The coming crisis of cultural engagement? Measurement, methods, and the nuances of niche activities

ABSTRACT Do ticketing data and national survey data on attendance tell the same story? This question is particularly important in the context of debates over the power of new forms of data to supplant the “traditional” survey methods that have underpinned our understanding of the social stratification of culture. This paper compares three data sources on attendance: the Active Lives Survey, the Taking Part Survey, and Audience Finder. We first compare self-reported attendance at events in each English local authority from the Active Lives survey with ticket sales data, finding a close relationship. We follow up by comparing the distributions of ticket buyers across the Indices of Multiple Deprivation with those from Taking Part, finding that for widely-ticketed and widely-attended art forms they track closely together, providing support for existing trends. Ticketing data does not seem to offer more information on social stratification than traditional social science sources. However, we extend the comparison through more detailed analysis of subcategories within less well-researched forms – literature and dance events – where numbers of attendees are lower, with accompanying uncertainty in survey sources. We find that the audiences for dance vary widely, with ballet attendance being heavily socially stratified but attendance at contemporary dance much more similar to the general population. However, we find that audiences for literature events are more heavily socially stratified than almost any other art form, almost regardless of the subcategory. The power of new datasets is in offering specificity about artforms, rather than overturning what we know about culture and inequality.


Introduction
Survey data have been central to our understanding of culture at least since Bourdieu's Distinction (1979/1984, if not before. The divisions, and associated social distinctions, displayed in the patterning of taste(s) across various European and North American societies would be impossible to fully substantiate without the techniques of the social survey designed to understand behaviour and preferences. The very idea of culture, in the way it has been understood by governments and policymakers, is inseparable from the social sciences that illuminate the distribution of cultural consumption.
However, for over a decade survey methods have been subject to concerns over a "crisis" (Burrows & Savage, 2014;Savage & Burrows, 2007, as these methods' claimsalong with the social scientific interviewto give specific and special insight into the social world have been challenged by new techniques and forms of data. Specifically, this challenge has been from two sources, what Burrows and Savage (2014) call "social transactional data" and "digital by-product data". Here, the information held by commercial organisations about consumer behaviours, particularly around consumption practices, seems to offer much more detailed, fine-grained, information on the social world compared with surveys and face-to-face interviews.
The "coming crisis" narrative has generated extensive debate (Crompton, 2008;Erola, Reimer, Räsänen, & Kropp, 2015;Goldthorpe, 2015), and served several useful purposes. First, it drew attention to datasets and practices that could provide useful information about the social world, in particular those parts of the social world seen to need rapid responses, or swift analysis not amenable to traditional methods (e.g. Procter, Vis, & Voss, 2013). Second, it focused critical analysis onto the creators and holders of both "big" datasets and transactional data, showing how the new modes of knowing the social reflected new flaws, or corresponded to existing lines of social inequality (Noble, 2018;O'Neil, 2016). Finally, it drew attention to the productive relationships between commercial or public bodies and social scientists for doing new types of social research (e.g. Lima & Bakhshi, 2018;Rae & Sener, 2016).
In this context, we interrogate the value of two approaches to understanding cultural attendance in the form of Audience Finder and Taking Part, augmented with Active Lives. This interrogation has two aims, and thus makes two contributions to the literature. First the analysis allows us to consider the extent to which new forms of transactional cultural data suggest a "coming crisis" for survey methods of knowing culture and the social world. Second, the analysis allows us to comment on the stratification of consumption in two areasdance and literaturethat have had less focus in the sociology of culture.
We begin by placing Audience Finder and Taking Part in the longer history of surveys of arts participation, commenting both on the value and limitations. In particular, we note the tendency, discussed extensively by Bryson (1996), Lizardo and Skiles (2015) and Savage and Gayo (2011) for the nuances within art forms to be flattened by the large-scale social survey. This point underpins the subsequent analysis. While a headline comparison of Audience Finder and Taking Part suggests Audience Finder data does not demonstrate a "coming crisis" for our understanding of the broad patterning and axis of inequality associated with cultural consumption, Audience Finder offers important new insights within art forms. We extend this discussion by comparing Audience Finder and Taking Part's understanding of dance and literary events. In doing so, we show the power of new forms of transactional data for giving detailed analysis of artforms, whilst at the same time reaffirming the accuracy and status of the large-scale survey approach instantiated in Taking Part. Methods, measurement, and the potential of new nuanced understandings combine to tell a familiar tale of the unequal nature of culture in England today.

