Perceptions of the Police

There is an extensive literature on determinants of citizen attitudes toward the police. Research in the United States finds that contact with the police, neighborhood, race, and age affect attitudes toward the police (see Reference Brown and BenedictBrown & Benedict 2002). Age is positively related to confidence in the police (Reference BensonBenson 1981; Reference Zamble and AnnesleyZamble & Annesley 1987; Reference CorreiaCorreia et al. 1996), with older citizens more likely to express confidence in the police than younger ones. In addition, much literature suggests that police community presence and public contacts affect public opinion of the police (Reference ReissReiss 1967; Reference Smith, Reiss and TonrySmith 1986; Reference TylerTyler 1988; Reference ShawShaw et al. 1998; Reference Reisig and ParksReisig & Parks 2000; Reference Terrill and ReisigTerrill & Reisig 2003; Reference Sunshine and TylerSunshine & Tyler 2003). Favorable-contact arguments are consistent with the neighborhood thesis (Reference WhyteWhyte 1943; Reference WilsonWilson 1963). Public perception of the police is a neighborhood or local phenomenon because the police are responding to local-oversight institutions and are subject to the approval of local public opinion (Reference Bordua and TifftBordua & Tifft 1971; Reference WeitzerWeitzer 2000; Reference SeronSeron et al. 2004). It follows necessarily, then, that public opinion must be based largely on the evaluation of the performance of the local police.

More recently, much empirical evidence shows that public opinion of the police is highly sensitive to events that are heavily publicized by the national media (see, e.g., Reference Albrecht and GreenAlbrecht & Green 1977; Reference Tuch and WeitzerTuch & Weitzer 1997; Reference ShawShaw et al. 1998; Reference SigelmanSigelman et al. 1997; Reference Sampson and BartuschSampson & Bartusch 1998; Reference Cao and HouCao & Hou 2001). The urban riots of the middle to late 1960s, police violence against civil rights protestors and African Americans, and other high-profile incidents have attracted the attention of the national media. In addition, a national agenda that stressed the need for greater law and order, tougher sentencing guidelines, and more police on the streets became a part of the national debate. Often at the center of these debates were issues related to race and ethnicity. Reference LasleyLasley (1994), Reference Tuch and WeitzerTuch & Weitzer (1997), and Reference SigelmanSigelman et al. (1997) demonstrate how short- and long-term effects of publicized police misconduct affect differently the opinions that majority and minority populations have concerning the national agenda of greater law and order.Footnote ² U.S. research provides considerable evidence for a significant relationship between the police and national governmental politics. This research, however, does not examine whether the determinants of national government/police relationships vary in some predictable and systematic fashion or provide a theory capable of explaining such systematic variations.

Police and Democratization in Latin America

There is reason to believe that national effects are not simply related to the political dynamics in the United States (Reference WeitzerWeitzer 1995; Reference Cao and HouCao & Hou 2001; Reference Cao and ZhaoCao & Zhao 2005).Footnote ³ Cao and Zhao show that “[T]rust in the [national] political system” has the strongest influence on police attitudes in nine Latin American countries and the United States (2005:409). They also show that support for police varies predictably, with citizens in the more stable democracies of the United States, Uruguay, and Chile expressing greater confidence in the police than citizens in less well–performing democracies of the Dominican Republic, Peru, Argentina, and Mexico (2005:408).Footnote ⁴

Regrettably, little public opinion work has been done to assess attitudes toward police in new Latin American democracies.Footnote ⁵ Given the dependent nature of Latin American police forces during authoritarian periods and the important role that police play in democratic governance, changes in public opinion about the police tell us a great deal about regime transformation. Beyond public opinion research about Latin American police departments, Cao and Zhao point out that there has been “little systematic discussion of evolution of police systems since the 1980s” (2005:404). Reference CallCall's (2002) study of new civilian security forces in Latin America is a noteworthy exception to this lacuna. Call shows that the mode of democratic transition best accounts for reforms to internal security structures. Still, reforms have been incremental in most Latin American countries, and protecting the government from its citizens remains the first mission of most Latin American police departments (2002:7–8).Footnote ⁶ Important to note, Call demonstrates that the transition from authoritarianism to democracy portends changes for police administration.

Cao and Zhao suggest that democratic development matters in attitudes toward the police, and they posit a direct relationship between national governmental institutions and police administration (2005:404). Nevertheless, the research does not examine whether specific governmental arrangements or levels of institutional development influence the national government/police relationship. In addition, the research fails to provide a theory that is capable of explaining cross-national variation in the relationship. Our theory is based on the idea that attitudes about the police are conditioned by the nature of the relationship between the national government and the police. Moreover, national elections focus attention on the partisan struggle for control of the democratic state (of which the police power is an essential part). This is the focus of our research.

