Public support for empowering police during the COVID-19 crisis: evidence from London

ABSTRACT

In the face of the COVID-19 pandemic, police services around the world were granted unprecedented new powers to enforce social distancing restrictions. In this paper, we present data from a rolling representative sample survey of Londoners (n = 3,201) fielded during the height of the first wave of the pandemic (April to June 2020). We examine the scale of public support for giving police additional powers to enforce the regulations, whether support for different powers ebbed and flowed over time, and which factors predicted support for police powers. First, we use interrupted time-series analysis to model change over time. Second, we pool the data to test the predictors of support for police powers. Aside from one lockdown-specific temporal factor (the easing of restrictions), we find that even in the midst of a pandemic, legitimacy, procedural justice and affective evaluations of pandemic powers are the most important factors explaining variation in public support for police empowerment.

KEYWORDS:

• COVID-19
• police empowerment
• procedural justice
• legitimacy

Introduction

The COVID-19 outbreak in early 2020 plunged much of the world into a public health crisis. Many countries introduced emergency powers that significantly curbed civic freedoms and liberties, largely through various forms of lockdown (ICNL 2020). As citizens around the world were told to stay home and avoid contact with others, the enforcement of new public health legislation often fell on the most visible representative of the criminal justice system: the police.

On 26 March 2020 the UK Government enacted new public health legislation that strengthened police enforcement powers in England, including on-the-spot fines to people caught flouting social distancing requirements and the ability to detain people suspected of having COVID-19 (Barber et al. 2021). The role of the police in enforcing these public health regulations immediately raised questions about the appropriate boundaries of police power during what was first and foremost a public health crisis. In a country where policing revolves around notions of consent and policing with the public (Bowling et al. 2019), the concern was that if officers used their pandemic powers in unfair and inappropriate ways then people may start to question whether the police are legitimate, and whether the State more broadly has the right to impose and enforce these new laws (Charman et al. 2023).

In this paper we shed light on public support for those enhanced police powers in April to June 2020. At a time when people’s basic freedoms of movement and assembly were being denied – and as the ‘hard’ Spring 2020 UK lockdown went through various phases of relaxation – we ask: what police powers did Londoners support, did support increase or decrease over time as lockdown was eased, and did police activity generate or threaten the trust and legitimacy they so fundamentally rely upon? Using data from a rolling representative sample survey of Londoners fielded during the height of the first wave of the pandemic (April to June 2020, with an average of 43 telephone interviews conducted each day), we measure support for new powers granted to police (e.g. the ability to issue fines and detain people) and support for surveillance powers that could realistically be used in the future to tackle non-compliance (facial recognition technology and mobile phone tracking).

In study one, we use interrupted time series analysis to track change in public support for pandemic powers over time. We consider a number of possible points of intervention, including the multiple easings of lockdown restrictions, and political (Dominic Cummings scandal) and police-related (Black Lives Matter) events. In study two, we test why people supported granting police new powers. UK-based research has looked at legitimacy and procedural justice in the context of compliance with rules and laws (Jackson et al. 2012, Pósch et al. 2021), as well as public support for enhanced powers and new surveillance technologies (Yesberg and Bradford 2019, Bradford et al. 2020). Drawing on this literature, we explore five sets of factors that may help explain support for police pandemic powers: (1) perceptions of police legitimacy; (2) affective responses to the police having pandemic powers (specifically, whether people feel safe or unsafe at the thought of police enforcing lockdown requirements); (3) perceptions of police procedural justice; (4); perceptions of police effectiveness; and (5) worry about crime. Each factor is likely to be correlated, so we examine not just the extent to which they directly and indirectly predict support for police powers, but also the relationships between these five potential predictors.

The paper proceeds as follows. We first outline the UK context and its legal response to the COVID-19 pandemic. We then set out the theoretical model and methods. After reporting findings from our two studies, we develop the idea that the police gain public support by generating popular legitimacy and being seen to act in procedurally just ways. Yet, people’s affective evaluations of specific powers and/or technologies may also be important – here, representing an affectively-laden desire for, of rejection of, a law-based State response to the pandemic. We conclude with some thoughts on future lines of research into the role that affect plays in public support for police empowerment.

Policing COVID-19 in the UK

It was on 23 March 2020 that the UK was placed into a nationwide lockdown, with people told to ‘stay home, protect the NHS, save lives’.¹ Just a few days later, the UK government enacted new public health regulations to strengthen police enforcement powers in England.² The Health Protection (Coronavirus, Restrictions) (England) Regulations 2020 granted police the power to issue on-the-spot fines to enforce new rules on social distancing, to ‘direct, remove or detain’ anyone suspected of having COVID-19 and to use reasonable force, if necessary, to exercise these powers (Health Protection Regulations, 2020).

As a side-effect of the new legislation, large numbers of people who would not usually come into contact with the police became, potentially, more likely to do so (Charman et al. 2023), with ‘ … the scope of the regulations suddenly [making] potential offenders out of large numbers of otherwise law-abiding people’ (Clements and Skidmore 2020). So the police had to weigh up the need to enforce compliance with a desire to not ‘stray too far from the tradition of policing by consent’ (Ghaemmaghami et al. 2021). Recognising the importance of not being heavy-handed, the National Police Chiefs’ Council (NPCC) and College of Policing advised officers to adopt a ‘four E’ approach if they believed the rules were not being adhered to: engage with the public; explain the social distancing regulations; encourage people to follow the regulations; and, as a last resort, enforce the law (College of Policing 2020).

In this context, what did Londoners – the site of the current empirical work – think about these new powers? Polling conducted during the height of the first lockdown indicated the ‘four E’ approach had been largely successful in the public eye, with the majority of people expressing support for the way the police had handled the lockdown (Desroches and Caluori 2020). Indeed, 15% of the public wanted the police to take more stringent action to ensure compliance with the rules, with only 6% considering the police to have been too heavy-handed. But what did Londoners think, at a more granular level, about pandemic police powers?

In this paper, we provide an empirical account of which particular powers the public supported and how support for different powers responded to key events, specifically the easing of lockdown in three phases, a high-profile political event involving the prime minister’s key adviser, and the police killing of George Floyd in the US. We also explore how legitimacy, affect, procedural justice and effectiveness perceptions ebbed and flowed over the same time period, and whether these perceptions helped explain support for police pandemic powers.

Prior research on support for empowering police

What is public support for the empowerment of the police? Generally speaking, it is to believe that existing and new powers and technologies are appropriate for officers to have and employ – that such powers and technologies will provide benefits and will not be misused. Working within this broad definition, scholars have variously framed empowerment as support for an overall expansion of police powers (e.g. ‘police should have the power to do whatever they think is needed to fight crime’), support for specific existing powers like stop-and-frisk/stop-and-search, support for new specific powers (e.g. ‘police should be able to fine people breaking lockdown’), support for use of new technologies (such as live facial recognition and algorithms to direct resources), and support for the use of military equipment (in the US). However it has been measured, studies have linked support for empowerment to (1) thinking the police are a legitimate authority that has the right to enforce the law, (2) positive affective evaluations of particular police powers, (3) believing that officers act and make decisions in procedurally just ways, (4) believing that officers are effective at fighting crime and supporting victims, and (5) worrying about the personal risk of crime (Sunshine and Tyler 2003, Moule et al. 2019, Bradford et al. 2020).

First, legitimacy has been shown to be important to police empowerment (Factor et al. 2014, Metcalfe and Hodge 2018, Pryce 2019). To view the police as legitimate is to believe that they rightfully hold power, and to recognise their moral authority to enforce the law, monopolise the use of violence in society, and decide what is proper, desirable and appropriate in a given context or situation (Tyler 1997, Tyler 2006a, 2006b, Jackson et al. 2013, Trinkner 2019, Pósch et al. 2021). According to work on empowerment, this authorisation process also includes deferring to the right of the police to judge for themselves the appropriateness of existing and new powers and technologies (cf. McLean and Nix 2022, Muibu and Olawole 2022). When people view the police as legitimate holders of power, they also tend to assume that new powers and technologies will be beneficial and will not be misused – that is, they defer to the judgment of the police in how to fulfil their function in normatively appropriate ways. As Bradford et al. (2017, p. 615) argue, ‘Legitimacy is based in an important sense in ‘right behaviour,’ but it may also serve to ‘make behaviour right’’.

