Fear and Legitimacy in São Paulo, Brazil:
Police-Citizen Relations in a High Violence, High Fear City
Jonathan Jackson, London School of Economics & Political Science
Krisztián Pósch, London School of Economics & Political Science
Thiago R. Oliveira, London School of Economics & Political Science
Ben Bradford, University College London
Silvia M. Mendes, University of Minho
Ariadne Lima Natal, Núcleo de Estudos da Violência da Universidade de São Paulo
André Zanetic, Universidade Federal da Grandes Dourados
In this paper we examine consensual and coercive police-citizen relations in São Paulo, Brazil.
According to procedural justice theory, legitimacy operates as part of a virtuous circle, whereby
normatively appropriate police behavior encourages public self-regulation and pro-active
cooperation, which then reduces the need for coercive forms of social control. Tests of the theory
in the US, UK, Australia and elsewhere typically pit normative versus instrumental accounts of
crime-control policy against one another. But can consensual and coercive police-citizen power
relations be so easily disentangled in a city in which many people fear crime, where some people
fear police but tolerate extreme police violence, and where the image of the police as “just another
(violent) gang” seems still to have significant cultural currency? Our analysis of the composition,
predictors and potential consequences of police legitimacy highlights points of similarity and
difference in police-citizen relations in this high violence, high fear city of the Global South.
Key words: Policing, legitimacy, Brazil, obligation to obey, compliance with the law, procedural
justice, distributive justice, bounded authority, police violence.
Word count: 13,730 not including appendix
The studies received financial support from the São Paulo Foundation (FAPESP) as part of
the project “Building democracy daily: human rights, violence, and institutional trust”
Brazil provides a context rather different to that of most of the liberal democracies in which research
on police legitimacy and legal compliance has been concentrated, such as the United States of America
(Reisig et al., 2007), the United Kingdom (Bradford, 2014) and Australia (Murphy et al., 2016). A range
of social problems combine in a possibly unique way in Brazil to shape the relationship between
individuals and state authorities, especially the police. There are the low levels of trust in public
institutions generally (Filgueiras, 2013). There are the high crime rates and high levels of fear of crime
(Cardia et al., 2014). There is the role of organized crime at multiple levels in society (something which
is a particular issue in our study site of São Paulo, see Dias, 2011; Ruotti, 2016; Willis, 2015). There
are high levels of police violence and a relatively high level of public support for at least some of this
violence (Peres et al., 2008; Caldeira, 2002). Brazil also stands somewhat apart from other Latin
American countries in terms of crime, fear and trust. According to the Americas Barometer Survey
fielded in 2016 and 2017, Brazil ranked 9
out of 28 countries in terms of victimization but 3
of crime, with almost a quarter of the respondents feeling that their neighborhood was “very unsafe”.
Brazilians have the lowest faith in their justice system, with almost 9 out of 10 respondents of the
opinion that they had little to no confidence in that the judiciary will punish the guilty (Cohen et al.,
In this paper we present one of the first empirical assessments of police legitimacy and legal
compliance in Brazil.
Drawing on data from two São Paulo-based surveys—one representative of eight
neighborhoods chosen to reflect area-level diversity, the other representative of the city population as a
whole—we test and extend procedural justice theory (PJT, see Tyler, 2006a, 2006b) in a city in which
some people can have ambivalent, even conflicted, attitudes towards police, and where people can fear
both crime and police. Our examination of which aspects of PJT translate—and which do not—
contributes to the growing international literature on instrumental and normative police-citizen relations
(Sunshine & Tyler, 2003; Papachristos et al., 2012; Tyler & Jackson, 2014; Murphy et al., 2016;
Trinkner et al., 2018). Our findings also add to ongoing debate about the meaning, measurement,
sources and consequences of legitimacy (Tyler, 2006a, 2006b; Reisig et al., 2007; Jackson et al., 2012;
Bottoms & Tankebe, 2012; Tyler & Jackson, 2013; Jackson & Gau, 2015; Hamm et al., 2017; Jackson
& Bradford, 2019; Trinkner, 2019; Posch et al., 2020).
Our two studies examine (a) the criteria that São Paulo residents use to judge the legitimacy of
the police and (b) the nature of legitimacy and its link to instrumental connections to the police and
legal compliance. Our conclusions center on the idea that the predictors of legitimacy in São Paulo are
consistent with work in settings like the US, UK and Australia—i.e. that a conceptualization of
primarily normative legitimating factors can be recovered even in high fear, low trust environments.
But contrary to existing work, instrumental (coercive) and normative (consensual) considerations can
usefully be framed in this new research setting as ranging along one single dimension, not as two
separate forms of police-citizen relations, as it is traditionally conceived. Through our empirically
supported argument that normative and non-normative police-citizen relations can be viewed of as two
ends of a single continuum, we reassess the standard conception of legitimacy and reveal the complex
nature of police-citizen relations in this major city of the Global South.
According to PJT, when police officers act in ways that accord with widely shared norms about how
power should be exercised, citizens tend to believe that the institution is legitimate (Tyler, 2006a,
2006b; Trinkner & Tyler, 2016; Tyler & Trinkner, 2018). Legitimacy of a legal authority is generally
defined along two connected lines: (a) the belief that the institution is moral, just and appropriate and
(b) the belief that the institution is entitled to enforce the law, make decisions and expect people to
willingly comply with their orders (Tyler, 2006a, 2006b; Tyler & Jackson, 2013; Jackson & Bradford,
2019; Trinkner, 2019). On the one hand, people judge the legitimacy of the police as an institution
against the societal norms that dictate what is appropriate conduct (e.g. do police officers make neutral
and objective decisions when dealing with citizens?). On the other hand, the content of legitimation (i.e.
the bases on which legitimacy is justified or contested) are an empirical question—i.e. they are not
assumed by an outside expert on the basis of political, moral, legal, religious or some other philosophy.
Indeed, what citizens of a particular social, political and legal context deem to be legitimating or
delegitimizing police conduct may vary from one country to another. For example, people in one context
might judge the legitimacy of the police most keenly on the extent to which officers respect principles
of fair process, while effectiveness might be more important in a different context.
PJT posits that procedural justice is the most important (legitimating) norm regarding the
appropriate use of power, and study after study has shown that procedural justice (in direct and indirect
encounters with the police with the police and in terms of general perceptions of police fairness) is the
strongest predictor of perceived police legitimacy (Tyler & Huo, 2002; Tyler & Fagan, 2008). Recent
US and UK work has also explored the role of respecting the limits of one’s rightful authority—
something that overlaps with procedural justice but is not reducible to it (Huq et al., 2017; Tyler &
Trinkner, 2017; Trinkner et al., 2018). Crucially, police effectiveness and whether police allocate
outcomes such as arrests, citations, protection, and service fairly across aggregate social groups (i.e.
distributive justice) are typically less important predictors of legitimacy in these contexts (Tyler, 2006a;
Jackson et al., 2013; Mazerolle et al., 2013; Bradford et al., 2014a; White et al., 2016; Mentovich et al.,
Legitimacy has also been shown to be more important predictor of legal compliance than
instrumental factors based on deterrence and fear of sanction (Tyler, 2006a, 2006b; Papachristos et al.,
2012; Jackson et al., 2012; Murphy et al., 2016; Trinkner et al., 2018). Overall, PJT posits that self-
regulation is encouraged through procedural-justice based approaches to rule enforcement (Tyler,
Overall, the findings of studies from an increasing number of countries across the world—
including Australia (Murphy & Cherney, 2012), Israel (Jonathan-Zamir & Weisburd, 2013), UK (Huq
et al. 2017), Ghana (Tankebe, 2009), South Africa (Bradford et al. 2014), Pakistan (Jackson et al. 2014),
Hong Kong (Cheng, 2015), Japan (Tsushima & Hamai, 2015), China (Sun et al., 2017) and Trinidad &
Tobago (Kochel, 2012)—generally support PJT’s two core predictions: (a) on the importance of
procedural justice in explaining variation in police legitimacy and (b) on the role that police legitimacy
plays in predicting people’s willingness to cooperate with legal authorities and comply with the law (for
reviews, see Jackson, 2018; Bolger & Walters, 2019; Walters & Bolger, 2019).
There is, however, evidence from some other countries that concerns about effectiveness and
overstepping the limits of their rightful authority are at least as important to legitimacy as procedural
justice in contexts that have (a) high levels of crime, (b) legal institutions that are not yet able to provide
a bare minimum of security to citizens, (c) police with a long history of abuse of power, and (d) where
people tend not to identify so strongly with the group that the police represent. This is backed up by
research in places like South Africa (Bradford et al., 2014b), Pakistan (Jackson et al., 2014), Ghana
(Tankebe, 2009) and China (Sun et al., 2017). In South Africa, for instance, Bradford et al., (2014b)
found that normative judgements about fair process may to some degree be crowded out by concerns
about police effectiveness and corruption, the sheer scale of the crime problem, and the association of
the police with a historically oppressive and underperforming state.
Crime and policing in Brazil
Brazil comprises a different setting in which to explore the role of police legitimacy and police
legitimation. In addition to the social issues outlined above, the history of the Brazilian police is
entwined with the history of slavery. During the period of European colonization in South America,
Brazil was the country with the largest number of slaves for the longest period, as well as the last country
in the Americas to abolish slavery (in 1888). In an echo of the history of the southern United States,
police throughout the period of slavery were essentially privately controlled by slave owners, and
policing was targeted at slaves. However, abolition did not secure effective redress to the emancipated
population, who experienced an incomplete and deficient form of citizenship (Schwarcz & Starling,
2015; Huggins, 2000; Guimarães et al., 2005) characterized by high levels of dependence on the state,
patrimonial-like relations with local power-holders and an absence of genuine civil rights (Schwarcz &
Brazil also lived through an unstable twentieth century, shifting between authoritarian
dictatorship (1937-45 and 1964-85) and democratic rule. Even the democratic regimes had a de facto
authoritarianism, with high levels of social control exerted over the marginalized poor and widespread
evidence of political repression and illegal physical violence (Pinheiro 1991; Holstein 2008). Although
the professionalization of the police was formally completed during the twentieth century (Batitucci,
2010), it remains associated with patrimonial power structures constituted by an overlap of the public
and the private spheres. These relations are characterized by a power imbalance and significant social
distance between a large excluded population and the State (dominated by a powerful minority).
