6. Network behaviour and bargaining power

Can partnership strategies bring advantages to participants in Horizon 2020? The obvious way to test for that would be to look at the impact of network metrics such as number of stable partners, or network centrality, on the probability of succeeding when presenting an application. Unfortunately, no information is available on non-funded applications.

Computing bargaining power in Horizon 2020

We had to fall back on an observable measures. We therefore built a bargaining power indicator as follows.

Consider a project with npartners and budget B. Denote by B(i) the budget allocated to partner i. We define the bargaining power of partner i for this project as

barPower(i) = ( N * B(i) - B ) / ( N * B )

The indicator is built to reflect the fact that it is harder to negotiate when there are many partners in this project. If itakes all of the budget in a two-partners project, barPower(i) = 0.5; if does the same when there are three partners, barPower(i) = 0.67.

The properties of this indicator are as follows:

• When the budget is equally distributed, barPower(i) = 0 for all ibecause N * B(i) = B
• When i takes more than 1/nth of B, barPower(i) is positive
• When i takes less than 1/nth of B, barPower(i) is negative.
• barPower(i) is limited between -1 and +1.
• barPower(i) = + 1 corresponds to i taking all of B across an infinite number of partners
• barPower(i) = - 1 corresponds to i taking nothing of a project with infinite budget.
• the sum of barPower(i) across all partners of a project is zero.

Next, we use this indicator to build another one, wBarPower, that represents the bargaining power of an organisation across all projects it participates in. This is simply a weighted average of the single project indicators for that organisation, using the budgets of different projects as weights.

Throughout the hackathon we use the European Commission's contribution as budget.

Bargaining power and network structure

We computed wBarPower for all organisations in the stable partnership network. Most of its frequency mass is around 0, with 50% of the observations between - 0.02 and 0.11 (mean = - 0.02, standard deviation 0.05). We observe a weak positive correlation (around 0.05) between wBarPower and both the number of projects and the number of stable partners, and a slightly stronger, negative one (- 0.09 ) between it and a dummy variable called inGC that takes value 1 if the organisation is part of the giant component (the "Horizon 2020 scene"), and 0 otherwise.

We next computed a linear regression model with wBarPower as the dependent variable. the number of projects, the number of stable partners, betweenness centrality in the stable partnership graph and a dummy variable taking value 1 if the organisation is in the giant component were taken as covariates. Results:

• When computing the model on all the 3,414 organisations in the stable partnership graph, the only highly significant coefficient is on inGC, and it is negative. Being connected to the giant component costs, on average, 2% of the organisation's bargaining power.

• When computing the model on the 1,375 private companies, the result is nearly identical to the above.

• When computing the model on the 631 universities, the coefficient on inGC becomes positive, and its significance disappears. The coefficient on the number of projects is now positive and statistically significant (but very small).

• R-square is low for all models, about 0.02.

We conclude that being a full participant in the Horizon 2020 scene in a network sense (that is, having at least one stable partnership to an organisation that is part of the giant component of the stable partnership graph) entails tougher bargaining. But not for everyone in the same way: Companies have to accept a reduction in their bargaining power. Universities and higher education establishments, on the other hand, do not see a similar reduction in bargaining power.

Interpretation

We can find no strong evidence of rent extraction related to network topology in Horizon 2020. Degree and betweenness centrality in the stable partnership network (the "Horizon 2020 scene") do not influence (weighted) bargaining power. This is as it should be.

However, there is a clue that the Horizon 2020 "scene" might skew power in favour of universities and away from private companies: participating in the network's giant component entails a loss of bargaining power for the latter, but not for the former.

The scatterplot below shows wBarPower vs. betwCentrality of organisations with at least 3 stable partners. Centrality is weighted by money amounts in the ecContribution variable. By this metric, the Fraunhofer Institut is far and away the most central organisation in Horizon 2020.