The Unequal Effects of the COVID-19 Pandemic on Political Interest Representation

The COVID-19 pandemic is viewed by many as the biggest global crisis since WWII and had profound effects on the daily lives of people and decision-making worldwide. Using the pandemic as a system-wide agenda shock, we employ a difference-in-differences design to estimate its causal effects on inequalities in political access, and social media prominence among business interests and NGOs. Our argument is twofold. First, the urgency and uncertainty of crises incentivized decision-makers to privilege providing access to business groups over securing inclusivity in the types of interests consulted. Second, NGOs compensated by increasing prominence in public communications. Our analysis of data from over 10,000 interest groups from over 100 countries registered in the European Union supports these hypotheses. Business interests successfully capitalized on the crisis in insider access, while NGOs increased prominence on social media. The results have wider implications for understanding how large-scale crises affect inequalities in representation. Supplementary Information The online version contains supplementary material available at 10.1007/s11109-022-09842-x.

A The composition of registered interest groups over time As noted in the main article, the number of business interests registered in the EU Transparency Register is larger than that of NGOs. Furthermore, in general, there are more business interests added each month to the registry than there are NGOs. To show this, we present in Figure A1 the number of interest groups from each group type registered with the EU over time. As the figure shows, the growth in the number of registered companies and businesses is outpacing that of its NGO counterpart. Growth in the number of business interests and NGOs is roughly proportional to size, as suggested by the parallel lines when these data are presented on the log scale in Figure A2. These differences in growth likely partly explain why there is a decreasing gap over time in the average number of meetings that business interests have with EU policy-makers over time relative to NGOs in Figure 2 in the article. The number of meetings that policy-makers have with business interests, in other words, has not kept pace with the growth in the number that register as lobbyists with the EU.

B Interest group type definitions and examples
In the main article, we examine interest groups defined as "Business interests" and "NGOs".
In the Transparency Register, each interest group is classified internally as belonging to one of fifteen sub-groups. These sub-groups classifications are themselves selected by each interest group when they register as a lobbyist with the EU. The classification of each interest group, therefore, is defined by the group itself, although subject to checks by the Registry secretariat. To examine differences in business interests and NGOs, we therefore collapse the relevant smaller categories into larger ones that define "Business interests" and "NGOs".
Our definition, based on these sub-categories, is presented in Table B1. 1 NGOs & identity groups Non-governmental organisations, platforms and networks and similar Organisations representing churches and religious communities Interest group sub-categories not included in these larger groups are "Academic institutions", "Other public or mixed entities created by law whose purpose is to act in the public interest", 'Trade unions and professional associations", "Other sub-national public authorities", "Regional structures", "Think tanks and research institutions", "Transnational associations and networks of public regional or other sub-national authorities", and "Other organisations".

C Timing of the pandemic
In the article, we code March, 2020 as the beginning of the pandemic. We do so because, first, March is the month in which the World Health Organization declared the pandemic as such (March 11) and, second, March clearly marks the start of widespread governmental responses to the pandemic across the EU, with restrictions on social and economic activities.
To show the latter empirically, we use data from the Oxford University Blavatnik School of Government's "Coronavirus Government Response Tracker" (Blavatnik School of Government, 2021). We aggregate the Tracker's "Stringency Index"-a measure of the intensity of government regulations to combat the pandemic-at the level of the EU, and present this measure graphically in Figure C3. As the figure makes clear, widespread governmental responses within the EU ramped up heavily in March, close to the WHO's declaration of the crisis as a pandemic.

