Investigative Resources and Crime Clearances: A Group-Based Trajectory Approach

Abstract

Past studies of crime clearance rates have largely ignored the role of investigators in the process. This omission is important because criminal investigation is essential to clearing crimes, particularly those in which offenders are not readily identifiable. Using data from 570 law enforcement agencies spanning a 13-year period from 2000 to 2012, this study developed group-based trajectories of violent and property crime clearance rates (an approach not taken in previous clearance rate research), then modeled group membership with two measures of investigative spending and one of the proportion of investigators in sampled agencies (“investigative resources”), while controlling for factors such as workload and crime type. Results from a series of multinomial regression models suggest investigative resources play only a marginal role in crime clearance trajectories. This finding is consistent with early detective research, including the classic RAND criminal investigative process study.

Keywords

group-based trajectory clearance rate investigation

Along with crime rates, arrests, citations, and response times, crime clearance rates (usually defined as offenses cleared by arrest divided by offenses recorded) are used to measure police performance. For better or worse, police administrators are accountable to policymakers for them (e.g., Navratil, 2015). The media use clearance rates to rebuke agencies whose performance falls short of expectations (e.g., Isackson, 2013). Criminologists have sought to explain them (e.g., Paré, Felson, & Ouimet, 2007). Just recently, National Public Radio assembled an online database wherein concerned citizens can readily access crime clearance rates for their own communities (Kaste, 2015a).

Crime clearance rates remain a key indicator of police performance, even despite awareness of their limitations (discussed further in the Clearance Rates section). Experts have for years recommended alternatives (e.g., Alpert & Moore, 1993; Bonkiewicz, 2015; Davis, Ortiz, & Euler, 2015), and some agencies have responded by measuring police success with increasingly sophisticated methods (e.g., Van Meter, 2001). Nevertheless, clearance rates continue to factor prominently into budgeting and resource allocation decisions. Indeed, they are arguably more important now than in points past, as clearance rates have generally been declining at the same time crime has been declining (Ousey & Lee, 2010). Lackluster clearance rates have suddenly become conspicuous (Kaste, 2015b).

Nine in 10 homicides were cleared in 1960. Recently, that figure dropped to six in 10 (Kaste, 2015b). The most precipitous—and almost perfectly linear—drop occurred between 1970 and 2000 (Ousey & Lee, 2010), bucking crime trends in both the “get tough” era (S. Walker, 1998) and the “great American crime decline” of the 1990s (Zimring, 2006).

Clearance rates first fell at a time when incarceration was increasing, then they continued to fall as crime declined in the 1990s. Conventional wisdom suggests at least some of the imprisonment binge owed to police identifying and arresting more criminals, but clearance rates did not bear this out. Likewise, it seems plausible that the 1990s crime drop freed up resources for improved investigations, but clearance rates continued to slip. One study attributed the drop to “substantial changes in the nature of homicides, combined with insufficient organizational capacity within the criminal justice system” (Maguire, King, Johnson, & Katz, 2010, p. 373), but otherwise this strange pattern remains largely misunderstood.

Explanations for changes in crime clearance rates remain scant and underdeveloped. The clearance rate literature pales in comparison with the crime rate literature, plus it is increasingly dated. The classic RAND Corporation study of detectives (Greenwood, Chaiken, & Petersilia, 1977), perhaps the most comprehensive to date, was published nearly 40 years ago. In addition, what little research is available either explains clearance rates in terms of (a) factors over which the police have almost no control, or (b) crude metrics that bear little intuitive connection to what crime clearances actually measure. Concerning the former limitation, researchers have explored the relationship between community size and clearance rates (e.g., Cordner, 1989; Mesch & Talmud, 1998; Paré et al., 2007), but it is not within police departments’ purview to alter the sizes of their communities. Similarly, researchers have explored whether community socio-demographic characteristics are associated with clearances (e.g., Borg & Parker, 2001; Liska & Chamlin, 1984), but police departments are in no position to alter the demographic composition of their respective communities.

Other studies have turned attention inward toward law enforcement organizations, exploring the association between crime clearance rates and police-related variables, including workload (Bayley, 1994; Sullivan, 1985), agency size (Cameron, 1987; Eck & Maguire, 2000; Levitt, 1997), case type (Cordner, 1989; Mesch & Talmud, 1998; Paré et al., 2007), agency accreditation (Doerner & Doerner, 2012), and access to information technology (Garicano & Heaton, 2010). Absent from this literature, however, is a study of whether money spent on investigators and/or the proportion of detectives in an agency (hereafter, “investigative resources”¹) are associated with trends in crime clearances. Prior studies have found that criminal investigators exert varying degrees of influence on crime clearances (see, for example, Greenwood et al., 1977, for one perspective on this, and Wellford and Cronin, 1999, for another), but no studies have explored the direct association between investigative resources and crime clearance rates in a large sample of law enforcement agencies. The present study seeks to fill this gap.

Using Uniform Crime Reports (UCR) offense and clearance data from 2000 to 2012, this study developed group-based trajectories of agency-level clearance rates. This approach represents an improvement over traditional research in this area, as it is important to consider whether an aggregate model obscures meaningful variation over time and between different types of agencies. Group-based trajectory modeling has recently been applied to crime rates (e.g., Schupp & Rivera, 2010; Weisburd, Bushway, Lum, & Yang, 2004), but this is the first study to model clearance rates in such a fashion. The clearance rate data were then merged with three waves of Law Enforcement Management and Administrative Statistics (LEMAS) data to explore whether investigative resources, as measured by specific questions in the LEMAS surveys, were associated with crime clearance rates.

Clearance Rates: Strengths and Weaknesses

Crime clearances are associated with the ability of the police to solve crimes (Alpert & Moore, 1993; Bayley, 1994; Mesch & Talmud, 1998). The clearance rate is generally measured by dividing the number of offenses “cleared” by the number of offenses known, then multiplying the resulting figure by 100. Crimes are “cleared” either by arrest or “exceptional means.” For an offense to be cleared by arrest, at least one person must be arrested, charged with the offense, and turned over to the courts for prosecution (Federal Bureau of Investigation [FBI], 2014). For clearance by exceptional means, the agency must have identified the offender, gathered sufficient evidence against him or her to support arrest, ascertained the offender’s location, and “encountered a circumstance outside the control of law enforcement that prohibits the agency from arresting, charging, and prosecuting the offender” (FBI, 2014). If the offender committed suicide, for example, the underlying offense would be considered cleared by exceptional means. Both types of clearances are fundamentally similar, as they require as a precondition that the police not only have identified the offender, but gathered enough evidence to support arrest and subsequent prosecution.