Surveys on arts participation
Large surveys of cultural participation have been vital in drawing a picture of the audiences for the main cultural institutions in European countries and beyond. They have identified and reinforced the findings that the cultural activities traditionally defined as "highbrow", such as visiting museums, going to the opera, to the theatre, or to a ballet, have remained the privilege of the highly educated (Bennett et al., 2009;Callier & Hanquinet, 2012;Campbell, O'Brien, & Taylor, 2019;Eurostat, 2016;Reeves & de Vries, 2016;Willekens & Lievens, 2016). The recent Panic! report (Brook, O'Brien, & Taylor, 2018) does not show any signs of change in these patterns for the UK context. However, these surveys have limitations, as their main focus is to be able to draw consistently precise estimates of general trends. Internationally, it is unusual for surveys of cultural participation to have sample sizes beyond around 5000 (Bennett et al., 2009;Callier & Hanquinet, 2012;Donnat, 2009;Willekens & Lievens, 2016). This means that understanding the dynamics of the audiences for well-attended events, and popular art forms, can be done in some detail. However, given these sample sizes, it is more difficult to draw conclusions about the audiences for niche cultural forms.
Moreover, the categories in social surveys of culture are generally limited. As Beer argues, [p]rominent forms of cultural sociology, including field analysis, have a tendency to work with quite rigid and fixed notions of genre and, therefore, tend to gloss over the creative and mobile drawing, re-drawing and imbrication of genre boundaries as they are created and formulated within the context of everyday cultural engagement. (2012, p. 146) Although most quantitative analysts would recognise the plurality of interpretation given to the same answer categories in a survey, the modes of surveying culture risk attributing a fixed value to artform and genre categories. In turn, this attaches a dominant interpretation, for instance, hip hop has been mostly articulated as a popular (as opposed to "highbrow") taste in the literature (Bryson, 1996), even if its appreciation has increased over time (Lizardo & Skiles, 2015) and some sort of connoisseurship of the genre may, in particular contexts, be a distinctive resource. Indeed, the categories of hip hop and rap can be seen as counter-posed with classical music, occupying opposite ends of scales (Veenstra, 2015).
Beyond the limited interpretations of individual categories is the range of categories in the first place. The nature of surveys means that respondents can't be asked about every single possible variety of art form, and path dependency has meant that some genres and activities that only small numbers of people like and attend are asked about in several different surveys, while other activities with much larger audiences are only rarely asked aboutfor example, attendance at stand-up comedy.
Alongside the question of what can be done using surveys of cultural participation is the question of what is done using surveys of cultural participation. Until recently, little work investigated spatial dynamics, in spite of the major differences in cultural venues and programming in different parts of the country (with some significant exceptions in recent years, e.g. Brook, 2017;Cutts & Widdop, 2016;Delrieu & Gibson, 2017). This more recent work has been illuminating in addressing supply-side accounts: inequalities in cultural participation are not simply due to what is on offer nearby. However, analysis of surveys of cultural participation has disproportionately focused on particular art forms, reflecting both the content of the survey to be analysed, and analysts' choices about which variables to focus on. Music, for instance, especially music preferences, has received a lot of attention (e.g. Savage, 2006;Veenstra, 2015). Such focus is perhaps due to Bourdieu's own emphasis: "nothing more clearly affirms one's 'class', nothing more infallibly classifies, than tastes in music" (1979/1984, p. 10). Yet, overall, we know much less about other areas such as dance or literature, and this is even truer when we move from an enquiry on people's tastes to one on their practices, such as their willingness to attend.