The Police and Political Government: A Model

Most models of attitudes toward the police are based solely on local factors. The most prominent among these is provided by Reference WilsonWilson (1963). He offers a theory of police behavior and public support for the police at the local level that with modification has applicability to the national level. Wilson argues, “[i]f a professional police force can only exist in large cities as part of a set of political and civil institutions of certain character, the desirability of professionalism cannot be considered apart from the desirability of these institutions as a set” (1963:213). Wilson's system contains three components: the police must attain a level of professionalism characterized by institutional autonomy; the other institutions of the system must attain a certain (nonpartisan) character; and the public must have sufficient representation and access to support the police.

At the national level, Wilson's professionalism component can be viewed as part of the usable state bureaucracy that operates within professional norms (Reference Linz and StepanLinz & Stepan 1996:55). This means the police are part of the state but not a part of political society's electoral contestation. In turn, political society must allow the police bureaucracy to operate without unwarranted encroachment from national political actors (usability). This is consistent with Wilson's notion that political institutions in the set of political and civil institutions attain a “certain nonpartisan” character.

This dynamic should hold for both federal and unitary countries. That is to say, there is a relationship between the national government and the local police regardless of the structural division of power. It can be represented by the following equation:

$P\left(PS\right)={\beta }_1\left(PGS\right)+{\beta }_j{Z}_j,$

where P is the probability of support for the police (PS), β ₁ is the coefficient estimated effect of support for the political government (PGS) on support for the police, Z _j is a set of control variables that include the traditional determinants of police attitudes (age, race, neighborhood [local factors], party identification, and political ideology), and β _j is the set of coefficient estimates for these control variables.

At this point, we reiterate that all democratic governments are not the same. Reference Cao and ZhaoCao and Zhao (2005) show that these differences matter in a country's aggregate level of police support. Measures of democracy such as those of Freedom House and Polity IV align well with aggregate levels of support for the police. Freedom House and Polity IV measures are built on traditional theories of democracy that emphasize regime-level factors such as free and fair elections and observance of civil and political rights (Reference SchumpeterSchumpeter 1975; Reference DahlDahl 1989).Footnote ⁷

Why might democratic development affect the level of police support? O'Donnell argues that a theory of democracy must go beyond the regime level and include aspects of the legal system of the state (2001:25). In turn, political (full) democracy differs from all other political types and has “four unique differentiating characteristics” (2001:25): (1) fair and institutionalized elections, (2) an inclusive and universalistic institutionalized wager (this is what Linz and Stepan call a working consensus about procedures of governance [1996:10]), (3) a legal system that enacts and backs the rights included in the democratic regime, and (4) a legal system that prevents anyone from being above and beyond the law. This typology divides these four characteristics at two levels: Conditions 1 and 2 are located at the regime level, and Conditions 3 and 4 are located at the legal system of the state level. In addition, this typology not only distinguishes political democracy from other regime types but also distinguishes political democracy from “diminished kinds” of democracy (Reference O'DonnellO'Donnell 2001:25).Footnote ⁸ Even if new democracies meet regime-level conditions as prescribed by theories of democracy, failure to meet the legal system conditions relegates the nation-state to “consolidating” status.

We use this full/consolidating dichotomy to differentiate democracies. In new democracies, the set of political and civil institutions has not attained the same character as the set of institutions in full democracies: most prominent, an autonomous legal system of the state. Nevertheless, Wilson's model should apply in new democracies where the legal system has been subordinated to the executive government. This legal system subordination includes the subordination of the police. Wilson's model applies because the police and the national political government are linked politically. That is, individuals who support the national political government are likely to support the police organization that they believe the national political government controls. However, we should expect the police in developing democracies to generate less public support because they are subordinate to the dominant political interest. Still, persons who politically support the government should also politically support the police. Accordingly, this model yields the following hypothesis for both full and consolidating democracies:

Elections as Focusing Events

Wilson's theory helps us understand how attitudes toward the police are associated with attitudes toward the national government. Still, it cannot account for idiosyncratic or periodic fluctuations in the strength of this association. Idiosyncratic fluctuations may relate directly to crises in the national government. Certainly, the urban riots of the middle to late 1960s were connected directly to shifting confidence in the U.S. government. Kingdon argues that crises are focusing events that call attention to a problem (1995:94). For example, the Rodney King beating was a focusing event that called attention to police brutality.

Kingdon and others are interested in the effect of focusing events on agenda-setting and policy outcomes.Footnote ⁹ We contend that national elections are periodic focusing events that affect the national government/police confidence relationship. Kingdon argues that focusing events are associated with powerful symbols (1995:97). Political leaders use symbols (such as patriotism) to focus attention on a subject that is normally in the background of people's minds. Elections often serve as focusing events that call attention to the police as a symbol of crime prevention and law and order. As a focusing event, the rhetoric of campaigns focuses public attention on both the partisan nature of the national government and issues of law and order, as well as crime and punishment. Political rhetoric therefore is different in presidential election and nonpresidential election years, and this rhetoric intensifies the relationship between the symbol (the police) and the national political government.

