Bringing the lab back in: Personnel composition and scientific output at the MIT Department of Biology

Highlights

• We introduce a novel dataset on MIT Biology labs from 1966 to 2000.

• We document trends in laboratory composition over more than three decades.

• We document that postdocs (vs. grad students or technicians) account for the majority of laboratory publication output.

• Relative to postdocs, grad students and technicians are equally important for breakthrough publications.

• We propose a methodology to reconstruct laboratories from bibliometric data.

Abstract

We study the link between a laboratory's personnel composition, its number and types of graduate students and postdocs, and the laboratory's productive output. Building upon a fine-grained dataset with full personnel lists from the MIT Department of Biology from 1966–2000, we find that while postdocs account for the large majority of publication outputs, graduate students and postdocs with external funding contribute equally to breakthrough publications. Moreover, technicians are key contributors to breakthrough publications, but not to overall productivity. Taken together, this study contributes to our understanding of knowledge work, as well as reinforcing the importance of a laboratory's personnel composition.

Introduction

“… if the scientists we shadow go inside laboratories, then we too have to go there, no matter how difficult the journey.” (Latour, 1988, p. 63)

The past two decades have witnessed an unprecedented advancement in a researcher's ability to collect and analyze large datasets. An exponential rise in computer storage and power, coupled with ready access to an ever increasing array of online data sources has enabled researchers to analyze datasets numbering in the millions of data points. Large-scale patent data have been used to study knowledge spillovers (Jaffe et al., 1993, Audretsch and Feldman, 1996, Breschi and Lissoni, 2001), inventor mobility (Marx et al., 2009, Singh and Agrawal, 2011), and inventor networks (Fleming and Sorenson, 2004), to name a few recent examples. More recently, a parallel easing of access to data on academic publications (e.g., Azoulay et al., 2006) has enabled the study of collaborative teams (Wuchty et al., 2007) and spillovers across individuals (Azoulay et al., 2010). Whether using patent or publication data, large-scale datasets allow the documentation of temporal trends across multiple fields, as well as the discovery of exogenous variation or the use of matched samples to aid in causal inference. Lastly, these empirical changes have been particularly pertinent for scholarship focused on the innovation economy, where the highly skewed distribution of productive individuals has been recognized for some time (Lotka, 1926). In short, the ability to access large datasets has engendered a revolution in the social studies of innovation, changing what questions social scientists might ask, as well as the way in which these new questions might be answered.

Despite the incontrovertible advantages of larger datasets (we doubt that any scholars would argue for fewer, rather than more data points), we fear that during this shift in the size and scope of data, critical supporting structures that underpin scientific productivity – in the context of this paper, the scientific laboratory – have fallen by the wayside. This oversight is remarkable given the central role of the laboratory in foundational studies ranging from the social construction of technology (Latour and Woolgar, 1979), mentorship and training (Zuckerman, 1977, Dasgupta and David, 1994), organizational structure and boundary spanning (Allen, 1984), as well as the coordination of innovative activities (Pelz and Andrews, 1976). More recently, burgeoning literatures on differences across scientists (Roach and Sauermann, 2010, Pezzoni et al., 2012) and incentive structures within firms (Cockburn et al., 1999, Liu and Stuart, 2014), as well as scientific careers and differences between graduate and postdoctoral stages (Stephan and Levin, 1992, Azoulay et al., 2009) reinforce the notion that there is considerable heterogeneity across scientists. As Stephan (2012) notes in her recent book, “Collaboration in science often occurs in a lab. The lab environment not only facilitates the exchange of ideas. It also encourages specialization…” (p. 67).

This paper's central goal is not to overturn laudable advances in data collection and analysis, but to urge greater attention to the study of scientific laboratories. As one avenue of motivation, in this paper we examine the personnel composition within laboratories. Specifically, we focus on laboratory members of varying characteristics and link changes in the number of these personnel types to the laboratory's scientific output. In particular, we ask the following set of related questions. First, in the biological sciences, how have laboratory personnel compositions changed as this field has grown in prominence through the twentieth century? Second, to what extent do laboratory members with different scientific experience (e.g., graduate students vs. postdocs), funding, or position (e.g., trainee vs. technician) affect the laboratory's scientific output? And in our examination of the laboratory's personnel composition, what might we learn about how different types of personnel members affect incremental versus breakthrough publications?

To answer these questions, we examine the laboratory compositions and scientific outputs for one elite set of scientists: principal investigators (PIs) running laboratories at the MIT Department of Biology. Using a complete personnel roster, we document an increase in the prevalence of postdoctoral scientists for the period 1966–2000, while the number of graduate students and technicians remained largely constant. Consistent with prior research, our analysis suggests that personnel are a critical determinant of laboratory productivity: larger laboratories have more publication outputs. Moreover, we find that experienced scientists (i.e., postdocs), particularly those with external funding (i.e., postdocs with fellowships), make greater contributions to the laboratory's publication outcomes, suggesting that both experience and funding are critical determinants of laboratory productivity.

However, when we focus solely on high-profile publications (i.e., publications in Science, Nature, or Cell), we present three unexpected findings. First, graduate students, who make only nominal contributions to overall publication counts, contribute as much to breakthroughs as postdocs with external funding. Second, postdocs without fellowships have no observable impact on breakthrough publications. In a final intriguing finding, technicians are instrumental to high-profile publications, but have no observable impact on lower impact publication output.

These results speak to the importance of composition, not just size, in a laboratory manager's consideration of potential laboratory members. Although larger laboratories result in more publications, only a subset of these personnel types appears to contribute to breakthrough publications. Moreover, our results have implications for the use of large-scale bibliometric data to study productivity. Our results suggest that the sole use of publication author lists to construct laboratory size may lead to severe biases in estimating a laboratory's productive resources. Coupled with recent, exponential advances in the collection of large datasets, our findings on laboratory composition provide motivation to revisit these ubiquitous social groups. We believe that the time is ripe for a large-scale examination of scientific laboratories.

This paper proceeds as follows. Section 2 reviews the literature on laboratories and their importance to knowledge production, as well as posing our research questions. In Section 3, we describe our setting and data, and Section 4 describes our measures and empirical strategy. Section 5 presents our findings. A final section concludes and discusses the implications of our findings for the current trend toward large bibliometric datasets.