More recently, a form of penal populism in Brazilian society is reflected partly by the expanding
prison population, where the per capita incarceration rate doubled between 2000 and 2014 (Iturralde,
2018). This punitive demand is matched by use of lethal force by the police (Anuário de Segurança
Pública, 2017). Cases of excessive violence by the São Paulo Military Police are common in 2018 alone,
there were 6,160 confirmed police killings (Monitor da Violência, 2019
), which is more than 25 times
higher per capita than in the United States (The Guardian: The Counted) – and official statistics likely
fall short of the real number (Willis, 2015). While it is impossible to ascertain the proportion of these
killings that were illegal, nor fully document the extent of unreported cases of killings committed by
officers, this is a large enough number to characterize police violence in São Paulo as a significant
societal problem. The election of Jair Bolsonaro to the Brazilian presidency is a further confirmation of
these tendencies—Bolsonaro is a former army captain who speaks nostalgically about the 1964-1985
military dictatorship and has openly advocated police killings of criminals, promising iron-fisted
policies and a crackdown on crime.
Within this context of multi-layered authoritarianism, the police remain highly militarized.
Policing in each Brazilian state is carried out by two independent organizations: the Military Police is
responsible for day-to-day policing and order maintenance; the Civilian Police, also known as the
“judiciary police”, comprise less than one third of overall police numbers (Lima et al., 2016) and
oversee crime registry and investigations. The Military Police retains a particularly strong historical
link to the period of slavery, indeed it has remained essentially the same organization since the
, when its primary purpose was repressing insubordinate slaves (Batitucci, 2010).
As all of its officers are part of the army reserve force, they must wear uniforms and carry weapons at
all times, and their training and deployment involves a number of military features, e.g. strong
The role and behavior of the police in Brazil suggests: (1) that the state does not guarantee an
appropriate level of security for a significant portion of society and (2) that even when security is
provided, social elites benefit more (Caldeira and Holston, 1999; Pereira and Ungar, 2004; González,
2017). The police are also more likely to use excessive force against people living in poor areas, who
are more likely to be from ethnic minority background (Mitchell and Wood, 1999; Paes-Machado and
Noronha, 2002; Goldstein, 2003; Garmany, 2014; Willis, 2015). Black residents seem to be less likely
to be involved in community policing programs, which tend to be concentrated in middle-class and
upper-middle-class areas (Alves, 2014). There is some evidence that elites and sections of the middle-
class support police violence that seems to them to be directed at maintaining existing class, race and
gender hierarchies, and while they are unlikely to view the police in a favorable light, they are still
supportive of state action that seeks to uphold the established order (Briceño-León et al., 1999).
Crucial to our argument in this paper is that the relationship between police and the policed in
Brazil is marked by conflicting attitudes toward the police, including fear of being mistreated by the
police, and relatively high levels of tolerance of, or at least ambivalence towards, what some might
deem to be excessive police violence against certain out-groups. Studies show that a fair amount of
Brazilians distrust the police (Silva and Beato, 2013), and some fear them (Cleber, 2015; Oliveira Junior
and Alencar, 2015), yet also that a considerable number do not necessarily condemn police violence
when it is targeted at the marginalized and excluded (Cleber, 2017; Garmany, 2014; Paes-Machado and
Noronha, 2002; Caldeira, 2002; Briceño-León et al., 1999). There have also been calls in upper-middle-
class neighborhoods to reinstate the edicts of the military dictatorship, which disproportionately
targeted people from ethnic minority background (González, 2017).
São Paulo presents, then, a fractured social, political, and economic policing climate—at least
compared to cities and countries in which PJT have been thus far tested. The police represent the
proximate face of a paternalistic and authoritarian state; low levels of trust may have had a negative
impact on the democratic legitimation of the State and even satisfaction with the general idea of
democracy (Filgueiras, 2013; González, 2017).
Study one: What legitimates the police in São Paulo?
What might the context of Sao Paulo, Brazil, mean for the construal of legitimacy, its antecedents, and
its consequents? By way of contribution, study one assesses whether people’s perceptions of police as
a moral, just and appropriate institution—operationalized as normative alignment between police and
citizen values (Jackson et al., 2012, 2013; Tyler & Jackson, 2013, 2014)—are founded primarily in
judgements of procedural justice, as has been found in the US (Sunshine & Tyler, 2003; Tyler & Fagan,
2008; White et al., 2016), UK (Jackson et al., 2012, 2013), Australia (Murphy & Cherney, 2012;
Mazerolle et al., 2013; Bradford et al., 2014a) and indeed Israel (Jonathan-Zamir & Weisburd, 2014;
Mentovich et al., 2018), or whether São Paulo residents place greater emphasis on bounded authority
(the restrained use of power, which extends beyond procedural justice), distributive justice and
effectiveness in the fight against crime. We also assess whether those who see police violence against
criminal out-groups as acceptable also view this as a normatively justifiable use of power, above and
beyond traditional perceptions of police conduct, i.e. we test whether tolerance of excessive police
violence predicts normative alignment, adjusting for procedural justice, distributive justice,
effectiveness and bounded authority.
The first posited predictor of normative alignment is procedural justice. Procedural justice is
about police officers treating people with respect and dignity, making neutral decisions, and allowing
citizens the chance to give their ‘side of the story’. The second potential predictor of normative
alignment is distributive justice: namely, judgements concerning the fair allocation of the outcomes of
power exercised across salient social groups in society. Distributive justice concerns the fair allocation
of ‘goods’ and ‘bads’, and it was measured by asking research participants whether they thought that
police ‘treated people equally’ whether those people were ‘rich or poor, white or black’. Distributive
justice may be a more important legitimating factor in São Paulo than in more traditional settings of
tests of PJT. We find in study two of the current paper that public concerns about the distributive justice
of the police are relatively strong in the current context, with around two-thirds of research participants
(63%) saying that the police either ‘never’ treated people equally, ‘rarely’ treated people equally, or
only ‘sometimes’ treated people equally. This compares to Tyler & Jackson’s (2014) nationally
representative study of US citizens, which found lower levels of perceived distributive injustice (45%
of US citizens said the police treated people differently according to their wealth and 40% said the
police treated people differently according to their ethnicity). Distributive justice may be more
important to the legitimation of the police in São Paulo because distributive justice is a more salient
perceived problem in the eyes of citizens.
The third potential predictor of normative alignment is effectiveness. Effectiveness in the fight
against crime (dealing with drug dealing and armed robberies, investigating crime, keeping people safe,
responding to emergency calls, and providing general crime-related services) may be more important
to police legitimacy in São Paulo compared to the US. On the one hand, levels of crime—and fear of
crime—are relatively high, so people may place special importance on the ability of the police to keep
them safe. On the other hand, perceived police effectiveness is relatively low. If special importance is
placed on this aspect of policing, then low levels of perceived police effectiveness may damage the
legitimacy of the institution. In study two of the current paper, for instance, we find that a relatively low
proportion (just over two-fifths, 42%) of respondents said that local police were doing a good or very
good job at responding to emergency calls. This compares to Tyler & Jackson’s (2014) finding that
around two-thirds of respondents (63%) reported that the police would arrive quickly if a violent crime
were to occur in their neighborhood. As in South Africa (Bradford et al., 2014b, found just over two-
fifths of people said that the police would arrive quickly in an emergency), it may be that the police in
São Paulo have not yet established the baseline, minimum ability to provide basic levels of security to
citizens, raising the importance of perceived effectiveness in the legitimation of the police.
The fourth potential predictor of normative alignment is bounded authority. Trinkner and
colleagues (Tyler and Trinkner, 2016, 2017; Trinkner et al., 2018; Huq et al., 2017) have argued that
the situations and powers that citizens believe that police should or should not have the right to operate
within (in terms of situation) and exercise (in terms of power) represent an essential element of how
people define and understand their relationship with legal authorities. Referencing the belief that
officers ‘should not be doing this thing, in this place, at this time, in this way’, the notion of bounded
authority goes above and beyond traditionally conceived concerns about fair treatment and decision-
making (Trinkner et al., 2018). It can also be distinct from whether officers have a legal right to
intervene, since citizens tend to have little understanding of law in this regard, yet nevertheless make
judgements about the appropriateness of police action.
We assess whether people place significant importance on not overstepping the limits of
rightful authority in the appropriate exercise of police power. Could São Paulo residents reject the police
as legitimate authorities partly because they feel they are encroaching on their own private domains of
autonomous behavior and acting outside of their rightful sphere of activity, e.g. protecting drug dealers,
independent of whether the police do this in a fair, or effective manner? As with distributive justice and
effectiveness, concerns about police overstepping their boundaries are relatively high in São Paulo
compared to the US. In study two’s city-wide representative survey, less than half of respondents said
that the military police and the civil police (47% and 49% respectively) always or very often acted
according to the law, while Tyler & Jackson (2014) found that more than two-thirds of respondents
(68%) agreed or strongly agreed with the statement that ‘when the police deal with people they almost
always behave according to the law’. Again, it may be that more widespread concerns about a particular
element of policing raises its importance in terms of legitimation (cf. Tankebe, 2009; Bradford et al.,
The fifth (and final) potential predictor of normative alignment is tolerance of—or ambivalence
towards—police violence. Many people in São Paulo may be aware, via extensive mass and social
media coverage, that the police use excessive violence in certain situations (González 2017; Willis
2015). Study one presented research participants with three scenarios of actual cases that occurred in
São Paulo over the last few years, involving murder of a suspect, torture of a person in custody and
disproportionate violence against protesters. They were then asked what they thought about the officer
behavior. We find, for example, that just under one-quarter (22%) thought that the police torturing a
drug dealer to get information was either ‘good’ or ‘excellent’. Even though people may generally
expect the police to act in fair and effective ways, some might make an exception in the sort of situations
where police can sometimes be extremely aggressive. Of particular interest is whether people who
approve of or tolerate this kind of behavior are more likely to believe that the police is a normatively
appropriate institution (adjusting for perceptions of procedural justice & distributive justice,
effectiveness and bounded authority) compared to people who view this kind of behavior in
disapproving ways, because they see it as an appropriate (legitimating) use of power.
Data, measures and analytical strategy
To test the predictors of normative alignment with the police, we draw on data from the second wave
of a three-wave longitudinal survey that
was designed to be representative of eight neighborhoods of
São Paulo (we use the second wave because it contained the variables required for the current analysis).
The first wave was conducted in 2015 and 150 citizens were selected in each area based on demographic
quotas (gender, age, and education), producing a sample of 1,200 respondents. A total of 928 of those
respondents then took part in the second wave of the study in 2017 (an attrition rate of 22.7%). All
interviews were conducted face-to-face at the respondents’ place using Tablet-Assisted Personal
Table 1 provides full details of the measures. Normative alignment with the police was captured
using items covering expectations, shared values and perceived appropriate behavior; we assume that
positive answers to these questions reflect ascribed normative justifiability of power (Hough et al.,
2013a, 2013b), i.e. when people believe that officers act in ways that align with their own expectations
and values, this lends the institution they represent the right to power and the authority to enforce the
law (Jackson et al., 2012, 2013; Bradford et al., 2014a, 2014b). We find significant variation in levels
of normative alignment. For instance, around one-third of respondents said that the police in their
neighborhood ‘never’ or ‘rarely’ act according to what the respondent believe is right, while around
one-third said they did so ‘always’ or ‘very often.’ Similarly, just over one-quarter of respondents said
that the police in their neighborhood ‘never’ or ‘rarely’ defended values that are important to a person
like the respondent, while around two-fifths said they do so ‘always’ or ‘very often.’ We then address
the factors that explain this variation.