As a measure of police performance, clearance rates are advantageous in several respects. First, they are viewed as more objective than some alternative indicators, such as attitudes toward police and fear of crime (Davenport, 1999; Leishman & Mason, 2003). Second, crime rates measure workload, but clearance rates shed light on how effective police are in apprehending lawbreakers. It is well-known that crime rates are at least partially affected by social and economic conditions beyond law enforcement’s control (Maguire & Uchida, 2000; Mastrofski, 1999), whereas clearance rates can be driven largely by offender identification and apprehension. As Weisburd, Hasisi, Jonathan, and Aviv (2010) observed, the clearance rates “. . . focus on ‘crimes’ rather than arrests means that it is linked directly to police investigations rather than specific offenders” (p. 729). As an alternative to clearance rates, arrest rates have been a popular metric of police performance, but arrest rates can be driven by crime rates—and, of course, police policy decisions.

Clearance rates are also beset by limitations that cannot be ignored (e.g., Brodeur, 1998; Hoover, 1996; Loveday, 1999; Maguire, 1997). First, although the FBI’s definition of a crime clearance is straightforward, it is by no means certain that police departments count clearances similarly across jurisdictions (Weisburd et al., 2010). Second, because considerable weight is attached to clearance rates at the jurisdiction level, police commanders are put under pressure to increase their clearance rates, possibly with overt manipulation (e.g., Cordner, 1989; Gill, 1987; M. A. Walker, 1992). Some experts have gone so far as to distinguish between two types of clearance rates: primary and secondary. The former measures police activity in the field whereas the latter measures clearances of old offenses resulting from helpful arrestees. Indeed, police and prosecutors routinely use as leverage over current offenders promises of leniency for helping identify other offenders (Skolnick, 1966). Another limitation lies in the calculation of the clearance rate itself; the numerator can include prior year, not just current year clearances. Despite these limitations, the clearance rate is still commonly used to measure police performance (e.g., Reiner, 1998). According to Paré et al. (2007), “Its use, in part, reflects the lack of alternative measures” (p. 244).

Factors Associated With Crime Clearances

With some exceptions, most crime clearance research was published prior to 2000. This research can be organized into three categories. The first, beginning with the classic RAND study of police detectives (Greenwood et al., 1977; Greenwood & Petersilia, 1975), generated a series of inquiries as to the role of investigators in crime clearances. The second explores community-level correlates of crime clearance rates, and the third looks at general police-related variables, such as workload and crime type.

Investigation Practices and Clearances

The landmark RAND Corporation study of detectives (Greenwood et al., 1977) examined (among other issues) whether investigative effort was associated with crime clearances.² The researchers found that only about 3% of crimes were cleared as a result of investigative work. Some 30% were cleared as a result of “pickup arrests” by patrol officers responding to crime scenes (Chaiken, Greenwood, & Petersilia, 1977, p. 192; see also Conklin, 1972; Greenwood, 1970). In another 50%, the perpetrator was known when the crime report was first taken. According to the authors,

This means that around 20 percent of cleared crimes could possibly be attributed to investigative work, but our own study showed that most of these were also solved by patrol officers, or by members of the public who spontaneously provided further information, or by routine investigative practices that could also have been followed by clerical personnel. (Chaiken et al., 1977, p. 192)

Subsequent research, however, has found that the role of detectives in crime clearances is more complex than previous studies let on—and argued that the RAND study was fraught with methodological limitations (e.g., Eck, 1992). For example, Eck’s (1992) study of burglary and robbery cases in three jurisdictions distinguished between information gathered by patrol officers in preliminary investigations and information gathered by detectives during later follow-up investigations. He found that information gathered at both stages was important in terms of clearances, that “. . . the work of both patrol officers and detectives is extremely important with respect to the subsequent making of follow-up arrests” (Eck, 1992, p. 243). His research also uncovered an “implicit triage system” (p. xvii) in which detectives divided cases into three categories: those that could not be solved, those that have already been solved, and those that, “with a reasonable amount of effort” (p. xviii), can be solved. Not unexpectedly, the last group of cases received the most investigative effort, and the role of detectives in them was important.

More recently, Brandl and Frank (1994) studied detectives in a medium-sized police department, focusing primarily on the amount of time detectives spend on follow-up investigations and the evidentiary strength of the cases. They found that in cases with moderate suspect information, the likelihood of an arrest increased with the amount of detective time spent on the case. In their words, “. . . the relationship between time spend and case outcome was the strongest in the cases with moderate suspect information” (p. 149). Similarly, Wellford and Cronin’s (1999, para. 13) multistate analysis of homicide clearances concluded that “[t]he number of detectives assigned to a case, and their actions, had a significant effect on closing the case.” Both studies contradicted previous research in which detectives were found to have little influence on crime clearances (e.g., Chaiken et al., 1977; Greenwood, 1970; Greenwood et al., 1977). At the least, recent studies suggest detectives are an important piece of the crime clearance puzzle, offering further justification for the present study’s focus on investigative resources and their role in crime clearance rates.

Community Factors

Crime clearances may be difficult in large urban areas. The more populous a given location, the easier it is for offenders to maintain anonymity (e.g., Willmer, 1970). Likewise, witnesses may have a more difficult time identifying perpetrators given the sheer volume of people in crowded locations (Felson, 1998). These assertions have not been consistently supported in prior research, however—at least with respect to clearance rates. Some research has found that city size is positively associated with crime clearances (e.g., Greenwood et al., 1977), but most have found a negative effect (Ahlberg & Knutsson, 1987; Cordner, 1989; Paré et al., 2007; Pogue, 1975). That is, clearance rates improve in smaller communities. Still other studies have found no effect of community size on clearance-related outcomes (e.g., Borg & Parker, 2001; Litwin, 2004).

