Clement, Shipilov, & Galunic (2018). Brokerage as a Public Good: The Externalities of Network Hubs for Different Formal Roles in Creative Organizations


Julien Clement – INSEAD

Andrew Shipilov – INSEAD

Charles Galunic – INSEAD


Tianyu He – INSEAD

Tom Taiyi Yan – University of Maryland

Article link:

Question 1. On the mechanism of hubs’ negative externalities
In the article, you proposed that a hub on the team can negatively disrupt group efficiency (hypothesis 2a), and theorized that their unsatisfactory performance might be tolerated because of their brokerage position. One may argue the opposite, that hubs in this highly interconnected industry, should be very motivated to avoid negative impressions given bad news travels quickly and can do considerable damage to their future employment. What is your opinion on this pushback? In other words, why would hubs predominantly experience the buffering of reputation-cost from their brokering roles but not suffer at all from the words of mouth? Additionally, if hubs only constitute a small part of the team (in terms of number or functional participation), would their negative externalities still be an issue?

Julien: That’s an interesting point, and it’s worth unpacking a little bit what is going on in the data. The results seem to favor our explanation rather than the alternative you are suggesting: we found a negative effect of hubs on the producers’ contributions to show performance, while your argument would suggest no effect. But why might that be? First, it isn’t clear that the network is highly interconnected: from 1998 to 2012, the yearly networks include more than 200 executives, and the average number of ties for each executive is 11 (with very heavy clustering within communities and few bridging ties, as shown in Figure 1 of the paper). Second, we define hubs as executives who bridge across individuals from different communities who are not connected to each other. This makes it quite plausible that their reputations would remain contained within communities, because the individuals they work with have few opportunities to share information with each other. But if we ran the analyses again and defined hubs only as individuals who bridge across individuals who know each other, perhaps the mechanisms you are suggesting would come into play!

As for your last question, we defined our network by only considering executives who held any of the seven most important functions in production teams (which sometimes include hundreds of individuals in total). By definition, any individual identified as a hub constitutes a very important part of the team—these executives are in direct contact with the producer and handle very significant parts of the shows. Intuitively, there is little reason to think that hubs would generate negative externalities if they are not involved in significant aspects of the shows (for instance, one screenwriter alone might not cause significant issues even if she has connections across the whole industry). What is interesting to me is to think whether hubs need to fulfill important roles in order to generate positive externalities. Perhaps even employees who don’t fulfill high-level executive roles may bring in good ideas from other shows they’ve worked on (assuming they can bring that information up to the creative directors).

Charlie: To add to Julien’s comments, your logic makes sense where there are good channels for Hub neglect to be widely known, but not many such channels in this context. More generally, because Hubs tend to deliver on the creative side, this probably gives them some powerful get-out-of-jail cards on the production side. The key to understanding the “how/why” on negative externalities is to view the character holistical, and so also take into account the positive externalities.

Question 2. Interpretation vis a vis the effect of top executives
In your analysis, you controlled for top executives’ (creative director and producer) access to brokerage. Meanwhile a number of research (e.g. Venkataramani et al, 2014; Balkundi et al., 2011) indicate that leader’s network position is influential to team performance. In retrospect, how do you see your findings fit with this line of research?

Julien: That’s a good question! The results suggest that being connected to hubs is more important to producers and creative directors than their own ego-network. We should be careful in generalizing this result, especially since much research has shown that ego-networks do matter. In our context, it isn’t all that surprising that producers are hurt by being connected to hubs more than by being themselves hubs: their incentives are typically much better aligned than the hubs’ with the production team. Producers are often remunerated in direct proportion to a show’s success, while other team members are mostly remunerated per episode produced. Hence, even if a producer has many commitments across different shows, she is likely to make sure that all shows are well managed. A hub in one of the five roles below producer and creative director, on the other hand, may be more opportunistic (or simply have a harder time dividing attention across shows, because these jobs require more presence on the production set). This points to some potential implications for incentive design: organizations might want to design contracts with high-powered incentives for employees whose attention is most likely to be divided across projects.

The non-result is more surprising for creative directors: since their job revolves mostly around gathering creative insights from around the industry, we would expect them to benefit from being hubs themselves! It’s worth noting that, once we make efforts to account for endogeneity in Model 4, the effect of creative director’s brokerage across communities gets closer to being positive and significant, though not quite. This may simply be an issue with our sample being relatively small. Also, the results suggest that being central in the industry (betweenness centrality) has a positive impact on creative directors. So, overall, our results certainly don’t contradict prior findings that executives’ networks do have an impact on their organization’s performance.