Second, studies have showed that support for empowerment is also related to people’s affective evaluations of particular police powers and technologies. What is affect? If emotions are discrete psychological states like anger, fear, disgust and happiness, affect is what Slovic and Peters (2006, p. 322) call the ‘the specific quality of ‘‘goodness’’ or ‘‘badness’’ (a) experienced as a feeling state (with or without consciousness) and (b) demarcating a positive or negative quality of a stimulus’. Applied to the current topic, affect refers to pleasant or unpleasant feelings that people have about a particular police power and/or technology (such as feelings of safety, or of unsafety). For instance, Yesberg and Bradford (2019) linked support for the routine arming of the UK police to positive feelings of safety rooted in the ‘more visible, firmer assertion of order’ that was baked into their imagery of an armed police force.

Third, empowerment has also been linked to people’s perceptions of the procedural justice of the police (McLean and Nix 2022). Procedural justice is about officers being seen to wield day-to-day authority in ways that respect widely shared normative principles of respectful interpersonal treatment and fair and neutral decision-making (Tyler 2006a, 2006b). Studies have shown that when people believe that officers respect principles of fair treatment and fair decision-making they are more likely to believe that officers will use new powers and technologies in right and proper ways. For instance, Nix et al. (2021) found that perceptions of police procedural justice predicted support for the enforcement of social distancing ordinances during the COVID-19 pandemic.³ Sargeant et al. (2022) showed that people who were in favour of additional lockdown-related powers also tended to believe that police enforced social distancing restrictions in procedurally just ways.

Fourth, people’s perceptions of police effectiveness in the control of crime and administration of justice also seem important (Pryce 2019). When people believe that officers act in effective ways – whether this be by cutting crime, supporting victims or catching criminals – they are more likely to believe that new powers and technologies will bring benefits to community safety and law enforcement. For example, Sunshine and Tyler (2003) found that perceptions of police effectiveness predicted police empowerment as public support for officers to have the power to stop and question citizens on the street, to decide which areas of the city should receive the most police protection, to do whatever they think is needed to fight crime, and so forth. Yesberg and Bradford (2019) showed that effectiveness predicted support for arming the UK police, with the technical competence to realise the benefits of carrying guns seemingly working through making people feel safer at the thought of armed police.

Finally, to the extent that people believe granting police additional police powers will make ‘fighting crime’ and ‘keeping communities safe’ easier, individuals who are worried about crime are more likely to accept a development that grants police more power. Fox et al. (2020) found that fear of crime was a factor in public support for police acquisition and use of surplus military weapons and equipment. Worry about crime has also been linked to the acceptance of police use of live facial recognition (Bradford et al. 2020) and the perceived benefits of body worn video cameras (Crow et al. 2017). Relatedly, worry about COVID-19 health risks has been found to predict support for extra lockdown-related powers (Sargeant et al. 2022).

Taken as a whole, the literature suggests that legitimacy, affect, perceptions of police procedural justice and effectiveness, and worry about crime are all associated with public support for police empowerment. In the specific context of public support for pandemic powers, procedural justice, effectiveness, perceived COVID-19 risk, and altruistic fear all seemed to be important (one study conducted in the US, Nix et al. 2021, and the other study conducted in Australia, Sargeant et al. 2022). Yet, no study has explored all the factors together, especially in the context of pandemic powers. Moreover, no study has examined the temporal trajectory of public support for police empowerment. Arguably, the COVID-19 pandemic and the timing of various Governmental non-pharmaceutical interventions provide a fascinating context in which to examine the predictors of empowerment, and the extent to which empowerment and key correlates ebb and flow over time.

The current study

In study one, we use interrupted time series to examine the extent to which attitudes towards the police changed over a three-month period during the first 2020 UK lockdown. We draw on data from a rolling representative sample survey comprised of around 43 telephone interviews a day on average, conducted between 1 April and 30 June. We differentiate between four powers related to pandemic policing (specifically, issuing fixed penalty notices, detaining people to enforce Government Coronavirus measures, using facial recognition technology to track people, and tracking people’s mobile phones) and a general measure of whether people support giving the police extra powers to deal with Coronavirus. We test whether support for police powers granted by an emergency legislation shifted in accordance with the prevailing situation on the ground (i.e. the pandemic), changes to the pandemic legislation, and other significant events (for an examination of changes in public perceptions of police legitimacy over time in Chile in 2020, see Vilalta-Perdomo et al. 2023). Additionally, looking at data from the height of the first wave of the pandemic in London allows us to provide some indication about whether people changed their minds, and if so, which event(s) (if any) seemed to track attitude change. We also test whether specific events that occurred during the study period (e.g. easing of the lockdown, the death of George Floyd) act as ‘tipping points’ that alter the trend in support for police powers.

In study two, we use structural equation modelling to test the theoretical model outlined in Figure 1. Pooling data over all time points, we investigate potentially mediating relationships between various predictors of support for police pandemic powers (e.g. legitimacy mediates some of the statistical effect of procedural justice on empowerment). Yesberg and Bradford (2019) found that people tended to use their sense of police trustworthiness (to be effective, procedurally fair, and engage with the community) as a heuristic through which to affectively appraise the potential risks, benefits, and appropriateness attached to officers being routinely armed. Building on this, we test whether the perceptions of procedural justice and effectiveness, and worry about crime, predict legitimacy and affect directly, and whether they predict empowerment indirectly through legitimacy and affect as well as directly. We also test the model before and after what we find, in study one, to be the key event: namely, the first easing of lockdown on May 13. This allows us to assess whether there was something specific about those first two months that meant the factors that predicted public support in normal times did not pan out quite so simply during that extraordinary time, when people’s freedoms were being curtailed to an extent not seen since World War II.

Figure 1. Conceptual map.

Before we turn to the two studies, we should say that the survey only included one measure of legitimacy (felt obligation to obey the police). We would ideally have had at our disposal multiple indicators of legitimacy, covering not just the belief that officers generally exercise their powers in normatively appropriate ways, and thus have the right to power, but also the belief that officers have the right to dictate appropriate behaviour, and thus have the authority to govern (Jackson et al. 2012, 2013, 2022). Given that space was tight in the survey, our interpretation can only focus on the aspect of legitimacy that is about a moral duty to obey police commands. We return to this limitation in the closing part of the paper.

Data and methods

Data

Data are drawn from a sample of 3,201 respondents to the Mayor’s Office for Policing and Crime’s (MOPAC) Public Attitudes Survey (PAS). The PAS is conducted on a rolling basis and includes a representative sample of residents from across London with data collection that follows stratified random sampling with the goal of being representative of smaller geographical areas of London (i.e. boroughs) each quarter. Approximately 12,800 Londoners are interviewed, normally face-to-face, annually at pre-selected addresses aiming to achieve 400 interviews in each of the 32 London boroughs by the end of each year (and approximately 100 interviews in each borough every quarter). Statistical weights are computed to compensate for differences in selection probabilities and unit-level non-response. Participants in this study were interviewed during one quarter of the year, between April and June 2020. The PAS asks about people’s experiences of crime and anti-social behaviour, and their perceptions of the police. From April 2020, a range of questions related to the COVID-19 crisis were included in the survey. Also from April 2020, to comply with the social distancing regulations, the interviews were conducted via telephone instead of in-person. See Table 1 for sample characteristics.

Table 1. Sample characteristics.

	N (unweighted)
Gender
Female	1794
Male	1405
Age
15–24	306
25–44	745
45–64	1546
65+	603
Ethnic group
Asian	444
Black	390
Mixed	116
White	2035
Other	65
Victim of crime related to covid
Yes	189
No	2999
Stopped by police during covid
Yes	144
No	3050
Total	3201

Constructs and measures

The PAS contains a range of questions employing Likert-type scales, which are used here to measure the key constructs of interest. Unless otherwise stated, all measures were coded in such a way that higher values indicate more positive evaluations of the construct measured. The items used to measure the key constructs are presented below. Latent variables for analysis were generated using Mplus 7.11 (see Appendix B for factor loadings and model fit statistics).