Table 1. Measures used in study one
How often do police in your neighborhood:
• Act according to what you believe is right?
• Have the same expectations you have about your
• Defend values that are important to a person like you?
How often do police in your neighborhood:
• Explain clearly why they pull over or arrest someone?
• Make impartial and just decisions?
• Pay attention to the information that people provide
• Treat people with respect?
How often do police in your neighborhood:
• Treat all people, rich or poor, black or white, equally?
How often do police in your neighborhood:
• Accept bribes?
• Protect drug dealers?
• Act as if above the law?
How effective are the police in your neighbourhood at:
• Reducing drug trades?
• Reducing armed robbery?
• Responding to emergency calls (190)?
• Providing Police Station services?
• Criminal investigation?
• Marches and protests?
• Keeping neighborhood safe?
very good (1-
The police are called after a motorcycle is robbed. Officers
identify the suspects and chase one of them down. Suspect
tries to hide in a dark alley, but officers catch him and get
him handcuffed. By radio, officers learn the suspect in
custody was already a justice fugitive who had previously
been condemned for drug trafficking. Officers then release
him, tell him to run away, and then shoot him in the back.
He dies immediately.
During a demonstration with thousands of people demanding
improvements for the city, some protesters start to destroy
storefronts’ glass doors and throw litter bins on fire. Officers
who were standing by the protest intervene using rubber
bullets and tear gas on all protesters. Protesters begin to run
away, and underneath the smoke a young woman is shot in
the eye with a rubber bullet, thereby losing her vision for
Officers catch a man in the action and arrest him for drug
trafficking. Before taking him to the police station, they
decide to go to the arrestee’s place with no warrants in order
to look for more evidences. There, officers torture him so
that he would tell them where he keeps the rest of the drugs
and give his partners’ names.
Procedural justice was measured using items that cover fair interpersonal treatment, neutral
decision-making, explaining decisions and allowing the citizen voice in interactions with the police
(Table 1). Distributive justice was measured using a single indicator of equal police treatment of
different social groups. Bounded authority items address corruption and acting as if they are above the
law. Effectiveness indicators capture crime reduction, responding to emergencies, investigating crime,
and keeping order on the streets. Tolerance of ‘excessive’ police violence was measured using three
vignettes: the first involving the killing of a suspect, the second involving the shooting of a protestor
using rubber bullets, and the third involving the torturing of a drug dealer to gain information.
Our analysis has three stages. First, we use confirmatory factor analysis (CFA) to assess the
empirical distinctiveness and scaling properties for procedural justice, distributive justice, effectiveness,
bounded authority and normative alignment. Second, we use latent class analysis to assess the
underlying structure of people’s responses to the three excessive police violence scenarios. We test
whether two, three or four classes best represent the data structure; we choose the model with the most
appropriate fit; and we derive a variable based on modal probabilities for the final stage of analysis.
Third, we use structural equation modeling to estimate which factors (procedural justice/distributive
justice, effectiveness, bounded authority and the two dummy variables for the police violence latent
classes) most strongly predict normative alignment.
We should note that we focus only on explaining variance in normative alignment (and not duty
to obey, the other dimension of legitimacy). By assessing the predictors of normative alignment, we
can estimate empirically what defines normatively appropriate action in the minds of São Paulo
residents. If, for example, procedural justice is a strong positive predictor of normative alignment, then
we infer from this is an important dimension of appropriate (i.e. legitimate) police conduct—it is, we
infer, a key element that defines what exercising authority in appropriate ways means in practice (for
discussion, see Huq et al., 2017; Jackson & Bradford, 2019; Trinkner, 2019). We turn to duty to obey
in study two.
Confirmatory factor analysis
To fit a four factor (plus distributive justice as a single indicator) CFA model, we use MPlus 7.2, setting
all indicators as categorical (ordinal). All latent constructs and the single indicator of distributive justice
were allowed to covary. The approximate fit statistics suggest that this model fits the data adequately
=[0.33, 0.045]; CFI=0.980, TLI=0.987). The factor loadings and R
statistics are fairly strong for all the measures, indicating reasonable scaling properties (procedural
justice: λ=0.71 to 0.78, R
=0.51 to 0.61; effectiveness: λ=0.67 to 0.83, R
=0.45 to 0.69; bounded
authority: λ=0.61 to 0.72, R
=0.37 to 0.52; normative alignment: λ=0.85 to 0.90, R
=0.72 to 0.80). The
constructs were all positive correlated, e.g. procedural justice and distributive justice (r=.76), procedural
justice and effectiveness (r=.66), distributive justice and effectiveness (r=.52), procedural justice and
bounded authority (r=.27), and bounded authority and effectiveness (r=.24).
Latent class analysis
Appendix A summarizes the analysis for the latent class analysis of people’s responses to the three
police violence vignettes. Briefly, the modeling implies that:
• 50% have mixed views about police use of force, albeit with a slight negative skew (the
• 42% are strongly against it (the ‘intolerant’ group); and,
• 8% are supporters of excessive police use of force (the ‘tolerant’ group).
In preparation for the next stage of analysis, the most likely (i.e. expected) latent class membership was
derived for each respondent given their scores on the various indicators.
Structural equation modelling
Figure 1 reports the results of a fitted structural equation model (SEM) that specifies normative
alignment as the outcome variable and procedural justice, distributive justice, effectiveness, bounded
authority and the two dummy variables for perceptions of police violence (“ambivalent towards police”
and “intolerant of police violence”) as predictor variables. The model controlled for age, gender and
each of the eight areas (the coefficients for these variables are omitted for visual ease).
Figure 1: Predicting normative alignment with the police
Fitted structural equation model with standardized coefficients. The reference category for the two attitudes towards police
violence dummy variables is ‘tolerant of police violence’. Controls were gender, age and seven dummy variables regarding
the eight areas.
Despite theoretical expectations derived from our reading of the social, political and legal
context of São Paulo, our results are consistent with those from Anglo-Saxon countries (e.g., Sunshine
and Tyler, 2003; Tyler & Fagan, 2008; Jackson et al., 2012, 2013; Murphy et al., 2016). Procedural
justice was the strongest positive predictor of normative alignment (β=.81, p<0.001). People who
believed that officers tended to treat people with respect and dignity, make fair and neutral decisions,
listen to people and explain their decisions also tended to be normatively aligned with the police (and
by inference, thought the police has the moral right to power). Effectiveness (β=.12, p<0.001), bounded
authority (β=.06, p=0.051) and distributive justice (β=.02, p=0.586) were relatively weak positive
predictors. Intolerance of and ambivalence towards excessive police violence was unrelated to
normative alignment in the fitted model (‘intolerant’ β=-.01, p=0.727, ‘ambivalent’ β=.02, p=0.484, the
reference category was ‘tolerant’). We should note that we also fitted a SEM without procedural justice,
distributive justice, effectiveness and bounded authority, and found similar results: ‘intolerant’ β=.05,
p=0.440 and ‘ambivalent’ β=.07, p=0.226.
On this basis, it seems that tolerance of excessive police violence is not a source of legitimation
in São Paulo, at least in the sense that those who tolerate police violence are not more likely to believe
Intolerant of police
RMSEA=.049, 95% CI [.045, .053]
that officers generally act in normatively appropriate ways. Our analysis suggests that what is most
important is procedural justice, which aligns with most research evidence from other social, political
and legal contexts (Jackson, 2018).
Study two: Examining the nature of legitimacy and its link to legal compliance in São Paulo
Study two turns to the obligation to obey part of the legitimacy concept and its connection to normative
alignment, fear of police and legal compliance. We show in the descriptive part of our findings that São
Paulo is a city in which a fair amount of people fear the police; some would obey an officer because of
fear of the consequences of disobedience; and some would disobey an officer for moral reasons. People
express a range of different ambivalent attitudes between the police.
This is relevant because PJT distinguishes between instrumental and normative police-citizen
relations, based on a fundamental distinction between value-based motivations to comply (that imply
consensual modes of policing based on fairness and legitimacy) and deterrence-based motivations to
comply (that imply coercive modes of policing on effectiveness and deterrence). PJT predicts that a
normative model of crime-control policy works better than an instrumental model. But can instrumental
and normative motivations be so easily separated in a city in which a significant number of people can
simultaneously fear police and view them as legitimate? Are, in other words, instrumental and
normative factors distinct in people’s minds, or should instrumental and normative factors be placed on
a continuum that ranges from instrumental at one end to normative at the other end, with some mixture
of the two intermingling in the middle? And what does this mean for the link between police legitimacy
and offending behavior?
Our analysis proceeds in four stages: (1) we test whether people in São Paulo have different
motives to feel and not feel obliged to obey the police, (2) consider the correlations between different
motivations, (3) we employ latent variable modeling, and (4) we use the resulting scale(s) to model
(1) Are there different motives to feel (and not feel) obligated to obey the police?
PJT frames the obligation to obey part of the legitimacy concept as comprising the presence or absence
of a normatively-grounded internalized moral value, i.e. that one should obey the commands of officers
because that is the right thing to do. But as Bottoms and Tankebe (2012), Tankebe (2009: 1279-1281,
2013: 105-106) and Johnson et al., (2014: 970) argue, people could feel an obligation to obey police for
non-normative reasons, including pragmatism, dull compulsion, and fear of the consequences of non-
compliance. Where there are high levels of police violence, malpractice and corruption, and where
police-community relations are often tense and conflictual, people could feel obligated to obey because
they feel defiance is dangerous and/or that they have little choice but to be obedient. We employ a
combination of closed- and open-ended survey questions to probe the issue—we use survey
respondents’ own assessments of why they might or might not obey the police to gauge the extent of
normative and non-normative obedience.
We also assess whether there is a normative obligation not to comply. Some people might
imagine police giving out illegitimate orders when answering the question “Do you think you have a
duty to obey the police when you think they are wrong?” They might imagine the content and/or context
of the order, and they might report a duty not to obey the police out of an active sense of moral protest.
This links back to the boundary concerns that we investigated in study one—where people assess
whether officers stray into spaces that they have no right to be—but it references a situation in which
respondents imagine themselves being ordered by the police to do something. We thus use the combined
closed- and open-ended survey questions to test the existence of normative and non-normative
(2) Are motivations to comply with the law distinct from each other?