Charlie: There is some other research (Galunic, Ertug, Gargiulo, AMJ, 2012) that looks at the leader-as-broker (and senior people more generally), and the spillover effects that may be possible to those under them (rather than the project, as in this ASQ article). As you suggest, it does show that positive externalities are possible in the leader role, that is that social capital under some circumstances can have secondary effects (improve the performance of subordinates). In general, we see these papers as examples of social capital as a public good, which is our broader theme.

Question 3. On network evolution
Given hubs can inflict negative externalities to team efficiency, we are curious about your take on how the network might evolve. What do you think will happen to these hubs with time? Would they be exiled or would their informal roles as hubs keep them as valuable asset in a team despite their underperformance in their formal roles? Would their access to brokerage decline?

Julien: In a way, this is linked to your first question: whether hubs will be punished for creating inefficiencies should depend on whether their reputation eventually gets hurt across the network. In a highly clustered network, this might take some time to happen. A producer did mention to me during an interview that the industry has been taking measures in recent years to reduce the power of influential television presenters. It used to be that only a few presenters worked on a large portion of all game-shows airing in France. This led to significant issues, because some presenters started making many demands and disrupting shows when those demands weren’t met. According to this producer (and this seems to be true based on my own observation of recent shows), game shows now tend to feature less experienced presenters to avoid such issues—the star should be the show itself, not the presenter. This suggests that hubs may indeed lose access to brokerage over time if they disrupt their neighbor’s projects, but this process might take quite a bit of time to unfold.

Question 4. On working with a substantive industry and constructing complicated dataset
a).Your paper zeroed in on French game show industry – a novel context, previously untouched by scholars – through which you constructed an impressive network dataset that fits your research agenda neatly. Was this your plan from the very beginning? How did you come to this context initially (is any of you a heavy game show fan)?

Julien: I was a first-year PhD student when I started working on the project, and I had just attended the “Network Evolution Conference” at INSEAD. There was a presentation by David Stark (Columbia University) on collaboration networks in the video-game production industry, which I thought was really exciting. I went to Andrew right after and told him “I want to work on something like that!”. He was interested, but told me it would be nice to find a novel dataset (many people were working on the video-game production industry) so I tried to think about possible data sources. Coincidentally, my father works in the machinery industry, and he used to work on the making of camera cranes for the game-show industry when I was a kid. This is how the idea came to me—I basically remembered going on TV sets when I was a 6-year old and thought “maybe there’s data on this”. It turns out, there is! It was actually very helpful to know a bit about the industry beforehand and to remember a few of the people who work in the French game-show industry (though I don’t think any of them remembered the 6-year-old me who used to run around their production set!). This helped me interview industry members, which is what made me realize the neat distinction between producers and creative directors which we use in the paper. And while I was starting to collect the dataset, I took Charlie’s Network Analysis class at INSEAD, which featured two papers on second-hand brokerage. The “plan” came together as we combined all these pieces into an authorship team and a paper.

Andrew: This is a great example of building unique datasets. When I work with doctoral students, I always ask them to think of data that reflects something unique about them. Maybe their background, hobbies or the background and hobbies of their relatives. This makes working on the paper much more interesting down the road: if someone has personal interests in the data, the person will be more curious to work with it. Datasets that originate from the unique experiences also have built-in “barriers to entry”, that is a PhD student has a lower probability of finding a new article where someone else has already answered the same research question in the same dataset.

Of course, the challenge then is to develop a compelling argument for how the findings in a unique dataset generalize to other contexts. Fortunately, when things really happen in one organizational context, there are always similar other contexts in which the same things can happen. And the development of boundary conditions surrounding a particular relationship is a good way to develop a theory.

b). For other scholars looking to contextualize their theories (especially phd students hoping to convert their projects into publications efficiently), what advice would you give regarding choosing and studying substantive contexts?