D Examination of pre-treatment parallel trends
Difference-in-differences models rely on an assumption of parallel trends: that prior to an intervention, the outcome variable for the groups of interest move in sync and that, counterfactually, these trends would continue in parallel were it not for the intervention of interest.
This counterfactual is, by definition, unknowable. However, it useful to examine whether there are parallel trends in the pre-intervention period: doing so does not provide direct evidence that trends in outcomes would have evolved similarly between groups in the absence of an intervention, but it provides indirect evidence that this assumption is likely reasonable (Cunningham, 2021).
To examine this empirically, we fit difference-in-differences models that include time period leads, such that we calculate separate difference-in-difference estimates for each month prior to the pandemic (Angrist and Pischke, 2009;Cunningham, 2021). If the assumption of parallel trends holds, we should observe no systematic difference in the differences between NGOs and business interests month-over-month prior the pandemic.
We include leads in a baseline difference-in-differences model, and one that is more flexible with respective to time trends through the inclusion of additional interest group-level time trends. More formally, our estimating equations are the following: where y it is the outcome variable for group i in month t; δ i and φ t are interest group and month fixed effects; and λ i (Equation D2) are interest group-level time trends. In these models, our parameters of interest are β t , which capture the differences in differences between NGOs and business interests per month prior to the pandemic. Estimating separate β t per month prior to the pandemic allows us to compare whether the per-month differences between NGOs and business interests differ from each other relative to a baseline month, chosen here as the month immediately prior to the pandemic. If trends between NGOs and business interests are parallel, we should observe no meaningful differences across the range of the estimates of β t . As noted above and as shown in Equation D1 and Equation D2, we fit these models both with and without group-level time trends, the latter of which flexibly accounts for trends among each interest group in the number of meetings or social media posts over time.
Results from the model for the number of meetings with policy-makers are presented in Figure D4. As Panel A shows, there is evidence that, in the pre-pandemic period, NGOs had less access to meetings with policy-makers (relative to business interests) as compared to later months. Estimates from the first months of the data, for example, show significant differences in access to policy-makers of NGOs relative to business interests that were larger relative to the baseline month immediately prior to the onset of the pandemic. This is also observable descriptively in Figure 1 from the main article, in which the gap between the average number of meetings between NGOs and business interests is decreasing over time. In other words, there appear to be deviations from parallel trends. We can adjust for this, however, by including interest group-specific time trends (Angrist and Pischke, 2009;Cunningham, 2021), as in the model specified in Equation D2. Accordingly, Panel B of Figure D4 presents results for pre-pandemic difference-in-differences from the model with interest group-specific time trends. As Panel B shows, the inclusion of these time trends results in pre-pandemic differences that show no clear changes month-over-month. As we note in the main article, we therefore use as our model for the political meetings data one that includes interest group-specific time trends to flexibly adjust for these differences over time.
We conducted similar tests for the model fit to the data on the number of tweets sent by NGOs and business interests in the pre-pandemic period. Results from these models are This figure presents estimates of per-month differences between the number of tweets sent by NGOs relative to business interests prior to the pandemic, where the baseline for comparison is t = 0 (i.e. February, 2020).
presented in Figure D5. Unlike with the political meetings data, in Panel A of Figure D5, we see no systematic differences in trends that suggest an absence of parallel trends. In Panel B, which presents estimates for a model with interest group-specific time trends, we also observe no clear pattern. Indeed, estimates in both panels are extremely similar. The more flexible model that includes interest group-level trends, in other words, is performing minimal adjustment. In the main article we include interest group-level time trends when investigating the effect of the pandemic on differences in posting behavior about NGOs and business interests. However, as is consistent with the results in both panels of Figure D5, the results are effectively equivalent in models that do no include interest group-level time trends (not shown).

E Regression results from event study model
In Figure 4 in the main article, we show graphically the results of an event study model specified as follows: where y it denotes the outcome variable for group i in month t; δ i and φ t are interest group and month fixed effects; and λ i are interest group-level time trends. As we note in the Research Design section of the article, the set of parameters, β t , capture differences in the outcome variable per month after onset of the pandemic (t ∈ {1, 2, . . . , 7}) relative to the time period prior to the pandemic (t ∈ {−13, −12, . . . , 0}). In Table E2, we present the relevant regression table, where each parameter represents the difference-in-differences for NGOs relative to business interests in a given month. As shown in Figure 4 in the main article, these parameters demonstrate the dynamics of the effect over time.

F Differential effect of the pandemic on high-and lowresource interest groups
In the main article, we discuss findings concerning the pandemic's differential effects on access to policy-makers and social media among interest groups with 'high' resources (upper tercile) and 'low' resources (lower terciles). We present the complete regression tables in Table F3.
As shown in Model (1), we find no evidence of a differential effect of resources on interest groups' access to policy-makers in general (p = 0.48). In other words, when pooling data from NGOs and business interests, we find no difference in access to policy-makers among interest groups with low and high levels of resources in general.
In Models (2) and (3), we examine the role of resources within interest group types (NGOs and business interests). In Model (2), we find that the pandemic caused an increase in access to policy-makers among business interests with higher resources relative to business interests with lower resources. In Model (3), we find that among NGOs, the pandemic caused a decrease in access to policy-makers among high-resource interest groups relative to lowresource interest groups. This result can be viewed in light of the fact that low-resource NGOs obtain very few meetings with policy-makers to begin with. High-resource NGOs, in other words, became more similar to low-resource NGOs; high-resource business interests, by contrast, gained even greater access relative to their low-resource counterparts.
Finally, we examine the overall role of resources on social media activity in Models 4-6.
We find no evidence that the pandemic differentially caused differences in the frequency of posting among business interests and NGOs collectively (Model (1)), or whether comparing high-resource and low-resource interest groups among business interests (Model (2)) and NGOs separately (Model (3)).