Our use of this agenda-setting theory is justified because symbols and focusing events only “rarely carry a subject to a policy agenda” (Reference KingdonKingdon 1995:98). This means that subjects and symbols may be deployed strategically during a focusing event with little concern for an accompanying policy agenda. While the goal of such rhetoric is to elect the politicians who use it, there is evidence that elections do have important policy consequences for the police as well. Police receive greater resources during election years (Reference LevittLevitt 1997). Here we examine whether elections also influence attitudes about the police.

If the police are a symbol, why might attitudes toward the police shift during a presidential election? Kessel argues that there are three attitudinal properties to voting behavior: salience, partisan valence, and importance (1980:197–8). These three properties also help explain the shift in the determinants of public attitudes toward the police during presidential elections. Salience is the prominence of an attitudinal object. The more salient a topic (or symbol), the more likely the citizen is to have an attitude about it. Partisan valence concerns how the attitudes about the symbol sum to form a partisan valence; the attitude gives one party an electoral advantage over others. Importance concerns the extent to which votes depend on a given attitude. This means that “an attitude that is salient and quite favorable to one party is usually important” (Reference KesselKessel 1980:198).

In short, attitudes toward the police are more strongly linked to the national government during presidential elections because the attitudes are salient, because they can be used for partisan advantage by a political party, and because they are related to an individual's voting decision. Elections therefore are focusing events that are associated with shifts in perceptions of the police.Footnote ¹⁰ This argument yields an enhanced predictive model:

$P\left(P{S}_{py}\right)=\left(\delta \right){\beta }_1\left(PGS\right)+{\beta }_i{Z}_j,\hspace{0.17em}\delta \hspace{0.17em}>\hspace{0.17em}1,$

where P is the probability of support for the police during a presidential year (PS _py), β ₁ is the coefficient estimated effect of support for the political government (PGS) on support for the police, Z _j is a set of control variables that include the traditional determinants of police attitudes (age, race, neighborhood, party identification, and political ideology), and β _j is the set of coefficient estimates for these control variables. We add the δ tuning parameter to capture the intensified relationship between the national government and the police. δ is a ratio that measures the strength of the political-government-support effect on police support in the presidential year divided by the strength of the average political-government-support effect on police support in nonpresidential years. In short, δ=[(β _1(py)/σ _PGS(py))/(β _1(bar)/σ _PGS(bar))].Footnote ¹¹ As β ₁ is multiplied by the δ tuning parameter, δ takes a value greater than 1 to account for the increased effect of national government support on police support in presidential election years. This argument yields our second hypothesis.

Volatility in Partisanship

We expect Hypothesis 2 to hold for both consolidating and full democracies. At the same time, “changing the cast of characters” may have different implications in these two types of democracy. These implications connect to Reference O'DonnellO'Donnell's (2001) argument about the legal system of the state. Because citizens in consolidating democracies are more likely to view the police as dependent on the political government (because the police have been historically), the partisan struggle for control of political government affects more greatly the attitudes that citizens hold about the police. This is because control of the executive branch infers partisan control of the coercive force of the democratic state (control of the legal system of the state). Because the set of national civil and political institutions have not reached the necessary level to infer police professionalization, individual-level attitudes about the police fluctuate due to changes in the partisan composition of the national political government. This means partisans are likely to shift their support for the police based on their relationship with the party that controls the national political government. In full democracies such as the United States, the national political government approximates the nonpartisan character that Reference WilsonWilson (1963) posits. Accordingly, partisans are not likely to shift their support for the police based on their relationship with the party that controls the national political government.

The Police and Government in Three Countries

We test our focusing-event thesis in three presidential democracies (Costa Rica, Mexico, and the United States). We do so because we employ a most different systems approach to comparative research (Reference Przeworski and TeunePrzeworski & Teune 1970). We choose these three countries because we want to assess the national government/police support relationship in the context of varying governmental conditions. These countries vary in governmental (structure) division of power, level of democratic and party-system development, military/police relations, and level of governmental centralization. Costa Rica is a unitary country, and its police department is subject to national-level oversight. The Costa Rican national police structure is housed in two national institutional structures. The Ministry of Interior oversees the Fuerza Pública (the National Police), while the Supreme Court of Justice oversees the Organismo de Investigación Judicial (OIJ). The Fuerza Pública oversees disorder patrol, while the OIJ oversees investigative duties.