Data sources on English arts attendance
In an English context, the key surveys on cultural participation are Taking Part (DCMS, 2018a) and Active Lives (Sport England, 2018). Active Lives' key benefit is in its large sample size, allowing local-level estimates on cultural participation, which have been used to identify "cold spots" (Gilmore, 2013), and investigate relationships with social and political behaviour (Romer, 2018). However, its questions on cultural participation are very limited. By contrast, Taking Part, owing to the much longer interview, provides detail on a wide range of cultural activities and forms of participation, and is widelyused in the academic and policy communities. This comes at the expense of the sample size, which at around 10,000 a year is still vastly above relevant comparators internationally. (For more detail on the sample composition and survey design, see DCMS, 2016;Taylor, 2016).
However, social surveys are not the only source of data on arts attendance. Here, we focus on Audience Finder. It closely represents what Savage and Burrows describe as the alternative to the social survey, with large volume transactional data with postcodes included. These serve as powerfully predictive proxies for a range of relevant variables (Webber & Burrows, 2018). In this context, it crucially includes far more detail on the range of activities it covers.
With these data sources, we can not only benchmark them against one another, we can also use them to investigate art forms that have previously received less focus in the academic study of cultural consumption. For example, we know a lot more about the social patterning of musical tastes than of tastes for dance, both because of the data that are collected and its analysis. In what follows, we describe what we might expect in the social patterning of two such fields: dance, and literature events.

Dance audiences: what do we know?
A taste for dance is usually surveyed by the activity of going to a dance performance. For instance, the "Cultural Capital and Social Exclusion" survey (Silva, Bennett, Savage, & Warde, 2008) does not contain specific data on this, apart from going to musicals; by contrast there are detailed questions on TV, book reading, and music. In comparison, Taking Part contains a broader range of questions: it distinguishes between ballet and contemporary dance but also African people's dance or South Asian and Chinese dance, and the general category of "other dance". Another example of a very thorough questionnaire is the "2014 Participatiesurvey" (Lievens, Siongers, & Waege, 2014) in the Dutch-speaking part of Belgium that includes questions on ballet, contemporary dance, folk and ethnic dance, street and urban dance, or musicals. Yet, the sample does not allow much exploration of the audience profile of these niche activities. For instance, 97% declare not to have attended a street/ urban dance performance over the last six months, compared with 95.5% for contemporary dance. 1 In England, the numbers are similar, with the most widely-attended type of dance event being contemporary dance, with 4.9% of the population having attended in the previous 12 months (DCMS, 2018b).
In 1998, Sussmann offered the critique that our knowledge of the modern dance audience was very limited, because of the emphasis the national surveys in the US had put on ballet. DiMaggio and Mukhtar (2004) identified a decline in attendance at the ballet, while contemporary art and jazz (in contrast to classical music) had increased in popularity. Moreover, market research seemed to suggest that the audience of contemporary dance in the late 1990s might have been more elitist than that of ballet (Sussmann, 1998). Here, one can contend that contemporary dance and other forms of non-traditionally classic dance may draw a slightly different public than ballet. The field of production of these two types of dance performances is certainly marked by stark contrasts; the field of reception is likely to be as well. In any case, the different types of dance performances cannot simply be expected to be connoted by the same degree of social distinction and, therefore, likely play a different role in our cultural hierarchies.