F.1 Sensitivity to alternative codings of 'high' and 'low' resources
As noted in the article and above, we define interest groups with 'high' resources as those in the upper tercile (above the 66.6th percentile) of all business interests and NGOs, and those with 'low' resources, those interest groups in the bottom two terciles. This choice, however, is nevertheless relatively arbitrary. To test the extent to which the results concerning resources above are sensitive to the coding of groups with 'high' and 'low' resources, we recode these 'high' and 'low' resource groups at different cut-offs and re-estimate the models in Table F3.
We first code 'high' resource group as those above the median (in the upper 50th percentile), and those in the 'low' resource group as those below the median. We then fit each of the six models shown in Table F3 with 'high' and 'low' resources defined as such. These models estimate the differential effect of the pandemic on access to EU policy-makers and tweet frequency between 'high' and 'low' resource group among (1) all interest groups, (2) businesses specifically, and (3) NGOs specifically. We then recode 'high' and 'low' resources at the 51st percentile, and refit the models. We estimate these models with resources defined from the median to 90th percentile by 1 percentile increments, to capture an wide range of possible codings.
The results are presented in Figure F6. Each panel presents point estimates and 95% confidence intervals for difference-in-differences models fit to the two main outcomes for the full dataset and data from businesses and NGOs specifically. The top panels correspond to the Models (1), (2), and (3) in Table F3 respectively; the bottom panels, Models (4), (5), and (6). As the figure shows, the results from Table F3 are generally insensitive to how 'high' and 'low' resources are coded. For estimates from Table F3 that are not significantly different from zero, the estimates are also not different from zero for estimates for any coding of resources across the full range of cut-offs (top-left panel of Figure F6, and bottom row). For the estimates of the pandemic's effect on differential access to policy-makers within business interests and within NGOs, the estimates are significantly different from zero across nearly the full range of resource codings (second and third panels in Panel A), as consistent with Models (2) and (3) in Table F3. In sum, the results in Table F3 are not an artifact of how interest groups are coded as having 'high' and 'low' resources.
F.2 Sensitivity analysis of results to comparison of the richest (upper quartile) and poorest (lower quartile) interest groups Above, we examined the sensitivity of the results concerning resources to differences codings of 'high' and 'low' resources across a wide range of cut-offs. Here, we also test whether the pandemic affected interest group access to policy-makers and social media behavior when comparing the highest-resource interest groups to the lower-resource interest groups. To do so, we subset the data to include only interest groups in the lower quartile of resources (defined as 'low') and those in the upper quartile (defined as 'high'). Using these data, we then fit the same models as included in Table F3. Results are presented in Table F4 and are effectively equivalent to those in Table F3: all point estimates are similar and are similarly statistically (in)significant to those in Table F3. This figure shows the estimated effect of the pandemic on differences in access to meetings with policy-makers and differences in the number of tweets sent by interest groups depending on their access to resources. Each point estimate and 95% CI represents the estimated effect of the pandemic on the difference in meetings and tweets between "high" and "low" resource interest groups by defining "high" and "low" resources at different cutoffs. Points in white indicate confidence intervals that do not cross zero. G Sensitivity of results stratified by resources (Table 3) to alternative codings of 'high' and 'low' resources In Table 3 of the main article, we examine whether the resources that are available to NGOs and business interests drive the results. We do so by examining the differential effect of the pandemic on access to policy-makers and social media behavior by stratifying interest groups by their available resources. In Table 3 of the main article, interest groups with 'high' resources are defined as those in the upper tercile (upper 67th percentile) of lobbying resources, and interest groups with 'low' resources are defined by those in the lower two terciles. These definitions of 'high' and 'low' resources, however, are relatively arbitrary.
We thus test whether the results in Table 3 are sensitive to how high-resource and lowresource groups are coded. To do so, we recode 'high' and 'low' groups at a wide range of cut-offs-from the median thru the 90th percentile-and refit the models from Table 3 for each potential cut-off.
Estimates of the differential effect of the pandemic on access to meetings with policymakers and social media posts among NGOs and business interests among high-and lowresource groups are presented in Figure G7. Panel A corresponds to Models (1) and (2) in Table 3 in the main article; Panel B, Models (3) and (4). The figure demonstrates that the results in Table 3 are insensitive to how 'low' and 'high' resource interest groups are coded. The left figure of Panel A shows that the pandemic caused a decrease in NGOs' access to meetings with policy-makers relative to business interests among high-resource groups, regardless of how 'high' resources is coded (all estimates are significantly different from zero). By contrast, the right figure of Panel A shows very little evidence that the pandemic caused a similar decrease in NGOs's access to meetings with policy-makers about low-resource groups, regardless of how 'low' resources is coded (all but two estimates are no significantly different from zero).
In Panel B of Figure G7, we see similarly that the results from Table 3  This figure shows the estimated effect of the pandemic on differences in access to policymakers and differences in the number of tweets sent by interest groups, among NGOs relative to business interests stratified by lobbying resources. Each point estimate and 95% CI represents the estimated effect of the pandemic on differences in meetings and tweets for NGOs relative to business interests when subsetting the data at different codings of "low" and "high" resources. Points in white indicate confidence intervals that do not cross zero.
is estimated to have a caused an increase in the frequency of social media behavior by NGOs relative to businesses, both among high-and low-resource groups, regardless of how 'high' and 'low' resources are coded.