Mexico and the United States are two federal cases. The United States is the model of the federal policing configuration. Local police are generally accountable to the mayor and city council. Wilson points out that many cities have established nonpartisan local elections to aid in the oversight of the police (1963:172). Nevertheless, national political leaders assert a national connection to the police. Meanwhile, the Mexican police are also subject to some local oversight. Nevertheless, the police, like much of the Mexican state prior to 1997, have been subjugated to Mexico's hegemonic party system (Reference CampCamp 1999), and the jurisdiction of the Mexican federal police overlaps with those of state and local police. The limited legitimacy of the Mexican police was related strongly to the legitimacy of the Partido Revolucionario Institucional (PRI), which lost in the 2000 presidential election.

The relationship between the PRI and the Mexican police suggests that partisan identification should contribute to a person's attitude about the police in the developing Mexico democracy. Mexico represents our developing democracy case, while we consider the United States and Costa Rica full democracies (see Reference Linz and StepanLinz & Stepan 1996 concerning Costa Rican democratic exceptionalism). Theoretically, we contend in Hypothesis 3 that police attitudes in developing democracies are likely to be sensitive to changes in partisan control of the national government. We expect that partisan attitudes in Mexico will change because the PRI lost control of the national government after the 2000 presidential election victory of Vicente Fox, the National Action Party (PAN) candidate. Alternatively, we do not expect changes in attitudes in the police attitudes of partisans in Costa Rica or the United States. That is, we do not expect National Liberation Party (PLN) partisans in Costa Rica to be less likely to support the police than Christian Social Unity Party (PUSC) partisans because the PLN lost control of the national government after the 1998 presidential election. Miguel Angel Rodríguez of the PUSC replaced José María Figueres of the PLN after the 1998 election. Meanwhile, we do not expect U.S. Republicans to become less likely to support the police than U.S. Democrats because the Democratic Party controlled the national executive government in 1998, 1999, and 2000.

The military is also an important source of variation across the three countries. The subordination of the police to the military during previous authoritarian periods contributes to a lack of autonomy and legitimacy for many Latin American police departments (Reference CallCall 2002). However in the two Latin American cases in this study, the military has a lesser role. Costa Rica eliminated its military in 1949. Consequently, Costa Rica has the least militarized police force in Latin America (Reference CallCall 2002). For Mexico, the military has been responsible for and accountable to civilian control. Since the breakdown of the hegemonic party system in 1997, the Mexican military has had an expanded role in Mexican internal security.Footnote ¹² Increasingly, the military has assumed positions of command over police officials during this nationalization. For its part, the U.S. military is firmly under civilian control and only functions in U.S. internal security in times of national emergencies.

Meanwhile, local (see Reference WilsonWilson 1963) and state government have important relationships with the police. While the Costa Rican police are subject to national oversight, Costa Rica has begun efforts to decentralize state structures. Indeed, 2002 marked the first municipal elections for Costa Rican mayors. This infers that battling local crime should become an important campaign issue at the local level, and mayors and city councils should seek greater control of local police departments. Due to federal governmental structures in Mexico and the United States, local and state governments play the most important role in police oversight. A summary of important differences between the three countries is presented in Table 1, which also contains country-level expectations for our three hypotheses. We expect Hypotheses 1 and 2 to hold for individuals in the three countries. We expect Hypothesis 3 (a significant partisanship shift) to occur only in Mexico, because of Mexico's consolidating status.

Table 1. Police and the Government in Three Countries

Data

We use four data sets to examine citizens' evaluations of the police: the 1998 Hewlett data set, the 1998 and 1999 University of New Mexico's Institute for Public Policy (IPP) National Public Opinion Survey, the 2000 World Values System Survey, and the 2003 Latinobarómetro. The Hewlett data set comes from a study that examines democratic citizenship in Costa Rica and Mexico. Market and Opinion Research International in Mexico and the research firm Dichter and Neira in Costa Rica performed the surveys in July 1998. Total respondents for the two samples are 1,002 for Costa Rica and 1,200 for Mexico, and the same precoded questionnaire was used. The IPP at the University of New Mexico survey of public opinion in the United States used the same precoded questionnaire for its two waves: September to October 1998, and October to November 1999. Total respondents for the two samples are 1,085 for 1998 and 918 for 1999.

The World Values System surveys were performed by the following polling agencies during the following time periods: for Mexico, the Instituto Tecnológico Autónomo de México from January 28 to February 7, 2000; and for the United States, the Institute for Social Research at the University of Michigan in two phases: November 19 to December 23, 1999, and August 4 to September 25, 2000. Total respondents for the two samples are 1,535 for Mexico and 1,200 for the United States. The Latinobarómetro Corporation compiled the 2003 Latinobarómetro data. The same precoded questionnaire was used in the two countries. Total respondents for the two samples are 1,004 for Costa Rica and 1,200 for Mexico. The Costa Rican and Mexican samples represent 100 percent of the national population.Footnote ¹³ All samples have a margin of error between 3 and 3.5 percent.

