists in Nazi Germany. Review of Economic Studies, 79 (2), 838-861.
Weinberg, B.A. (2007) Geography and innovation: Evidence from Nobel physicists, Ohio State University
Wooldridge (2002a), Introductory Econometrics: A Modern Approach, second edition, South Western, Cincinnati, OH.
Wooldridge (2002b), Econometric Analysis of Cross Section and Panel Data, MIT Press, Cambridge, MA.
Ziman, J. (1991) Public understanding of science. Sci Technol Human Values, 16 (1), 99–105
Zubieta, A.F. (2009) Recognition and weak ties: Is there a positive effect of postdoctoral position on academic performance and career development? Research Evaluation, 18, 105-115.
Zucker, L.G., Darby, M.R. and Marilynn, B. (1998) Intellectual human capital and the birth of U.S. biotechnology enterprises. American Economic Review, 88, 290-306.
Zucker, L.G., Darby, M.R. and Torero, M. (2002) Labor mobility from academe to commerce. Journal of Labor Economics, 20, 629-660.
Appendix A: Robustness checks addressing endogeneity
A.1. Instrumental variable approach
Some papers have addressed the endogeneity arising from reverse causality between performance and mobility by using natural experiments and quasi-randomised assignment (Moser et al. 2013; Borjas and Doran, 2012). However, these represent rare events that are of little policy relevance (see Appendix A.2. for a discussion). As mobility cannot be randomised in practice we need to discuss other research designs that can deal with reverse causality and selection processes, i.e. instrumental variable approach (cf. Wooldridge, 2002a).[19] However, finding plausible instruments is very difficult, especially in the case of mobility and productivity, where one researcher’s instrument may be another researcher’s hypothesised cause of publications. Addressing the causality between inventor mobility and productivity, Hoisl (2007) proposes a simultaneous relationship and considers “city size” as instrument for mobility in the productivity equation. Toole and Czarnitzki (2010) use lagged regional variables as an instrument for joining or founding a firm in their productivity equations. While these instruments might be able to explain mobility opportunities to business firms without effecting productivity, they are not very convincing in an academic context. Researchers in larger or more dynamic cities may not only have more employment opportunities, but proximity to more peers may also affect their productivity. A positive effect of mobility on publications may then simply be a spurious relation caused by access to larger networks, but, instruments beyond regional indicators are difficult to identify and measure.
Dahl and Sorenson (2010) showed for a sample of Danish scientists and engineers that also the highly skilled value proximity to family and friends and are willing to forgo parts of their income to live closer to home. Franzoni et al. (2012) confirm that family ties play an important role in motivating the return of academics to their home country. We thus propose the distance to ones place of birth as instrument. Researchers that live further away from home are more likely to move as they have less social costs associated to the move. Distance from home should not affect productivity though of course this cannot be ruled out as close family may provide help with child care which in turn could affect productivity.[20]
In robustness checks, we thus introduce an instrument for mobility in the regressions comparing pre- and post-mobility performance. We estimate the following model:
[pic] (B1)
[pic] (B2)
where sp’it is the count variable representing scientific performance. [pic] denotes the predicted mobility event and Xit represents a set of control variables. distit-1 is the distance from home, our instrument.
The instrument, distance from home, is measured as the distance from place of birth.[21] The distance is measured in km with a ceiling of 1,000km for overseas researchers. Only 118 researchers in the sample provided this information. For these researchers, average distance is 363 km. Of the 43 mobile researchers included in regressions 3 and 4 in Table 3, the instrument is only available for 27, reducing the number of observations significantly.
The results of the instrumented model are presented in Table A1. The results show that for the instrumented post-mobility indicator we find a positive effect on academic performance (88% significance in the publication equation and 95% in the citation equation). The effect of the instrumented mobility dummy is positive and significant in years one and three. This indicates that there is no trade-off and no decrease in short-term performance, but that mobility results in increased publications. We test for endogeneity and the validity of the instrumental variables approach based on the two-step model described in Wooldridge (2002b). The test shows that while distance is significant in the first stage regression, the exogeneity test based on residuals is not rejected. Thus, our model does not seem to suffer from endogeneity bias. We therefore regard the results of the original model without instrument as more reliable.