H Interest group staff size as an alternative measure of resources
In Table 3 in the main article and Table F3, we measure the resources available to interest groups by their lobbying budget, as defined in the EU Transparency Register. As a robustness, check, we also replicate these two tables using an alternative measure from the EU Transparency Register: the full-time staff size of each interest group dedicated to lobbying activities. As with the lobbying budget, we define 'high' and 'low' resource interest groups as those in the upper tercile ('high') and lower two terciles ('low') of staff sizes.
Results are presented in Table H5 and (3) is nearly identical, its level of statistical significance a somewhat lower (p = 0.11)) than when stratifying by lobbying budget. In Table H6, the estimated effects are substantively equivalent, and there are no differences in statistical significance across all six models relative to estimates using lobbying budget as a measure of resources in Table F3.

I Regression results for the log number of meetings and tweets
In the main article, we present difference-in-differences regression models for the outcomes defined as (1) the number of meetings that each interest group has with EU policy-makers, and (2) the number of tweets sent by each interest group. As a robustness check, we also fit the main regression models to the log number of meetings and tweets.
We begin by investigating the effect of the pandemic on differences in the number of meetings that businesses and NGOs have with policy-makers and the number tweets sent by each interest group. Results for the log count of meetings and tweets (analogous to Table 1 in the main article) are presented in Table I7. The results are effectively equivalent to those in the article. Onset of the pandemic is associated with a decrease in the number meetings that NGOs had with EU policy-makers relative to business interests (p < 0.001). By contrast, the pandemic is associated with an increase in the frequency of tweets sent by NGOs relative to business interests (p < 0.01). 0.288 0.864 * p<0.05; * * p<0.01; * * * p<0.001. Standard errors, in parentheses, are clustered at the level of the interest group. The outcome variable is defined as the log number of meetings or tweets from each interest group aggregated at the month level from January 1, 2019 to October 1, 2020. 0.266 0.861 * p<0.05; * * p<0.01; * * * p<0.001. Standard errors, in parentheses, are clustered at the level of the interest group. The outcome variable is defined as the log number of meetings or tweets from each interest group aggregated at the month level from January 1, 2019 to October 1, 2020. Data included are those meetings and tweets that are not classified as being related to the COVID-19 pandemic.
We then fit difference-in-differences models equivalent to those in Table 2 in the main article, where the outcome is the log number of meetings and tweets for meetings and tweets that are not classified as being related to COVID-19. Results are presented in Table I8.
Similar to the results presented in Table 2 of the main article, we find no strong evidence that onset of the pandemic is associated with differences in the log number of meetings that NGOs or business interests had with EU policy-makers, or the frequency of tweets sent by each class of interest group when explicitly COVID-related meetings and tweets are removed.
Finally, we stratify by the resources available to each interest group and fit models to estimate the effect of the pandemic on the log number of meetings that interest groups have with EU policy-makers and the number of tweets they send. We fit a difference-in-differences model equivalent to that used in the main article (Table 3) to the logged outcomes. Results are presented in Table I9. Compared to the analogous table in the main article (Table 3), the results are effectively equivalent. Among high-resource interest groups, the pandemic caused  (1)), an effect that is not observed among low-resource interest groups (Model (2)).
Finally, similar to the results presented in the main article, we find that the pandemic caused an increase in the social media frequency of NGOs relative to business interests, both among low-resource and high-resource interest groups (although not significantly so in the latter).