Support for police powers

To measure support for police powers to deal with the COVID-19 crisis, participants were asked the following questions: ‘To what extent do you support or oppose the Metropolitan Police Service … ’ (1) getting extra powers to deal with Coronavirus; (2) issuing fixed penalty fines to enforce Government Coronavirus measures; and (3) detaining people to enforce Government Coronavirus measures; (4) using facial recognition technology to track people to ensure they are only out of their homes for a good reason; (5) tracking people’s mobile phones to ensure they are only out of their homes for a good reason. Participants responded on a 5-point scale from strongly oppose to strongly support. Note that this list includes powers police actually had at the time (1 to 3) and ones they did not (4 and 5); although it seems possible many respondents thought all these powers were in fact in place.

Affective response to police powers

One item measured people’s affective response to police powers: ‘Does the Metropolitan Police Service having extra powers to deal with Coronavirus make you feel more safe, less safe or does it make no difference?’. Responses were coded such that 1 = less safe, 2 = no difference and 3 = more safe. Building on Yesberg and Bradford (2019), who amongst other things measured affect towards armed police by asking people whether they would feel safer if they saw armed officers, we assume that when people say they would feel safer if the police had extra powers to deal with COVID, they are expressing an emotion generated in response to the particular object here – the affective aspect of the attitude.

Police legitimacy

Police legitimacy was measured using only one item, representing respondents’ perceived obligation to obey the police: ‘I feel an obligation to follow police orders’. Participants responded on a 5-point agree/disagree scale. As mentioned earlier, we would have liked to have been able to draw upon data from multiple indicators of legitimacy that capture the twin beliefs that the police is (a) a moral, just and appropriate institution that (b) has the right to dictate appropriate behaviour. But we were limited by survey space.

Procedural justice

Three items measured procedural justice (specifically interpersonal treatment and voice provision): ‘They would treat you with respect if you had contact with them for any reason’, ‘The police in your area treat everyone fairly regardless of who they are’, and ‘The police in your area listen to the concerns of local people’. Participants responded to these items on a 5-point agree/disagree scale. It would have been optimal to have a full set of indicators, particularly items that capture neutral decision-making, but these were not available.

Effectiveness

Participants were also asked how confident they were in the London police service to handle the pandemic: ‘How much confidence do you have in the ability of the Metropolitan Police Service to handle the coronavirus outbreak?’. Participants responded on a 4-point scale, from no confidence at all to a lot of confidence.

Worry about crime

Worry about crime was measured using two PAS items: ‘To what extent are you worried about crime in your area’ and ‘To what extent are you worried about anti-social behaviour in your area’. Participants responded on a 4-point scale, from not at all worried to very worried.

Control variables

In our models below, we control for a number of demographic variables, including age, gender, and ethnicity. We also control for whether respondents had experienced any crime or incident motivated by or related to the coronavirus outbreak, and whether respondents had been stopped or questioned by the police to check whether they were complying with government COVID-19 measures.

Results

Study one provides an overview of the scale of support for police powers before using interrupted times series analysis to explore change in support over the study period. Study two tests our theoretical model (Figure 1) by pooling data across the full time period.

Study one

Support for police powers

We start by presenting respondents’ general level of support for giving police additional powers to deal with the COVID-19 crisis. A clear majority of respondents supported the police getting extra powers to deal with the crisis (83%), including issuing fixed penalty fines to enforce the restrictions (79%). There was slightly less (but still majority) support for giving police the power to detain people (60%). Less than half of respondents supported giving the police the power to use facial recognition technology (48%) and to track people’s mobile phones (38%). It seems, therefore, that the overwhelming majority of Londoners supported giving the police some additional powers to deal with the COVID-19 crisis.

Interrupted time series analysis

The dataset included the day the interview was conducted, allowing us to track support for police powers over time. We used interrupted time-series analysis to consider when change or changes in the trend-line of opinions could have occurred during the study period. A key consideration with interrupted time-series analysis is choosing the point of intervention, which is sometimes referred to as the cut-off or change point. The point of intervention is usually decided a priori and is informed by changes in policy or important events that transpired at that particular moment in time. In our analysis, we considered five points of intervention⁴:

• 13 April: The day after the news broke regarding the prime minister’s key adviser, Dominic Cummings’, apparent breaking of lockdown regulations.

• 13 May: The first easing of the lockdown in England.

• 26 May: The day after the death of George Floyd reignites the Black Lives Matter protest movement.

• 1 June: The second easing of the lockdown in England.

• 15 June: The third easing of the lockdown in England.

For the Dominic Cummings and George Floyd news stories, we picked the day after the event because we reasoned most people would only have a chance to learn about these events when they had been in the headlines for a little while. Conversely, each easing of the lockdown in England was accompanied by a public information campaign by the government in the days leading up to the event.

We used interrupted time series analysis as suggested by Linden (2015) and STATA 15 to derive the estimates. Interrupted time series analysis presumes that there is one or multiple interventions that meaningfully change the trend-line of the variables of interest. Hence, one or multiple pre- and post-intervention trends are compared to each other and the changes to these trends are estimated.⁵ In our models, the five items measuring support for particular police powers were the outcome variables. To mitigate the chances that the changes in views were due to sampling variation, we added one explanatory variable in the model, which was a product of the sampling weight and a propensity score which was calculated using the demographic characteristics of each individual (i.e. gender, age, ethnicity). For each day, we took the average of the outcome and explanatory variables, thus generating one observation for each outcome and explanatory variable on each date.

The model results are presented in Table 2. We found that support for the use of live facial recognition and mobile tracking remained largely unchanged across the study period, with weak significant pre- and post-intervention trends for the former, and no significant effects for the latter. These findings imply that people in London either did not change their opinions (mobile tracking), or even if they did (live facial recognition), this change could not be attributed to any specific intervention during the study period.

Table 2. Interrupted times series analysis of changes in support for police powers (Newey-West standard errors and one lag).

	General support	Fixed penalty	Detaining people	Live Facial Recognition	Mobile tracking
Dominic Cummings scandal (13 April)
Pre-intervention trend	−0.023* [0.009]	−0.006 [0.027]	−0.015 [0.025]	0.033 [0.023]	0.024 [0.017]
Day of intervention	0.329** [0.104]	0.117 [0.156]	0.229 [0.188]	−0.279 [0.150]	−0.258 [0.182]
Change in trend vs pre-intervention	0.013 [0.009]	−0.001 [0.027]	0.006 [0.069]	−0.037 [0.023]	−0.027 [0.017]
Post-intervention trend	−0.010*** [0.002]	−0.007*** [−0.001]	−0.009*** [0.002]	−0.004* [0.002]	−0.002 [0.002]
Propensity score*Sampling weight	−0.029 [0.041]	0.029 [0.045]	0.009 [0.069]	0.085 [0.093]	−0.016 [0.063]
Constant	4.606*** [0.080]	4.269*** [0.197]	3.778*** [0.167]	2.926*** [0.197]	2.645*** [0.115]
First easing of lockdown (13 May)
Pre-intervention trend	−0.001 [0.002]	−0.001 [0.003]	−0.003 [0.003]	−0.009 [0.004]	−0.006 [0.005]
Day of intervention	−0.137 [0.099]	−0.200* [0.093]	0.040 [0.135]	0.019 [0.164]	0.132 [0.184]
Change in trend vs pre-intervention	−0.011* [0.005]	−0.004* [0.002]	−0.009* [0.004]	0.070 [0.006]	0.002 [0.007]
Post-intervention trend	−0.011* [0.005]	−0.005* [0.002]	−0.012*** [0.003]	−0.002 [0.005]	−0.004 [0.004]
Propensity score*Sampling weight	−0.014 [0.042]	0.039 [0.044]	0.020 [0.073]	0.085 [0.095]	−0.024 [0.061]
Constant	4.49*** [0.064]	4.232*** [0.092]	3.742*** [0.091]	3.186*** [0.110]	2.797*** [0.095]
Death of George Floyd (26 May)
Pre-intervention trend	−0.005** [0.002]	−0.003 [0.002]	−0.001 [0.003]	−0.007 [0.004]	−0.003 [0.003]
Day of intervention	−0.054 [0.106]	−0.273** [0.092]	−0.246* [0.122]	0.080 [0.170]	0.006 [0.145]
Change in trend vs pre-intervention	−0.009 [0.008]	0.005 [0.004]	−0.006 [0.006]	0.005 [0.007]	−0.001 [0.006]
Post-intervention trend	−0.0136 [0.008]	0.002 [0.003]	−0.007 [0.005]	−0.002 [0.005]	−0.003 [0.006]
Propensity score*Sampling weight	−0.013 [0.043]	0.039 [0.047]	0.018 [0.074]	0.085 [0.095]	−0.025 [0.061]
Constant	4.557*** [0.070]	4.278*** [0.086]	3.711*** [0.092]	3.166*** [0.113]	2.746*** [0.099]
Second easing of lockdown (1 June)
Pre-intervention trend	−0.005*** [0.001]	−0.006** [0.002]	−0.004 [0.003]	−0.006* [0.003]	−0.002 [0.002]
Day of intervention	−0.066 [0.123]	−0.141 [0.102]	−0.211 [0.135]	0.042 [0.159]	−0.030 [0.139]
Change in trend vs pre-intervention	−0.009 [0.012]	0.007 [0.005]	−0.002 [0.007]	0.006 [0.008]	0.001 [0.008]
Post-intervention trend	−0.014 [0.012]	0.002 [0.004]	−0.002 [0.006]	−0.001 [0.007]	−0.002 [0.008]
Propensity score*Sampling weight	−0.013 [0.043]	0.038 [0.045]	0.017 [0.073]	0.085 [0.096]	−0.025 [0.061]
Constant	4.573*** [0.066]	4.324*** [0.084]	3.756*** [0.090]	3.150*** [0.104]	2.744*** [0.102]
Third easing of lockdown (15 June)
Pre-intervention trend	−0.008*** [0.001]	−0.007*** [0.001]	−0.006*** [0.002]	−0.005** [0.002]	−0.003 [0.001]
Day of the intervention	0.185 [0.199]	0.033 [0.101]	0.050 [0.122]	−0.082 [0.154]	−0.065 [0.175]
Change in trend vs pre-intervention	−0.038 [0.038]	0.010 [0.009]	−0.018 [0.014]	0.022 [0.019]	0.010 [0.022]
Post-intervention trend	−0.046 [0.039]	0.003 [0.009]	−0.025 [0.014]	0.017 [0.019]	0.008 [0.022]
Propensity score*Sampling weight	−0.013 [0.042]	0.042 [0.045]	0.023 [0.071]	0.084 [0.095]	−0.023 [0.060]
Constant	4.623*** [0.068]	4.350*** [0.079]	3.809*** [0.090]	3.120*** [0.098]	2.750*** [0.104]