Conflict theory maintains that the criminal justice system discriminates against minorities and/or the poor (Chambliss & Seidman, 1982; Turk, 1969). To the extent this occurs, it could have one of two effects on crime clearances. On one hand, greater scrutiny of minority and/or poor communities may be associated with greater crime clearance rates; more police patrolling these areas gives rise to more opportunities to catch lawbreakers (e.g., Smith, 1986). On the other hand, the system may also discriminate against minority and/or poor victims, not taking their offenses seriously, and thus compromising the ability of police to clear crimes. In other words, the law may be “less available” to people in poor status groups (e.g., Gross & Mauro, 1989; LaFree, 1989). Both perspectives have received support in the clearance rate literature. Some studies have found a positive association between clearance rates and poverty (e.g., Paré et al., 2007; Paré & Ouimet, 2004) or stratification/inequality (a composite measure in Borg & Parker, 2001, of income, education, unemployment, and racial segregation). Others, though, have found lower crime clearance rates in poor or minority communities (Sullivan, 1985). Still others have found mixed evidence within the same study (Litwin, 2004).

Police and Workload Factors

The effects on crime clearances of police- and workload-related factors have also been examined. The evidence is most clear with respect to crime types. It is well-known that crime clearance rates are higher for serious offenses, particularly murder (e.g., Ahlberg & Knutsson, 1987; Cordner, 1989; Paré & Ouimet, 2004). For example, Cordner (1989) found that clearance rates were lower in departments that handled disproportionate amounts of property crime. Other studies have delved into specific types of single crimes. Riedel and Jarvis (1998; see also Puckett & Lundman, 2003) found that spousal homicides were easier to clear than organized crime-related homicides. More recently, Paré and his colleagues (2007) concluded “. . . any assessment of the performance of police departments based on clearance rates must be adjusted for the types of crimes the police handle” (p. 256).

Police workload, generally defined in terms of the number of offenses a police department handles (as opposed to the types), has also been linked with crime clearance rates.³ Specifically, several studies have uncovered an inverse relationship between workload and clearance rates (e.g., Bayley, 1994; Greenwood et al., 1977; McClintock & Avison, 1968; Sullivan, 1985). Other, but comparatively fewer, studies found no association between both variables (Lattimore, Trudeau, Riley, Leiter, & Steven, 1997; Weisburd et al., 2010). Related to workload, a number of studies have explored whether the number of officers or number of officers per citizen (or per 1,000 citizens, etc.) is associated with clearance rates. Jang, Hoover, and Lawton (2008) found that the number of officers had no effect on clearance rates. Similarly, Weisburd et al. (2010) found no effect on clearances of officers per 1,000 residents. Doerner and Doerner (2012), however, found that clearance rates decreased as agency size increased.

A hodgepodge of more recent studies have explored all manner of factors besides crime type and workload that may be associated with clearance rates. For example, utilizing National Incident Based Reporting System (NIBRS) data, Roberts and Lyons (2009) found that homicides with non-White offenders were more likely to be cleared than those with White offenders, irrespective of victim race. Jang et al. (2008) explored whether broken windows law enforcement was associated with clearance rates. Whereas broken windows enforcement improved burglary and auto theft clearances, the opposite effect was observed for larceny. Terror alert levels have also been linked with police performance. In a study of Israeli communities, Weisburd et al. (2010) found that high threat levels lead to reduced clearances in Jewish communities, but improved clearances in majority Arab communities. Researchers have even explored whether agency accreditation bears on clearance rates, but the sole study in this area found no such effect (Doerner & Doerner, 2012). Finally, it seems information technology spending alone is not associated with clearance rates, but when combined with specific management styles, namely COMPSTAT, improved productivity may result (Garicano & Heaton, 2010).

The Present Study

This study addressed two shortcomings in the crime clearance literature. First, no multijurisdictional models of crime clearance rates have incorporated measures of investigative resources, despite the fact that investigators have a nonzero (and perhaps significant) effect on crime clearances. This study incorporated three distinct measures of investigative resources and explored their association with clearance rates.

Second, prior work in this area has examined crime clearance rates at distinct periods in time. In contrast, this study developed group-based trajectories (Nagin, 2005) of crime clearance rates, thus permitting a more nuanced understanding of the role investigative resources play in crime clearances. Trajectory analyses have not yet been brought to bear on crime clearances, but there is arguably something to be gained from doing so. Just as criminal careers follow distinct trajectories, police departments surely follow similar paths with their clearance rates; some are steadily improving, while others may be stagnating or performing inadequately.

Method

Agency-level UCR “Offenses Known and Clearances by Arrest” data for the years 2000 through 2012⁴ were merged with LEMAS data for the years 2000, 2003, and 2007.⁵ Several manipulations were performed to complete a successful merger of all 16 data sets. First, with the UCR data, all observations with zero offenses reported were dropped, as a clearance rate with a zero in the denominator is undefined. Second, some agencies reported crimes cleared and offense counts in such a manner as to generate clearance rates in excess of one. These agencies were dropped (there were approximately 10 such agencies). These adjustments, coupled with the merger of all 13 UCR waves, resulted in an initial sample of 8,188 law enforcement agencies.

The next steps involved combining LEMAS data sets. First, 2000 and 2003 LEMAS data were merged on a common variable, the Federal Information Processing Standards (FIPS) code. There were 2,985 agencies in the 2000 LEMAS data set, and 2,859 in the 2003 data set. Merging both resulted in a sample of 1,248 agencies. The combined 2000/2003 data were then merged with 2007 LEMAS data on agency name, agency type, and state (the 2007 data did not contain FIPS codes). This resulted in a sample of 689 agencies in the combined 2000/2003/2007 LEMAS file. After the LEMAS and UCR data were merged, 570 agencies remained in the final sample. The sample shrunk from 689 to 570 because not all LEMAS agencies reported their crime statistics to the FBI. Finally, some agencies failed to provide complete data in the LEMAS surveys, so some models (see summary statistics below, and particularly Table 1) were estimated on fewer than 570 agencies, but no fewer than 482 in any given model.

Table 1.

Summary Statistics.

Variable	Observations	M	SD	Minimum	Maximum
Dependent
Violent Crime Clearance Rate^a	496	52.68	17.29	0	89.71
Property Crime Clearance Rate^a	496	16.09	7.12	0	48.02
Investigative Resources
Total Spending	482	954,416.1	810,683.8	243,741	8,133,670
Investigative Spending	496	5,008,928	9,628,371	197,561.3	134,000,000
Proportion Investigators	496	15.88	5.04	2.29	34.63
Controls
Workload	496	29.98	14.23	0.82	67.69
Officers per 1,000	496	2.48	2.45	0.70	42.19
Crime Type	496	0.45	0.18	0.002	1.20
Population	496	178,730.2	286,601.6	0	3,801,382

Averages 2000-2012.