We then investigate the correlations between fear of the police, different motives to obey or disobey,
and normative alignment with the police. If people can simultaneously fear the police and view them as
legitimate, then this would complicate the standard PJT account, summarized in the left hand-side of
Figure 2. The standard approach is to measure instrumental police-citizen relations using items like ‘I
only obey police because I am afraid of them’ (Posch et al., 2020) and ‘What is the likelihood that you
be caught and punished if you bought something you think might be stolen?’ (Sunshine & Tyler, 2003)
and normative police-citizen relations using items like ‘You should do what the police tell you, even if
you disagree’ (Jackson et al., 2012), ‘Obeying the law ultimately benefits everyone in the community’
(Gur & Jackson, 2020), ‘The police usually act in ways consistent with your own ideas about what is
right or wrong’ (Peyton et al., 2019), ‘I trust the leaders of the NYPD to make decisions that are good
for everyone in the city’ (Tyler & Fagan, 2008), and ‘How wrong is it to buy something you think might
be stolen?’ (Trinkner et al., 2018).
Figure 2: Standard and new approaches to instrumental and normative motivations to comply
with the law
Can such a clean differentiation between normative and instrumental police-citizen relations
(and motivations to comply with the law) be identified in São Paulo? To foreshadow the findings, we
find that a sizeable minority of people (around 20%) also report a normatively grounded obligation to
obey were also afraid of the police, and a sizeable minority of people (around 15%) who report a strong
sense normative alignment (i.e. they are also afraid of the police). If attitudes towards the police can be
ambivalent—if instrumental and normative police-citizen relations cannot be so easily differentiated—
then perhaps they are better represented not as two unipolar scales (the standard approach in Figure 2)
but as one bipolar scale that moves from instrumental to normative (the new approach in Figure 2)? At
the positive end of such a continuum could be normative alignment with the police, low fear of the
police, and normative forms of obligation to obey the police. At the negative end of the continuum
could be the lack of normative alignment, high fear of the police, and a felt obligation to obey the police
based on fear of non-compliance. In the middle could be ambivalent mix of instrumental and normative
(3) What are the predictors of legal compliance?
Taking the standard approach depicted in Figure 2, a number of studies have found that legitimacy
(defined as the presence or absence of consensual authority) is a stronger predictor of legal compliance
than fear of being caught, supporting the idea that legitimacy encourages value-based forms of self-
regulation (chiefly, cooperation and compliance) and instrumental motivations are both separate to
legitimacy and less important (Bolger & Walters, 2019). What about the relationship between
legitimacy and legal compliance in São Paulo? To foreshadow the results once more, we find that a one
latent trait model fits the data better than a two latent trait. By using the latent trait as a predictor of
offending behavior, and compare its predictive power against alternate measures of legitimacy, we can
test the idea that the more coercive the relationship between police and citizens (from the citizens’
perspective), the more likely they are to having reported committing a crime, and the more consensual
the relationship is between police and citizens, the less likely they are to having reported committing a
Coercive power and
instrumental motivators (high
scores = high fear of police
and being caught)
Consensual authority and normative
motivators (high scores = high legal
and police legitimacy and perceived
immorality of the criminal acts)
Coercive versus consensual
relations (single motivational
Data and methods
A representative survey of adults in the city of São Paulo was conducted between the months of June
and July 2015. A two-stage cluster sampling design was used. During the first stage, clusters were
randomly selected—census tracts of the city were polled, taking into consideration a systematic
Probability Proportional to Size (PPS) at the region level. In the second stage, after randomly selecting
houses, a fixed number of respondents was specified in each census tract, following specific quotas on
demographic variables defined considering gender, age, education, and occupation. All interviews were
conducted face-to-face in respondents’ homes using Tablet-Assisted Personal Interviewing (TAPI) (for
details, see NEV-USP, 2017). The final sample consisted of 1,804 respondents aged 16 and over living
in 96 districts of the city.
Measures and analytical strategy
To measure duty to obey, we use a traditional closed-ended measure (“Do you think you have a duty
to obey the police even when you believe they are wrong?”, with a binary “yes” or “no” response) and
an open-ended follow-up question (“Why do you think you have/do not have a duty to obey the police
even if you believe the police is wrong?”). We use thematic analysis on the open-ended question (for
further details see Appendix B). Normative alignment with the police is measured using a single
indicator: “The police act according to what I believe is right”, (1) never to (5) always. Personal and
general fear of the police was also measured (personal: “Are you afraid of the police?”, with three
responses of “no”, “undecided” and “yes”; general: “How often would you say that people are afraid
of the police?”, “never” to “always” on a 1 to 5 Likert-scale).
We use latent trait analysis to assess the scaling properties of the different motives to obey or
disobey the police, normative alignment and fear of police. In addition to answers to the closed-ended
obligation question (binary) and the categories derived based on the open-ended follow-up (nominal),
normative alignment (ordinal) and general and personal fear of the police (ordinal) are included in the
latent trait modelling. We test whether a two or a one latent trait model fits the data best (see Figure
2), with the two trait model distinguishing between instrumental and normative police-citizen relations,
while the one trait model places instrumental factors at one end (fear of the police, normative motives
to disobey the police, instrumental motives to obey the police, and the absence of normative alignment)
and normative factors at the other end (fear of police, normative motivations to obey the police, and
the presence of normative alignment).
We then use the resulting measurement model to predict self-reported compliance with the
law, adjusting for personal morality. To measure personal morality, five questions were asked: “Do
you think it is right or wrong that people: (i) try to bribe a traffic warden to avoid a fine?; (ii) buy
counterfeit goods?; (iii) use cable TV signal without paying for it?; (iv) buy goods without a receipt to
pay less?; and (v) pay for a private doctor or a private dentist without a receipt?”. Answers were either
“right” or “wrong”. To measure compliance with the law, these same five items were asked, but
considering the question: “Have you ever…?”, and answers could be either “yes” or “no”. For both
personal morality and offending the “no” and “wrong” answers were coded as 0, and the “yes” and
“right” answers as 1. For the final scales of personal morality and offending, the responses from these
five variables were summed, such that two six category variables representing offending behavior and
(associated) personal morality judgements.
We assess whether legitimacy predicts offending behavior, adjusting for geographical region,
gender (self-identified male or female), age, income (with regards to the minimum wage, six
categories), education (seven categories), ethnicity (White, Black, ‘Mulato’, Asian, Native Brazilian
and personal morality. Measures of duty to obey or the coercive power to consensual authority
continuum trait scores were included with personal morality and demographic variables (gender, age,
income, education, and ethnicity). Regional clustering was considered when estimating the standard
errors. In the current data, missing data is an issue—in some cases pursuing complete case analysis
would have resulted in losing almost quarter of the data. Multiple imputation (Lall, 2016; Lang & Wu,
2017) was carried out using chained equations and thirty-seven variables with partially missing values
from the dataset Because the highest proportion of missingness among the variables of interest was
approximately 13%, thirteen imputations were done following von Hippel’s (2009) recommendation.
Based on the nature of the outcome variable, ordered logistic regression models
were fitted with
clustered robust standard errors for the regions considering the clustering of the sampling design. We
specify legitimacy in the fitted regression models according to the 3 stages outlined above: (1) as a
simple yes/no binary variable of duty to obey, (2) as the different categories created based on
combining the closed-ended and open-ended responses of duty to obey, and (3) as a single latent trait,
with normative considerations at one end, and instrumental considerations on the other.
Are there different motives to feel obligated (and not feel obligated) to obey the police?
Some 74% of respondents who reported feeling a duty to obey police. Of these, just under half (34%
of all respondents and 46% of those who said they would obey the police) gave a reason that indicated
a sense of normatively-grounded obligation (Table 2 contains illustrative examples of the responses
given to the open-ended question). This makes up the “acceptance of authority” group. The remainder
of those who reported feeling obligated to obey (40% of all respondents and 54% of those who said
they would obey the police) indicated that they obeyed for instrumental reasons. This makes up the
“coercive obligation” group. Turning to the 26% of respondents who reported not feeling obligated to
obey, just over two thirds of them (18% of all respondents and 69% of those who said they would
disobey the police) believed that the police lack moral authority, making up the “rejection to authority”
category. The rest of the respondents’ (8% of all respondents and 31% of those who said they would
disobey the police) answers implied some sense that officers were acting in an immoral way,
potentially giving illegitimate orders, and that not complying is therefore the right thing to do, making
up the “disobedient protest” category.
Table 2: Answers to the closed-ended duty to obey question and the four categories derived by
the content analysis of the open-ended follow-up question
Do you think you have a duty to obey the police even when you
believe they are wrong?
Acceptance of authority
"It is the citizen's duty to act in
accordance with the law";
"The police officer represents
the law and must be
"We must obey the police,
whether it's right or wrong"
"They are wrong and if I obey I
will be going against my beliefs";
"Laws are for everyone and I can
call internal affairs";
“I must make the officer
understand my point of view”
Rejection of authority
"Because you cannot trust them";
"Because they wear uniforms and
have the prerogative to say what
the law is, and then engage in
abuse of authority";
"Because they are corrupt and
are worse than many outlaws"
"So that I do not suffer the
"I must obey because if I do not
I can be imprisoned and
accused of contempt of
"If you do not obey, you can get
A significant minority of respondents reported being afraid of the police (around 30%). Table 3 shows
that levels of fear were lower among those individuals who reported a normative sense of obligation to
obey (20%) and a normatively-grounded duty to disobey the police (25%). Levels of fear were higher
among individuals who reported they would not comply because they viewed the police as illegitimate
(32%) and those who said they would comply because they feared the consequences of non-compliance
(39%). However, it is noticeable (a) that a surprisingly high number of those who expressed normative
forms of obedience or disobedience also feared the police, and (b) a surprisingly high number of those
who expressed an instrumental form of obligation to obey did not also fear the police.
Table 3: Fear of the police and forms of obligation to obey/not to obey the police
Are you afraid of the police?
Normative acceptance of authority 385 (80%) 99 (20%) 484 (100%)
Disobedient protest (normative) 68 (75%) 23 (25%) 91 (100%)
Coercive obligation (instrumental) 321 (61%) 209 (39%) 530 (100%)
Normative rejection of authority 178 (68%) 83 (32%) 261 (100%)
Total 952 (70%) 414 (30%) 1366 (100%)
The association between fear of the police and normative alignment with the police is a little
stronger (Table 4). Of those who strongly agreed with the statement ‘The police act according to what
I believe is right’, 15% were afraid of the police, compared to 48% of those who strongly disagreed
with the statement. Again, though, there does seem to be a certain amount of ambivalence—on the face
of it, it is puzzle why would people who believe that the police act in normatively appropriate ways
would also fear them.
