Fig 1.
The ratio of paper vs patent output, aggregated to the department and individual levels.
Each circle represents a department, and output per department / per person determines position along the axes. Many departments tend to have a bias toward one output type, but a notable group produces high relative numbers of both papers and patents. Here EECS, MechE and ChemE stand for Electrical Engineering and Computer Science, Mechanical Engineering and Chemical Engineering respectively.
Table 1.
Summary of data after pre-processing.
This dataset, including MIT tenure and tenure-track faculty between the years 2004 and 2014, will be used for the remainder of the study. The table lists unique documents as Observations; the average annual output of the MIT community over the time frame; observations with more than one MIT faculty involved; total number of publishing and patenting faculty (not mutually exclusive).
Fig 2.
Annual A) papers and B) patents output by MIT faculty, between 2004 and 2014, with trend lines describing growth. A) Papers are fit with R2 = 0.96 p-value <0.0001 and t-value 14.12. B) A) Patents are fit with R2 = 0.77 p-value of 0.0004 and t-value of 5.45. The trends show that number of patents (a = 0.04) are growing faster than papers (a = 0.0075).
Fig 3.
Patent and paper output per department per year.
A) Paper output remains fairly constant at the department level, and as a trend across all departments. B) Conversely, some departments exhibit a strong increase in patenting while others do not–there is a notable increase in the disparity of output over the course of the decade (excepting a nearly universal dip in 2008). Here EECS, MechE, ChemE, Chem, DMSE and MAS stand for Electrical Engineering and Computer Science, Mechanical Engineering, Chemical Engineering, Chemistry, Dept of Material Science and Engineering and Mathematics respectively.
Fig 4.
Paper and patent output per building between 2004 and 2014.
On these choropleth maps, buildings are color coded by output volume and labeled with their name (the facility code). Colors are assigned using a Jencks algorithm with five buckets. (For a detailed list of the buildings see https://whereis.mit.edu).
Fig 5.
Intra-Building and Intra-Department activity.
Each observation that has more than one MIT faculty collaborator is assigned a value (from 0 to 1) such that higher values indicate a greater incidence of co-affiliation, for either a building or department. Notably, rates of collaboration within buildings and departments have increased for patents, and decreased for papers. The latter is far more variable, and exhibited an intra-departmental peak in 2009, followed by an intra-building peak in 2012. A) For papers we observed an average of 0.55 for building and 0.67 for Departments. B) For patents we have an average of 0.23 and 0.42 for intra-building and intra-departments activity respectively.
Fig 6.
A) Campus buildings, coded according to heterogeneity, as calculated using the Shannon measure of information entropy. This shows variation in faculty departmental affiliations per building. Values range from 0 to 2.5, classified with a Jencks algorithm. Buildings are labeled by name (facility code). B) MIT campus buildings, coded according the average total area of lab and office space per faculty member. There is a distribution of values from 145ft2 to 2,065ft2 allocated per faculty member. Buildings are labeled by name (facility code).
Fig 7.
A) Co-authorship and B) co-inventor networks. Nodes represent faculty members, and node size is determined by its degree. A link indicates one or more collaborative papers or patents, and its thickness is the weight–a function of number of papers between the given authors, and the total number of authors on those papers. Nodes are colored according to department, and some notable departments are labeled. Here EECS, MechE, ChemE, BioE and MAS stand for Electrical Engineering and Computer Science, Mechanical Engineering, Chemical Engineering, Biological Engineering, and Mathematics respectively. Visualization has been done using Cytoscape.
Table 2.
Networks properties.
Fig 8.
The relative frequency of collaborations between MIT faculty, plotted against their spatial distance on campus.
A) Papers and B) Patents. As distance between two faculty members increases, the likelihood of their collaboration decreases according to a negative exponential function. The same pattern holds true for patents and papers. Papers are fit with R2 = 0.98 while patents with R2 = 0.99.