Literature event audiences: what do we know?
In contrast to attending a dance performance, reading is a more common practice. In the UK it is a widespread activity, compared to other forms of cultural participation. Yet it can be argued it is declining and marked by distinct levels of disengagement. Bennett and colleagues showed that about a fifth of their sample have not read a book in the last 12 months preceding their survey, with two thirds reading less than 10 books (p. 94). As with music, book reading plays a role in the social stratification of tastes, given its association with the educated middle class (Wright, 2006; see also Kraaykamp & Dijkstra, 1999;van Rees, Vermunt, & Verboord, 1999), and this relates both to genre and to volume. This is in line with the findings of Hanquinet (2017) who analysed the degree of legitimacy of reading genres in French-speaking Belgium based on those who have read at least one book over the year preceding the survey (about 66% of the sample). Among the genres she investigated, the proportion of "legitimate" ones (based on the ratio between higher educated and lower educated across three generations) is relatively large (more than 50%). In comparison, historical fiction, bestsellers, and historical books are, amongst others, recognised as legitimate. Amongst the "not legitimate" genres, there are science-fiction novels, 2 romantic novels, and DIY books. Similarly, Atkinson indicated that men and women with high cultural capital disproportionately read more books "demanding symbolic mastery", including classic and contemporary fiction, art/photography books, and poetry (2016, p. 259). Yet, without challenging this view, Hanquinet argues some genres, such as poetry, are appreciated by such a small group that it is hard to characterise it as culturally distinctive. For instance, only 4% of people who have read at least one book in the year preceding the survey have chosen poetry as their most-read type of literature. Similar patterns with regard to poetry can be found in France (Donnat, 2009).
Literary culture is here understood through book readership, rather than attendance at literary events. These are widespread, and often supported by national arts councils, although they often skew towards discussion of the more "elite" literary forms. Where studies of literary events exist, they do not focus on the profile of the audiences (Childress, 2017). Craig and Dubois contended though that the audience at poetry readings is likely to be part of the poetry world or social contacts there in support to the poet. We are then left with many unanswered questions about one area of the literary field that has become increasingly important to sustaining the publishing's economy (Driscoll, 2014).

Data and methods
Taking Part's picture of culture and stratification in England, along with the relative absence of detailed knowledge on "niche", yet culturally significant forms of culture, presents an obvious opportunity for the analysis of Audience Finder data. Our two core concernsabout what transactional data tells us about culture and stratification, and the stratification of specific artforms-are thus addressed by a four-stage analysis. In the first two, we compare Audience Finder with each of the other key surveys on cultural attendance; in the second two, we analyse Audience Finder in more detail.
First, we compare Active Lives and Audience Finder to benchmark attendance at a local level, to identify whether local authorities with high numbers of transactions in Audience Finder are also the areas which Active Lives finds high levels of cultural attendance. Second, we move from analysing areas to analysing different forms, through the comparison of the makeup of audiences at different types of cultural events in each of Taking Part and Audience Finder, to benchmark the two data sources. Third, we look into Audience Finder in more detail, in order to compare audiences across different types of events in each of dance and literary events, to identify how similar audiences are at different types of events within each of these categories. Finally, we compare attendances at each type of dance and literary events between venues that form part of the Arts Council England National Portfolio (NPO). We are especially interested in the comparison between NPOs and other cultural organisations because of ACE's specific mission of ensuring "Everyone has the opportunity to experience and to be inspired by the arts, museums and libraries" (ACE, 2013, p. 39). This policy aim means we should expect NPOs to focus resources on those groups who are traditionally underrepresented in artform audiences. By contrast, those without the specific focus on this form of audience development, and the associated allocation of resources, might have audiences that reflect expected patterns of inequality in attendance.
In analysing the makeup of audiences, we focus particularly on the 2015 Index of Multiple Deprivation (DCLG, 2016). The IMD is a location-based measure of deprivation at the Local Super Output Area (LSOA) level, with areas of up to 3000 people or 1200 households, and is based on seven dimensions, including income, employment, and health deprivation and disability. There are disadvantages to using a location-based measure: living in an area of relatively high deprivation does not mean each resident within it is deprived, and there is debate within the literature about how well the IMD captures area-level deprivation (Deas, Robson, Wong, & Bradford, 2003;Rae, 2009). However, it has the crucial advantage of comparability across both data sources: Taking Part includes information on the IMD decile in which its respondents live, while Audience Finder's data on ticket buyers' home addresses allows IMD data to be added.