Dependent Variables

The dependent variables were the violent and property crime clearance rates. The violent crime clearance rate for a particular year was calculated by adding the total number of clearances for murders, rapes, robberies, and aggravated assaults, then dividing by the sum of murder, rape, robbery, and aggravated assault offenses.⁶ Offenses used in the calculation of the property crime clearance rate were burglary, larceny, and motor vehicle theft. The author opted for violent and property crime clearance rates in lieu of clearance rates for the individual offenses because several jurisdictions reported no offenses and/or clearances for specific offenses during the observation period. This would have further eroded the sample size. Summary statistics for these and other variables included in the analysis appear in Table 1.

Independent Variables

Seven independent variables were included in the analyses. All were obtained from the LEMAS data and were averaged together over the three waves.⁷ First, “Total Spending” was calculated by dividing each agency’s total operating budget by its number of full-time sworn investigative officers. Presumably, an agency that receives ample funding for all its officers would be better prepared to fund and/or staff investigations. This measure does not specifically address investigations and is, therefore, somewhat limited, but prior studies of the relationship between police levels and crime (e.g., Marvell & Moody, 1996; Worrall & Kovandzic, 2010) have employed similar measures.

Second, “Investigative Spending” was calculated by dividing the number of full-time sworn investigative officers by the total number of full- and part-time sworn personnel in the agency and then multiplying the quotient by the agency’s total operating budget. This afforded a more refined measure of resources devoted to investigations. It improved on the first measure by focusing squarely on the proportion of investigators in the agency and, by extension, resources devoted to them. Third, “Proportion Investigators” was calculated by dividing the number of full-time investigative officers by the number of full-time sworn officers. Together, these three variables provided different measures of investigative resources in law enforcement agencies. Similar measures have been used in studies of the association between police levels and crime (e.g., Marvell & Moody, 1996; Worrall & Kovandzic, 2010), but without the investigative focus.

Four control variables were also included. “Workload” was calculated by dividing the number of index crimes by the number of full-time sworn personnel. A similar measure has been used in several models of crime clearances (e.g., Puckett & Lundman, 2003; Weisburd et al., 2010). Officers per 1,000 citizens were also included, as they have been deemed an important measure of police resources likely to bear on clearance rates (e.g., Doerner & Doerner, 2012; Jang et al., 2008). As discussed earlier in the literature review, however, their effects on clearance rates remain unclear. Next, “Crime Type” was calculated by dividing the number of violent crimes by the number of property crimes. A number of prior studies support the notion that violent crimes are cleared more readily than property crimes (Ahlberg & Knutsson, 1987; Cordner, 1989; Paré & Ouimet, 2004), justifying inclusion of this variable. Finally, “Population,” particularly the “population served,” as measured by the LEMAS survey, was included to address prior findings that crimes may be more (e.g., Greenwood et al., 1977) or less (Ahlberg & Knutsson, 1987; Cordner, 1989; Paré et al., 2007; Pogue, 1975) “solvable” in crowded urban centers.

Analytical Strategy

Group-based methods are useful for modeling unobserved heterogeneity in the population. They have proven especially helpful with identifying and explaining the behaviors of subpopulations of offenders (e.g., Nagin, 2005; Nagin & Piquero, 2010; Piquero, 2008). Group-based trajectory modeling (GBTM) has only recently started to gain traction outside the offender context (e.g., Schuck, 2013; Stewart, Morris, & Weir, 2014). It has not been used in the modeling of crime clearance rates, but it has been applied in the crime rate context (e.g., Schupp & Rivera, 2010; Weisburd et al., 2004).

Consistent with prior practice, a censored normal distribution was selected for the analyses. Although it is debatable whether a clearance rate is censored in the traditional sense, the censored model can be used for continuous data (Nagin, 2005). All models were estimated with the user-written “traj” plugin in Stata (Jones & Nagin, 2013).

The first step in the analyses was to identify trajectories of crime clearance rates. To evaluate fit, the Bayesian Information Criterion (BIC) was used, as were posterior probabilities and the percentage of observations assigned to each group. Beginning with violent crime, the author specified a one-group model, then continued up through as many as five groups. In addition, all combinations of “constant only,” linear, quadratic, and cubic parametric forms were specified. In the end, the optimal violent crime model consisted of four groups (a five-group model assigned a mere 1% of cases to the fifth group).⁸ Posterior probabilities of group assignment averaged no less than .93, and all exceeded the .70 cutoff suggested by Nagin (2005). Trajectories (including 95% confidence intervals) for this model appear in Figure 1. Note that the confidence intervals are tightly bound around each plot, indicating minimal variability within each group.

Figure 1.

Violent crime clearance rate trajectories.

As Figure 1 depicts, there were four groups of agency-level clearance rate trajectories. Clearance rates improved over the observation period for Groups 1 and 3, but they stagnated for Groups 2 and 4. Moreover, Figure 1 shows that clearance rates in the lowest performing group were increasing between 2000 and 2012, whereas the best performing agencies maintained steady clearance rates. Finally, contrary to some recent accounts, clearance rates, at least for violent crime on the whole, did not decline during the observation period. For want of a better description, the groups depicted in Figure 1 can be labeled thusly: Group 1 = low and improving; Group 2 = moderate and stable; Group 3 = average⁹ and improving; and Group 4 = high and stable.

The same approach was taken for property crime clearance rates, but a two-group linear “1 1” model resulted in the best fit. Figure 2 shows two distinct groups of agencies (both linear). In Group 1, clearance rates were low but improving (i.e., low and improving). In Group 2, clearance rates were better in 2000 and improving at a more pronounced rate between 2000 and 2012 (i.e., moderate and improving). Note further that the maximum overall clearance rate for property crime fell short of 25%. This reinforces the notion that property crimes are more difficult to clear than violent crimes (e.g., Ahlberg & Knutsson, 1987; Cordner, 1989; Paré & Ouimet, 2004).

Figure 2.