Table 4: Fear of the police and normative alignment with the police
Are you afraid of the police?
Normative alignment with the
Never act appropriately 150 (52%) 140 (48%) 290 (100%)
Rarely act appropriately 126 (55%) 103 (45%) 229 (100%)
Sometimes act appropriately 259 (72%) 101 (28%) 360 (100%)
Almost always act appropriately 206 (77%) 60 (23%) 266 (100%)
Always act appropriately 285 (85%) 50 (15%) 335 (100%)
Total 1026 (69%) 454 (31%) 1480 (100%)
Are instrumental and normative police-citizen relations two ends of a single continuum?
Appendix C provides complete results of the latent variable modelling, but briefly, we find that a two
latent trait model that distinguishes between instrumental and normative factors fits the data, but a one
latent trait model (with instrumental factors at one end of the continuum and normative factors at the
other) fits the data even better. We proceed with the one trait solution for reasons of parsimony (there
is value in going with the simplest solution) and because the descriptive findings above show
ambivalence in the overlapping nature of instrumental and normative sentiments.
To get a better sense of what a certain score of the coercive to consensual continuum
corresponds to, it is worth noting that—as with any latent trait—its mean is close to zero (M=-0.001)
and it is normally distributed (SD=0.84, Min=-2.21, Max=2.29). Table 5 shows how the continuum is
constituted via fitted probabilities of different levels of fear of the police, obligation to obey and
normative alignment, conditional on continuum placement, holding all other variables constant at their
mean/reference category. First, levels of personal and general fear of the police are high at the -2SD
(coercive) end of the continuum, low at the +2SD (consensual) end of the continuum, and below the
middle at the mean (the probability of being personally afraid of the police for people at the mean of
the continuum is 0.21) and the probability of saying that people are ‘almost always’ or ‘always’ afraid
of the police is 0.66).
Second, people at the -2SD (coercive) end of the continuum are likely to either feel a coercive
obligation to obey the police (probability of .48) or no obligation to obey the police because they do not
think officers have rightful authority (probability of .41). People at the +2SD (consensual) end of the
continuum are likely to feel an obligation to obey the police based on consent (probability of .73),
although interestingly, there is a .16 probability of feeling a coercive obligation to obey the police.
People at the mean of the continuum had a 0.31 probability of feeling a consensual obligation to obey
the police and a 0.41 probability of feeling a coercive obligation to obey the police.
Third, people at the -2SD (coercive) end of the continuum are highly likely to feel that the
police ‘never’ act according to what they believe is right (probability of .88). People at the +2SD
(consensual) end of the continuum are highly likely to feel that the police ‘always’ act according to
what they believe is right (probability of .90). People at the mean of the continuum had a 0.17
probability of saying ‘rarely’, a 0.39 probability of saying ‘sometimes’ and a .22 probability of saying
On this account, it does not seem to be the case that people in São Paulo have two distinct sets
of attitudes toward police, one marked by normative concerns and the other by instrumental concerns.
Rather, we find these attitudes exist in a symmetrical relationship at the two extremes and are
intermingled in the middle of the continuum.
Table 5: Fitted probabilities to illustrate the composition of the coercive-consensual continuum
Fitted probabilities Personal fear of the police – Are you afraid of the police?
No Don’t know Yes
-2SD 0.091 0.082 0.827
-1SD 0.297 0.171 0.533
Mean 0.639 0.148 0.213
+1SD 0.882 0.058 0.060
General fear of the police – How often would you say that people are afraid of the
Never Rarely Sometimes Almost always Always
-2SD 0.008 0.012 0.060 0.160 0.760
-1SD 0.020 0.028 0.127 0.259 0.567
Mean 0.046 0.061 0.230 0.310 0.352
+1SD 0.105 0.121 0.325 0.265 0.184
Duty to obey content analysis categories
resistance Coercive Rejection of authority
-2SD 0.060 0.056 0.475 0.409
-1SD 0.148 0.078 0.474 0.300
Mean 0.312 0.094 0.406 0.189
+2SD 0.730 0.071 0.159 0.039
Fitted probabilities Normative alignment – The police act according to what I believe is right
Never Rarely Sometimes Almost always Always
-2SD 0.878 0.083 0.032 0.006 0.002
-1SD 0.469 0.280 0.192 0.043 0.016
Mean 0.098 0.172 0.390 0.220 0.119
+1SD 0.013 0.030 0.151 0.282 0.524
Predictors of legal compliance
Finally, we turn to the question of whether duty to obey and coercive to consensual continuum predict
self-reported compliance. The results of the regression models can be found in Table 6. Model 1 uses
the original binary duty to obey variable, Model 2 the four content analysis categories, while Model 3
the coercive power to consensual authority continuum latent trait. In the first model, only personal
morality is significant; ceteris paribus, a one-unit increase in the score of personal morality is associated
with 46% decrease in the odds of reporting higher levels of offending behavior. In contrast, the original
duty to obey variable does not seem to have an impact on offending behavior, controlling for all else.
Table 6: Ordinal regression analysis with odds ratios, multiple imputation and clustered robust
standard errors for the sampling regions in squared brackets
Model1 Model2 Model3
Duty to obey 1.107
Content analysis categories
Acceptance of authority 0.700***
Rejection of authority 0.809
Coercive/consensual continuum 0.795***
Personal morality 0.540***
1-2 times the minimum wage 1.043
2-5 times the minimum wage 1.455**
5-10 times the minimum wage 1.565**
10-20 times the minimum wage 1.539
20+ times the minimum wage 1.001
Literate, no schooling 0.308***
Incomplete primary school
Complete primary school 0.608
Incomplete elementary school 0.873
Complete elementary school
Incomplete high school 1.018
Complete high school 1.192
Incomplete college 1.268
Complete college 0.928
Black (African/Caribbean descent) 1.069
Mulato (mixed descent) 0.989
Yellow (Asian descent) 0.739
Native Brazilian 0.794
Cutoff1 0.015 0.013 0.016
Cutoff2 0.063 0.055 0.068
Cutoff3 0.227 0.200 0.247
Cutoff4 1.081 0.954 1.184
N 1804 1804 1804
*p<0.05, **p<0.01, ***p<0.001 Reference categories: minimum wage, illiterate, white, male, coercive obligation (Model 2)
Model 2 finds a similar partial association for personal morality, where a one-unit increase in
morality decreases the fitted odds of higher levels of offending by 46%. From the content analysis
categories with coercive obligation as reference category, only acceptance of authority emerges as a
significant predictor. All else equal, belonging to the acceptance of authority category instead of the
coercive obligation category reduces the fitted odds of higher levels of reported offending by 30%.
Neither belonging to the disobedient protest category, nor belonging to the rejection of authority
category, are significant predictors of offending behavior. Finally, Model 3
finds comparable statistical
effects for personal morality, where holding all else constant, a single unit increase in morality decreases
the fitted odds of higher levels of offending by 45.4%. Yet, here, the latent trait of the coercive power
to consensual authority continuum is also significant: a one-unit increase in the latent trait reduces the
fitted odds of higher levels of offending by 21%.
Another way of interpreting the results of Model 3 is to derive the fitted probabilities of the
outcome, conditional on the changes in the coercive power to consensual authority continuum latent
trait, while holding all other variables constant at their mean/reference category. Results are presented
in Table 7, and show that a one standard deviation increase in the coercive power to consensual authority
continuum latent trait corresponds, approximately, to a 5 percentage-point expected increase in the
probability of reporting not to have committed any crime. In other words, the more normatively
appropriate respondents felt their relationship with the police was, the less likely they were to report
committing crimes. Conversely, those whose relationship with police was based on fear and a lack of
consent were more likely to report breaking the law.
Table 7: Fitted probabilities of number of crimes committed for Model 3
Fitted probabilities Number of crimes committed
-2SD 0.251 0.310 0.242 0.141 0.045 0.010
-1SD 0.306 0.321 0.216 0.114 0.034 0.008
Mean 0.367 0.322 0.188 0.091 0.026 0.006
+2SD 0.500 0.293 0.132 0.056 0.015 0.003
The past decade has seen an increasing number of tests of procedural justice theory in the US, UK,
Australia, as well as Israel, China, South Africa, Japan, Ghana, Pakistan and countries across continental
Europe (for a review of the international literature, see Jackson, 2018). Most of these studies have shown
(a) that the main factor explaining variation in perceived legitimacy (procedural justice) is relatively
constant across multiple contexts, (b) that normative motivations of legitimacy and personal morality
are distinct from instrumental motivations to comply and cooperate that focus on fear, risk and reward,
and (c) that legitimacy is more strongly associated with self-reported compliance with the law (Walters
& Bolger, 2019) and cooperation with legal authorities (Bolger & Walters, 2019) than instrumental
factors like the perceived risk of sanction (although see Glöckner et al., 2019).
São Paulo is a city with a long history of crime, police violence, and relatively widespread fear
of the police. The starting premise for our two studies was that the social, political and legal history and
context may have important implications for how we think about the sources of legitimacy, the nature
of instrumental and normative police-citizen relations, and the link between legitimacy and legal
compliance. We sought to address a few ways in which some of the central tenets of procedural justice
theory (PJT) might need to adapt to the reality of policing in this city of the Global South
In PJT, legitimacy is typically defined along two connected dimensions: (1) normative
justifiability of power in the eyes of citizens and (2) willing citizen consent to the legal demands
associated with an authority. Our two studies map onto each of these two constituent parts. Study one
contributed to the increasingly international literature on the dimensions of police conduct that
legitimate (or delegitimize) the police in the eyes of citizens. We found in study one that a fair amount
of individuals in São Paulo tolerate—or at least are ambivalent towards—police use of excessive force
against certain out-groups, and moreover that significant numbers believe that the police are ineffective
against crime and believe the police lack distributive justice. We asked, given these findings, whether
tolerance of excessive violence partly legitimates the police in the eyes of citizens (for discussion of the
theoretical and empirical links between legitimacy and attitudes towards citizen and police violence,
see Jackson et al., 2013, Gerber & Jackson, 2017, and Bradford et al., 2017), and whether effectiveness
and distributive justice are also important legitimating factors.
Contrary to expectations—and consistent with previous research—we found that procedural
justice was key to police legitimation in São Paulo. PJT distinguishes between normative sources of
legitimacy (procedural justice and bounded authority) and instrumental sources of legitimacy
(effectiveness and distributive justice) and we found that procedural justice had by far the strongest
partial association with normative alignment, followed by police effectiveness and respect for
boundaries, both of which were relatively weak predictors. Our results imply that even in a low trust
and high fear city like São Paulo, where certain forms of excessive use of force by the police are not
firmly rejected, the key normative component of appropriate police behavior (procedural justice) is the
best predictor of perceived appropriateness of the police as an institution.