Data: Active Lives
Active Lives is an annual survey covering England, run by Sport England, which first ran in 2015/16, but which is the successor to the earlier Active People survey (Sport England, 2018). The data are largely collected online, with invitations sent by post to addresses randomly selected through the Postal Address File, and with some cases consisting of postal submissions. Its very large sample (with a target of 198,250) reflects the fact that it aims to provide estimates for each of England's 326 local authorities: in most cases 500 in each.
Here, we use the 2015-17 18-month estimates collated by Arts Council England (ACE, 2018), which incorporate survey weights; these include slightly larger sample sizes for each local authority, in most cases 750 in each, with accompanying lower uncertainty around each estimate. We focus on whether people reported they have "Attended an event, performance or festival involving creative, artistic, dance, theatrical or musical activity": 3 respondents first answer yes or no, and those respondents who answer yes are asked if they have done it once, twice, or three or more times. Using this, we generate estimates for the average number of attendances per head, with the conservative assumption that people who report having attended three or more times have in fact attended three times.

Data: Taking Part
Taking Part is an annual survey covering England, run by the UK's Department for Digital, Culture, Media and Sport, whose first wave took place in 2005/06, and now consists of both a cross-sectional face-to-face survey of around 10,000 people per year and a smaller (but increasing) number of people in a longitudinal web panel. We use the most recent available wave of the face-to-face, cross-sectional data, collected 2016-2017 (DCMS, 2018a), and focus on two key variables. The first is the IMD decile respondents' homes are classified into, described above. The second is their responses to the questions classified in the survey as "arta": whether they have attended various different types of events in the previous twelve months. We then use these variables to identify the social stratification of each category of event, as defined through the IMD: what percentage of people going to the theatre are in the most deprived decile, and so on.
The estimates of attendance at different events in Taking Part vary widely, from 31% of the 16+ for "Other live music" to 1.7% for "African people's dance or South Asian and Chinese dance". The uncertainty around estimates for the activities fewer people attend is therefore greater than for those activities more people attend.
Taking Part is a complex survey design, with accompanying survey weights. However, both the unweighted and weighted data provide slightly different proportions of households in each IMD decile. We therefore weight the observations to adjust for these different proportions, to draw the comparison with Audience Finder.

Data: Audience Finder
The data from Audience Finder is derived from transactions at cultural venues, collated by The Audience Agency. This data comes to a total of 33,910,657 observations. Postcodes were then combined with a lookup table 4 to be nested within LSOAs; to this, we added data on the IMD decile each LSOA corresponds to, using a further lookup table. 5 As not all observations include complete UK postcode datafor example, a transaction might be conducted at the box office for a production about to start, for which the ticket buyer doesn't offer a postcode at all, or a transaction may be associated with an address outside England, meaning the IMD do not applythe eventual sample consists of 28,176,606 observations. We can then use this data to observe the distribution of ticket sales across different art forms, both corresponding to the relevant categories within Taking Part, and across the variety of second-tier art forms within Audience Finder.

Reconciling categories
The categories used across Taking Part and Audience Finder are not identical. First tier categories within Audience Finder are generally broader than in Taking Partfor example, "dance" compared with four dance categorieswhile categories within Taking Part are generally broader than second tier categories within Audience Finder. We therefore group second tier categories together in order to compare patterns across the two datasets. Full details on how groups have been classified together are present in Appendix A.
Results: comparing Active Lives with Audience Finder Figure 1 shows the distribution of transactions in the Audience Finder dataset in each of the local authorities in England. Local authorities are grouped into six categories based on transactions, with equal numbers of local authorities in each area.
These results are largely consistent with what is seen elsewhere. The areas with the largest numbers of tickets sold per head are disproportionately wealthier areas in the south of England, possibly reflecting the distribution of cultural spending across the country (see e.g. Dorling & Hennig, 2016). We can draw this comparison more directly by comparing the number of transactions per head from Audience Finder with the number of estimated attendances per head in Active Lives, in order to identify how close the relationship is between the two datasets. This is shown in Figure 2. Figure 2 shows a reasonably strong relationship between the two datasets. While not perfectly correlated (R = 0.53), most areas with more transactions in Audience Finder are areas with more estimated attendances in Active Lives. There is, of course, uncertainty around the estimates for the events per head in Active Lives, reflecting both the uncertainty of point estimates based on samples of 750 observations, but also based on the assumption that people reporting three or more attendances in a year have in fact attended exactly three times. This is reflected in the very wide spread in transactions per head among those areas with above-average events per headthe right-hand side of the graphcompared with the lower variance on the left hand side of the graph. This suggests that the number of transactions in each local authority in Audience Finder, per head, is broadly similar to the estimated fraction of people attending cultural events derived from Audience Finder. Figure 3 shows the differences between the proportions of people attending different types of events in Taking Part, and the proportions of ticket sales from those same categories of events in Audience Finder, by IMD decile. In all cases, these proportions are adjusted for the numbers of observations in each category. Activities are ordered by the fraction of the Taking Part sample who reported attending in the previous twelve months, from film at the highest to African people's dance and South Asian and Chinese dance at the lowest.