Public Perception of Police and Government

Our first task is to examine the correlation between perceptions of the national government and citizens' attitudes toward the police. Table 2 shows the percentages of respondents who express “some” or “much” confidence in the police, the national government, the military, local government, and state government. For the 1998 and 1999 University of New Mexico surveys, the table shows the percentages of respondents who believe that the particular institution is doing a “good” or “excellent” job. Table 2 also shows the Pearson correlation between perceptions of the police and perceptions of the other institutional structures.

Table 2. Confidence in Police, Government, and Sub-National Government

Notes: Correlation with confidence in the police is shown in first parenthesis line below the confidence percentage. Correlation with confidence in the government is shown in the second parenthesis line below the confidence percentage. Percentage Expressing Confidence is individuals who express some or much confidence in the institution. For the 1998 and 1999 U.S. samples, confidence is the citizens who respond that the institution is doing an excellent or good job. NA=Not asked. Data sources:

^a 1998 Hewlett Survey;

^b 2000 World Value Survey;

^c 2003 Latinobarómetro;

^d 1998 and 1999 University of New Mexico Public Policy Survey.

Three of the surveys examine attitudes in Mexico. Mexicans' confidence in the government increases from 30 percent in 1998 to 36 percent in 2000 and then declines severely to 23.5 percent in 2003, while their confidence in the police declines in a more linear but equally dramatic fashion for the same time period (33, 28, and 16 percent, respectively).Footnote ¹⁴ Moreover, governmental confidence correlates most highly with confidence in the police in 2000 (r=0.542), while the Mexican police/government correlations for 1998 and 2003 are significantly lower (r=0.326 and 0.222, respectively). Mexico held an historic presidential election in 2000, and this correlation surge in 2000 offers some support for Hypothesis 2.

A similar correlation pattern occurs among the three U.S. distributions. While U.S. citizens express far greater and more consistent confidence in the police (73 percent in 1998, 71 percent in 1999, and 71 percent in 2000) than they do in the government (45 percent in 1998, 36 percent in 1999, and 38 percent in 2000), U.S. governmental confidence correlates with confidence in the police most highly in the 2000 presidential year (r=0.320). The governmental/police correlation reaches only 0.154 in 1998 and 0.210 in 1999. The Costa Rican distribution also provides evidence for this possible presidential year surge. Costa Ricans' confidence in the government is correlated with confidence in the police at 0.397 in the presidential year of 1998, but declines to 0.304 during the nonpresidential year of 2003.Footnote ¹⁵ Bivariate findings suggest that public attitudes toward the police are responsive to national presidential elections. A more rigorous test of this thesis is presented in the next section.

Analytical Model of Citizen Evaluation of Police

In this section, we more fully test Hypotheses 1 and 2 by creating a multiple regression model that approximates the theoretical model that we presented earlier in this discussion. To account for our hypotheses and the previous acknowledged determinants of attitudes toward the police, we posit confidence in the police to be a function of these variables: government (national) support, support for other governmental institutions, age, race, partisanship, and political ideology. Confidence in the police is our dichotomous outcome variable, with citizens who express some or much confidence coded as confidence (1) and all other responses coded as no confidence (0).Footnote ¹⁶

For the 1998 and 1999 U.S. samples, confidence is the response that the police are doing an “excellent” or “good” job. The University of New Mexico questionnaire does not use the confidence-in-police question to evaluate citizens' attitudes toward the police. Instead this survey asks respondents to assess the performance of the police with responses of poor, fair, good, and excellent. While this assessment question differs from the confidence question, we believe our equating of these two questions is justified because both measures are predominantly measures of short-term satisfaction with the performance of the police.Footnote ¹⁷ Moreover, as evidence of face validity, Table 2 shows that confidence-in-the-police and evaluation-of-police questions produce similar results for U.S. respondents.

The Explanatory Model

Given the dichotomous outcome variable (confidence in the police), we use logistic regression to examine the determinants of citizens' confidence in the police and posit the following analytical model:

Government support (confidence) is the primary independent variable. This variable is an ordinal variable with values of 1 to 4: 1 = no confidence, 2 = little confidence, 3 = some confidence, and 4 = much confidence. Once again, the University of New Mexico questionnaire does not use the confidence-in-government question to evaluate citizens' attitudes toward the national government. This survey asks respondents to assess the performance of the federal government with responses of poor, fair, good, and excellent. As we mentioned for the police, Table 2 shows that confidence-in-the-government and evaluation-of-government questions produce similar results for U.S. respondents. Moreover, the 1998 and 1999 IPP use of these evaluative questions is consistent across the police and governmental variables. We include party identification as our second important independent variable and to test Hypothesis 3. We specify primary party-dummy variables in the following manner: Costa Rica—first party = PUSC, second party = PLN; Mexico—first party = PRI, second party = PAN, third party = PRD; United States—first party = Republican, second party = Democrat. Party codes are explained in fuller detail in the Appendix.