Study two then turned to the obligation to obey part of the legitimacy concept: namely, the
belief ‘…that it is appropriate and right for some external authority to make decisions about law and
legal policy and that they ought to voluntarily follow those decisions, without concerns about reward
and punishment’ (Tyler & Trinkner, 2018: 39). PJT scholars make a sharp distinction between
consensual and coercive policing and police-citizen motivations to comply with the law (and the police).
The first assumes that rule-following is rooted in norms and values, whereby people follow rules and
commands because they think it is the right thing to do. The second assumes that rule-following is
rooted in instrumental judgements of risk and punishment, whereby people obey the law and police
because they fear the consequences of non-compliance.
Contrary to work from the US, UK and Australia, we showed in São Paulo that legitimacy can
be usefully conceived as existing as one pole on a coercive to consensual continuum, moving from an
instrumental relationship at one end to a normative relationship at the other. We found in São Paulo that
people’s obligation to obey the police are more complex and varied: some people reported feeling a
duty to obey the police ‘even if they think the officer is wrong’, suggesting a voluntary deference that
flows from internalization of group norms and values; others reported feeling a duty to obey because
they were concerned about the consequences of non-compliance. Our analysis indicated that the
instrumental police-citizen relationship is based on fear of the police and the absence of normative
appropriateness and normative obligation. By contrast, the normative police-citizen relationship is
based on a lack of fear of the police, an absence of instrumental obligation, and the presence of
normative appropriateness and normative obligation to obey police. People who fall in the middle of
the continuum exhibit both instrumental and normative connections to the police.
Prior tests of PJT have found that legitimacy predicts legal compliance, adjusting for fear of
being caught by the police and personal morality (Sunshine & Tyler, 2003; Jackson et al., 2012; Murphy
et al., 2016; Trinkner et al., 2018). We found that the resulting coercive to consensual continuum was a
positive predictor of legal compliance, adjusting for people’s views on the immorality of the illegal acts
under investigation. People who were more likely to engage in criminal activities were also more likely
to fear the police, more likely to feel obligated to obey officers for instrumental reasons, and more likely
to not believe that the police lacks normative appropriateness. This contrasts with extant work in the
US, for instance, that approaches something akin to fear of the police (or at least fear of being caught
by the police if one were to commit a crime) as conceptually distinct from legitimacy (indeed orthogonal
to) and find that legitimacy is a stronger separate predictor of legal compliance than instrumental factors
(Sunshine & Tyler, 2003; Tyler, 2006a, 2006b; Trinkner et al., 2018).
This is striking for four reasons. First, it suggests that the sense of normative obedience toward
authority should not in São Paulo be considered as fundamentally distinct from an understanding that
obedience is necessary, a rational (instrumental) response to an often brutal police agency. A fair
number of respondents in study two felt that they ‘should’ obey police instructions, but they had two
distinctly different reasons for doing so. In some cases, obedience was normative, and thus premised in
and representative of a relationship with police marked by legitimacy. In other cases, though, obedience
was prudential and essentially coerced.
Second, we found that only normatively grounded obedience of police was linked to greater
compliance with the law; those who indicated they would obey police through fear were less likely to
say they would obey the law. This finding may have important implications for how we think about
instrumental compliance in criminal justice settings. It is often assumed that even if people do not
comply with the law for legitimacy-based reasons, then instrumental motivations may still be in play,
at least in some times and places. On this account, if all else fails, aggressive policing that demonstrates
the risks of non-compliance can hold the line. Yet our results suggest that when people obey police
through fear this may not only fail to increase legal compliance but actually diminish it. Policing styles
oriented toward increasing prudential compliance, at least as far as this is generated by fear of the police,
may therefore be actively counter-productive, and indeed criminogenic.
Third, this last point may have relevance for policing beyond Brazil because, we suspect, the
notion that legitimacy exists on a continuum with prudential or coerced compliance will be relevant in
many other contexts as well. There are two somewhat contrasting ways to position this argument. First,
to the extent that legitimacy is founded in norms of fairness and due process, to say one would obey
police out of fear of the consequences of disobedience is in some senses inevitably to indicate the
absence of legitimacy. A relationship based on fear is antithetical to legitimacy, not something that
merely runs alongside it, and this may be the case outside Brazil as well as within it. Second, ‘obedience’
may be better positioned on continuum representing ‘reasons for obedience’ because, in the end, most
people will comply with police instructions (i.e. they will obey). Bittner argues that police are “a
mechanism for the distribution of nonnegotiably coercive force” (1990:131). Non-compliance with
police instructions will always risk violence, and if the policed do not comply through the mechanism
of legitimacy they will be coerced, ultimately at the end of a gun (in most countries at least). To put
it another way, compliance with police instructions will be forthcoming in the vast majority of cases,
whether it stems from normative or instrumental concerns, and this brute fact should be taken into
account in our models of public obedience toward police. What is at stake in PJT, of course, is
precisely the reason for obedience, the claim being that normative compliance (willing consent) is
both ethically more desirable and a more sustainable model for police authority and the exercise of
power. In as much as they suggest that coerced compliance may be actively criminogenic, our results
here support this latter claim.
Finally, our findings speak to the difficulty of measuring police legitimacy in contexts outside
advanced liberal democracies. But despite such difficulties we have demonstrated that ‘the moral duty
to obey’ can be measured in countries where the question might prompt ambiguous responses, provided
that proper methodological care is taken. In study two we used a combination of methods, content
analysis followed by latent trait analysis, to capture people’s understanding of legitimacy of the police.
We would encourage other researchers to employ similar techniques. At the same time, we also want
to emphasize the need for more thorough question testing. It is possible that there are still better
measures of duty to obey that could be successfully fielded both in high legitimacy and low legitimacy
Batitucci, Eduardo Cerqueira. 2010. “A Evolução Institucional da Polícia no Século XIX: Inglaterra,
Estados Unidos e Brasil em Perspectiva Comparada.” R. Brasileira de Segurança Pública, 4: 30-
Bittner, Egon. 1990. Aspects of Police Work. Boston: Northeastern University Press.
Blader, Steven L. and Tom R. Tyler. 2009. “Testing and Extending the Group Engagement Model:
Linkages between Social Identity, Procedural Justice, Economic Outcomes, and Extrarole
Behavior.” Journal of Applied Psychology 94 (2): 445–64.
Bolger, Colin P. and Glenn D. Walters. 2019. “The Relationship between Police Procedural Justice,
Police Legitimacy, and People’s Willingness to Cooperate with Law Enforcement: A Meta-
analysis.” J. of Criminal Justice, 60 (1): 93-9.
Bottoms, Antony and Justice Tankebe. 2012. “Beyond Procedural Justice: A Dialogic Approach to
Legitimacy in Criminal Justice.” J. of Criminal Law & Criminology 102(1):119–70.
Bradford, Ben, Jenni Milani and Jonathan Jackson. 2017. “Identity, Legitimacy and ‘Making Sense’ of
Police Use of Force.” Policing: An International J. of Police Strategies & Management, 40 (3):
Bradford, Ben, Kristina Murphy and Jonathan Jackson. 2014a. “Officers as Mirrors: Policing,
Procedural Justice and the (Re)production of Social Identity.” British J. of Criminology, 54 (4):
Bradford, Ben, Aziz Huq, Jonathan Jackson and Ben Roberts. 2014b. “What Price Fairness When
Security Is at Stake? Police Legitimacy in South Africa.” Regulation & Governance 8 (2): 246–
Bradford, Ben. 2014. "Policing and social identity: Procedural justice, inclusion and cooperation
between police and public." Policing & Society 24 (1): 22-43.
Briceño-León, Roberto, Leandro Piquet Carneiro, and José Miguel Cruz. 1999. "O Apoio dos Cidadãos
à Ação Extrajudicial da Polícia no Brasil, em El Salvador e na Venezuela." Pandolfi, Dulce
Chaves et alii (orgs.), Cidadania, Justiça e Violência, Rio de Janeiro, Fundação Getúlio Vargas.
Caldeira, Teresa P. R. 2002. "The Paradox of Police Violence in Democratic Brazil." Ethnography 3
Caldeira, Teresa P. R. and James Holston. 1999. "Democracy and Violence in Brazil." Comparative
Studies in Society and History 41 (4): 691-729.
Cardia, Nancy, Frederico Castelo Branco, Viviane de Oliveira Cubas, Renato Alves, and Gustavo Higa.
2014. "PercePções Sobre a Justiça entre Moradores da Cidade de São Paulo: 2001-2013." Revista
USP 101: 159-72.
de Oliveira Junior, Almir and Rafael Augusto da Costa Alencar. 2015. “A Predisposição para Chamar
a Polícia: Um Estudo sobre a Percepção do Desempenho e da Confiabilidade das Instituições
Policiais.” Rev. Brasileira de Segurança Pública 9 (1): 158-70.
Dias, Camila Caldeira Nunes. 2011. "Da pulverização ao monopólio da violência: expansão e
consolidação do Primeiro Comando da Capital (PCC) no sistema carcerário paulista." PhD diss.,
Universidade de São Paulo.
Filgueiras, Fernando. 2013. "Perceptions on Justice, the Judiciary and Democracy." Brazilian Political
Science R. 7 (2): 62-87.
Forum Brasileiro de Segurança Pública. 2017. "11º Anuário Brasileiro de Segurança Pública." São
Paulo: Fórum Brasileiro de Seguranca Publica.
Garmany, Jeff. 2014. “Space for the State? Police, Violence, and Urban Poverty in Brazil.” Annals of
the Association of American Geographers 104 (6): 1239-255.
Gerber, Monica M. and Jonathan Jackson. 2017. “Justifying Violence: Legitimacy, Ideology and Public
Support for Police Use of Force.” Psychology, Crime & Law 23 (1): 79-95.
Glöckner, Andreas, Berenike Waubert de Puiseau, and Emanuel V. Towfigh. 2019. “Integrating
Theories of Law Obedience: How Utility-theoretic Factors, Legitimacy, and Lack of Self-control
Influence Decisions to Commit Low-level Crimes. Judgement and Decision Making 14 (3): 318-
Goldstein, Donna M. 2003. Laughter Out of Place: Race, Class, Violence, and Sexuality in a Rio
Shantytown. Berkeley: University of California Press.
Guimarães, Juliany Gonçalves, Ana Raquel Rosas Torres and Margareth RGV De Faria. 2005
“Democracia e Violência Policial: O Caso da Polícia Militar.” Psicologia em Estudo 10 (2): 263-
Huggins, Martha K. 2000. “Legacies of Authoritarianism: Brazilian Torturers´ and Murderers´
Reformulation of Memory.” Latin American Perspectives 27 (2): 57-78.