J Placebo intervention
Is the observed effect of the pandemic on interest group meetings with policy-makers and tweet frequency related to the month that it occurred (in March, 2020)? To test this, we fit a models equivalent those presented in Table 1 of the main article, but use a placebo intervention of one year previous, in March 2019. Results are presented in Table J10. Unlike in the main article in which we find differential effects of the pandemic on NGO access to meetings relative to business and tweet frequency, we find neither using a placebo intervention.

K Meetings results among groups with Twitter accounts
In the article, we estimate the effect of the COVID-19 pandemic on business interests' and NGOs' meetings with policy-makers, and tweet frequency among all available interest groups in the sample. However, because not every interest group has a Twitter account, the samples used to estimate the effect of the pandemic for meetings and for tweet frequency are different.
In this section, we estimate the effect of the pandemic on business interests' and NGOs' meetings with policy-makers only among those interest groups that have a Twitter account (and thus are included in the tweet frequency results).
Results are presented in Table K11. The estimate from Model (1) presents the estimate of the effect of the pandemic on meetings with policy-makers among NGOs relative to business interests using all groups in the sample. This result is equivalent to that presented in Table 1 of the main article. Model (2) presents the effect only among interest groups that have a Twitter account. As the estimate shows, the result is robust to confining the sample only to interest groups with Twitter accounts, with the estimate being even greater in magnitude. 0.295 0.288 * p<0.05; * * p<0.01; * * * p<0.001. Standard errors, in parentheses, are clustered at the level of the interest group. The outcome variable is defined as the number of meetings from each interest group aggregated at the month level, with data from January 1, 2019 to September 30, 2020.

L Differential effects of the pandemic by sub-category
In the main article, we examine the differential effect of the pandemic on business interests relative to NGOs in aggregate. In this section, we investigate this relationship across social and economic interest group sub-categories. We classify interest groups according to their self-described categories of interest as captured by the Transparency Register data. 2 Specifically, we aggregate these categories of interests (e.g. "Trade", "Competition", "Public health") into larger categories representing 6 overall groups: "Business and Industry", "Economic (general)", "Public health", "Environment", "Development & social affairs", and "Agriculture". Because interest groups can select multiple categories, these groupings are not mutually exclusive: an interest group might, for example, lobby in the areas of both trade and environmental regulation.
To examine differences in the effect of the pandemic on business interests and NGO among these sub-categories, we fit a series of difference-in-differences models to the subset of data from each grouping. As in the main article, the outcomes are the frequency of meetings with EU policy-makers and the frequency of social media posts on Twitter.
Results are presented in Table L12 (meetings) and Table L13 (social media posts). As Table L12 suggests, NGOs were differentially affected by the pandemic relative to business interests in terms of less access to policy-makers across all sub-categories: the differential effect of the pandemic is negative across all models, with statistically significant effects in all, but one case. The magnitude of the differential effect is largest among interest groups that concern public health, although we note that because subgroup analysis necessarily reduces the sample size for each case, comparisons across sub-categories noisy.
In Table L13, we find similarly consistent results with respect to the frequency of social media posts: in each case, the differential effect of the pandemic is positive, suggesting that the pandemic caused an increase in social media posting behavior by NGOs relative to business interests across all sub-categories. The largest statistically significant effects are those among interest groups concerned with public health and the economy. We caution, however, that the relatively large standard errors that results from sub-group analysis mean that no differences across interest group sub-categories are themselves statistically significant.

M Differential effects on trade unions and public authorities
The main article focuses on NGOs and business interests. However, it is useful to also examine whether system-wide shocks such a pandemic can have effects on other categories of interest groups. While trade unions represent specific economic interests, they can also be seen as representatives of broader societal interests providing them with a capacity to mobilize the wider public. Sub-national public authorities similarly reflect societal interests by representing citizens within a given geographical area at large.
To examine the differential effects of the pandemic on trade unions and sub-national public authorities, we fit difference-in-differences models, comparing these two sets of interest groups both to NGOs and business interests. Results for trade unions are presented in Table M14 and those for sub-national public authorities in  Tables M14 and M15). In sum, the effect of the pandemic on meetings with policy-makers and social media posts for trade unions and sub-national authorities was similar to that of NGOs. The pandemic caused an increase in meetings for business interests relative to each interest group type, and caused an in-crease in social media communications among NGOs, trade unions, and sub-national public authorities relative to business interests.