Property crime clearance rate trajectories.

Once the best-fitting violent and property crime models were identified, ANOVA was used to check for significant differences in the independent variables (“risk factors” in GBTM parlance) across groups. Last, multinomial logistic regression models were estimated to explore the association between investigative resources and crime clearance rates across the different trajectories.

Results

Table 2 reports ANOVA results comparing group means for all seven variables across the four violent crime clearance rate groups. Significant group differences existed for three variables, namely, Investigative Spending, Workload, and Crime Type. The remaining means did not differ significantly across groups. This is not particularly problematic, however, as the groups were identified based on clearance rate trajectories alone.

Table 2.

ANOVA Results and Group Means: Violent Crime.

Variable	G1	G2	G3	G4	F
Investigative Resources
Total Spending	1,031,114	895,297.5	918,492.4	1,046,154	0.99
Investigative Spending	6,456,330	6,721,404	4,602,994	3,138,527	3.43*
Proportion Investigators	16.31	15.68	16.28	15.26	1.22
Controls
Workload	31.26	31.90	30.55	26.35	3.65*
Officers per 1,000	2.50	2.60	2.59	2.15	0.95
Crime Type	0.45	0.49	0.44	0.41	4.44**
Population	238,171.3	202,678.9	161,440.5	131,215	2.24

p < .05. **p < .01.

Next, Table 3 reports the results from a series of multinomial regressions. Separate models were estimated for each variable measuring investigative resources. Starting with Model 1, we see that Total Spending (an agency’s total operating budget by its number of full-time sworn investigative officers) was not predictive of group membership (“low and improving” agencies served as the reference category). However, and consistent with prior research (e.g., Bayley, 1994; Greenwood et al., 1977; McClintock & Avison, 1968; Sullivan, 1985), Workload was inversely predictive of membership in the “high and stable” group, as was Officers per 1,000 (Jang et al., 2008, also uncovered such an association).

Table 3.

Multinomial Regression Results: Violent Crime.

	Model 1		Model 2		Model 3
Variable	b	SE	b	SE	b	SE
Investigative Resources
Total Spending
Group 1	—	—	—	—	—	—
Group 2	0.004	0.028	—	—	—	—
Group 3	0.003	0.027	—	—	—	—
Group 4	0.033	0.027	—	—	—	—
Investigative Spending
Group 1	—	—	—	—	—	—
Group 2	—	—	0.013*	0.006	—	—
Group 3	—	—	0.005	0.005	—	—
Group 4	—	—	−0.007	0.008	—	—
Proportion Investigators
Group 1	—	—	—	—	—	—
Group 2	—	—	—	—	−0.039	0.039
Group 3	—	—	—	—	−0.009	0.035
Group 4	—	—	—	—	−0.053	0.039
Controls
Workload
Group 1	—	—	—	—	—	—
Group 2	0.006	0.014	0.007**	0.013	0.008	0.014
Group 3	−0.001	0.13	−0.003	0.012	−0.002	0.013
Group 4	−0.035*	0.015	−0.040**	0.014	−0.038*	0.015
Officers per 1,000
Group 1	—	—	—	—	—	—
Group 2	−0.004	0.083	−0.018*	0.073	−0.012	0.076
Group 3	0.016	0.080	0.010	0.067	0.013	0.073
Group 4	−0.409**	0.153	−0.317*	0.149	−0.428**	0.151
Crime Type
Group 1	—	—	—	—	—	—
Group 2	1.881^†	1.027	1.185	0.999	1.215	1.006
Group 3	−0.153	0.983	−0.693	0.958	−0.641	0.967
Group 4	−0.437	1.061	−1.043	1.037	−0.874	1.040
Population
Group 1	—	—	—	—	—	—
Group 2	−0.021	0.041	−0.438	0.183	−0.026	0.044
Group 3	−0.070	0.049	−0.202	0.139	−0.058	0.050
Group 4	−0.182^†	0.099	−0.014	0.197	−0.147^†	0.085

Note. Group 1 served as the reference category. Spending and population variables divided by 100,000 for interpretation—small coefficients otherwise).

†

p < .10. *p < .05. **p < .01.

Model 2 in Table 3 explored the relationship between Investigative Spending (number of full-time sworn investigative officers divided by the total number of full- and part-time sworn personnel, then multiplied by the total operating budget) and group membership with respect to the violent crime clearance rate. Spending per investigator was positively associated with membership in the moderate and stable group. As for the controls, results were similar to Model 1. Finally, Model 3 showed that the Proportion Investigators (number of full-time sworn investigators divided by total full-time sworn personnel) was not predictive of group membership. Workload and Officers per 1,000 were once again inversely associated with group membership, but only for the top performing group (i.e., “high and stable”).

Turning attention to property crime, Table 4 presents ANOVA results from a comparison of means across both groups. Total Spending, Investigative Practices, Officers per 1,000, Crime Type, and Population all differed significantly across the two groups. Proportion Investigators and Workload, however, did not.

Table 4.

ANOVA Results and Group Means: Property Crime.

Variable	G1	G2	F
Investigative Resources
Total Spending	1,063,245	813,456.3	11.50**
Investigative Spending	6,596,546	2,933,947	18.24**
Proportion Investigators	15.58	16.28	2.35
Controls
Workload	29.58	30.50	0.50
Officers per 1,000	2.72	2.18	6.0*
Crime Type	0.47	0.42	6.94*
Population	222,038.8	122,126.8	15.23**

p < .05. **p < .01.

The results from three multinomial property crime clearance regressions appear in Table 5. Several significant relationships appear in the Controls category, and all are consistent with findings reported in Table 2. For example, Officers per 1,000 was inversely predictive of membership in the top performing group. Somewhat contrary to expectations, however, are the findings related to investigative resources. First, Total Spending was inversely predictive of membership in the “moderate and improving” group. In other words, increased spending was less likely to put an agency in the top performing group. This finding is not altogether unexpected because agencies may devote additional resources to the investigation of violent crime. Second, Investigative Spending was also inversely predictive of membership in the top performing group. Again, though, this finding simply compares one group with another and does not constitute evidence that investigative spending has no overall effect on property crime clearance rates.

Table 5.

Multinomial Regression Results: Property Crime.

