1. Data, networks and experimentation
Hasan Bakhshi
Nesta Policy & Research Unit
IRC Annual Summit,
26th November, 2013
http://www.nesta.org.uk/
2. “Innovation policy would
work better, we suggest, if
modelled on experimental
science and directed to the
task of minimising the
uncertainty that
entrepreneurs face in the
discovery of opportunities
and constraints”
3. “…uncertainty is a defining
feature of emergent areas
subject to persistent
structural change like the
creative industries, and
should be dealt with in a
systematic way.”
5. Innovation policy as a process
Test a
hypothesis
Discover
what was
unknown
Test a
further
hypothesis
6. Data and evidence-based policy
Data
Programme
Ex post
Evaluation
Ex ante
evaluation
Programme
Data
7. CASE 1: CREATIVE CREDITS
Innovation
voucher
SME
Innovation project
RCT
Creative
SMEs receiving Credit 78% more
likely to undertake their project
✓
Strong evidence of S/T output
✓
additionality in terms of increased
innovations after six months Source: Bakhshi
et al (2011)
8. CASE 1: CREATIVE CREDITS
Innovation
voucher
SME
Innovation project
RCT
Creative
But no significant output
additionality after 12 months
X
X
No significant network or behavioral
additionality after 12 months Source: Bakhshi
et al (2013)
10. CASE 2: DIGITAL R&D FOR THE ARTS
Arts
organisations
Funding
Technology
companies
DIGITAL R&D
FUND
Digital R&D
Projects
Academic researchers
£7 million, 2012-15
50-60 R&D projects?
Sector-wide learning
12. 1736 new Twitter following
connections between attendees
after LeWeb’12 London
24% ↑ in total number of
following connections between
attendees
8% ↑ in total number of
following connections made by
attendees with non-attendees
HASANGreen lines are the following connections between attendees at the LeWeb12 London conference formed in the three months after the conference. The light blue lines are the new following connections between attendees and speakers. The dark blue lines are the new following connections involving speakers.Within that three-month period, 1736 new following connections formed between attendees. After allowing for un-following activity, this represents a 24% increase in the number of connections between attendees. This compares with an 8% increase in the number of their following connections with non-attendees. Can we attribute all of this to the event? Unfortunately not! There is massive self-selection in event attendance – people go along tend to have common interests which means that they are more likely to form new connections with each other than with others. Running a controlled experiment where you randomly decide which events people can go to is not realistic! But we are exploring ways of using other data, including other properties of individuals’ Twitter networks, to control for their propensity to follow each other.
HASANAdditional connections are valuable insofar as they lead to information flows that would not otherwise flow so directly and greater awareness. But on their own the connections are weak. We also want to know if they trigger stronger connections; content analysis of the tweets may give proximate indications. The frequency with which words like ‘meeting’ and ‘email’ appear in this wordcloud trivially illustrates what I’m getting at, but what we’re looking at in much greater depth.