***p < .001 **p < .01 *p < .05.

For the other three powers, the change in the trendline appeared to happen on 13 May: the first easing of lockdown in England. These changes in the trendline compared to the pre-intervention trend were weak but significant, which is unsurprising given we were considering one data point for each day (i.e. a relatively small segment of time in the scheme of the pandemic). For general support for police powers, the expected change in the trendline compared to pre-intervention was β = −0.011 (p < 0.05) each day, resulting in a post-intervention trend of similar magnitude (β = −0.011, p < 0.05). The second biggest change emerged for support for detaining people, where the average daily change compared to the pre-intervention trend amounted to β = −0.009 (p < 0.05), with a post-intervention trend of β = −0.012 (p < 0.001). Finally, the estimated daily change compared to pre-intervention trends was −0.004 (p < 0.05) for fixed penalty fines, with a β = −0.005 (p < 0.05) post intervention trend.

Figure 2 plots the changes in the trendline, providing a visual guide to the effect sizes. The figure shows that (in line with the above findings) the change in the trendline was larger for general support for police powers and support for detaining people: both produced steeper slopes compared to the slope for support for fixed penalty notices. By contrast, the figure for live facial recognition indicates a fairly linear negative trend, while the by-and-large horizontal line for mobile tracking implies that attitudes toward that technology barely changed during the study period.

Figure 2. Interrupted time series analysis charts of support for police powers after the first easing of the lockdown in England (13 May 2020).

Apart from the results regarding the overall trend, our analysis also revealed some significant findings that emerged at the point of the intervention. First, on the day after the Dominic Cummings event, general support for police powers significantly increased (β = 0.329, p < 0.01). Second, support for fixed penalty notices significantly decreased on the day of the first easing of the lockdown (β = −0.200, p < 0.05), as well as the day after George Floyd’s death (β = −0.273, p < 0.01). Finally, George Floyd’s death also had a significant negative impact on support for detaining people (β = −0.246, p < 0.05). Crucially, however, these effects were short-lived: apart from the first easing of the lockdown, the other points of intervention did not appear to change the overall trend in support for police powers.

Procedural justice, effectiveness, legitimacy and affect

As a point of comparison, and using the same interrupted time-series approach, we also considered how trends changed in four important explanatory variables: perceived effectiveness of the police, procedural justice, police legitimacy, and affective response. From the four variables, the trendline for police legitimacy was closest to attitudes towards facial recognition, with no statistically significant change over time (Table 3 and Figure 3).

Figure 3. Interrupted time series analysis charts of procedural justice, effectiveness, police legitimacy, and affect after the first easing of the lockdown in England (13 May 2020).

Table 3. Interrupted times series analysis of changes in support for effectiveness, procedural justice, police legitimacy and affect (Newey-West standard errors and one lag).

	Effectiveness	Procedural justice	Police legitimacy	Affect
Dominic Cummings scandal (13 April)
Pre-intervention trend	−0.049 [0.042]	0.011 [0.017]	0.011 [0.010]	0.002 [0.007]
Day of intervention	0.297 [0.318]	0.074 [0.113]	−0.001 [0.112]	0.063 [0.068]
Change in trend vs pre-intervention	0.041 [0.042]	−0.018 [0.017]	−0.016 [0.010]	−0.008 [0.007]
Post-intervention trend	−0.007* [0.003]	−0.007*** [0.002]	−0.005 [0–004]	−0.006*** [0.001]
Propensity score*Sampling weight	−0.085 [0.089]	−0.011 [0.028]	−0.034 [0.024]	0.016 [0.021]
Constant	3.681*** [0.330]	4.07*** [0.143]	4.636*** [0.075]	2.509*** [0.055]
First easing of lockdown (13 May)
Pre-intervention trend	−0.003 [0.009]	0.002 [0.002]	−0.001 [0.001]	−0.001 [0.002]
Day of intervention	−0.369 [0.277]	0.009 [0.095]	0.175 [0.148]	0.001 [0.071]
Change in trend vs pre-intervention	0.004 [0.011]	−0.015** [0.004]	−0.011 [0.009]	−0.008* [0.003]
Post-intervention trend	0.001 [0.007]	−0.013** [0.004]	−0.012 [0.009]	−0.008* [0.003]
Propensity score*Sampling weight	−0.065 [0.087]	−0.008 [0.025]	−0.040 [0.028]	0.014 [0.019]
Constant	3.414*** [0.236]	4.100*** [0.074]	4.687*** [0.047]	2.522*** [0.039]
Death of George Floyd (26 May)
Pre-intervention trend	−0.011* [0.005]	−0.001 [0.002]	0.001 [0.001]	−0.001 [0.001]
Day of intervention	−0.004 [0.320]	−0.053 [0.99]	0.107 [0.185]	−0.046 [0.064]
Change in trend vs pre-intervention	0.012 [0.013]	−0.016* [0.007]	−0.019 [0.017]	−0.008 [0.004]
Post-intervention trend	0.001 [0.013]	−0.016* [0.006]	−0.019 [0.017]	−0.009* [0.004]
Propensity score*Sampling weight	−0.088 [0.088]	−0.013 [0.024]	−0.035 [0.025]	0.014 [0.019]
Constant	3.576*** [0.193]	4.133*** [0.059]	4.664*** [0.040]	2.537*** [0.032]
Second easing of lockdown (1 June)
Pre-intervention trend	−0.011* [0.005]	−0.001 [0.001]	−0.001 [0.001]	−0.002** [0.001]
Day of intervention	0.123 [0.0377]	−0.098 [0.118]	0.221 [0.213]	−0.036 [0.068]
Change in trend vs pre-intervention	0.009 [0.019]	−0.015 [0.009]	−0.028 [0.024]	−0.009 [0.006]
Post-intervention trend	−0.002 [0.019]	−0.016 [0.010]	−0.029 [0.024]	−0.011 [0.006]
Propensity score*Sampling weight	−0.091 [0.089]	−0.017 [0.024]	−0.033 [0.024]	0.014 [0.020]
Constant	3.583*** [0.190]	4.157*** [0.057]	4.683*** [0.037]	2.552*** [0.032]
Third easing of the lockdown (15 June)
Pre-intervention trend	−0.009* [0.004]	−0.004** [0.001]	−0.001 [0.001]	−0.004*** [0.001]
Day of intervention	0.501 [0.335]	0.040 [0.138]	0.420 [0.353]	0.071 [0.097]
Change in trend vs pre-intervention	−0.040 [0.030]	−0.038 [0.026]	−0.105 [0.081]	−0.029 [0.018]
Post-intervention trend	−0.050 [0.030]	−0.042 [0.027]	−0.106 [0.081]	−0.032 [0.018]
Propensity score*Sampling weight	−0.086 [0.085]	−0.034 [0.026]	−0.041 [0.029]	0.002 [0.021]
Constant	3.532*** [0.171]	4.246*** [0.058]	4.701*** [0.039]	2.596*** [0.030]