	Model 1		Model 2		Model 3
Variable	b	SE	b	SE	b	SE
Investigative Resources
Total Spending
Group 1	—	—	—	—	—	—
Group 2	−0.048*	0.020	—	—	—	—
Investigative Spending
Group 1	—	—	—	—	—	—
Group 2	—	—	−0.013*	0.006	—	—
Proportion Investigators
Group 1	—	—	—	—	—	—
Group 2	—	—	—	—	0.006	0.021
Controls
Workload
Group 1	—	—	—	—	—	—
Group 2	−0.007	0.008	−0.000	0.008	−0.002	0.008
Officers per 1,000
Group 1	—	—	—	—	—	—
Group 2	−0.211*	0.101	−0.154^†	0.086	−0.213*	0.092
Crime Type
Group 1	—	—	—	—	—	—
Group 2	−1.221*	0.572	−1.310*	0.565	−2.041*	0.041
Population
Group 1	—	—	—	—	—	—
Group 2	−0.361*	0.094	−0.045	0.163	−0.388**	0.094

Note. Group 1 served as the reference category. Spending and population variables divided by 100,000 for interpretation—small coefficients otherwise).

†

p < .10. *p < .05. ** p < .01.

Discussion

This study improved on prior clearance rate research in two respects. First, it incorporated investigative resource measures. This is important because clearance rates, relative to crime rates at least, contain a fundamental investigative component. It also provides policy-relevant and practical information to law enforcement administrators. Prior studies of police clearance rates have focused heavily on factors such as officers per 1,000. To the extent this measure is associated with clearance rates, it is not particularly easy for departments to just hire more officers with the goal of improving crime clearances. Internally, though, agencies may be able to reallocate or redistribute funds to enhance investigations, without the need to hire more sworn personnel.

Second, this study brought group-based trajectory models to bear on crime clearances in a large sample of law enforcement agencies. This technique has been confined primarily to offender samples, but recently has started to expand into the realm of crime rates (e.g., Schupp & Rivera, 2010; Weisburd et al., 2004). The present study, thus, constitutes the first attempt to apply GBTM methodology to crime clearance rates. As much is known about the relative strengths and weaknesses of different agencies’ crime clearance capabilities, such a methodology seems most appropriate. The assumption that all agencies follow similar paths with respect to clearance rates is inappropriate and misguided. Slipping clearance rates may make great headlines (Kaste 2015a, 2015b), but there are several agencies, as this study bears out, that were steadily improving during the observation period.

Findings from this research offered a less than ringing endorsement for allocating more resources to criminal investigations. With respect to violent crime, spending per investigator was modestly and positively associated with group membership. The opposite effect, however, was observed in the property crime context. Early agency-specific studies of police detectives were critical of their role in crime clearances (e.g., Folk, 1971; Greenwood, 1970; Isaacs, 1967; Willmer, 1970). The famous RAND study of police detectives (Greenwood et al., 1977; Greenwood & Petersilia, 1975) was particularly critical of police detectives, claiming that only 3% of crimes were cleared because of investigative handiwork. Subsequent studies argued detectives were more important than previously thought (e.g., Brandl & Frank, 1994). The current study’s finding offers more support for the RAND perspective than that presented in subsequent police detective research.

Limitations

This study was not without limitations. While the sample was sizable and covered most large law enforcement agencies throughout the United States, the vast majority of small agencies were excluded.¹⁰ Another problem pertains to cross-jurisdictional comparisons of crime clearance rates. There is no need to revisit the problems inherent in crime clearance rates, except to say that their limitations should be acknowledged. They do, though, remain a key metric of police performance, and perhaps a more objective one than alternative measures (Davenport, 1999; Leishman & Mason, 2003).

Another issue is that the data sets analyzed as part of this project did not match perfectly. UCR data were available in yearly increments from 2000 through 2012. LEMAS data, however, were only available for three of those years (2000, 2003, and 2007). The author averaged the LEMAS data together, essentially treating them as constant across the observation period. While this did not affect the results (separate models using each LEMAS year were estimated), it is also a stretch to assume that investigative spending remained constant during a time that clearance rates may or may not have been changing. Unfortunately, yearly data on agencies’ investigative spending are not available.

Conclusion

Studies of aggregate crime rates are legion. Studies of crime clearance rates are exceptionally rare in comparison, yet law enforcement agencies must continue to answer for both. This study sought to add to the crime clearances literature with a focus on investigative resources in a large sample of agencies, while also bringing GBTM to bear on a problem to which it has not yet been applied. It found only a weak association between investigative resources and clearance trajectory group membership.

What, then, drives crime clearance rates? Perhaps more important, what is behind the largely downward trend in (particularly murder) clearance rates over the past several decades? Some speculative answers can be proffered. One is that crime itself is changing; some offenses, especially those involving computers and the Internet, are difficult to solve. Another is that the public is more aware of declining crime rates, and wants them to continue on a similar path, meaning clearances are less of a priority now than they once were. To the extent law enforcement agencies respond to such public sentiment, they may be focusing more on crime prevention than crime clearance. Indeed, solving crime could be secondary to blocking it from occurring in the first place. Future research should, therefore, explore whether public priorities, police strategies, and changes in the nature of the crime problem (or a combination of all three) help explain variation in crime clearance rates.

Footnotes

Appendix

Correlation Matrices for LEMAS Variables.

LEMAS	Total Spending			Investigative Spending			Proportion Investigators
LEMAS	2000	2003	2007	2000	2003	2007	2000	2003	2007
2000	1.0000	—	—	1.0000	—	—	1.0000	—	—
2003	.5508	1.0000	—	.9591	1.0000	—	.4462	1.0000	—
2007	.4246	.6031	1.0000	.9025	.8914	1.0000	.3713	.4627	1.0000

Note. LEMAS = Law Enforcement Management and Administrative Statistics.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biography

John L. Worrall is professor of criminology at the University of Texas at Dallas. He has published articles and book chapters on a wide range of topics, ranging from legal issues in policing to crime measurement. He is also the author of several books, including Crime Control in America: What Works? (3rd ed., Pearson) and Criminal Procedure: From First Contact to Appeal (5th ed., Pearson). He currently serves as editor of the journal Police Quarterly.

References

Ahlberg

Knutsson

(1987). Why do clearance rates vary? Stockholm, Sweden: National Council for Crime Prevention.