Huq, Aziz Z, Jonathan Jackson and Rick J. Trinkner. 2017. “Legitimating Practices: Revisiting the
Predicates of Police Legitimacy.” British J. of Criminology 57 (5): 1101–22.
Jackson, Jonathan, Aziz Z. Huq, Ben Bradford and Tom R. Tyler. 2013. “Monopolizing Force? Police
Legitimacy and Public Attitudes toward the Acceptability of Violence.” Psychology, Public Policy
& Law 19 (4): 479–97.
Jackson, Jonathan, Ben Bradford, Betsy Stanko and Katrin Hohl. 2013. Just Authority? Trust in the
Police in England and Wales. Oxon: Routledge.
Jackson, Jonathan, Ben Bradford, Mike Hough, Andy Myhill, Paul Quinton and Tom R. Tyler. 2012.
“Why do People Comply with the Law? Legitimacy and the Influence of Legal Institutions.”
British J. of Criminology 52 (6): 1051-71.
Jackson, Jonathan, Ben Bradford, Sarah MacQueen and Mike Hough. 2020. “Truly Free Consent?
Clarifying the Nature of Police Legitimacy Using Causal Mediation Analysis of Field Experiment
Data”. Available at SSRN: https://ssrn.com/abstract=2620274 or
Jackson, Jonathan, Muhammad Asif, Ben Bradford and Muhammad Zakria Zakar. 2014. “Corruption
and Police Legitimacy in Lahore, Pakistan.” British J. of Criminology 54 (6): 1067-88.
Jackson, Jonathan and Ben Bradford. 2019. “Blurring the Distinction between Empirical and Normative
Legitimacy? A Methodological Commentary on ‘Police Legitimacy and Citizen Cooperation in
China’.” Asian J. of Criminology 14 (4): 265-89.
Johnson, Devon, Edward R. Maguire and Joseph B. Kuhns. 2014. “Public perceptions of the legitimacy
of the law and legal authorities: Evidence from the Caribbean.” Law & Society Rev. 48 (4): 947-
Jonathan-Zamir, Tal and David Weisburd. 2013. “The Effects of Security Threats on Antecedents of
Police Legitimacy: Findings from a Quasi-experiment in Israel.” J. of Research in Crime and
Delinquency 50 (1): 3-32.
Kuha, Jouni. 2004. “AIC and BIC: Comparisons of Assumptions and Performance.” Sociological
Methods & Research 33 (2): 188–229.
Lall, Ranjit. 2016. “How Multiple Imputation Makes a Difference.” Political Analysis 24 (4): 414-33.
Lang, Kyle M. and Wei Wu. 2017. “A Comparison of Methods for Creating Multiple Imputations of
Nominal Variables.” Multivariate Behavioral Research 52 (3): 290-304.
Lind, E. Allan and Tom R. Tyler. 1988. The Social Psychology of Procedural Justice. New York:
Mazerolle, Lorraine, Emma Antrobus, Sarah Bennett and Tom R. Tyler. 2013. "Shaping Citizen
Perceptions of Police Legitimacy: A Randomized Field Trial of Procedural Justice." Criminology
51 (1): 33-63.
Meares, Tracey. 2016. “Policing and Procedural Justice: Shaping Citizens' Identities to Increase
Democratic Participation.” Northwestern University Law R. 111 (6): 1525-35.
Mentovich, Avital, Guy Ben‐Porat, Natalie Levy, Phillip A. Goff and Tom Tyler. 2018. "Policing
Alienated Minorities in Divided Cities." Regulation & Governance
Mitchell, Michael J.,and Charles H. Wood. 1999. "Ironies of Citizenship: Skin Color, Police Brutality,
and the Challenge to Democracy in Brazil." Social Forces 77 (3): 1001-20.
Murphy, Kristina, Ben Bradford and Jonathan Jackson. 2016. “Motivating Compliance Behavior among
Offenders: Procedural Justice or Deterrence?” Criminal Justice & Behavior 43 (1): 102-18.
Nylund, Karen L., Tihomir Asparouhov and Bengt O. Muthén. 2007. “Deciding on the Number of
Classes in Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo Simulation
Study.” Structural Equation Modeling 14 (4): 535-69.
Oliveira Júnior, Almir. 2011. “Dá para Confiar nas Polícias? Confiança e Percepção Social das Polícias
no Brasil.” Rev. Brasileira de Segurança Pública 9 (2): 6-22.
Oliveira, Thiago R., André Rodrigues de Oliveira and Sergio Adorno. 2019. “Legitimidade da Polícia:
Um Modelo de Mensuração.” Rev. Brasileira de Ciências Sociais, 34 (100): 1-25.
Oliveira, Thiago R., André Zanetic and Ariadne Lima Natal. Forthcoming. “Preditores e Impactos da
Legitimidade Policial: Testando a Teoria da Justeza Procedimental em São Paulo.” Dados –
Revista de Ciências Sociais.
Paes-Machado, Eduardo and Ceci Vilar Noronha. 2002. “Policing the Brazilian Poor: Resistance to and
Acceptance of Police Brutality in Urban Popular Classes (Salvador, Brazil).” International
Criminal Justice R. 12 (1): 53-76.
Papachristos, Adrew V., Tracey L. Meares and Jeffrey Fagan. 2012. “Why do Criminals Obey the Law?
The Influence of Legitimacy and Social Networks on Active Gun Offenders.” J. of Criminal Law
& Criminology 102: 397-440.
Pereira, Anthony W. and Mark Ungar. 2004 “The Persistance of the Mano Duro: Authoritarian
Legacies and Policing in Brazil and the Southern Cone,” in Authoritarian Legacies and
Democracy in Latin American and Southern Europe, edited by Katherine Hite and Paola Cesarini.
Notre Dame: Notre Dame University Press.
Peres, Maria Fernanda Tourinho, Nancy Cardia, Paulo de Mesquita Neto, Patrícia Carla dos Santos and
Sérgio Adorno. 2008. “Homicídios, Desenvolvimento Socioeconômico e Violência Policial no
Município de São Paulo, Brasil.” Revista Panamericana de Salud Publica 23: 268-76.
Peyton, Kyle, Michael Sierra-Arévalo, and David G. Rand. 2019. “A Field Experiment on Community
Policing and Police Legitimacy.” Proceedings of the National Academy of Sciences, 116 (40):
Pinheiro, Paulo Sérgio. 1991. “Autoritarismo e Transição.” Rev.evista Universidade de São Paulo, 9:
Pósch, Krisztián. 2019. “Testing Complex Social Theories with Causal Mediation Analysis and G-
computation: Toward a Better Way to do Causal Structural Equation Modeling. Sociological
Methods & Research, 0049124119826159.
Reisig, Michael D., Jason Bratton and Marc G. Gertz. 2007. “The Construct Validity and Refinement
of Process-based Policing Measures. Criminal Justice & Behavior 34 (8): 1005-1028.
Ruotti, Caren. 2016. “Pretensão de Legitimidade do PCC: Justificação e Reconhecimento de suas
Práticas nas Periferias da Cidade de São Paulo.” PhD. Dissertation. Universidade de São Paulo.
Schwarcz, Lilia Moritz and Heloisa Murgel Starling. 2015. Brasil: Uma Biografia. São Paulo, Brazil:
Silva, Geélison F. and Cláudio Beato. 2013. “Confiança na Polícia em Minas Gerais: O Efeito da
Percepção de Eficiência e do Contato Individual.” Opinião Pública, 19 (1): 118-53.
Sturgis, Patrick, Ian Brunton‐Smith and Jonathan Jackson. 2019. “Regression-based Response Probing
for Assessing the Validity of Survey Questions’, in Advances in Questionnaire Design,
Development, Evaluation and Testing, edited by Beatty, P., Willis, G. and Padilla, J-L. Wiley, pp.
Sunshine, Jason and Tom R. Tyler. 2003. “The Roles of Procedural Justice and Legitimacy in Shaping
Public Support for Policing.” Law & Society R., 37 (3): 513-48.
Tankebe, Justice. 2009. "Public Cooperation with the Police in Ghana: Does Procedural Fairness
Matter?" Criminology 47 (4): 1265-93.
Trinkner, Rick, Herbert Rodrigues, Debora Piccirillo, Faith E. Gifford and Aline Morais Mizutani
Gomes. 2019. “Legal socialization in Brazil: Examining the generalizability of the procedural
justice model.” International J. of Comparative and Applied Criminal Justice,
Trinkner, Rick, Jonathan Jackson and Tom R. Tyler. 2018. “Bounded Authority: Expanding
“Appropriate” Police Behavior Beyond Procedural Justice.” Law & Human Behavior, 42 (3): 280-
Trinkner, Rick. 2019. “Clarifying the Contours of the Police Legitimacy Measurement Debate: A
Response to Cao and Graham.” Asian J. of Criminology 14 (4): 309–35.
Trinkner, Rick, and Tom R. Tyler. 2016. "Legal Socialization: Coercion versus Consent in an Era of
Mistrust." Annual Review of Law & Social Science, 12: 417-39.
Tyler, Tom R. 2006a. Why People Obey the Law. New Haven, CT: Yale University Press.
Tyler, Tom R. 2006b. “Psychological Perspectives on Legitimacy and Legitimation.” Annual R. of
Psychology 57: 375-400.
Tyler, Tom R. and Jonathan Jackson. 2013. “Future Challenges in the Study of Legitimacy and Criminal
Justice” in Legitimacy and Criminal Justice: An International Exploration, edited by Justice
Tankebe and Alison Liebling. Oxford: Oxford University Press, pp. 83-104.
Tyler, Tom R. and Jonathan Jackson. 2014. "Popular Legitimacy and the Exercise of Legal Authority:
Motivating Compliance, Cooperation, and Engagement." Psychology, Public Policy and Law 20
Tyler, Tom R. and Rick Trinkner 2017. Why Children Follow Rules: Legal Socialization and the
Development of Legitimacy. New York: Oxford University Press.
Tyler, Tom R. and Steven Blader. 2000. Cooperation in Groups: Procedural Justice, Social Identity,
and Behavioral Engagement. Philadelphia: Psychology Press.
Tyler, Tom R. and Yuen Huo. 2002 Trust in the Law: Encouraging Public Cooperation with the Police
and Courts. New York: Russell Sage Foundation.
Tyler, Tom R., and Jeffrey Fagan. 2008. “Why do People Cooperate with the Police?” Ohio State J. of
Criminal Law 6 (1): 231-75.
Tyler, Tom R., and Steven L. Blader. 2003. “The Group Engagement Model: Procedural Justice, Social
Identity, and Cooperative Behavior.” Personality and Social Psychology Review 7 (4): 349–61.