[Table A1 about here]
A.2. Addressing endogeneity through experimental designs
In an ideal setting economists would want to study the effect of mobility in an experiment that randomly assigns researchers into mobile and immobile groups, thereby exploiting an exogenous source of variation in the explanatory variable to analyse the effect of mobility on productivity. A natural experiment sets an exogenous and abrupt change in the group under scrutiny. For example, a university department that closes unexpectedly forcing all its researchers to move, assuming that the closure was not due to their (lack of) productivity. Then, the possible causal link of productivity on mobility can be controlled for using this external shock as an instrument, or by defining an unaffected control group. Moser et al. (2013) use the dismissal of Jewish scientists from Nazi Germany as a natural experiment to address the endogeneity problem in an analysis of mobility to the US. Similarly, Borjas and Doran (2012) use the collapse of the Soviet Union as an external shock, to measure mobility and productivity. Alternatively, the regression discontinuity (RD) design (Imbens and Lemieux, 2008) can be used if treatment effectively is random. It assumes that the value of a treatment predictor is on either side of a fixed threshold. This design can therefore be applied to situations where an administrative body sets transparent rules for treatment and defines a cut-off point according to resource restrictions. The RD design was originally implemented by Thistlewaite and Campbell (1960) which analysed the impact of a scholarship award and was based on observed test scores and on future outcomes, comparing a group that just passed with the group that just failed. In the case of mobility, RD designs can be useful to measure student mobility if places in higher education institutions are determined by a placement test, or for participation in research visit programmes if these are determined by a stringent set of criteria. In the case of job to job mobility, the RD design is difficult to implement since assignment to treatment is not based on participation in a programme, or on observable selection criteria.
Appendix B: Robustness checks using alternative dependent variables
Tables B1 and B2 consider upward and downward mobility for two additional measures of publication quality that take account of different citation windows or other citation related idiosyncrasies. First, we consider the quality weighted variable based on the total number of citations received before April 2013 (the date of data download) by each year’s papers. Thus, we allow for longer (at least 8 years and up to 21 years) time periods of citation accumulation. We observe a positive sign for the upward mobile, but the effect is insignificant (Table B1), while the signs for downward mobile (Table B2) remain negative in the most complete specification. However, none of the estimates is significant. Thus, we find no strong difference between mobile and immobile or between pre- and post-mobility. Overall the results corroborate those using a five-year citation window.
In columns 6 to 10 of Tables B1 and B2 we examine the effect of mobility on average research quality (average number of citations received by each article in the first 5 years after publication). This denotes the average quality of a researcher’s publications regardless of their quantity. Interestingly, the signs in the regression for upward mobile are reversed. We observe a negative sign for upward mobile, indicating that researchers that move to a better institution produce publications of lower average quality. For downward mobile we still observe a negative sign in column 8 which compares pre- and post-mobility productivity. The previously consistent negative effect of the downward shock, however, does not persist. All the coefficients are insignificant suggesting that there is no significant effect of downward or upward mobility on average publication quality.
Overall, we observe that, in relation to quantity, while researchers that move to a higher ranked university perform better ex-ante, and researchers that move to a lower ranked university perform worse ex-post, the quality of their publications does not change significantly.