Alpert

G. P.

Moore

M. H.

(1993). Measuring police performance in the new paradigm of policing. In Performance measures for the criminal justice system (pp. 109-133). Washington, DC: Bureau of Justice Statistics. Retrieved from http://www.bjs.gov/content/pub/pdf/pmcjs.pdf

Bayley

(1994). Police for the future. New York, NY: Oxford University Press.

Bonkiewicz

(2015). Bobbies and baseball players: Evaluating patrol officer productivity using Sabermetrics. Police Quarterly, 18, 55-78.

Borg

M. J.

Parker

K. F.

(2001). Mobilizing law in urban areas: The social structure of homicide clearance rates. Law & Society Review, 35, 435-466.

Brandl

Frank

(1994). The relationship between evidence, detective effort, and the disposition of burglary and robbery investigations. American Journal of Police, 13, 149-169.

Brodeur

J. P.

(1998). How to recognize good policing: Problems and issues. Thousand Oaks, CA: Sage.

Cameron

(1987). A disaggregated study of police clear-up rates for England and Wales. Journal of Behavioral Economics, 16, 1-18.

Chaiken

J. M.

Greenwood

P. W.

Petersilia

(1977). The criminal investigation process: A summary report. Policy Analysis, 3, 187-217.

10.

Chambliss

W. J.

Seidman

R. B.

(1982). Law, order, and power. Reading, MA: Addison-Wesley.

11.

Conklin

(1972). Robbery and the criminal justice system. Philadelphia, PA: J. B. Lippincott.

12.

Cordner

G. W.

(1989). Police agency size and investigative effectiveness. Journal of Criminal Justice, 17, 145-155.

13.

Davenport

D. R.

(1999). Environmental constraints and organizational outcomes: Modeling communities of municipal police departments. Police Quarterly, 2, 174-200.

14.

Davis

Ortiz

C. W.

Euler

(2015). Revising “measuring what matters”: Developing a suite of standardized performance measures for law enforcement. Police Quarterly, 18, 469-95.

15.

Doerner

W. M.

Doerner

W. G.

(2012). Police accreditation and clearance rates. Policing: An International Journal of Police Strategies & Management, 35, 6-24.

16.

Eck

J. E.

(1992). Solving crimes: The investigation of burglary and robbery. Washington, DC: Police Executive Research Forum.

17.

Eck

J. E.

Maguire

E. R.

(2000). Have changes in policing reduced violent crime? An assessment of the evidence. In Blumstein

Wallman

(Eds.), The crime drop in America (pp. 207-265). Cambridge, UK: Cambridge University Press.

18.

Federal Bureau of Investigation. (2014). Crime in the United States 2014. Washington, DC: Author. Retrieved from https://www.fbi.gov/about-us/cjis/ucr/crime-in-the-u.s/2014/crime-in-the-u.s.-2014/offenses-known-to-law-enforcement/clearances/main

19.

Felson

(1998). Crime in everyday life (2nd ed.). Thousand Oaks, CA: Pine Forge Press.

20.

Folk

J. F.

(1971). Municipal detective systems: A quantitative approach (Technical Report No. 55). Cambridge, MA: Operations Research Center, Massachusetts Institute of Technology.

21.

Garicano

Heaton

(2010). Information technology, organization, and productivity in the public sector: Evidence from police departments. Journal of Labor Economics, 28, 167-201.

22.

Gill

(1987). Clearing up crime: The big “con.” Journal of Law and Society, 14, 254-265.

23.

Greenwood

P. W.

(1970). An analysis of the apprehension activities of the New York City Police Department. New York, NY: RAND Corporation.

24.

Greenwood

P. W.

Chaiken

J. M.

Petersilia

(1977). The criminal investigation process. Lexington, MA: Lexington Books.

25.

Greenwood

P. W.

Petersilia

(1975). The criminal investigation process, Volume 1: Summary and policy implications. Santa Monica, CA: RAND Corporation.

26.

Gross

S. R.

Mauro

(1989). Death and discrimination: Racial disparities in criminal sentencing. Boston, MA: Northeastern University Press.

27.

Hoover

L. T.

(1996). Quantifying quality in policing. Washington, DC: Police Executive Research Forum.

28.

Isaacs

H. H.

(1967). A study of communications, crimes, and arrests in a metropolitan police department (Appendix B in Task Force Report: Science and Technology). Washington, DC: President’s Commission on Law Enforcement and Administration of Justice.

29.

Isackson

(2013, March, 25). Chicago’s criminals are getting away with murder. Chicago Magazine. Retrieved from http://www.chicagomag.com/Chicago-Magazine/May-2013/Getting-Away-with-Murder/

30.

Jang

Hoover

L. T.

Lawton

B. A.

(2008). Effect of broken windows enforcement on clearance rates. Journal of Criminal Justice, 36, 529-538.

31.

Jones

B. L.

Nagin

D. S.

(2013). A note on a Stata plugin for estimating group-based trajectory models. Sociological Methods & Research, 42, 608-613.

32.

Kaste

(2015a, March 30). How many crimes do your police “clear”? Now you can find out. NPR. Retrieved from http://www.npr.org/2015/03/30/395799413/how-many-crimes-do-your-police-clear-now-you-can-find-out

33.

Kaste

(2015b, March 30). Open cases: Why one-third of murders in America go unresolved. NPR. Retrieved from http://www.npr.org/2015/03/30/395069137/open-cases-why-one-third-of-murders-in-america-go-unresolved

34.

LaFree

(1989). Rape and criminal justice: The social construction of sexual assault. Belmont, CA: Wadsworth.

35.

Lattimore

P. K.

Trudeau

Riley

K. L.

Leiter

Steven

(1997). Homicide in eight U.S. cities: Trends, context, and policy implications. Washington, DC: National Institute of Justice.

36.

Leishman

Mason

(2003). Policing and the media: Facts, fictions, and factions. Devon, UK: Willan Press.

37.

Levitt

(1997). Using electoral cycles in police hiring to estimate the effect of police on crime. American Economic Review, 87, 270-290.

38.

Liska

Chamlin

(1984). Social structure and crime control among macrosocial units. American Journal of Sociology, 90, 383-395.

39.

Litwin

K. J.

(2004). A multilevel multivariate analysis of factors affecting homicide clearances. Journal of Research in Crime & Delinquency, 41, 327-351.