We include other independent variables as statistical controls and as representations of the acknowledged determinants of police attitudes. To account for support for other governmental institutions, we include measures of local government, state government, and military support. Unfortunately, we lack measures of attitudes toward local government and the military for all eight distributions of the study.Footnote ¹⁸ Nevertheless, we can control for the effect of attitudes toward the military on police confidence in four of our eight distributions (Mexico 1998, 2000, and 2003; and United States 2000). Military confidence is measured on a four-point scale with 1 = none, 2 = little, 3 = some, and 4 = much.

We can also control for attitudes toward local and state government in four of our eight distributions (United States 1998, 1999; Costa Rica 2003; and Mexico 2003). We use respondents' evaluations of the mayor or city manager to model local government evaluation and respondents' evaluations of the governor to model state government evaluation in the United States.Footnote ¹⁹ For the 2003 Costa Rican and Mexican distributions, we use confidence in local officials as a measure of local government evaluation.Footnote ²⁰

Age is included in the model because it is a proven and resilient predictor of police attitudes in the United States, with older respondents more supportive of the police. Because we use national data, we lack measurements of neighborhood demographic effects. We do, however, include attitude about crime, which captures the concern that some individuals have about the local crime problem. While attitudes about crime may be related to local concerns, national candidates also seek to nationalize the issue. Recall that McCann and Lawson show that Mexicans' attitudes toward crime control were responsive to campaign effects from the 2000 Mexican presidential campaign (2003:69). Interestingly, this panel study shows that attitudes moved from the “tough-on-crime” position to the “fight-crime-with-job-creation” position across the five months of the study.

We also include attitude about order and size-of-respondents'-city (city size) in our model. Attitude about order taps into general attitudes about the democratic regime's function and concerns about civil and political liberties. Moreover, Reference Cao and HouCao and Hou (2001) find a significant negative relationship between the flip side of order (what they term a deviant subculture) and confidence in the police. We believe that municipality size should be negatively related to confidence in the police because citizens in smaller municipalities have a greater opportunity to have a personal relationship with police officers (Reference Cao and ZhaoCao & Zhao 2005:408). In addition, we control for attitude about corruption, political ideology, gender, education level, income, and political efficacy (see Appendix for a description of measurements).

Because the variance is a function of the mean for the binomial distribution and susceptible to over-dispersion, we simulate the standard errors and use the z-score for the government support coefficient estimate as our primary test statistic and measure of the strength of the government/police support relationship.Footnote ²¹ The z-score is the MLE coefficient estimate for the variable of interest divided by the simulated standard error estimate of the MLE coefficient estimate. The z-score has an approximate standard normal distribution, and a z-score value of 2 is fairly strong evidence of an effect. We are also interested in the overall strength of association of the police confidence models. The Pseudo R ², while not an actual measure of model fit in the generalized linear model (GLM) context, is a good comparative measure of the strength of association that approximates the function that the R ² plays as a measure of linear association in the ordinary least squares model. We use Reference NagelkerkeNagelkerke's (1991) version, which allows the Pseudo R ² to achieve a maximum value of 1. This approach to the R ² creates an R ² measure that is comparable across the eight logistic models of this study.

Findings

Table 3 shows that even when we control for a variety of other factors, confidence in the national government (government support) is a significant factor in citizens' assessment of the police in the three countries and all eight samples; that is, even as we control for attitudes toward other governmental institutions, race, age, attitude about crime, and attitude about order. Since past research finds that these factors are related to citizens' attitudes toward the police, this finding offers solid evidence for our first hypothesis. Citizens positively and significantly link their support for the police to their support of the national government. The conditional effect of government support on police support is strongest for the Mexico 2000 distribution (z-score = 19.4). Meanwhile, the strength of the government/police support relationship for Mexico 1998 and Mexico 2003 are similar (z-scores = 8.08 and 6.31, respectively), but significantly lower, when we control for these other factors. Recall that 2000 is a presidential year. The Mexican distribution therefore offers support for our second hypothesis. The focusing event provided by the presidential election intensifies the connection between the national government and evaluation of the police when we control for other factors.

Table 3. Logistic Regression Model of Police Support: Institutional Support Variables

Pres=Presidential Year Model. Estimates for party identification, race and ethnicity, corruption and citysize are shown in Table 4. Model also controls for political ideology, gender, education, income, and efficacy. These variables are not included due to lack of space (see Appendix). Coef=Coefficient. Standard errors of coefficient estimates are simulated in the R statistical environment using Design Librarys lrm (logistic regression model) function in the R statistical environment. (se)=are simulated standard errors of MLE estimates. Data sources: 1998 Hewlett Survey; 2000 World Value Survey; 2003 Latinobarómetro; and 1998 and 1999 University of New Mexico Institute for Public Policy Survey.