***p < .001 **p < .01 *p < .05.

Police effectiveness was somewhat different: it showed some erosion over time, with the post-intervention trend negative compared to the first time point on 13 April (β = −0.007, p < 0.05) and the pre-intervention trend also becoming significant and negative from 26 May onwards (26 May: β = −0.011, p < 0.05; 1 June: β = −0.011, p < 0.05; 15 June: β = −0.009, p < 0.05), but no single time point (‘interruption’) could be identified which could explain the decline.

Conversely, the cut-off point for affect had its change point after the first easing of the lockdown, with an estimated change of −0.008 (p < 0.05) pre- and post-intervention. Procedural justice also had a cut-off point on the same date: the change in trend compared to pre-intervention amounted to −0.015 (p < 0.01) with a −0.013 post-intervention trend (p < 0.01). It, however, had a second inflection point also after the death of George Floyd with pre- and post-intervention trend change estimates of −0.016 (p < 0.05).

The above results indicate that perceptions of police legitimacy did not change during the observed period, and while attitudes towards effectiveness declined over time, no inflection point could be identified. By contrast, affective responses started to decrease after the easing of the first lockdown. In the case of procedural justice, the picture was a bit more muddled. We carried out robustness tests (see Appendix A), which implied that the most likely change point for that variable was 10 May and that multiple cut-off points could not be detected. In other words, the 13 May date seems a reasonable (although not perfect) choice for this variable as well.

Study two

We next consider predictors of support for police powers during the pandemic. The dependent variable – support for police powers – was a latent variable consisting of three items: general support for police powers, support for fixed penalty fines, and support for detaining people (we excluded the two surveillance powers due to the lack of change in these variables in study one⁶; see Appendix B for factor loadings and model fit statistics).

Table 4 presents pairwise correlations between support for police powers and the explanatory variables used in the analysis. As the table shows, support for police powers was most strongly correlated with affect, procedural justice, and legitimacy.

Table 4. Pairwise correlations.

	1	2	3	4	5	6
Procedural justice (1)	1
Effectiveness (2)	.57***	1
Worry about crime (3)	−.18***	−.12***	1
Affect (4)	.40***	.35***	.05	1
Police legitimacy (5)	.45***	.27***	.04	.28***	1
Support for police powers (6)	.44***	.35***	.07**	.62***	.42***	1

***p < .001 **p < .01 *p < .05.

We used Mplus 7.11 to estimate a structural equation model that tests the theoretical model contained in Figure 1. We used a robust maximum likelihood approach (MLR), which is robust to non-normally distributed data and has been shown to be superior to multiple imputation in handling missing data patterns (Yuan et al. 2012). We applied sampling weights computed to make the data representative of adults living in London. We used clustered robust standard errors for each day of data collection (standard errors were assumed to be correlated within but not between clusters). We derived standardised estimates, so the relative magnitude of the associations (and the contribution made by each variable) could be comparable across the model. Finally, for the estimation of the standard errors of the indirect effects, we took a Bayesian approach with two independent Markov Chain Monte Carlo (MCMC) chains and 100,000 iterations for each chain. Prior literature suggests using Bayesian methods for standard error estimation in mediation analysis to compensate for the non-normal distribution of the indirect effects, thus providing more credible estimates (MacKinnon et al. 2013).

Predicting support for police powers

Figure 4 presents the results, where police legitimacy and affect were regressed onto procedural justice, effectiveness and worry about crime, and support for police powers was regressed onto all other variables (controlling for gender, age, ethnicity, deaths from COVID-19, a variable representing the number of COVID-19 deaths in respondents’ boroughs at the time of interview, and the first easing of the lockdown in England as a dummy variable because this event was shown in Study 1 to be a likely tipping point for changes in support for police powers).⁷ Only paths significant at the p < .05 level are shown.

Figure 4. Results from a structural equation model predicting affect, police legitimacy and support for police powers. Fit statistics: Chi-square = 293 df = 92, p = <.0005, RMSEA = .027, RMSEA 90% CI = .024; .030, CFI = .967, and TLI = .933. Control variables for all parts of the model were gender, age, ethnicity, deaths from COVID-19, a variable representing the number of COVID-19 deaths in respondents’ boroughs at the time of interview, and easing of lockdown.

Starting from the left-hand side of Figure 4, we first found that procedural justice was an important factor, with statistically significant paths from procedural justice to affect (β = .30), legitimacy (β = .47), and support for police powers (β = .11). In other words, people who believed that officers act in procedurally just ways also tended to report positive feelings of safety at the thought of the police having pandemic powers, tended to view the institution as having legitimate right to dictate appropriate citizen behaviour, and tended to support them having pandemic powers. Less important were (a) effectiveness, which predicted affect (β = .17) and support for police powers (β = .06), and (b) worry about crime, which predicted affect (β = .09) and legitimacy (β = .10).

Second, affect (β = .48) and legitimacy (β = .20) both predicted support for police powers. Those who felt safer at the thought of police having pandemic powers also tended to be more supportive of them having them. Equally, people who felt a moral duty to obey officers also tended to support their empowerment. Looking at the indirect effects, we found that affect and legitimacy mediated some of the partial associations between procedural justice, effectiveness and worry about crime when predicting support for police powers. Table 5 presents the direct, indirect and total statistical effects, as well as the percentages of total effects mediated. The total effects of procedural justice, effectiveness and worry about crime were partly mediated by affect and legitimacy (just over half of the statistical effects). Effects decomposition showed that for procedural justice, the pathways ran through both legitimacy (β = .10) and affect (β = .14); for worry about crime, the pathways ran through both legitimacy (β = .02) and affect (β = .04); and for effectiveness, the pathway ran through affect (β = .08). If people believed the police are effectively dealing with the pandemic, then the thought of them having specific powers to enforce lockdown requirements may make them feel safer, in part because of the enhanced sense of the benefits of pandemic powers. Equally, those more worried about crime also tended to have more positive affective responses towards the new powers, leading to a tendency to report greater support for police powers.

Table 5. Direct, indirect and total effects calculated from path model in Figure 3.

	Support for police powers
	β	SE(B)
Procedural justice
Direct	.11***	[0.01]
Total indirect	.24***	[0.02]
% mediated	68%
Total	.35***	[0.04]
Effectiveness
Direct	.06*	[0.02]
Total indirect	.09***	[0.02]
% mediated	60%
Total	.15***	[0.03]
Worry about crime
Direct	.04	[0.03]
Total indirect	.06***	[0.01]
% mediated	60%
Total	.10***	[0.03]
Easing of first lockdown
Direct	−.06**	[0.02]
Total indirect	−.05**	[0.02]
% mediated	45%
Total	−.11***	[0.03]

***p < .001 **p < .01 *p < .05

Finally, we tested the SEM (using Montecarlo integration) before the first easing of lockdown and after. Given that the only real ‘break’ we found in study one pertained to police issuing fines, we focus only on public support for fines. The final outcome variable was therefore a single (ordinal categorical) indicator rather than a latent variable. The standardised regression coefficients for the key predictors of support for police issuing fines were consistent during the first two months of lockdown and after the first easing of restrictions: affect was β = .34 before and β = .30 after; legitimacy was β = .18 before and β = .17 after; procedural justice was β = .08 before and β = .10 after, and effectiveness was β = .04 before and β = .05 after.⁸ The standardised regression coefficients for the key predictors of affect were also consistent (procedural justice was β = .32 before and β = .28 after; effectiveness was β = .16 before and β = .19 after; and worry about crime was β = .09 before and β = .09 after), as was the case for key predictors of legitimacy (procedural justice was β = .41 before and β = .54 after; and worry about crime was β = .11 before and β = .09 after).

Discussion

In the face of the COVID-19 pandemic, police services around the world were granted unprecedented new powers to enforce social distancing restrictions to try to get the virus under control. Drawing on data from a representative survey of Londoners conducted during the height of the first wave of the pandemic, we measured the scale of public support for enhanced police powers, we considered changes in support over the study period, and we explored the factors that predict support for police powers.

We found, first and foremost, that the overwhelming majority of Londoners supported giving the police some additional powers to deal with the COVID-19 crisis. But there seemed to be some boundaries to this support. While the majority of respondents were supportive of the police getting the power to issue fixed penalty fines and detain people, only a minority were supportive of police having the power to use facial recognition technology and track people’s mobile phones. Previous research in pre-COVID-19 times has shown that concerns around privacy are a strong predictor of the rejection of new police powers related to surveillance (Heen et al. 2017, Sakiyama et al. 2017, Bradford et al. 2020); it could be that privacy concerns are also driving the lower levels of support for surveillance powers amongst our sample.

Second, interrupted time series analysis showed that support for most of the pandemic powers declined over time. Importantly, the first easing of the lockdown in England on 13 May appeared to act as a catalyst for change in general support for police powers, detaining people and particularly fixed penalty notices, but not for live facial recognition or mobile phone tracking. These findings suggest that Londoners were willing to temporarily support police powers that were related to the pandemic effort, particularly the ability of the police to fine people who broke lockdown; however, this support started to wane after the first easing of the lockdown. Compliance with lockdown restrictions in England during the first wave of the pandemic was generally high (Jackson et al. 2020), so the decline in support for police powers (especially the ability to fine people) following the first easing of the restrictions could simply be due to respondents realising that more intrusive powers were not necessary. Crucially, police legitimacy remained unchanged during the study period and the easing of the first lockdown was linked to declines in perceptions of procedural justice and the affective evaluation of the safety (or unsafety) that pandemic powers represented to them.

Another notable finding from study one was that some other events resulted in short-term changes in support (e.g. the Dominic Cummings scandal increased general support for police powers and George Floyd’s death decreased support for police enforcement powers); however, these effects were short-lived and did not seem to impact the overall trend in attitudes. These findings indicate that events which seem influential at the time might not have lasting impact on people’s perception of the police, at least for the perceptions studied here.

Third, turning to study two, we found that police legitimacy, affect and procedural justice predicted support for police powers. Research on police empowerment has shown that people who grant police legitimacy are also more willing to empower them (Moule et al. 2019, Fox et al. 2020). Legitimacy has also been shown to predict support for the use of new police technologies (Crow et al. 2017, Heen et al. 2017, Sakiyama et al. 2017, Bradford et al. 2020, Hobson et al. 2023) and police acquisition of surplus US military equipment (Moule et al. 2019, Fox et al. 2020). Echoing this, we found that people who felt a moral duty to obey police also tended to be more willing to grant police the right to determine what powers are appropriate and acceptable, here enforcing pandemic regulations. This is about authorisation and deference; in the words of Gerber and Jackson (2017, p. 84):

Duty to obey is about allowing the judgements, order and rules of an authority to supersede and replace one’s own judgement. When citizens feel a duty to obey, they grant officers the right to dictate their own [appropriate] behaviour – here the right to use force, to decide when it is appropriate, and to be supported (not questioned) by citizens in such use. This might be considered to be a subset of a broader process of granting a power-holder the right to determine what is proper, desirable and the right thing to do in a given context or situation (Kelman 1973, Kelman and Hamilton 1989).

We also found that affect was an important predictor of empowerment. It seems that, even in the midst of a pandemic, both during the first two months of ‘hard lockdown’ and in subsequent periods of the relaxation of restrictions, people relied on how they felt about giving police more power to inform their decision of whether or not to support police powers. People may in part have replaced a difficult question ‘should the police be able to fine and detain people for breaking lockdown and use facial recognition and the tracking of mobile phones to check people’s movements?’ with an easier question ‘does the thought of such powers in the hands of the police feel good or bad?’ This would fit with research into public acceptance of new technologies which shows that when knowledge and experience are lacking, people’s emotional response to an issue drives their stance towards it (Siegrist et al. 2007, Midden and Huijts 2009, Merk and Pönitzsch 2017). Importantly, this finding raises questions about the imagery lying behind feeling safe or unsafe at the thought of police being able to issue fines for lockdown breach and so forth. Yesberg and Bradford (2019) found that people who tagged the mental image of an armed police officer with positive affect, specifically feelings of safety, were more likely to support the routine arming of the police in the UK. They were attracted to the idea of a more forceful police that exerted its authority in stronger, more potentially coercive ways. In the current study, we may be capturing something analogous, albeit that the attraction may be to the idea of this arm of the State being able to use its power and position to fight the pandemic and exert social control on (and on behalf of) the population. In both cases, part of the attraction could be general (people with certain political and cultural ideologies and values are generally attracted to a tougher police force and a stronger criminal justice system) and specific (to arming officers or having pandemic powers).

Procedural justice was also important, with support for pandemic powers seeming to be partly motivated by the belief that officers will treat people fairly and make fair decisions. In terms of indirect statistical effects, people may have felt more safe or less safe when thinking about pandemic powers, in part because they believed that officers will be fair in terms of interpersonal style and decision-making. They also tend to view the police as legitimate (which then motivates support for police empowerment) when they believe that officers treat people fairly and make fair decisions. These findings fit with research conducted in normal times. For example, Yesberg and Bradford (2019) found that affect fully mediated the relationship between trust in police fairness and support for a policy to arm more police officers. Bradford and colleagues (2020) found that police legitimacy fully mediated the relationship between trust in police fairness/engagement and acceptance of police use of live facial recognition.

Effectiveness and worry about crime played roles, albeit lesser ones. Prior research into police empowerment in normal times has found that normative judgements about police are more important predictors of support for police powers than instrumental concerns (Yesberg and Bradford 2019, Bradford et al. 2020); trust in the people being empowered (i.e. police) seems to be more important than the ends toward which the power is oriented (i.e. a reduction in crime). Our findings lend support to this assertion and show that, even in the context of the pandemic, normative judgements play a superior role.

Limitations

Some drawbacks to this study must be acknowledged. First, we could only consider data from the height of the first wave of the pandemic. Data preceding and following these three months would have been useful to provide information about how quickly the increased support in police powers materialised and when these attitudes finally levelled off or picked up again during future waves.

We were also limited by the variables available in the dataset. We only had one measure of police legitimacy (duty to obey); we were missing normative alignment, which is another component of the construct (for discussion, see Jackson 2018). Although duty to obey represents an internalised acceptance of the rightful authority of police to dictate appropriate behaviour, some research has shown that normative alignment is more important for predicting support for police power (Gerber and Jackson 2017, Bradford et al. 2020). We were also missing an open-ended question exploring the imagery behind people’s feelings of safety or unsafety at the thought of the police having pandemic powers.

Other limitations include the use of cross-sectional data, which limits our ability to make causal claims based on these results. Longitudinal data, where changes in attitudes of the same individuals are tracked over time, would have been instructive in this regard. Lastly, although our findings provide a unique snapshot into support for police powers during what is an unprecedented public health crisis, we do not know whether the same predictors will generalise to police empowerment in normal times. Our findings do, however, fit with most previous research on the topic of police empowerment, indicating that people rely on similar thought processes when considering their attitudes about the police during crises and normal times.

