Following a recent a survey commissioned by the Arts Council England, the Arts and Humanities Research Council and Nesta to track how arts and cultural organisations in England are adopting digital technologies, Nesta's Hasan Bakhshi reports how an application of big data helped to precisely quantify the impact of live broadcasting on theatre attendance.
A simple enough question you might think, but one that turns out to be very tough to answer. Theory doesn’t help much. On the one hand, live broadcasts might increase local theatre attendances insofar as the broadcasts serve to promote live shows at theatres. But on the other hand, they might ‘cannibalise’ theatre attendances if the live broadcasts are so good in the eyes of the theatre-going public as to substitute for local theatre. Given the uncertainties is it any surprise that people have concerns? As Sir Alan Ayckbourn has put it: “One's fear, which may be groundless, is that eventually we and our equivalent theatres will stop doing plays and they'll all be streamed live from these centres of excellence".
We might survey audiences for the answer, but what people say they will do can be a bad guide to what they would do in practice. We might observe over time the theatre-going habits of members of the public who have attended live broadcasts, but for that we’d need to match cinema and theatre transactions at the level of the individual – that data is not easy to put together for multiple venues, even if we could overcome the customer confidentiality constraints. More fundamentally, even if we can access such data, how can we separate out the impact of live broadcasts on theatre attendance from other drivers of theatre attendance – such as education and income levels – that correlate with live broadcast attendance? Increasingly, researchers can bring ‘big data’ sets into play to allow them to answer really tough questions such as these, and happily the performing arts are no exception.
More and more performing arts companies are broadcasting their shows live into cinemas as a means of increasing their overall audience reach. In 2006, the Metropolitan Opera in New York began broadcasting HD performances of its operas into digital cinemas. In 2008, the Berlin Philharmonic launched its Digital Concert Hall streaming service, and in 2009, London’s National Theatre began broadcasting live its theatre productions.
Outside of the performing arts, last year the British Museum began live broadcasts of its blockbuster exhibitions to cinemas. In a recent survey
of how arts and cultural organisations in England are adopting digital technologies, we found that live broadcasting was the fastest growing technology.
In an earlier paper
, we looked at the question of the impact of the National Theatre’s live screenings on the National Theatre’s box office and concluded that they had if anything boosted bookings. But that was the impact of just one live screening on one theatre. What impact have the live screenings had over time on theatre attendance more generally?
The gold standard in answering this question would be to select a group of identifiable individuals and to track their theatre attendance over, say, a year. We’d then randomly select one-half of them and provide an incentive for them to attend National Theatre Live screenings. We would then track their theatre attendance for a further year, and compare it with theatre attendance of the remaining half that was not given incentives – the control group. While it would certainly be possible to run such a randomised controlled trial
– Nesta would be more than happy to do so if arts funders were willing to fund it – it would be a significant and costly undertaking.
An alternative approach, which does not require researchers to have access to data at the level of the individual, would involve tracking the theatre attendance of a group of similar individuals for a year. We’d then try and estimate over the following year whether they were likely to have attended National Theatre Live screenings. We’d also track their theatre attendance as a group over this period. This is the approach we follow in a new paper
, where we identify individuals as a ‘group’ if they live in the same locality, and estimate their likelihood of attending a live screening by the proximity of their district to live screening venues.
We can do this because we have been given access to a daily box office data set of around 16 million transactions for 54 performing arts organisations on an aggregated basis drawn from The Audience Agency’s Arts Council-funded Audience Finder data set. These records represent 44 million tickets and span a period from early 2009 through to late 2013, which coincides with the introduction and expansion of National Theatre Live.
After cleaning the data and coding each performance as either theatre or non-theatre, we end up with around 12 million records, of which 32% are for theatre performances.
The performing arts organisations in the data set are shown in Figure 1, along with a 30km radius around each venue (an area intended to be big enough to capture ‘local theatre’ for the purposes of these results – between 70 and 100% of tickets are sold within this radius, depending on the venue). Figure 2 plots the sample in Greater London. The 54 organisations for which we have data are by no means a representative sample of England’s performing arts organisations, but they are a reasonable spread by region and size.
Fig 1: Map showing all performing arts organisations in the sample. Venues close together are shown as a circle with a number, indicating the number of venues. Darker shading indicates the 30km radius catchment area.
Figure 2: Map showing all performing arts organisations in the sample, in London. Venues close together are shown as a circle with a number, indicating the number of venues.
In addition, the National Theatre has provided us with the data set of all National Theatre Live screenings. To the end of 2013 these cover a total of 37 screened productions (including encore performances) from 2009 involving 482 unique cinemas.
It’s worth noting that more than half of our data is for non-theatre productions. This is useful for two reasons: first, because we can examine the impact of National Theatre live screenings on attendances for other performing art forms. But also because they serve as a useful control group – insofar as we might expect any impact of the live screenings on non-theatre attendance to be weaker than for theatre.
Our modelling strategy is to ask: is an individual over time more or less likely to attend their local theatre if, other things being equal, they live closer to a cinema screening a National Theatre Live performance?
Any simple econometric model of the level of theatre ticket sales by district on geographic proximity of that district to the theatre would be riddled with what modellers called endogeneity biases. For example, it is highly likely that screening venues that opt into National Theatre Live are more likely to be areas with high theatre-going demographics. To control for such biases, we look at the change over time in the geographic distribution of ticket sales, exploiting the time variation in National Theatre Live.
There are multiple ways of constructing this ‘time-varying’ NT Live ‘treatment’ variable – should we consider proximity to only the nearest or all screening venues, for example? Should we impose some sort of a distance threshold or allow the treatment to dissipate continuously the further a district is from a screening venue? Do we consider people to have been ‘exposed’ to National Theatre Live if their local cinema has broadcast a screening at any time in the past or only in the previous year, for example?
We investigate the sensitivity of the models to all these options, but we present the baseline results here where only the nearest screening venue is considered, where individuals are considered exposed to a live screening only if a venue broadcast a performance in the previous year, and where exposure is assumed to dissipate continuously with distance.
So, what do we find? On average, areas in England within a 3km radius of a cinema showing National Theatre Live saw a 5% increase in local theatre-going over the next 12 months, compared with otherwise similar areas that were not within this radius. The national result is driven by London, which has seen a 6.4% increase in local theatre attendance following a National Theatre Live screening. Outside of London there is no significant impact either way. Moreover, in a slight variation on this model, we see that this effect declines with distance, as we’d expect: that is, areas within 1km see the largest boost, then 2km, then 3km. When we look at the impact of the live broadcasts on attendances at non-theatre performances we also find strong results, albeit weaker than for theatre productions. What do we take away from these findings?
First, there is no evidence in any of our models that the National Theatre Live screenings have on average harmed attendances at local theatre, at least in our sample of venues.
Second, there is evidence of a boost in theatre attendances in London for populations near cinemas participating in National Theatre Live.
Third, that the impact on non-theatre performances is almost as strong is puzzling and requires further research.
Fourth, our research illustrates the value of public agencies collating and making available to researchers big data sets in the arts (open data), just as is the case in other areas.
Hasan Bakhshi leads Nesta's creative and digital economy policy and research. His recent work includes co-authoring the Next Gen skills review of the video games and visual effects industries, which has led to wholesale reforms of the school ICT and computing curriculum in England, and the Manifesto for the Creative Economy, which sets out ten recommendations by which governments can help the creative economy grow. For a more complete biography, please visit here.
This article was originally published on Nesta's blog and the original version is available here.