Figure 1 plots the predicted probability of police support for centrist male individuals who identify with the first political party and have median support for other governmental institutions, median education, median income, and median efficacy by government support for the three respective countries. The first political party is the PRI in Mexico, the PUSC in Costa Rica, and the Republican Party in the United States. The “Mexico” plot of Figure 1 shows graphically the effect government support has on support for police and represents the predicted probabilities produced by the police confidence models. In 1998 (the solid line), centrist-PRI partisans' support for the police is linearly related to their support for the government. In the 2000 presidential year (the dashed line), centrist-PRI partisans' support for the police is more strongly related to their support for the government. By 2003 (the dotted line), centrist-PRI partisans' support for the police is linearly related to their support for the government as it was in 1998. Note that the two lines are roughly parallel. Still, centrist-PRI partisans are far less likely to support the police than they were in 1998. This is convincing evidence that the intensity of the 2000 presidential election affected the government/police support relationship.

The figure plots predicted probability of police support for a centrist male individual who identifies with the first political party and has median local support (3 in the United States and 2 in Costa Rica and Mexico), median state support (3 in the United States), median military support (3 in the United States and 2 in Mexico), median education, income, and efficacy by government support in the three respective countries.

Figure 1. Predicted Police Support by Government Support.

Yet the 2000 Mexican presidential election was historic, and the increased connection between the government and the police could simply be a manifestation of the importance of this particular presidential election. On the other hand, the 2000 U.S. and 1998 Costa Rican models corroborate the thesis that presidential elections in general influence attitudes toward the police. For U.S. citizens, the governmental confidence z-score value increases from 4.62 in 1998 to 13.52 by the 2000 presidential election year. The “United States” plot of Figure 1 shows that in 1998, the probability that centrist U.S. male Republicans support the police increases linearly based on their level of support for the government (the solid line).Footnote ²² By 2000, the probability that these same centrist U.S. male Republicans support the police increases more strongly as their level of support for the government increases (the dotted line). For Costa Rican citizens, the governmental confidence z-score value decreases from 11.17 in the 1998 presidential year to 7.65 in the nonpresidential year of 2003.Footnote ²³ The “Costa Rica” plot of Figure 1 shows the increase in the government support slope in the 1998 presidential year (the solid line). In both the U.S. and Costa Rican cases, the increase of the presidential year slope is consistent with the more dramatic Mexican presidential year increase.

In addition, the δ tuning ratio, which measures the strength of the conditional government support/police support relationship in the presidential year divided by the strength of the average conditional government support/police support relationship in nonpresidential years, is 1.3 for Costa Rica, 1.8 for the United States, and 2.3 for Mexico.Footnote ²⁴ As we stipulated in our articulation of the presidential-year police support model, this δ ratio (the presidential-year z-score/the average nonpresidential year z-score) exceeds 1. In addition, the 2000 U.S., 2000 Mexican, and 1998 Costa Rican models explain 51, 47, and 25 percent of the variation in police confidence probabilities, respectively. This is significantly greater than the variation explained by the nonpresidential-year models, which include a local support variable. Even with the local support variable included in the 2003 Costa Rican model (an additional variable), the 1998 presidential-year Costa Rican model explains far more of the variation in police attitudes. Taken together, this is solid evidence in favor of Hypothesis 2.

In Hypothesis 3, we hypothesize that partisans' attitudes toward the police in developing democracies are more likely to shift due to change in partisan control of the national government. Given Mexico's developing democracy status, we find support for our thesis in the “Mexico” plot of Figure 1. Note that PRI partisans are far less likely to support the police in 2003 than they were in 1998. In addition, Table 4 shows that none of the Mexican political-party dummies is related significantly to police confidence in 2000. By 2003, all other partisan identifiers, including the PRI, are less likely than PAN partisans (the new party of the government) to have confidence in the police. The change in the party of the government affects the attitudes that the PRI and other partisans have toward the police. In 2003, the PRI still had significant political power at the state and local levels. Nevertheless, PRI partisans have negative attitudes toward the police because the PAN controls the national-political government.

Table 4. Police Support Models: Party Identification and Race and Ethnicity

Notes: Party Codes: Costa Rica—First Party=PUSC; Second Party=PLN; Mexico—First Party=PRI; Second Party=PAN; Third Party=PRD; US—First Party=Republican; Second Party=Democrat. These variables are not included due to lack of space (see Appendix). Coef=Coefficient. Standard errors of coefficient estimates are simulated in the R Statistical environment using Design Library's lrm (logistic regression model) function in the R Statistical environment. (se)=are simulated standard errors of MLE estimates. Data sources: 1998 Hewlett Survey; 2000 World Value Survey; 2003 Latinobarómetro; and University of New Mexico Institute for Public Policy Survey.