Tyler, Tom, Jonathan Jackson and Avital Mentovich. 2015. “On the Consequences of Being a Target
of Suspicion: Potential Pitfalls of Proactive Police Contact.” Journal of Empirical Legal Studies
12 (4): 602-36.
Von Hippel, Paul T. 2009. “How To Impute Squares, Interactions, and Other Transformed Variables.”
Sociological Methodology 39 (1): 265–91.
Walters, Glenn D., and P. Colin Bolger. 2019. “Procedural Justice Perceptions, Legitimacy Beliefs, and
Compliance with the Law: A Meta-analysis.” Journal of Experimental Criminology 15 (3): 341-
Willis, Graham Denyer. 2015. The Killing Consensus: Police, Organized Crime, and the Regulation of
Life and Death in Urban Brazil. Oakland. University of California Press.
Zanetic, André, Bruno Paes Manso, Ariadne Lima Natal, and Thiago Rodrigues Oliveira. 2016.
“Legitimidade da Polícia: Segurança Pública para além da Dissuasão.” Civitas: Rev. de Ciências
Sociais 16 (4): 148-73.
Appendix A: Latent class analysis of public attitudes towards excessive police violence
We use latent class analysis (LCA) to assess the underlying structure of approval/disapproval of
excessive police use of force. We set the indicators as ordered categorical variables. When deciding the
number of classes, researchers generally rely upon three considerations (Nylund, Asparouhov, and
Muthén, 2007): (1) model fit statistics, in particular the likelihood-ratio (LR) tests and model fit indices
gauging how well the model represents the data, (2) the entropy of the model and the average latent
class probabilities, establishing whether clear delineation of classes is feasible, and (3) substantive
considerations, or in other words, whether the derived classes make sense.
From the three models, the 4-class solution is the least preferable option (Table A1). The
bootstrapped LR-test is not significant, implying that the model does not improve upon the 3-class
solution, and the BIC (Bayesian Information Criterion) value is the highest, which indicates the worst
fit. Entropy represents how well the LCA model differentiates between the latent classes, with values
closer to one representing better solutions. In comparison, the average latent class probabilities quantify
how precisely units can be assigned to their classes. On these two statistics, the 4-class solution does
not perform well either, with the entropy being lower compared to the 3-class solution, and the average
latent class probabilities being the lowest across the models.
Table A1: Model fit comparison of the latent class analysis
Number of classes Bootstrapped LR-test BIC Entropy Average latent class
2-class solution LR=-2570.616, p<0.001 5204.419 0.387 0.757-0.844
On the one hand, then, from the 2- and 3-class solutions, the 2-class solution has the better
model fit statistic according to the BIC and the 3-class model does not appear to be significantly better
than the 2-class one (although it comes close with p=0.072). On the other hand, both the entropy and
latent class probabilities favor the 3-class approach with much higher entropy and slightly higher
average class probabilities. These apparently contradictory test statistics can be explained by looking at
the emergent latent classes in the 3-class model (Table A2), where one of the classes is much smaller
than the other two. Tests of global fit and model comparisons tend to be very sensitive to imbalances in
class sizes and the presence of a small, but distinctive class can make it appear as if the 2-class solution
performed better. All things considered, therefore, we find the 3-class model to be more appropriate,
particularly because it also makes more sense from a theoretical point of view, as discussed above.
Table A2: Latent Class Analysis results in probability scales for each variable
LCA results in probability scale
Variable Response alternatives “Ambivalent towards
Murder of a
suspect in custody
Very poor 0.563 0.171 0.958
Poor 0.240 0.171 0.006
Neither good or bad
Excellent 0.032 0.158 0.000
Very poor 0.310 0.155 0.733
Poor 0.294 0.209 0.097
Neither good or bad
Good 0.148 0.261 0.036
Excellent 0.000 0.239 0.030
Search without a
warrant and torture
of a suspect
Very poor 0.186 0.000 0.757
Poor 0.369 0.000 0.051
Neither good or bad
Good 0.135 0.453 0.083
Excellent 0.000 0.507 0.085
N 468 74 385
% 50.5% 8.0% 41.5%
The best way to understand the three emergent latent classes is to juxtapose each variable’s
categories to the probability of belonging to a certain class. We named the biggest group (n=468) in the
3-class model “ambivalent towards violence” because most people belonging to this class disapproved
of excessive police use of force, although a minority were either undecided or willing to approve of
excessive police violence. We named the second and smallest (n=74) class “tolerant of violence”
because in this group the majority of people were either undecided regarding the use of excessive police
violence or approved of it. Finally, we named the third group “intolerant of violence” (n=385). In this
group, in each variable at least approximately three-fourths of the respondents found the way the police
handled the described scenarios ‘very poor’.
Appendix B: Details of the thematic analysis
For the thematic analysis, an elaborate coding scheme was devised with six distinct steps as shown by
the flowchart (Figure B). Further details regarding the coding are available from the authors upon
Figure B: Outline of the thematic analysis carried out responses to the open-ended question
Appendix C: Latent trait analysis
We first fit a two-trait model. This is based on the idea that people differentiate between instrumental
and normative considerations when they think about whether they would obey police instructions. The
model for normative considerations included the original duty to obey variable, the two consensual
content analysis categories (i.e. acceptance of authority and disobedient protest), and an item tapping
into normative alignment with the police. The instrumental latent trait contained the original duty to
obey variable, the two prudential content analysis categories (rejection of authority and coercive
obligation), and two items regarding personal and general fear of police. The normative and
instrumental latent traits were correlated with one another. To make the relative contribution of each
variable comparable, the coefficients reported here were standardized by the corresponding continuous
latent variable’s variance, quantifying relative influence of the nominal, binary, and ordinal variables
for each trait (this, however, does not permit cross-model comparison).
Figure C1: Two latent traits for normative and instrumental considerations
The results for the two-trait solution are shown in Figure C1.
For the normative considerations
trait, acceptance of authority had the strongest positive loading (λ
=1.68), then normative
=1.53), and disobedient protest (λ
=1.26). The trait loading of the
original duty to obey variable was weaker (λ
=-1.01). For the instrumental considerations trait,
all of the trait loadings were negative, as an indication of the negative association between the two latent
traits. General fear of police has the strongest trait loading (λ
=-2.45), followed by personal
fear of police (λ
=-1.24), the coercive category (λ
=-1.10), rejection of authority
=-0.75), and finally, the original duty to obey variable (λ
=0.44). As expected,
being afraid of the police was the strongest indicator of instrumental considerations, followed by the
duty to obey related constructs. Yet, it is still notable that general fear of the police contributes almost
twice as much to the latent trait compared to personal fear of the police, twice as much compared to the
coercive category, and more than three times as much compared to the rejection of authority category,
and more than five times as much as the original duty to obey variable.
General fear of
Personal fear of
-2.45 -1.10 -0.75
Duty to obey the
1.53 1.68 1.26
Figure C2: Single latent trait, the coercive to consensual continuum
Note, however, the high correlation between the two latent traits (r=-0.72). This suggests that
instrumental and normative considerations might indeed be two sides of the same coin (i.e. part of a
single underlying dimension). We next fit a single trait model, incorporating all the observed variables
described above, and named it coercive to consensual authority continuum. The results from this
modified model are shown in Figure C2. General fear of police has the strongest trait loading
=-2.60), followed by normative alignment (λ
=1.58), personal fear of police
=-1.12), acceptance of authority (λ
=1.07), disobedient protest (λ
duty to obey (λ
=-0.57), and rejection of authority (λ
=0.28). As with the instrumental
model, general fear of the police made the greatest contribution to the latent trait, but the contribution
of normative alignment is also almost three times that of duty to obey. Overall, the latent trait of coercive
power to consensual authority continuum comprises mainly the presence or absence of fear of the police
and the recognition (or not) of shared values, whilst obligation related deliberations only appear to play
a secondary role.
A straightforward way to compare the latent trait models is to rely on the penalised model
selection criteria, the AIC (Akaike Information Criterion) and the BIC (Bayesian Information
Criterion). These indices can be used to compare non-nested models, lower values, suggesting a better
fit (Kuha, 2004). For both the AIC and BIC, the single-trait solution (AIC=21024.49, BIC=21150.94)
outperformed the two-trait one (AIC=22560.85, BIC=22714.79). This implies that consensual and
instrumental relationships with the police can be reflected along the same, single dimension which we
call normative compass. Crucially, while high scores on this scale indicate that people feel that the
police are normatively appropriate, share their values and are not to be feared, and thus, they ought to
obey them (i.e. legitimacy); low scores do not indicate simply an absence of legitimacy but rather the
presumption of an antithetical, illegitimate relationship (that the police do not share their values, are to
be feared, and are not justified in commanding obedience).
General fear of
Personal fear of
Duty to obey the
1.58 -1.12 -2.60
1.07 0.71 0.28
See Trinkner et al., (2019) in English and in Portuguese see Oliveira et al., (2019), Oliveira et al., (forthcoming) and Zanetic
(2017). Trinkner et al., (2019) analyzed the second wave of the São Paulo Legal Socialisation Study of adolescents, finding a
strong association between procedural justice and legitimacy, and that procedural justice has a moderately strong relationship
with crime perception. Oliveira et al., (2019) discussed how to measure police legitimacy among São Paulo citizens based on
respondents’ normative alignment and duty to obey the police, as well as tested procedural justice theory. Oliveira et al.,
(forthcoming) studied the extent to which a normatively grounded duty to obey the police mediates the association between
perceived fairness and self-reported compliance with the law. Zanetic (2017) analyzed the association between perception of
procedural fairness and both trust in the police and legitimacy of the law. All four studies found empirical evidence that
supports Tyler’s (2006a, 2006b) process-based model of self-regulation.
We should note that a recent development of PJT predicts that bounded authority concerns are also important. The idea is
that concerns about process and about officers overstepping the limits of their rightful authority are both important criteria by
which people judge the legitimacy of the police as an institution, with emerging evidence in the UK (Huq et al., 2017) and the
US (Trinkner et al., 2018).
cresce-em-2018-assassinatos-de-policiais-caem.ghtml> (access on 08.28.2019).
Though their name has changed a few times since then, “Military Police” has been the organization’s name since 1970.
Self-reported ethnicity was measured following the official ethnical distribution of the Brazilian Institute of Statistics and
Alternative link functions (i.e. negative binomial and poisson) were also tested. The ordinal regression model fitted the data
better according to AIC and BIC.
Alternative models were also tested with higher order effects of coercive power to consensual authority continuum (e.g.,
squared, cubic) and interactions with other variables (e.g., personal morality, gender), but none of these were significant.
For both Figure 3 and Figure 4, all pathways were significant on the 0.1% level and the stars were left out for visual ease.