40.

Loveday

(1999). The impact of performance culture on criminal justice agencies. International Journal of the Sociology of Law, 27, 351-377.

41.

Maguire

E. R.

King

W. R.

Johnson

Katz

C. M.

(2010). Why homicide clearance rates decrease: Evidence from the Caribbean. Policing & Society, 20, 373-400.

42.

Maguire

E. R.

Uchida

C. D.

(2000). Measurement and explanation in the comparative study of American police organizations. In Duffee

(Ed.), Criminal justice 2000: Vol. 4. Measurement and analysis of crime and justice (pp. 491-558). Washington, DC: National Institute of Justice.

43.

Maguire

M. R.

(1997). Crime statistics, patterns, and trends: Changing perceptions and their implications. In Maguire

Morgan

Reiner

(Eds.), The Oxford handbook of criminology (pp. 135-219). Oxford, UK: Oxford University Press.

44.

Marvell

T. B.

Moody

C. E.

(1996). Specification problems, police levels, and crime. Criminology, 34, 609-643.

45.

Mastrofski

S. D.

(1999). Policing for people: Ideas in American policing. Washington, DC: Police Foundation.

46.

McClintock

F. H.

Avison

N. H.

(1968). Crime in England and Wales. Heinemann Educational Publishers. London.

47.

Mesch

G. S.

Talmud

(1998). The influence of community characteristics on police performance in a deeply divided society: The case of Israel. Sociological Focus, 31, 233-248.

48.

Nagin

D. S.

(2005). Group-based modeling of development. Cambridge, MA: Harvard University Press.

49.

Nagin

D. S.

Piquero

A. R.

(2010). Using the group-based trajectory model to study crime over the life course. Journal of Criminal Justice Education, 21, 105-116.

50.

Navratil

(2015, January 15). Pittsburgh police call killings “public health emergency,” promise better solve rate. Pittsburgh Post-Gazette. Retrieved from http://www.post-gazette.com/local/city/2015/01/15/Pittsburgh-police-call-killings-public-health-emergency-promise-better-solve-rate-homicides/stories/201501150287

51.

Ousey

G. C.

Lee

M. R.

(2010). To know the unknown: The decline in homicide clearance rates, 1980-2000. Criminal Justice Review, 35, 141-158.

52.

Paré

Felson

R. B.

Ouimet

(2007). Community variation in crime clearance: A multilevel analysis with comments on assessing police performance. Journal of Quantitative Criminology, 23, 243-258.

53.

Paré

Ouimet

(2004). A measure of police performance: Analyzing police clearance and charge statistics. Canadian Review of Policing Research, 1, 23-42.

54.

Piquero

A. R.

(2008). Taking stock of developmental trajectories of criminal activity over the life course. In Liberman

(Ed.), The long view of crime: A synthesis of longitudinal research (pp. 23-78) New York, NY: Springer

55.

Pogue

T. F.

(1975). Effect of police expenditures on crime rates: Some evidence. Public Finance Review, 3, 14-44.

56.

Puckett

J. L.

Lundman

R. J.

(2003). Factors affecting homicide clearances: Multivariate analysis of a more complete conceptual framework. Journal of Research in Crime & Delinquency, 40, 171-193.

57.

Reaves

B. A.

(2015). Local police departments, 2013: Personnel, policies, and practices. Washington, DC: Bureau of Justice Statistics.

58.

Reiner

(1998). Process or product? Problem of assessing individual police performance. In Brodeur

J. P.

(Ed.), How to recognize good policing: Problems and issues (pp. 55-72). Thousand Oaks, CA: Sage.

59.

Riedel

Jarvis

(1998). Decline of arrest clearances for criminal homicide: Causes, correlates, and third parties. Criminal Justice Policy Review, 9, 279-306.

60.

Roberts

Lyons

C. L.

(2009). Victim-offender racial dyads and clearance of lethal and nonlethal assault. Journal of Research in Crime & Delinquency, 46, 301-326.

61.

Schuck

A. M.

(2013). A life-course perspective on adolescents’ attitudes to police: DARE, delinquency, and residential segregation. Journal of Research in Crime & Delinquency, 50, 579-607.

62.

Schupp

Rivera

(2010). Identifying imprisonment patterns and their relation to crime among New York counties 1990-2000. Criminal Justice Policy Review, 21, 50-75.

63.

Skolnick

(1966). Justice without trial: Law enforcement in democratic society. New York, NY: John Wiley.

64.

Smith

(1986). The neighborhood context of police behavior. Crime and Justice: A Review of Research, 8, 313-341.

65.

Stewart

D. M.

Morris

R. G.

Weir

(2014). Youth perceptions of the police: Identifying trajectories. Youth Violence and Juvenile Justice, 12, 22-39.

66.

Sullivan

P. S.

(1985). Determinants of crime and clearance rates for seven index crimes. Washington, DC: National Institute of Justice.

67.

Turk

(1969). Criminality and legal order. Chicago, IL: Rand McNally.

68.

Van Meter

D. J.

(2001). Evaluating dysfunctional police performance: A zero-based approach. Springfield, IL: Charles Thomas.

69.

Walker

M. A.

(1992). Do we need a clear-up rate? Policing & Society, 2, 293-306.

70.

Walker

(1998). Popular justice: A history of American criminal justice (2nd ed.). New York, NY: Oxford University Press.

71.

Weisburd

Bushway

Lum

Yang

(2004). Trajectories of crime at places: A longitudinal study of street segments in the city of Seattle. Criminology, 42, 283-321.

72.

Weisburd

Hasisi

Jonathan

Aviv

(2010). Terrorist threats and police performance: A study of Israeli communities. British Journal of Criminology, 50, 725-747.

73.

Wellford

Cronin

(1999). An analysis of variables affecting the clearance of homicides: A multistate study. Washington, DC: Justice Research and Statistics Association. Retrieved from http://www.jrsa.org/pubs/reports/Clearance_of_Homicide.html

74.

Willmer

M. A. P.

(1970). Crime and information theory. Chicago, IL: Aldine.

75.

Worrall

J. L.

Kovandzic

T. V.

(2010). Police levels and crime rates: An instrumental variables approach. Social Science Research, 39, 506-516.

76.

Zimring

F. E.

(2006). The great American crime decline. New York, NY: Oxford University Press.