N Effects on the frequency of sharing social media posts from NGOs and business interests
In the main article, we find that the onset of the COVID pandemic increased the frequency of social media posts by NGOs relative to business interests. A useful follow-up question is whether this increase in posting also resulted in an increase among NGOs in sharing of these posts overall by ordinary users. To test this, we fit difference-in-differences models equivalent to those in the main manuscript, but where the outcome is the number of social media posts by NGOs and business interests that are "favorited" or "retweeted" (shared) by other users in a given period. Results are presented in Table N16. They show that the pandemic increased the number of favorites and retweets received by NGOs relative to business interests (Models 1 and 3, p < 0.01). As a robustness check, we also include interest group-level time trends to account for potential violations of the parallel trends assumption. When these trends are included, the effects of the pandemic on the number of favorites and retweets received by NGOs relative to businesses are no longer statistically significant. We thus find suggestive, albeit not conclusive evidence, that the increase in social media posting by NGOs resulted in increases in the reach of their public communications.

O Poisson regression results
In the main article, we use OLS regression models to estimate the differential effects of the pandemic on access to meetings with policy-makers and the frequency of social media posts.
Difference-in-differences models for quasi-experimental research designs typically use OLS models for both continuous and limited dependent variables (e.g. binary, count) (Angrist and Pischke, 2009, 94-99). One may be interested in estimates from generalized linear models when the outcome variable is not continuous, however. The count distributions for the number of meetings and number of tweets for each interest group aggregated across the time period of study are presented in Figures O8 and O9. The data are unsurprisingly right skewed: a large number of the over ten thousands interest groups in the data do not have meetings with policy-makers during the time period of interest, and relatively small number of interest groups send large numbers of tweets. 4 To show estimates from a count model, we fit fixed effects poisson models analogous to the OLS models in Table 1 from the main article. Results are presented in Table O17. They show that the pandemic caused a 45% decrease in meetings for NGOs relative to business interests (1 − e −0.590 = 0.45), and a 17% increase (e 0.154 = 1.17) in the frequency of tweets.
We note, however, that one drawback of count models with unit-level fixed effects is that units that are observed with all zeroes across time are dropped from the sample.

P Differences in meetings before and during the COVID pandemic between business interest sectors
In Figure  To examine potential differences more formally, we use a panel regression model that compares changes in meetings between the pre-COVID and COVID periods among business interests by sector. We note that a drawback to the sectoral data is that interest groups can indicate belonging to multiple sectors. Examining differences between groups is thus not a straightforward comparison between exclusive sets of groups. Nevertheless, regressing a set of interactions between a binary variable indicating the COVID pandemic period and each sector to which a business interest belongs can provide us with a rough idea of whether certain sectors benefited from the pandemic.
Results are presented in Table P18, where the group "Business and Industry" is used as a baseline category. The results suggest that business interests linked to public health and research and innovation may have benefited more from the pandemic in terms of access to policy-makers than did business interests associated with other categories.  To provide qualitative background for understanding the interest groups that may have benefited from the pandemic, we present in Table Q19 a list of the 50 interest groups that saw the largest increase in their average number of meetings with policy-makers per month between the pre-COVID and COVID periods.

R Differential effects of the pandemic by companies and other business interest groups
In the main article, we examine the overall effects of NGOs relative to business interests.
As shown in Table B1, business interests are defined by a variety of sub-categories. Here, we examine the effects of the pandemic on differences separately for NGOs compared to "Companies & groups", and NGOs compared to interest groups that are defined by the other categories ("Professional consultancies", "Self-employed consultants", "Law firms", and "Trade and business associations"). Results for the effect of the pandemic on NGOs' meetings with policy-makers and the frequency of communication on social media relative to business associations and companies are shown in Table R20. As the table shows, NGOs have fewer meetings as a result of the pandemic relative to both business associations and companies (Models 1 and 2), and send more tweets than business associations and companies (Models 3 and 4), although only significantly so for social media communications relative to companies.