[Tables B1 and B2 about here]
Figures
[pic]
Figure 1: Age in 2005
[pic]
Figure 2: Year of first tenured position
[pic]
Figure 3: First position after PhD
[pic]
Figure 4: Age of researcher in year of first promotion
[pic]
Figure 5: Average publication numbers
[pic]
Figure 6: Publication numbers in years since move
Tables
Table 1: Definition and Summary Statistics 1982-2005
| | |Full Sample | |Reduced Sample of UK-HEI |
| | |2367 observations | |1662 observations |
|VARIABLES |
| |
TABLE 4: Upward Mobility between UK-HEI
| |
TABLE 5: Downward Mobility between UK-HEI
| |
Table A1: Instrumental Variable Approach – Mobility between UK HEI
| |(1) |(2) |(3) |(4) |
|VARIABLES |PUB |PUB |CIT5yr |CIT5yr |
| | | | | |
|PostUNIMobit |1.525 | |2.692** | |
| |(1.002) | |(1.172) | |
|L. UNIMobit | |2.074* | |3.646* |
| | |(1.174) | |(2.078) |
|L2. UNIMobit | |-0.129 | |-0.401 |
| | |(0.567) | |(1.067) |
|L3. UNIMobit | |1.813* | |4.515*** |
| | |(1.050) | |(1.659) |
|AGEit |0.008 |0.012 |0.021 |0.168 |
| |(0.057) |(0.054) |(0.113) |(0.134) |
|AGEit 2 |0.000 |-0.000 |0.000 |-0.001 |
| |(0.001) |(0.000) |(0.001) |(0.001) |
|RANK2it-1 |-0.144 |-0.095 |-0.668* |-0.415 |
| |(0.325) |(0.234) |(0.388) |(0.453) |
|RANK3it-1 |-0.368 |-0.412 |-1.016* |-1.114* |
| |(0.492) |(0.350) |(0.548) |(0.573) |
|FEMALEi |-0.105 |0.138 |-0.600** |-0.013 |
| |(0.149) |(0.132) |(0.286) |(0.318) |
|POSTDOCi |-0.337 |-0.241* |-0.765** |-0.464 |
| |(0.208) |(0.147) |(0.357) |(0.337) |
|PATENTit-1 |0.396*** |0.298** |0.746*** |0.599*** |
| |(0.153) |(0.129) |(0.209) |(0.202) |
|PHYSICSi |0.295 |0.030 |0.551** |0.119 |
| |(0.229) |(0.106) |(0.245) |(0.268) |
|COMPUTERi |-0.575 |-0.846*** |-1.474** |-2.743*** |
| |(0.352) |(0.276) |(0.690) |(0.710) |
|MECHANICALi |-0.394* |-0.320 |-1.006*** |-0.563 |
| |(0.211) |(0.250) |(0.335) |(0.397) |
|UniRankingit-1 |0.216 |-0.003 |0.662 |-0.227 |
| |(0.293) |(0.270) |(0.471) |(0.632) |
|LONDONit-1 |-0.476 |-0.360 |-0.803 |-1.011 |
| |(0.434) |(0.251) |(0.645) |(0.835) |
|Pre-sample Average (PUB/CIT) |-0.491*** |-0.530*** |0.004 |0.000 |
| |(0.147) |(0.155) |(0.003) |(0.003) |
|Stock (PUB/CIT) |0.020*** |0.024*** |0.001*** |0.001*** |
| |(0.003) |(0.003) |(0.000) |(0.000) |
|Constant |0.892 |1.402 |2.604 |-0.358 |
| |(1.530) |(1.468) |(2.594) |(3.213) |
|lnalpha |-1.575*** |-1.870*** |0.148 |0.088 |
|log Likelihood |-976.472 |-494.490 |-2019.843 |-1029.951 |
|Observations |403 |204 |403 |204 |
|Clusters |27 |27 |27 |27 |
|Robust clustered standard errors in parentheses; Year fixed effects in all models; Chemistry is the omitted |
|category; |
|*** p ................
................
In order to avoid copyright disputes, this page is only a partial summary.
To fulfill the demand for quickly locating and searching documents.
It is intelligent file search solution for home and business.
Related download
- conceptests for science and math classes
- why general chemistry instructors are boston university
- a brief overview portland state university
- researchers mobility and its impact on scientific
- thomson cathode ray tube experiment new vcl lab
- regulations for graduate students in chemistry
- chemistry 101l boyd county public schools
- s 2 1 2 no
- auburn university
Related searches
- time and its importance
- language and its importance
- information system and its components
- management and its importance
- biodiversity and its conservation
- biodiversity and its conservation pdf
- technology and its negative effects
- education and its importance
- define management and its functions
- recovering intoxication and its aftermath
- heart and its parts function
- management and its importance articles