The “Mexico” plot of Figure 1 shows the dramatic decline in PRI partisans' support for the police from 1998 (the solid line) to 2003 (the dotted line). The police confidence probability of a PRI partisan with “much” support for the government declines from 0.67 in 1998 to 0.21 by 2003. By contrast, U.S. Republicans are more likely than their Democratic (the party of the president in 1998, 1999, and 2000) counterparts to support the police. In addition, we find no significant difference in the police support probabilities of Costa Rican PLN and PUSC partisans even when the party of the government changed from PLN in 1998 to PUSC in 2003. It is also worth noting that the PUSC candidate won the presidency in 2002. Nevertheless, there is little difference in the mean probability of PUSC support for the police across this five-year period. This is solid evidence in support of Hypothesis 3.

Why do the attitudes of Mexican partisans shift so dramatically, while the attitudes of U.S. and Costa Rican partisans remain in a fixed configuration? By fixed, we mean that Republicans have consistently stronger support for the police than do Democrats, even when Democrats controlled the national government. Reference WilsonWilson (1963) points out that political institutions at the local level must attain a certain (nonpartisan) character for the professional police force to exist. Our partisanship findings suggest that a similar type of system relationship must develop at the national level. The police in Costa Rica and the United States are a part of a usable state bureaucracy that operates within professional norms (Reference Linz and StepanLinz & Stepan 1996:55). The unconsolidated nature of the set of Mexican national civil and political institutions prompts Mexican partisans to change their level of support for the police based on the partisan composition of the national political government. Control of the executive branch infers partisan control of the coercive force of the democratic state. In short, many Mexican citizens believe that the legal system of the state (including the police) is a tool of the dominant political party. In the aftermath of the loss of control of the national government, PRI partisans become less likely to support the police than do PAN (the new party of the government) partisans. Our partisan findings suggest that police attitudes reflect on the character of the democratic regime and support Hypothesis 3.

Conclusion

We tend to think of the police as a local institution, and generally also as a nonpartisan one, which performs a specific function: to serve and protect its citizenry. We find strong cross-national attitudinal support for this local support thesis. Other research indicates that attitudes toward national governmental structures also influence attitudes toward the police. We extend this research and show that perceptions of the police are responsive to systematic fluctuations in attitudes toward the national government. We show that presidential elections, with their greater focus on crime-related issues, provide a focusing event at the national level that influences citizen perceptions of the police at the local level.

The determinants of public attitudes toward the police shift during the presidential election year because the attitude is salient, provides a partisan advantage for a political party, and relates to the voting decision. In short, the attitude toward the police is politicized by the national presidential election. A national presidential election is a periodic change in the character of the national government that impacts attitudes toward the police. The election as a focusing event intensifies the police support/government support relationship for citizens in both the developing Mexican democracy and the full democracies of Costa Rica and the United States. Still, we find an important difference in the police attitudes of citizens in these two types of democracy. Mexicans' support for the police shifts based on changes in the partisan composition of the national political government, while Costa Ricans and U.S. citizens do not shift their attitudes about the police based on changes in the partisan composition of the national government.

Nevertheless, candidates in full democracies believe that “borrowing” the legitimacy of the police offers an electoral advantage. Indeed, we see signs of an emphasis on law-and-order themes in the 2008 U.S. presidential election. The Republican presidential candidate Rudolph Giuliani adopted a “top cop” persona as the centerpiece of his primary election campaign. His emphasis was not only his experience in the aftermath of the September 11, 2001, terrorist attack, but also the idea that he alone of the candidates could keep America safe. In addition, former Tennessee Senator Fred Thompson ascended to the upper tier of candidates in the Republican primary election campaign partly on his notoriety as the “tough-on-crime” district attorney on the popular television series Law & Order. On the Democratic side, New York Senator Hillary Clinton also emphasized experience and toughness, in both foreign and domestic policy. Although tough rhetoric on foreign policy is at the forefront of campaigns in a nation at war, law-and-order themes at home remain the essential bread and butter of the campaign. Merely by earning the endorsement of a police organization or standing for a photo opportunity with a group of police officers provides support for the message; this candidate is tough on crime and is tough enough to be president. Therefore, the 2008 presidential race emphasizes the same themes, and we expect the same sort of bump in public support for the police, as a result.

For developing democracies, and building on Cao and Zhao's (2005) findings, our findings suggest that police attitudes are an important barometer of attitudes about the development of state apparatus autonomy. Moreover, we speculate that a similar effect should exist for attitudes about other areas of the bureaucratic state, most particularly the judiciary. While our study involves three countries, it suggests that a broader study that could include other Latin American nations, Canada, and Europe, for example, would provide a more thorough test of our focusing event thesis and its relation to state autonomy. In the meantime, the results here provide strong support for the idea that we should conceptualize the police not merely as local actors operating on a limited stage, but as political actors who perform on a broad national stage.