Future lines of research

On the basis of our findings, how might studies into public attitudes to police power(s) proceed? We recommend scholars pay greater attention to affect. First, people may exchange a difficult question ‘what are the risks and benefits of pandemic powers?’ with a simpler question ‘what’s my gut feeling about this?’. Risk perception studies have found that people who have ‘tagged’ an object with negative emotional content, such as feelings of dread or unease, also tend to see that object as high risk and low benefit; conversely, a risk object that evokes positive affect tends to be seen as low risk and high benefit (Alhakami and Slovic 1994, Finucane et al. 2000). On the one hand, in the absence of knowledge and direct experience, if people have a positive affective response to the idea of the police having enhanced pandemic powers, they may tend to see the associated risks of misuse to be low and the associated benefits to be high, so they may tend to accept and support the powers. On the other hand, if under the same conditions people have a negative affective response to police pandemic powers, they may tend to see the risks of misuse to be high and the benefits to be low, so they may tend to reject the powers. Future studies should explore this possibility.

Second, work on risk perception (Siegrist and Árvai 2020) has shown (a) that ‘risk objects’ like armed officers, crime, nuclear energy and climate change can evoke images that have both affective and cognitive elements, (b) that affective and cognitive imagery can both be important to risk acceptability and support for various policies, and (c) that affect can hold sway over cognition in certain circumstances (Slovic et al. 2002, cf. Jackson 2006). What might affective imagery look like in the current context? In their study of affective evaluations of armed UK police, Yesberg and Bradford (2019) concluded with the idea that support for armed police was linked to positive feelings of safety that were rooted in the ‘more visible, firmer assertion of order’ that was baked into their imagery of an armed police force. In study two of the current paper, affective feelings towards police pandemic powers may be rooted in an attraction to the idea of the State being able to use legal powers to fight the pandemic and exert social control both on and for the population. Future studies should dig into the cultural imagery, and their ideological underpinnings, that lie at the heart of affective evaluations of specific powers and technologies.

The affective imagery evoked when people think about police and their powers is likely to be shaped by both prior dispositions and individual and collective experiences. When some people are asked in a survey, or in a political discussion, to think about armed police they may reach for images such as the London Bridge/Borough Market attack of 2017, when armed police shot dead three terrorists in the Borough Market area of London, or perhaps, within the same context, of reassuring armed patrols at high profile locations such as Heathrow. Others, however, may reach for events such as the killing of unarmed Black man Chris Kaba by armed officers in 2022. Similarly, the extra surveillance powers granted to police may have evoked images of police working alongside NHS workers to ‘fight the virus’, or, conversely, of George Orwell’s Big Brother. Consideration of affective responses to policing – of the images people reach for, and why these are salient for them – enables contextualisation of perceptions of and judgements about policing within wider social and cultural experiences. Future studies should dig into the cultural imagery, and its ideological underpinnings, that lie at the heart of affective evaluations of specific powers and technologies.

Conclusion

In this paper we have shown that, even in unprecedented times, people’s support for police powers is rooted in procedural justice, legitimacy and affect. There was one notable ‘event effect’, where the first easing of lockdown seemed to bring with it a sense that the police no longer need to use fines to enforce lockdown requirements. But aside from that, our analysis suggesting that people rely both on their emotional response to new police powers and their perceptions about the people being empowered fits with previous research into police empowerment in normal times. Taken together, our findings highlight that people’s perceptions of police, particularly whether those officers act in respectful ways and make neutral and accountable decisions as they engage in everyday policing, can have important implications for the acceptance of new police powers. It seems that, even in the midst of a pandemic, people rely on how they feel about giving police more power to inform their decision of whether or not to support police powers.

Disclosure statement

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

Notes

1 NHS refers to the National Health Service.

2 Wales, Scotland and Northern Ireland have separate regulations.

3 In an online survey experiment, Nix et al. (2021) found a modest effect of providing information about COVID health risks to police on support for less general policing, and no effect on support for more policing of social distancing. We refer here to their modelling of the observational predictors of public support for social distancing policing.

4 Of course, there are a number of other events that occurred during the study period that we could have selected as our points of intervention (e.g. another government advisor resigning on 6 May after breaking lockdown rules, the London BLM protests in early June). We chose our points of intervention due to these events being the most widely publicised in the media and elsewhere.

5 To determine the appropriate lag for the autocorrelation of the adjacent error terms, we used a Cumby-Huizinga test which assesses whether the autocorrelation structure is correctly specified for a model (Baum and Schaffer, 2013). As we fitted a relatively large number of models, there was some variation in the number of suggested lags between 0 and 2. For the sake of consistency, we considered 1 lag for every model. Importantly, this specification did not meaningfully change any of the results presented in this paper.

6 We fitted the SEM using all five measures of public support for police powers and the findings were very similar.

7 These paths are not shown in the figure for visual ease.

8 Using the full dataset, we also tested the interaction between affect and the lockdown vs first easing of lockdown variable (the interaction effect was not statistically significant, p = 0.11) and the interaction between obligation to obey the police and the lockdown vs first easing of lockdown variable (the interaction effect was not statistically significant, p = 0.43).

Appendices

Appendix A. Robustness checks for interrupted times series analysis

As the first robustness check, we used a method proposed by Cruz et al. (2017), which they call ‘robust interrupted time series’ analysis. This approach takes a data-driven strategy in identifying the change point for the analysis which might come before or after the date of the intervention. According to this method, the estimated change point for general support for police powers, support for fixed penalty fines, procedural justice, and affective response fell on 11 May and on 14 May for support for detaining people. For direct comparison with the above analysis, we reran the models. As shown in Table A1 and Figure A1, the results and trends remained largely unchanged.

We carried out a further robustness check, where we considered multiple interventions simultaneously, such as the impact on the trendline of multiple easings of the lockdown (Linden 2015). We have fitted models in close to thirty different configurations, still, only 13 May appeared to produce any statistically significant changes when multiple cut-offs were considered. Due to the high number of models, they are not included in this paper, but the code and results for the analyses are available from the authors upon request. Finally, it is worth mentioning that considering multiple single cut-off points for an interrupted time series analysis (like we did in Table 2) is a robustness check commonly suggested for interrupted time series (Linden and Adams 2011). In that tradition, our results would be indicative that 13 May was likely an appropriate choice.

There are three conclusions which can be drawn from these robustness tests. First, it seems likely that the prime minister’s speech on 10 May could have started changing minds even before the regulations changed on 13 May. By contrast, the change point for support for detaining people only occurred a bit later, after people had a chance to potentially experience their increased freedoms. Second, these days were still close to the cut-off point we defined a priori. The proximity of these days and the largely unchanged results imply that adopting 13 May as the change point stands up to scrutiny. Finally, the analyses with multiple change points suggest that there was only a single event that changed attitudes towards police powers, which was the first easing of the lockdown. The trendline appeared to be unaffected before or after that event.

Table A1. Interrupted Time Series Analysis table of support for police powers with the date specified as suggested by Cruz et al. (2017)

	General support (11 May)	Fixed penalty (11 May)	Detaining people (14 May)	Procedural justice (11 May)	Affect (11 May)
Models at the change point suggested by Cruz et al. (2017, 2018)
Pre-intervention trend	0.001 [0.003]	0.001 [0.003]	−0.002 [0.003]	0.003 [0.003]	−0.001 [0.002]
Day of the intervention	−0.118 [0.100]	−0.201* [0.090]	−0.041 [0.136]	−0.058 [0.095]	0.012 [0.074]
Change in trend vs pre-intervention	−0.012* [0.005]	−0.006* [0.003]	−0.009* [0.004]	−0.017*** [0.005]	−0.008* [0.003]
Post-intervention trend	−0.011** [0.004]	−0.005** [0.002]	−0.011** [0.003]	−0.013*** [0.003]	−0.008** [0.002]
Propensity score * Sampling weight	−0.038 [0.049]	−0.054 [0.048]	−0.040 [0.045]	−0.006 [0.024]	0.014 [0.019]
Constant	4.510*** [0.074]	4.336*** [0.106]	3.787*** [0.110]	4.009*** [0.077]	2.521*** [0.042]

Figure A1. Interrupted Time Series Analysis charts of support for police powers, affect and procedural justice with the date specified as suggested by Cruz et al. (2017).

Appendix B. Latent constructs and measures

Table

Results from a three factor solution with no cross-loadings	Factor loading
Support for police powers (‘To what extent do you support or oppose the Metropolitan Police Service … ’ high = support)
Getting extra powers to deal with Coronavirus	0.777
Issuing fixed penalty fines to enforce Government Coronavirus measures	0.820
Detaining people to enforce Government Coronavirus measures	0.671
Procedural justice (‘To what extent do you agree with these statements about the police in your area?’ high = agree)
They would treat you with respect if you had contact with them for any reason	0.694
They treat everyone fairly regardless of who they are	0.810
They listen to the concerns of local people	0.629
Worry about crime (‘To what extent are you worried about … ’ high = more)
Crime in your area?	0.680
Anti-social behaviour in your area?	0.881
Fit statistics
Chi square	80
Degrees of freedom	17
p-value	<0.001
Root mean square error of approximation	0.034
CFI	0.982
TLI	0.970

Appendix C. Stationarity and autocorrelation

To assess the stationarity of our data we carried out the Augmented Dickey-Fuller-test with the null-hypothesis of a random walk (possibly with a drift), whilst the alternative hypothesis posited stationarity with a linear time trend. As shown in Table C1 and C2, we could reject the null-hypothesis across the board, implying that time-series analysis was appropriate for the modelling.

Table C1. Accompanying stationarity and autocorrelation tests for the interrupted times series analysis of changes in support for police powers.

	General support	Fixed penalty	Detaining people	Live Facial Recognition	Mobile tracking
Augmented Dickey-Fuller test	−5.912 p < 0.001	−7.396 p < 0.001	−8.091 p < 0.001	−8.959 p < 0.001	−9.189 p < 0.001
Cumby–Huizinga general test for autocorrelation – lag-1	19.355 p < 0.001	14.388 p < 0.001	4.578 p < 0.05	0.013 p > 0.05	0.001 p > 0.005
Cumby–Huizinga general test for autocorrelation – lag-2	1.970 p > 0.05	2.838 p > 0.05	1.605 p > 0.05	0.638 p > 0.05	0.014 p > 0.05
Durbin-Watson statistic	1.768	1.783	1.783	1.946	1.986

Table C2. Accompanying stationarity and autocorrelation tests for the interrupted times series analysis of changes in support for effectiveness, procedural justice, police legitimacy and affect.

	Effectiveness	Procedural justice	Police legitimacy	Affect
Augmented Dickey-Fuller test	−6.151 p < 0.001	−9.326 p < 0.001	−7.438 p < 0.001	−6.833 p < 0.001
Cumby–Huizinga general test for autocorrelation – lag-1	17.287 p < 0.001	0.031 p > 0.05	5.748 p < 0.05	16.858 p < 0.001
Cumby–Huizinga general test for autocorrelation – lag-2	2.943 p > 0.05	0.716 p > 0.05	1.215 p > 0.05	6.536 p < 0.01
Durbin-Watson statistic	1.778	2.024	1.721	1.741

A further test involved considering the autocorrelation function in the data. For this, we relied on the Cumby-Huizinga test for autocorrelation. As shown in Table C1 and Table C2, whilst autocorrelation was present at lag-1, this autocorrelation disappeared by lag-2 in most instances. For live facial recognition, mobile tracking, and procedural justice, it was suggested that no autoregressive function is needed, whilst for affect, two lags were recommended. As a robustness check, we considered the alternative model-specifications for these outcome variables, which are present in Table C3 but the substantive results remained largely unchanged. We also report the Durbin-Watson test, which is in line with the other test, indicating some low-level autocorrelation across the board.

Table C3. Interrupted times series analysis for mobile tracking, live facial recognition and procedural justice (no lag) and affect (two lags) (Newey-West standard errors).

	Mobile tracking	Live facial recognition	Procedural justice	Affect
Dominic Cummings scandal (13 April)
Pre-intervention trend	0.026 [0.034]	0.040 [0.034]	0.014 [0.026]	−0.001 [0.016]
Day of intervention	−0.268 [0.270]	−0.229 [0.265]	0.063 [0.209]	0.066 [0.129]
Change in trend vs pre-intervention	−0.028 [0.034]	−0.045 [0.034]	−0.022 [0.026]	−0.005 [0.016]
Post-intervention trend	−0.002 [0.002]	−0.005** [0.002]	−0.008*** [0.002]	−0.006*** [0.001]
Propensity score*Sampling weight	0.006 [0.050]	0.070 [0.054]	−0.008 [0.030]	−0.012 [0.021]
Constant	2.611*** [0.239]	2.840*** [0.243]	4.044*** [0.172]	2.575*** [0.110]
First easing of lockdown (13 May)
Pre-intervention trend	−0.006 [0.005]	−0.008 [0.005]	0.002 [0.003]	−0.001 [0.002]
Day of intervention	0.136 [0.151]	0.007 [0.148]	0.009 [0.104]	0.006 [0.067]
Change in trend vs pre-intervention	0.002 [0.006]	0.005 [0.006]	−0.015*** [0.004]	−0.007* [0.003]
Post-intervention trend	−0.004 [0.004]	−0.002 [0.004]	−0.013*** [0.003]	−0.008*** [0.002]
Propensity score*Sampling weight	−0.005 [0.050]	0.031 [0.053]	−0.007 [0.029]	−0.010 [0.020]
Constant	2.780*** [0.145]	3.228*** [0.148]	4.093*** [0.093]	2.560*** [0.062]
Death of George Floyd (26 May)
Pre-intervention trend	−0.002 [0.003]	−0.007* [0.003]	0.001 [0.002]	−0.002 [0.001]
Day of intervention	0.008 [0.156]	0.083 [0.150]	−0.056 [0.104]	−0.040 [0.068]
Change in trend vs pre-intervention	−0.001 [0.007]	0.003 [0.007]	−0.016** [0.005]	−0.008 [0.004]
Post-intervention trend	−0.003 [0.006]	−0.004 [0.006]	−0.016*** [0.004]	−0.009** [0.003]
Propensity score*Sampling weight	0.006 [0.049]	0.031 [0.052]	−0.011 [0.029]	−0.012 [0.020]
Constant	2.706*** [0.129]	3.218*** [0.129]	4.126*** [0.081]	2.575*** [0.055]
Second easing of lockdown (1 June)
Pre-intervention trend	−0.002 [0.003]	−0.006* [0.003]	−0.001 [0.002]	−0.003* [0.001]
Day of intervention	−0.029 [0.163]	−0.001 [0.159]	−0.087 [0.110]	−0.016 [0.072]
Change in trend vs pre-intervention	0.006 [0.009]	0.005 [0.008]	−0.016 [0.008]	−0.009 [0.005]
Post-intervention trend	−0.002 [0.008]	−0.001 [0.008]	−0.016 [0.009]	−0.011 [0.006]
Propensity score*Sampling weight	0.008 [0.049]	0.040 [0.051]	−0.014 [0.029]	−0.016 [0.020]
Constant	2.700*** [0.123]	3.182*** [0.124]	4.152*** [0.079]	2.598*** [0.053]
Third easing of the lockdown (15 June)
Pre-intervention trend	−0.003 [0.002]	−0.005** [0.002]	−0.004** [0.001]	−0.004*** [0.001]
Day of intervention	−0.069 [0.199]	−0.081 [0.194]	0.089 [0.143]	0.084 [0.084]
Change in trend vs pre-intervention	0.011 [0.024]	0.020 [0.023]	−0.046* [0.027]	−0.030 [0.016]
Post-intervention trend	0.008 [0.024]	0.015 [0.023]	−0.049 [0.033]	−0.034 [0.017]
Propensity score*Sampling weight	0.008 [0.047]	0.047 [0.050]	−0.024 [0.028]	−0.024 [0.019]
Constant	2.709*** [0.109]	3.156*** [0.110]	4.228*** [0.074]	2.633*** [0.046]

Finally, we also considered the partial autocorrelation. The results for these are presented in Figure C1 and it shows that for the most part, partial autocorrelation emerged at lag-1, diminishing afterwards and not reaching statistical significance for the most part. There were some exceptions, with small signs of partial autocorrelation at lag-2 for variables such as general support for police powers, procedural justice, or legitimacy, and at lag-3 for affect (where the earlier model specification suggested lag-2 autocorrelation). The greyed areas on the figures correspond to the Box-Ljung-tests, also referred to as Q- or white noise-tests. We have attempted multiple alternative model specifications, but (1) most of the findings remained by-and-large the same and (2) the new models provided a worse overall model fit compared to the analyses presented in the main part of the paper. Therefore, despite small signs of partial autocorrelation, our results remained consistent with well-fitting models.

Figure C1. Partial Autocorrelation graphs.