Page 66 - Technology and Innovation Journal - 19-1

404 DEMPWOLF & SHNEIDERMAN

new insights and frame better hypotheses. Event- product idea was first conceived. Limiting our anal-
Flow provides tools for visually simplifying event ysis to drugs and medical devices, we take the date of
sequences to reveal common and rare patterns (8,9). final FDA approval as the end date and a reasonable
Theory Formation (Modeling): A key goal for proxy for product launch date. Neither the dates that
researchers is to develop and test theories so as to commercial ideas were originally conceived nor the
guide future activities. The well-established linear actual product launch dates are reliably recorded or
model of innovation (basic research leads to applied made publicly available, thus the need for proxies.
research, then product development, culminating in The datasets available for modeling STI processes
commercialization) has its followers, as well as many (see Table 3) have several current limitations, and
critics. Comparisons with alternative models such much of the work yet to be done under this study
as the ABC principle (applied and basic combined) involves cleaning, matching, transforming, and link-
could advance understanding of what leads to more ing existing datasets. We present two preliminary
frequently successful outcomes (10). It is fairly com- examples that demonstrate some of the event ana-
mon practice in articles and presentations to show lytics capabilities of EventFlow (www.cs.umd.edu/
the linear model because of its simplicity and then hcil/ eventflow) and suggest the methods and kinds
immediately state that, in practice, innovation rarely of final results we might expect when all of the data
follows the linear model. The popular understanding cleaning and matching is completed.
of innovation might be improved by documenting the The first example models and analyzes the trajec-
prevalence of the linear model and its alternatives. tories starting with clinical trials and ending with last
Hypothesis Testing: Event analytics can be as FDA approval for 2,402 medical devices. Clinical
simple as seeing if event type A occurs more fre- trial success is typically a necessary input for final
quently before or after event type B. For example, do FDA approval. In certain cases, successful results in
patents precede or follow founding of companies? early-stage trials may be sufficient for provisional,
Another simple question is: How soon after found- temporary approval, allowing the drug or device
ing a company do companies release a product? A to be deployed prior to completion of the full set of
refined version of this question might examine the clinical trials. The preliminary results of this second
distribution of times between founding a company analysis demonstrate EventFlow’s ability to simplify
and releasing a product. There are more sophisticated the visualization of the dataset in ways that suggest
questions that can also be posed in event analytic overarching patterns in the data and allow researchers
tools, such as: Do companies with three or more to pose clear, simple questions for further investiga-
patents before product launches have more successful tion. In this case, the visualization shows two distinct
outcomes than companies with fewer patents? groups in the data: one in which the FDA approval is
received after clinical trials are completed and one in
MODELING AND MEASURING INNOVATION which FDA approval is received during the clinical
TRAJECTORIES: DATA AND EXAMPLES trials (see Figures 3 and 4). The visualizations suggest
The following examples are based on a dataset several additional research questions, demonstrating
comprising 34,331 records, each representing a spe- EventFlow’s usefulness as a tool for data exploration.
cific drug or medical device. Each record contains The second example analyzes drug innovation
the events—research, patents, clinical trials and FDA trajectories from first patent to last FDA approval for
approvals—associated with that product. In total, the 884 drugs, resulting in mean, median, and standard
model includes 85,690 events. The list of event types deviation metrics for these trajectories (see Figures
and the count of each type is shown at the bottom of 5 and 6).
the left EventFlow panel shown in Figure 2.
As a practical matter, answering the question of Data Gathering for Innovation Trajectories
how long innovation takes requires identifying start We use the EventFlow software to model innova-
and end points. In our first example, we take the tion trajectories in drugs and medical devices from
date of first patent application as the starting point multiple datasets (Table 3):
and a reasonable proxy for the date that the initial