Results: comparing Taking Part with Audience Finder
IMD deciles are ordered such that areas with the highest levels of deprivation are in the first decile, while areas with the lowest levels of deprivation are in the tenth decile. Activities that are socially stratified such that larger numbers of people in less deprived areas participate will therefore have patterns where the heights of the bars increase from left to right, while activities that are more popular in areas with more deprivation will have patterns where the heights of the bars decrease from left to right.
Each set of bars sums to a total of 100%. If there were no social stratification of a form, each of the black and grey bars in each panel would have a height of 10% for each decile, summing to 100% in each case. In this way, we adjust for the different numbers of reported attendances, and of sales, in each form.
We can assign the different categories here to three main groups: those activities that are both reasonably widely-attended, and mostly ticketed; those activities that are not widely-attended but are mostly ticketed, and those activities that are widely-attended but not mostly ticketed.
The distribution of attendance across both data sources is very similar for several activities. The distributions across IMD deciles, for example, musicals, classical music, and plays/ dramas for both Audience Finder and Taking Part all show similar gradients, with lower proportions of transactions coming from less deprived deciles than from more deprived deciles. Audience membership for these events largely takes place in ticketed venues, likely to be part of the Audience Finder dataset. This provides support for Taking Part: an alternative dataset, much larger but partial in some ways, tells an almost identical story of who is attending.
Other activities' distributions differ in various ways. For both film and live music, attendance is higher for people in less-deprived areas, but not hugely so, using the Taking Part data; social inequalities in attendance at these activities are much sharper in the Audience Finder data than in Taking Part. This likely reflects the distinctiveness of the venues covered by Audience Finder for these activities, with fewer large venues putting on popular music, for example. Perhaps the most striking difference is that for art exhibitions, where Taking Part shows secular increases in participation by IMD decile, while in Audience Finder rates of attendance are highest in deciles 2-5. Instead of reflecting differences in venues, this likely reflects differences in the ticketed and non-ticketed audience, with most attendance at art exhibitions being either non-ticketed, or tickets being bought on the door rather than in advance, with no data on postcode being collected.
Finally, there are the activities towards the bottom of the figure, where events are often ticketed, in the sorts of venues in Audience Finder, but where there are large discrepancies between Audience Finder and Taking Part. For example, Taking Part shows the most deprived 10% areas attending more contemporary dance than the subsequent two deciles, while Audience Finder shows this decile to be by far the one with the fewest transactions. In these cases, the issue is likely to be more with Taking Part than with Audience Finder: the numbers of attendances are so small that the uncertainty around estimates of the fractions of attendances in each decile are very large. In addition, as with any survey, Taking Part suffers from social desirability biasrespondents who feel like the kinds of people who go to the opera but who haven't been in the last twelve months may still claim to have done sowhile Audience Finder consists of directly measured behaviour.
The activities we expect to look similar across Taking Part and Audience Finder do; the activities we expect to look different do. This suggests that, where events are more niche and so difficult to investigate using Taking Part, but widely-ticketed and likely to be held by Audience Finder, Audience Finder represents a unique opportunity to find out more about the audiences for those activities in particular. For this reason, we move to analysing two such areas in more detail: dance, and literary events. Figure 4 presents the distribution of transactions from across the different second-tier categories across the IMD within the dance category in Audience Finder. Unlike Figure 3, Figure 4 records the overall numbers of transactions within each category, rather than adjusting for the fact that more tickets are sold for traditional ballet than for UK & Ireland traditional dance.

Dance audiences
Of the 1.6 million dance transactions in Audience Finder mapped to IMD, 83% fall into the four most popular categories: traditional ballet, contemporary ballet, contemporary dance, and community/amateur dance. As in Figure 3, some of the smaller numbers likely reflect the scope of the data as much as the overall popularity of the events: the overall scope of street dance/hip hop is far beyond what occurs in ticketed venues.
Among these four most popular categories, there are major differences in the distribution of ticket attendance across IMD. Both traditional and contemporary ballet display the kinds of social inequality seen elsewhere in "highbrow" art forms, with people living in areas with lower levels of deprivation vastly more likely to purchase tickets than people areas with higher levels of deprivation. Indeed, these gradients are very similar to each other; it is surprisingly not the case that the audience for contemporary dance, with its hypothetical tastes for more avant-garde and challenging work. While people living in the most deprived decile buy far fewer tickets, and the second most deprived decile only about twice as many, the remainder of the distribution is relatively even, with similar proportions of transactions coming from deciles 3-10. In this way, the audience for contemporary dance appears very different from the audience for both traditional and contemporary ballet.
Finally, within community and amateur dance, the gradient is flatter; while there are more transactions in less deprived areas, these differences are not as sharp as for ballet.
We next turn to differences between these distributions across venues that are part of Arts Council England's national portfolio, and those that are not. We focus on these four categories with overall volumes of transactions in the hundreds of thousands ( Figure 5).
While the balance between the overall volume of transactions between NPOs and non-NPOs varies -58% of transactions in Audience Finder for community/amateur dance are from NPOs, compared with 85% for contemporary dancethere are striking differences in the IMD distributions of purchasers between the two categories of venues. Except for community/amateur dance, non-NPOs have sharper gradients than NPOs do: to put it another way, the differences in the number of transactions between people in more and less deprived areas are even sharper in the non-funded sector than in the funded sector. Among NPOs, around 14% of tickets for contemporary ballet were sold to people in the least deprived 10% of areas, while for non-NPOs the figure is 20%; the equivalents for contemporary dance are 10% and 16%. Community/amateur dance trends in the other direction, with NPOs having slightly higher proportions of their audiences drawn from the least deprived four deciles than non-NPOs.
Non-NPOs may, of course, be charging more to reflect their lack of state support to keep ticket prices low, and thus be more difficult for those with lower disposable incomes to access. At the same time, the national portfolio may also be slightly ameliorating the social stratification of dance events. While it may be unrealistic to expect publiclyfunded organisations to have audiences that identically resemble the general population, here their audiences are closer to the general population than the audiences for venues outside of the Arts Council's national portfolio.
However, while this may be the case for dance, it is not necessarily the case that this applies for other types of events; we therefore now turn to literary events. Figure 6 presents the distribution of transactions from across the different second-tier categories across the IMD within the literature category in Audience Finder: as Figure 4, but with literature rather than dance. 50% of the events within the literary events are classified as literary talks: incorporating other literature, poetry/spoken word, and author readings covers 97% of the events in this broader category.

Literary events
The inequalities within the literary talks category are the most unevenly distributed by IMD decile of any activity addressed so far, with 21 times the number of transactions in the least deprived decile as in the most deprived decile. Crucially then the literary talk and the author reading, a crucial element of the consecration of literary value and a mechanism for supporting the literary field, is deeply socially exclusive. For those artforms that seek to have a broader public base and speak to a greater range of communities, our analysis shows the importance of fine-grained detail. The category "Other literature" is also striking but less extreme at 11 times, while poetry/spoken word is less extreme again at 3 times. Indeed, while the distribution within poetry and spoken word events shows inequalities more similar to other activities, the least deprived decile actually has fewer transactions than deciles 8 and 9, and around the same number as decile 7.
In this case, poetry and spoken word seem to be occupying a comparable position to contemporary dancemuch more evenly distributed across IMD deciles than other forms of literary events, although still markedly socially stratifiedwhile other forms of literary events have major differences between deciles. It is also noteworthy that for both literary talks and other literature there is a local peak at the sixth decile.
Next, we compare how attendance at these events differs within and outside Arts Council England's national portfolio, in Figure 7, for those activities corresponding to 97% of the transactions in this category. Unlike dance, the differences between literary events at Arts Council England-supported venues, and other venues, are much smaller; where there are differences, they skew towards events at NPOs being more socially stratified than events not at NPOs. 25% of transactions for literary talks at Arts Council-supported venues went to households in the least deprived 10% of areas; the equivalent figure for non-NPOs was 21%. This is even more striking for author readings, although the numbers of author readings outside of NPOs in Audience Finder is relatively small. Poetry and spoken word events seem to be a partial exception to this trend. Events in NPOs are reasonably evenlybalanced across IMD deciles, while less deprived areas tend to have more transactions. The exception within this category, though, is the top decile, where a relatively small fraction of tickets is sold in non-NPOs.

Conclusion
The "coming crisis" for cultural statistics seems, from this analysis, not to have materialised. This is not to say that the methods we use to understand culture are free from limitations and problematic assumptions. Neither does it ignore how the "social life" of these methods is an important subject for critical engagement and reflection by cultural researchers (O'Brien, 2013;Oman, 2016). Those points notwithstanding, by showing how a core form of social inequality, here geographic indicators of deprivation, exhibits similar patterns when asking who attends culture in England, we can see the continued relevance and accuracy of "traditional" survey approaches such as Taking Part and Active People. The need to understand differences within categories, along with the neglect of niche activity, is a longstanding problem for the study of culture. Here our analysis shows the power of new datasets that allow us to cut through the uncertainty associated with niche cultural interests in Taking Part. We have focused on two broad art forms that have received less focus in the quantitative sociology of culturedance, and literature eventsshowing that the social stratification within these activities varies substantially. While audiences for both traditional and contemporary ballet are similarly socially stratified, audiences for contemporary dance are drawn almost evenly across areas with higher and lower levels of deprivation. By contrast, audiences for literary events, particularly literary talks, are perhaps the event that is marked by the most striking social stratification. This is not a call for a withdrawal of focus from surveys, but a recognition that different data sources can serve different purposes, with transactional data being particularly useful for shedding light on activities that are hard to gain insight into from surveys.
For policy makers, there is much to consider from this analysis. In particular, what a comparison of artforms from the Audience Finder dataset suggests is that there may be valuable lessons on engagement and participation strategies across different artforms, rather than by asking organisations to look for best practice within their field. Moreover, it suggests NPO status is working to deliver "great art and culture for everyone" (ACE, 2013) by having an audience for dance and literary events that is more reflective of the general population that that of non-NPOs. However, at a more general level, our analysis suggests the continuing need for attention to the relationship between inequality and culture, in this case as understood using the geographic proxy of the Indices of Multiple Deprivation. Even with new datasets, and alternative criteria beyond the demographic categories of class, gender, disability, and race which are the usual means of understanding how culture is stratified, the same issues confront practitioners and policymakers alike.

Notes Funding
This work was supported by Arts and Humanities Research Council [grant number AH/S004483/1].

Notes on the contributors
Laurie Hanquinet is Senior Lecturer in Sociology at the University of York. Her work has focused on cultural institutions, socio-cultural inequalities, different dimensions of cultural participation and consumption, and on European identity. She is the co-editor of the "Routledge International Handbook of the Sociology of Art and Culture" (2015) with Mike Savage.
Dave O'Brien is Chancellor's Fellow in Cultural and Creative Industries at the University of Edinburgh. He has published extensively, as well as led and participated in a range of research projects, on cultural and creative industries.
Mark Taylor is Senior Lecturer in Quantitative Methods (Sociology) at the Sheffield Methods Institute, University of Sheffield, and is AHRC Leadership Fellow (Creative Economy) until 2021. His research interests are in the sociology of culture: in consumption, production, and education, and its relationship to inequality.