Andrew: Always make an effort to speak to people who represent your “data points”. Scholars doing qualitative work do this all the time, while scholars doing quantitative work don’t do this as often. It is true that there is nothing more practical than a good theory, but a good theory should be informed by the real-world puzzles, or at least by the real world events. Conversations with managers can help you test whether the theory you are developing make sense to them. These conversations can also help develop alternative explanations for your theory and suggest neat ways to test boundary conditions. For example, our dependent variable in this study is a show’s performance, operationalized as a viewership score. Both producer and director have to work hard so that the show is appealing to the audience, so this dependent variable picks up ability of executives to engage both in creativity and efficiency-related formal tasks. One of the Reviewers asked us to develop some alternative dependent variables that could allow to isolate either the effect of a producer’s network or the effect of a director’s network. Julien suggested that we use variation in show length as an indicator of the producer’s ability to coordinate tasks inside her team. He could think of this variable because he observed how the show was made. He realized that sometimes there are disruptions caused by the lack of commitments of some production team members and these disruptions force the production team to make shows with unequal episode lengths. As we hoped, this variable was affected by the producer’s interactions with the hubs but not with the creative directors’ interactions with the hubs. Without observing how the show was made, Julien would not have had this idea and we might not have had the paper published.

c). What did a typical day look like during the data collection? Are there any interesting stories that happened in the process?

Julien: The “typical day” evolved over time, for the better! The first step was to collect data on the TV shows (team members, viewership, …) from the French “INA” Institute, located in the basement of the French National Library. This institute is very good at keeping data, but not quite as good at making it available: I had access to a dataset containing millions of lines of data on French TV shows (not only game shows) since the 1980s, but I had to copy-paste this data manually onto a CD… and the whole system crashed whenever I copied more than 500 lines at a time! It took me a few months to collect everything (I was going to the institute on weekends or whenever I didn’t have class) and watch video recordings of all shows in all years to record the members of each production team from the credits. This was the least fun part; what came next was a lot more exciting. For the rest of the project, I iterated between working with data and interviewing industry members to understand what the data really meant. It was a lot of fun talking to them and getting to know about the industry! I was even lucky enough to be invited by a producer to follow him at the annual international conference on game-shows in Cannes, France. So, overall, I think I’ve been quite lucky with the data collection experience.

Andrew: We also shouldn’t forget the days watching all TV shows again after the second R&R decision came in. To respond to some valid Reviewer concerns, we wanted to code the content of the shows. This would allow us to determine whether the community-based network structure had any reflection in the shows themselves. That is, we wanted to see whether there were similarities in show contents when these shows were made by professionals from the same community as opposed to being made by professionals from different communities. It was a good opportunity to deeply immerse oneself into a French TV show watching culture: this is a very fun industry with lots of creativity and great show concepts. But I have never in my life watched so much TV in a single day. At least I had a good excuse: this was a work-related task now.

Question 5. On methods and visualization
Your paper implemented sophisticated visualizing tools and statistical methods. As a PhD student, how did you harness these skills? What recommendations do you have for PhD students at different stages/years to build up a sufficient skill set, especially with the increasing demand for big data and advanced coding techniques?

Julien: The visualizations are perhaps what I’m most proud of in the paper! I made them using a visualization library in JavaScript called “D3.js” which I find very powerful. I hope readers will look at the appendix where the color graphs are; I think they really capture the paper’s story. The graphs are also helpful in explaining the paper to people who don’t have the time to read the whole paper (like the people I interviewed… or my parents). I think it’s definitely worth spending time early in the PhD program to develop capabilities in handling data and graphs (perhaps D3.js is pushing it a bit too far, but at least learning Python is a must). I learned to code in Python as I was working on agent-based simulations for another paper, and then used it to manage data in this paper. I think it will get harder and harder to do well in a PhD program without learning to program. This paper uses a dataset of decent size, but every year it becomes easier to collect new datasets of much bigger size and granularity. Developing programming skills makes it possible to take advantage of this by building ambitious projects. Also, learning to program is quite fun!

More generally, this project was a great opportunity to develop my analytical skills. As I started working early on the project, it also made the PhD coursework a lot more appealing to me: for instance, our econometrics class didn’t look abstract to me at all because I had my own project to apply the techniques I learned. Andrew was also very helpful: he makes a lot of efforts to stay aware of the most advanced statistical methods and was always happy to share his knowledge with me. I suppose one conclusion from this could be that starting projects early during the PhD is a good idea, but I would advise caution: working early on projects can also create “tunnel vision” problems whereby we think too much about a specific project and not enough about all the other things we learn during PhD coursework. It probably took me more time than other students to develop ideas for my dissertation work and think of the identity I want to have as a scholar, because I had less time to put aside to think about those things. I’m very happy with how things turned out in my case, but that’s worth keeping in mind.

Thank you for interviewing us!

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: