Cultural consumption and Covid-19: evidence from the Taking Part and COVID-19 Cultural Participation Monitor surveys

ABSTRACT How did cultural consumption change during the Covid-19 pandemic? Whilst the impact of the pandemic on cultural production has been given significant attention, work on consumption has seen less attention. This paper addresses this gap in the literature by presenting a comparative analysis of two, nationally representative, surveys of cultural activity in England. The analysis demonstrates that, when cultural consumption moved online and to digital modes of delivery and engagement as a result of the pandemic, there was no discernible transformation in the stratification of cultural participation in England. The majority of the population, characterised by the absence of participation in formal, and often state-funded, cultural forms, saw no change to their patterns of engagement. Where cultural consumption did increase, this was among the small minority of people who were already highly engaged. This minority maps closely onto pre-existing inequalities identified by existing research on cultural consumption, in England and beyond. For cultural consumption and the stratification of taste, it seems that the ‘new normal’ of pandemic life was much like the ‘old normal’ of an art and cultural audience characterised by significant inequality.

The global pandemic that began in 2020 saw terrible damage to societies across the world. The medical crisis brought by Covid-19 went hand in hand with efforts by states and communities to slow viral transmission. For many nations, these efforts were some form of lockdown, shelter in place, or stay at home orders. For many cultural organisations, this meant an immediate stop in both production and consumption. TV and film production was stopped, and galleries, concert venues, and theatres were closed.
Consumers faced a restricted choice of options on how to spend their time, with some experiencing a sharp increase in leisure time although others, due to caring responsibilities, experienced a sharp decline (Roberts, 2020). During the first weeks of 'life under lockdown' many popular accounts emerged of people taking up new cultural interests or enhancing proficiency in cultural practices at home, for example, cooking, drawing, or playing a musical instrument.
In the UK, the initial lockdown of March 2020 demonstrated the agility of much of the cultural and creative sector. The sector saw a proliferation of new modes of engagement with culture, including theatre productions delivered online, virtual museum tours, and livestreamed gigs from musicians' homes. Arts Council England's first emergency response package of £160 million, released on 24 March 2020, was primarily designed to ensure the financial survival of subsidised A substantial body of subsequent research, using a wide range of research methods, has found similar findings in a number of different countries (e.g. Bennett et al., 2009;Lena, 2019), and via cross-national comparisons (Falk & Falk, 2011;Falk et al., 2016;Katz-Gerro, 2017;Lagaert et al., 2018;Reeves, 2019;Van Hek & Kraaykamp, 2013). Extensive, often highly-technical, academic debates have taken place since the first edition of Distinction regarding whether and how Bourdieusian theories can be tested. Yet the connection between tastes (as measured by attendance at, participation in, knowledge of, and preferences for, a range of cultural forms), cultural hierarchies, and social inequalities is clear (see, Hanquinet, 2017 for an overview).
For a fuller understanding of the impact of the pandemic on cultural consumption, two additional issues are critical. First, in addition to understanding which variables matter for variation in cultural consumption, researchers must also reflect on how cultural consumption can be measured. Several studies focus on tastes for different musical genres (Bryson, 1996;Katz-Gerro, 1999;Peterson & Kern, 1996), while others address attendance at different events, although limiting analysis to what are sometimes termed 'highbrow' activities, such as classical music concerts, ballets, and visual art museums (Christin, 2012;Kracman, 1996).
Secondly, researchers must consider analytic strategy: whether measures of cultural taste or engagement are treated as distinct from one another, for example, by investigating the social stratification of attendance at different types of event separately; or by analysing such measures jointly, for example, by generating clusters, scales or factors. With regard to cluster analysis, people are grouped together on the basis of shared patterns of cultural consumption. In some cases, this clustering is applied to variables within a broad artform, such as responses to questions about attending different settings exhibiting visual art (Chan & Goldthorpe, 2007); in others, individuals are classified on the basis of their activity across art forms (for example, Bunting et al., 2008). Specific technical debates and competing definitions notwithstanding (de Vries & Reeves, 2021), research findings within this field are remarkably consistent in finding regularities in the social stratification of cultural consumption, regardless of the way that cultural consumption itself is measured.
One consequence of these two issues is that 'cultural consumption' ends up being treated as shorthand for 'the subsidised arts sector'. Descriptors such as 'nonparticipants' or 'non-consumers' are used to describe people who do, in fact, have active lives filled with active leisure, although those activities are not conventionally captured in social surveys or included in formal models (Taylor, 2016). The standard questions used in surveys may not make meaningful sense to those answering the questions, particularly people from minority groups (Novak-Leonard et al., 2015). As a result, the translation to policy can omit key nuances, with groups of 'nonparticipants' constructed as a problem to be solved (Stevenson, 2013). Given these issues, care needs to be taken to ensure that as broad a definition of culture as possible is used to understand changes in participation as a consequence of the pandemic. The publicly-funded sector is only a subset of people's pandemic leisure activities (Roberts, 2020).

The impact of digital
As with the study of the stratification of cultural consumption, there are analogous issues in the digital realm. Much of the research on cultural consumption has, arguably, reflected institutional arrangements and consumption practices premised on a relatively limited range subsidised cultural forms, attendance modes, and broadcasting technologies. The rise of a variety of digital forms of culture, both in terms of consumption and production, suggests a different context from the one shaping Distinction's analysis. Indeed, for many initial theorists of digital culture (e.g. Barlow, 1996), digital engagement and activity offered the possibility of dissolving some of the hierarchies and divisions associated with art and culture. In the English context, from which this paper's data is drawn, various public policy and associated thinktank activity has hailed the potential of digital to transform the cultural sector (Department for Digital, Culture, Media and Sports, 2018).
However, there is a research consensus on need for caution, if not outright scepticism, regarding the impact of digital culture on distinctions, differences, and divisions. These range from critics analysing the limitations of algorithmic recommendations by digital platforms such as Netflix, YouTube, and Spotify (e.g. Beer, 2013;Wright, 2011), through to more general critiques of algorithmic decision-making (Noble, 2018), user experiences online (Carmi, 2020), and personalisation technologies (Kant, 2020). These more general critiques mirror concerns over the impact of digitalisation on cultural workers (e.g. Christin, 2020;Patel, 2020) and social inequalities more generally (Helsper, 2020, Sobande, 2020a, 2020b. Indeed, for scholars of leisure (e.g. Lupton, 2016;Redhead, 2016;Silk et al., 2016), optimistic perceptions of the possibilities offered by digital, such as greater citizen influence over media industries (McGillivray, 2014;McGillivray et al., 2016), are tempered by recognition of the need to attend to the reproduction of existing, alongside the emergence of new, digital inequalities such as barriers to internet access (Ofcom, 2021).
Replication of existing inequalities is clearly demonstrated by Mihelj et al. (2019). Using the UK's Department for Digital, Culture, Media, and Sport's Taking Part survey of cultural consumption in England (the same dataset used for part of the analysis in this paper), Mihelj et al. find that, for museums and art galleries, digital modes of engagement mirror 'offline' behaviours; existing inequalities are replicated. Whilst digital activities engage new audiences, these audiences have the sorts of demographic characteristics that form the bulk of the existing audience for museums and galleries. Indeed, as they point out 'Belonging to higher occupational classes, having a degree, not having a disability or longstanding illness, and residing in the capital city all remain strong predictors of participation, while ethnicity has a strong negative effect. Most worryingly, the gaps between the haves and the have nots are even wider online than in the case of physical visits. This means that rather than helping increase the diversity of audiences, online access seems to reproduce, if not enlarge, existing inequalities'. Mihelj et al. (2019).
This pattern is neither confined to museums and galleries, nor to England. Using Swiss data, Weingartner (2021) found most forms of digital media, excluding television, followed the pattern of reproducing the sorts of stratification patterns, and thus inequalities, commonly found in the cultural consumption literature. In Spain, Montoro-Pons and Cuadrado-García (2020) identified continuing cultural hierarchies in music consumption despite the use of seemingly democratising means of digital engagement.
What, then, should we expect in terms of change to cultural consumption patterns in the context of the Covid-19 pandemic? The unavailability of many in-person cultural and leisure activities, both highbrow and popular, can be expected to drive many consumers to enhance their activity within the limited selection of activities that are still accessible, such as reading and TV watching. They may also exploit the opportunity to take on new activities that were not pursued before, such as art creation or exercise.
Additionally, policy and practitioner voices have faith in digital to engage new audiences and enhance the consumption of existing, already engaged individuals (Department for Digital, Culture, Media and Sports [DCMS], 2018;Ofcom, 2021). Therefore, we could also expect to see a wide increase in online cultural activity. This would be driven by digital newcomers joining previously active consumers who are transferring their consumption habits from the physical to the online realm.
On the other hand, the academic research base suggests we should not expect to see dramatic changes in cultural consumption patterns. Rather we should expect the replication, and perhaps acceleration, of the inequalities and stratification patterns found in what Mihelj et al. (2019) call the 'off-line' cultural world. Cultural lifestyles are embedded behaviours that are not likely to shift abruptly due to changes in external conditions such as those brought by the pandemic. The adoption of new cultural behaviours may be more likely to happen among those already active, while remaining inaccessible to individuals with lower levels of pre-existing cultural engagements. Moreover, online fatigue and increased external constraints aggravated by the pandemic, such as parenting, health, or work, may lead to a reduction of cultural and leisure activity. These constraints are likely to affect those who are already less culturally active, thus exacerbating existing inequalities.
We study these questions empirically by looking at physical and online cultural and leisure activities that were available both before and during the pandemic. The following section describes how we analysed data on changes in consumption of these items between pre-pandemic and pandemic periods, in order to identify patterns of change and resilience in cultural consumption and their corresponding social profiles.

Data
In order to interrogate the impact of the pandemic, the paper uses two data sources: the DCMS Taking Part Web Panel COVID-19, and the Audience Agency COVID-19 Cultural Participation Monitor surveys. The analysis of the two datasets in parallel, helps us to strengthen the validity of our results and serves as a robustness check to the models presented below. The datasets do not have identical scopes, and each contains a slightly different set of questions. Due to the relative complexity of the method and data structure, this difference is beneficial in confirming that our results have general applicability.
The Taking Part Web Panel COVID-19 survey (TPCOVID) was conducted in three waves in May, June and July 2020 (DCMS, 2018). At the conclusion of their interviews for the standard faceto-face surveys, respondents to the Taking Part regular survey are asked if they are willing to participate in further surveys online in subsequent quarters. Our respondents here are drawn from this pool. The survey contains around 1000 respondents in each of the three waves. However, for the analyses in this paper, we pooled all the respondents together. Most of the respondents participated only in one wave of the survey. For respondents that participated in more than one wave, we kept only a single observation that was randomly selected, so as to not include any individuals more than once. The final data file contains 2175 distinct respondents.
We focus on responses to questions about engagement with 41 different activities, where respondents were furthermore asked about the amount of time they have spent on them since the introduction of restrictions by the UK Government in March 2020. These activities cover a wide range, including traditional highbrow cultural activities often studied as part of research into cultural consumption, popular cultural activities, and general leisure activities such as physical exercise or online calls with friends and family. The available response options are 'less time', 'about the same', 'more time', and 'I never do this'. We acknowledge that these categories are imprecise with regards to the magnitude of change, however they capture the most significant aspect of it -its direction (positive or negative). This allows us to investigate the change in respondents' cultural consumption occurring since the start of the pandemic. Table A1 in the online appendix summarises the set of activities by change in frequency of engagement.
Second, the Audience Agency COVID-19 Cultural Participation Monitor survey has been conducted via several waves over the course of the pandemic. Here, we report data collected in the second wave of the survey (TAAW2), which includes questions similar to those drawn from TPCOVID. For TAAW2, a sample of 1,503 respondents, representative of the UK population, was drawn on February 2021.
We analysed questions indicating whether they had done any of a list of 42 leisure activities during the pandemic. Where they provided a positive response for a particular activity, the respondents were also asked about their frequency of engagement. Response options comprised 'fewer times', 'same', 'more times', or 'I did this for the first time during the pandemic'. The category 'I did this for the first time during the pandemic' was, in most cases, very small. These categories were re-coded in order to match those of the TPCOVID data and eliminate inconsistencies in the responses (see figure A1 in the online appendix). TAAW2 also included 15 questions about engagement in in-person activities, such as going to a dance performance or a museum during the pandemic. Since these activities were mostly unavailable at the time, we did not use these questions in our analysis save for those pertaining to outdoor activities (e.g. visiting a historic park or garden) where analogues were available in Taking Part. Table A2 in the online appendix presents the activities and the frequencies of the different responses from TAAW2 after this recoding.
Although the questions that we analysed from both data sources were similar in many respects, they still possess significant differences that encouraged us to examine each dataset as complementary but individually-valuable sources. First, the activities asked about in the two surveys are not phrased in the same way, and many appear on only one of the surveys. Secondly, the TPCOVID questionnaire asks respondents to compare the amount of time dedicated to leisure activities during an unspecified period before the pandemic, whereas TAAW2 questionnaire asks about the number of times in a specific period of one year before the pandemic. Thirdly, TAAW2 respondents were asked about their change in an activity only if they indicated earlier that they had done this activity during the pandemic, whereas all TPCOVID respondents were asked to indicate change in level of activity for all items. Crucially, the data collection for the two surveys also took place at different time periods (summer 2020 and February 2021), when restrictions on activity in England were significantly different. While the amount of in-person cultural activity was similarly limited in both, TPCOVID took place in a context where restrictions were being lifted, while TAAW2 survey took place while extensive restrictions had been re-imposed.

Analysis
We ran the same analytic procedure on both data sources. First, we ran a hierarchical cluster analysis of the cultural consumption variables (41/30, depending on the dataset) as a dimension reduction exercise, in order to construct scales for analysis. Since both datasets contain subsets of related activities, the subsequent analysis steps would be affected by the size of those sets and may assign more weight to larger groups of related items. We ran the clustering analysis in the initial step to reduce that bias and ensure that each type of activity gets a similar weight in the following steps. Clustering of cultural activities using statistical methods such as Factor Analysis or Hierarchical Cluster Analysis prior to the analysis of consumption is a common practice in cultural consumption research (Leguina & Miles, 2017;Taylor, 2016;Van Eijck, 2001). In this study, we employ a hierarchical cluster analysis that is suited to handle categorical unordered data by using the ClustOfVar R package (Chavent et al., 2012).
Second, we ran latent class analysis (LCA) on these scales, in order to identify clusters of individuals with similar changes (or lack of change) in their cultural consumption by using the poLCA R package (Linzer & Lewis, 2011). The use of the LCA methodology in the paper is built on the premise that cultural lifestyles are represented in observable patterns of cultural consumption. LCA can specify these groups (classes) by identifying individuals who share the same cultural consumption patterns. For that reason, LCA is often used in cultural and leisure consumption research (e.g. Alderson et al., 2007;Chan & Goldthorpe, 2007;Widdop & Cutts, 2013). In the context of our research, the LCA method aims to identify changes in broad consumption patterns during the pandemic. We are interested in discerning between those that changed the volume of their cultural and leisure consumption during the pandemic, negatively and positively, and those whose consumption modes were not affected by the pandemic situation. Since the initial cultural activities clustering procedure yields several groups of distinct cultural activities, the LCA is more parsimonious and prevents the necessity of running a separate model for each group of activities that would result in a fragmented set of findings. Significant differences in consumers profiles between groups of cultural items will show up as separate latent classes. Since the data only indicates the direction of the consumption change (i.e. less, more, same), we cannot relate to the difference as a continuous scale but rather as categories of change.
Third, we used multinomial regression, with these latent classes as dependent variables, to test whether those variables associated with cultural consumption outside of a pandemic context are equally found to be associated with change in cultural consumption following the introduction of pandemic-related public health restrictions.

Taking Part (TPCOVID) data
We conducted a hierarchical cluster analysis of the activity items in the TPCOVID (see Figure A2 in the online appendix). Employing an adjusted Rand criterion on the resulting hierarchical tree suggested items segmented into 9 clusters of activities. Ranked from the clusters containing more common activities to the most infrequent, we label them as follows: Cluster 1 -TV, Cluster 2 -Outdoor activity, Cluster 3 -Media reading and listening, Cluster 4 -Sports watching, Cluster 5 -Online art and performance, Cluster 6 -Exercise, Cluster 7 -Video games, Cluster 8 -Online heritage and library, Cluster 9 -Creative activities. Table A3 in the online appendix presents the resulting clusters, assigned with labels according to their content. Items allocated to the first cluster all involved cultural creation: writing, performing music, drawing and painting and so on at home. TV watching of different forms -via synchronous broadcasts, or free or subscription-based on demand services -formed cluster 1. Cluster 3 grouped radio-listening with listening to podcasts and audiobooks, alongside items relating to reading books and magazines.
Since the pandemic encouraged seeking out online consumption modes and increased the offering of live online events, we might have expected that clustering of variables capturing change in consumption might yield clusters based on medium of consumption such as online activities, live-streamed events and so on. However, the cluster analysis demonstrated the importance of content over medium. For example, change in watching live music, dance or art events were clustered (i.e. correlated more strongly) with change in watching pre-recorded music, dance or art events cluster 5) rather than watching live TV (cluster 1) or participating in a live exercise or dance class (cluster 6). Cluster 2 encompassed visiting particular sites for exercise, such as parks, historic gardens, places of historic or archaeological interest. Cluster 8 captured taking virtual tours of museums and galleries and sites of historic interest, as well as engagement with historical research online and use of online library services. Cluster 7 grouped together different measures of video gaming, including online betting, and cluster 4 watching pre-recorded sports and eSports events online. The clusters suggest that reported change in engagement were primarily shaped by pre-existing cultural tastes rather than by new consumption opportunities provided as a result of the pandemic, both in terms of availability of consumers' time and the works and resources made newly-available online.
We next conducted a latent class analysis to identify classes of respondents that share similar patterns of change in their leisure activities. We used the clusters stemming from the hierarchical cluster analysis above to generate indicators of change in consumption. We did so by assigning each respondent the modal value regarding change in consumption for the items within each cluster. For example, a respondent that increased her viewing of live arts event and live music or dance performance online, while maintaining the same level of watching pre-recorded arts events and also reducing watching pre-recorded performances, would be classified as increasing consumption of cluster 5 activities. In the case of a multimodal distribution (more than a single mode value) we used the following criterion: if one of the values is 'I never do this', we used is the other value, that is indicating change as the mode, and if both values indicate change, we used the lower value (e.g. 'less' over 'same'). The number of incidents where two of the modes values were 'more' and 'less' at the same time is negligible (for example, 8 out of 2175 respondents in cluster 5 of TPCOVID data). Both BIC and cAIC criteria pointed to a four-class solution as best fitting the data (see Figure A3 online appendix). Figure 1 displays the four classes graphically, summarising the average reported activity pattern for each of the 9 clusters for these four classes in turn.
The first class, consisting of 14% of the sample, is characterised primarily by intensification of activity. The second class (18% of the sample) is characterised by maintaining similar rates of activity both before and during the pandemic: they demonstrate resilience of engagement to the shock. The third class (5% of the sample) is characterised by curtailment of leisure and cultural activity during the pandemic. The fourth class is the largest, consisting of 63% of the sample. Respondents here exhibit low rates of activity overall, and are most likely to not engage across the range of activities asked about, whether before or during the pandemic. The exception for this group was TV watching, which they were likely to increase or maintain during the pandemic, and outdoor activities, which they decreased. Accordingly, we termed them 'Enhancers', 'Resilient', 'Reducers' and 'Non-engaged'.
For the next stage of the analysis, we estimated a multinomial logistic regression model to examine the determinants influencing likelihood of being an 'Enhancer', 'Resilient', 'Reducers' or 'Non-engaged' cultural consumer during the pandemic. The dependent variable in the regression was class membership. We used a set of individual variables included in the data and known from the literature to affect cultural consumption and leisure activity as independent variables: • Sex -male/female • Ethnicity -White/BAME 1 • Age group -16-34/35-45/55 and over • Work status -Not working/Working/Stopped working • Parenting -Having children under 16 currently living in your household (yes/no) • Region -London/North/Midlands/South • Wave -May/June/July.
Notwithstanding the importance of income and social status for cultural consumption patterns, such variables were not collected in the TPCOVID19 data, and are unavailable for the respondents.

The audience agency (TAAW2) data
We first proceeded with a similar hierarchical cluster analysis of TAAW2 items (see Figure A4 in the online appendix). This source differs in its coverage of cultural engagement, with more detailed measures of consumption of livestreams or recordings of the performing arts, alongside measures of cultural production at home. However, its wide range of measures does allow us to provide a complementary analysis, one which again takes the breadth of potential engagement options seriously. In this case, according to the adjusted Rand criterion, the hierarchical cluster analysis resulted in a partition of 5 clusters of activities which we labelled: Cluster 1 -Outdoor, Cluster 2 -Reading, Cluster 3 -Other online,Cluster 4 -Online culture and art, Cluster 5 -Creative Activity. As before, these are ordered from the more common activities, to the most infrequent. Table A4 in the online appendix summarises these together. Due to the large number of activities surveyed, engaging only a small proportion of the respondents where the sample size was furthermore lower, the clustering proved to be less nuanced than for the TPCOVID source. Nevertheless, clusters identifying distinct segments of activities relating to the respondent's own creative activity, outdoor activity, and reading were replicated. The remaining online activities were clustered into two groups: culture and arts activities (including highbrow genres such as opera and contemporary dance, and more popular forms such as music gigs and family arts events), and 'other activities'a group likely to include activities such as watching films or standup comedy alongside others that were not asked about directly.
We again conducted a latent class analysis, treating the five clusters identified in the previous stage as manifest variables. Our procedure of selecting the number of classes was driven here by the attempt to find a parallel structure to that we found in the TPCOVID data that resulted in 4 classes. However, probably due to the smaller sample size, the imposed four-class solution failed to distinguish respondents characterised by being active 'fewer times' and those active at 'the same' rate. The four-class solution accordingly lacked face validity. Imposing a five-class solution yielded a more clear-cut separation between those reducing their activity and those maintaining it, while still displaying reasonable fit indices (see Figure A5 online appendix). This solution was more theoretically plausible and we accordingly retained in preference to the four-class solution. Figure 2 presents these five classes and the average change in engagement frequency for each cluster within each class.
Three of the classes share characteristics parallel to those identified in the TPCOVID dataset and were accordingly also termed 'Enhancers', 'Reducers' and 'Resilient'. These classes are very small, together comprising 7% of the sample. The majority belongs to one of the two classes characterised by very infrequent engagement with the activities asked about in the survey. The two classes differ in distinguishing those likely to do outdoor activities before the pandemic who had decreased engagement following its start, and those who were non-engaged outdoors even before the pandemic. For the purpose of this paper, we termed them 'Non-engaged hikers' and 'Non-engaged' accordingly. The difference in the overall size of the non-engaged groups in TAAW2 compared with TPCOVID might be largely explained by the fact that TAAW2asked about participation in the 12month period before the pandemic, while TPCOVID asked more generally about engagement in the past.
We then estimated a multinomial logistic regression model, similar to that for the TPCOVID data in including as similar a set as possible of explanatory variables. The dependent variable in the model was again class membership, with independent variables comprising: • Free time -The impact of Covid-19 on the respondent's amount of free time -more/ same/ less (included as a close substitute for the work status variable in TPCOVID) • Parenting -Having children under 16 currently living in your household -Yes/No • Region -London/North/Midlands/South/Scotland/Wales (Northern Ireland was omitted due to technical issues with the data) • Health -Whether the respondent is limited by a health problem or disability -No/ Yes (where 'a little' and 'a lot' were merged to create a binary variable)

Taking part (TPCOVID) data
Results of our second multinomial logistic regression model are presented in Table 1. The baseline comparison category in the model is the group of 'Resilient' consumers that showed continuity in their consumption patterns. Since the model presents only comparisons with the 'non-engaged' baseline category, we ran additional post-estimation tests that included comparisons between the other clusters too (see Figure A6 in the appendix). The findings show the stratification axes shaping change in leisure and cultural consumption in response to the pandemic. We found that women were more likely to be 'Enhancers' than 'Resilient' compared to men. On the other hand, among the two groups that did not change their consumption patterns during the pandemic, we found women were more likely to be non-engaged than resilient active consumers. Note: Standard errors in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001 Ethnicity plays a role too in determining consumption patterns. Notwithstanding the relatively small sample of ethnic minorities in the data, we find that members of ethnic minorities had a statistically significant higher propensity than White respondents to be 'less active' during the pandemic compared with membership of all other groups. Controlling for all the other covariates, ethnic minority respondents were more than twice more likely to be a 'Reducer' than an 'Enhancer' or 'Resilient'. However, they were also less likely to be a 'Non-engaged' compared to all other categories (see online appendix).
Most parents, especially those with younger children, had to care for their children more than usual during the lockdowns which may have affected cultural consumption in several ways. On the one hand, taking care of children consumes free time, but it is also plausible that parents will devote increased time spent with their children to cultural and leisure activities. We found that respondents with children in their household were more likely to be a 'Reducer' than members of any other category (see online appendix). However, these differences were not statistically significant.
Age had an intricate effect on the change in consumption patterns. Older respondents, compared to the 35-54 age group, had a much lower propensity (about a third) of increasing their leisure and cultural consumption levels during the pandemic. Younger respondents were more likely to see a change in their consumption pattern during the pandemic than being 'Resilient'. However, there was no statistically significant difference between their propensity of being a 'reducer' or 'Enhancer' (see online appendix). Younger respondents were also more likely be non-engaged than resilient. We can sum this up by saying that while young respondents were more likely to see change in their consumption level over the course of the pandemic conditional on being an active consumer, older respondents were more likely to sustain their consumption level or reduce it than increase it (see online appendix).
Another pandemic-related factor linked to change in time availability is the work status of the respondents. Respondents that were not working both before and during the pandemic were most likely to belong to the non-engaged group, although this difference was only statistically significant compared to the resilient group. However, respondents that had stopped working since the start of the pandemic showed a clear difference compared to working and non-working respondents. Those who stopped working were more likely to increase their level of consumption ('Enhancers'), although we found no significant difference between the propensity to increase or decrease that level. Lastly, in most cases, the geographical location did not have a significant effect on consumption changes. However, between the two non-change groups, residents of the north of the UK were more likely to be 'Non-engaged' rather than 'Resilient', reflecting lower levels of cultural participation in this region in place even before the pandemic.

The audience agency (TAAW2) data
We estimated a similar multinomial logistic model to the one estimated for TPCOVID using TAAW2. Setting aside differences in what was asked, the smaller sample size and smaller sizes of the three active consumption classes was an additional constraint on the power of the model. Nevertheless, we wished to test whether the model's results are generally comparable with those of the previous model. Results are presented in Table 2, with additional comparisons reported in Figure A7 in the online appendix.
Similar to what was found in the TPCOVID data, results here indicate that women are more likely to be non-engagers of both types (previously-active outdoors and previously-non-engaged) than resilient engagers. In this model, we additionally found women more likely to be enhancers than resilient as we had found in the TPCOVID model. In the TAAW2 model, parents are more likely to be 'Reducers' than non-engaged (both types) and 'Enhancers', a finding consistent with that for the TPCOVID model (although it was not statistically-significant there), which underscores parents' struggles during the pandemic (see online appendix). In the TAAW2 model, we did not find a significant effect for respondent ethnicity except for a higher propensity of those other than White to belong to the 'Non-engaged' category compared to the 'Non-engaged/previously hikers' group. These results are very likely affected by the small proportion of ethnic minorities in the data. The comparison with older did nor yield statistically-significant differences in this model, except for a higher propensity of belonging to the 'Non-engaged/previously hikers' group compared to most other categories (see online appendix). Younger respondents were significantly more likely to be 'Resilient' than 'Non-engaged/previously hikers', which is compatible with our findings in the TPCOVID data.
One advantage of the TAAW2 dataset is its inclusion of measures of self-reported health. Health problems and disabilities are usually linked to consumption barriers. We might therefore expect that during the pandemic consumption differences between those with poorer health and others were attenuated, and that the increase in online content might benefit those traditionally-excluded. Results (see online appendix) suggest that respondents with health issues were more likely during the pandemic to either reduce their consumption (compared to being a non-engaged of both types) or increase it (compared to non-engaged).
Finally, we can examine the effect of increased time availability as a resource for consumption. As found in the TPCOVID model (working status variable), having more free time is associated with change. Respondents that reported having more free time were more likely to increase their level of consumption compared to the two non-engaged categories. Standard errors in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001

Conclusion: cultural consumption and the Covid pandemic
Writing during the summer of 2020, as re-opening began in the UK (and across many other parts of Europe) Roberts (2020, p. 622) noted that 'the adjustments that people made during lockdown when most out-of-home leisure became inaccessible can be summarised as increasing the amounts of time spent doing things at home that they had done before'. Using time use data, Roberts demonstrated the continuities of leisure practices, rather than a wholesale revolution following the lockdown of March 2020. Drawing on surveys focused on cultural consumption, our analysis expands on this conclusion. We find that the prevalence of low and non-engagement with cultural consumption was not altered by the pandemic. A majority of respondents reported no involvement with almost any form of cultural consumption other than watching TV and other popular media. The social profile of this group is similar to the known characteristics of low cultural engagement consumers. An additional significant proportion of the population was able to retain their level of cultural consumption through the pandemic but, like the non-engaged, mostly did not exploit the new access opportunities brought by the pivot to digital.
Increasing and diversifying cultural consumption during the pandemic was associated mostly with young age groups and was contingent on having spare time. External constraints such as work and parenting reduced the chances of enhancing consumption. These were also among the risk factors of lowering cultural consumption during the pandemic, along with geographical and health inequalities.
Accordingly, we come to a similar conclusion to Mihelj et al.'s (2019) analysis of Taking Part, that finds that the 'offline' patterns of inequality continued into the predominantly digital cultural world of the pandemic. We can see the inequalities associated with race, gender, and age are all still influential on pandemic cultural consumption. There is also striking evidence of the impact of parenting on the change, or lack thereof, in cultural behaviour. Finally, the continuities in the stratification of cultural consumption during the pandemic raises a range of broad questions as to the value of culture and the likelihood of a transformative impact of the pandemic on key aspects of contemporary culture.
Having two sets of data allows for fruitful comparisons, each offering a distinctive perspective with its own strengths and weaknesses. What is striking is the consistency of the story being told by the analysis of both Taking Part and The Audience Agency survey data, collected at different stages of the pandemic. This consistency between the data sets mirrors the consistency of 'offline' and digital pandemic stratification, whereby the patterns of inequalities demonstrated by the existing literature were not disturbed even by such a major, global, social change.
There are, obviously, limitations to the analysis and the two datasets from which the research is drawn. The demographic variables available in the web panel version of Taking Part are limited. This absence makes it impossible to interrogate the role of social class and education in the social stratification of cultural consumption in this digital context. In addition, it is important to reinforce that our picture of digital participation is limited to the specific digital activities that the surveys ask about. While both surveys draw on a wide range of different activities, and our conclusions are not limited to the publicly-subsidised arts sector, we do not have information on genre within the activities in which people were engaging. We do not know, for example, whether people who increased the number of hours they spent watching on-demand streaming services spent time watching multiple episodes of critically acclaimed dramas in different languages, or if they revisited shows they had already watched multiple times.
These issues notwithstanding, the analysis has important implications for leisure studies and cultural sociology scholars, as well as for policymakers and practitioners. For researchers it reinforces the continuities of stratification and inequality in relation to culture, as well as the need for more detailed and more sophisticated approaches to 'non-engagement'. For policy and practice, as the comments that opened the paper indicated, the struggle for a reformed or reimagined cultural sector continues. The pandemic has, based on our analysis, not disrupted the arts audience, whether in terms of bringing in new engagement or in terms of key demographics of existing consumers. 'Digital' will not be enough to change the arts; rather, the core lesson of the pandemic is the need for real vision as to what a representative and fair cultural sector would look like in order to reach a truly representative, and thus diverse, audience. Note 1. We acknowledge the limitation of the BAME category and the fact that it represents an heterogenous group.
However, the sample size does not allow us a finer distinction between ethnic groups. Therefore we use of the BAME category congruently with other research works in the field (e.g. Oakley & O'Brien, 2015;Roberts, 2020).

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
This work was supported by the Arts and Humanities Research Council [AH/V00994X/1].

Notes on contributors
Tal Feder is a post-doctoral fellow at the Faculty of Architecture and Town Planning at the Technion -Israel Institute of Technology. Prior to this, he was a post-doctoral fellow at Indiana University and the University of Sheffield. He is a sociologist with interests in cultural policy, sociology of art and culture, inequality, consumption, and quantitative research methods. His current research studies cultural justice and cultural equality from a spatial perspective.