LANIC LANIC Home Page This page in English Esta página en Español Esta página em Português Lanic Site Map Search Lanic & Latin America Suggest a Link Information About Lanic


Session VIII:

"Innovative Methodologies & Statistical Analysis in Working with Latin American Crime Data"


Two Matters Concerning Data Collection In Latin American Criminology

Christopher Birbeck

Universidad de los Andes

 

Introduction

Techniques of data collection (which I define to include sampling strategies, access to information, and methods for eliciting and recording information) exist in an abstract sense to textbooks on methodology. They become concrete when applied in a field setting, and must always be adapted to the setting. Thus, data collection practices will vary to some degree in relation to the characteristics of the field setting. In the following brief remarks, I wish to identify two aspects of data collection in Latin American criminology that in my opinion warrant continued reflection, creative thinking, and methodological research. They are: a) the recent difficulties in generating representative samples of urban populations; and b) the validity of self-administered surveys. These are not the only aspects of data collection in Latin American criminology that might be considered (how to measure white collar crime; the effects of "telescoping" on the accuracy of respondent recall; how to improve institutional records; are other topics that immediately come to mind); nor are they related to each other in any particular way. My choice of these topics is dictated by personal research experience, and my hope is that other participants in this workshop will have ideas to contribute to the following reflections.

Generating Representative Samples of Urban Populations

A sample is representative if all elements of the population have an equal chance of inclusion in the sample. Generating representative samples requires, obviously, a complete listing of population members so that each has an equal probability of selection. When constructing representative samples of the general population in Latin American field research, such listings are rarely available (voter lists may be incomplete; a considerable proportion of the population does not have a driver's license; many households are illegally connected to utilities, etc.). However, this difficulty has typically been resolved by the use of multistage areal cluster sampling, based on the units (municipalities, parishes, segments, blocks, residences) defined and updated by national census offices. which gives households and individuals a quasi-equal probability of selection. Thus, representative general population samples in Latin American criminology research have to date (as far as I know) always used sampling frames provided by census offices.

But quasi-equal probability of selection, although necessary, is not sufficient to guarantee equal probability of inclusion in a data collection procedure because access to the selected population members must be successfully obtained. And access is markedly different in its distribution across jurisdictions and socioeconomic levels in Latin America (more so, I think, that in North America or Western Europe). Victim surveys are an excellent example of the problems caused by unequal access to population. Over the last 20 years, I have participated in the design and application of victim surveys in Colombia (Cali) and Venezuela (national, Caracas, Maracaibo, Mérida)1 and have witnessed some quite notable changes in access to households.

The process is ironic in nature (for criminologists, at least) and is simple to describe. Perceptions of crime levels and the risks of becoming a victim (lumped together in Venezuela under the term inseguridad personal) have encouraged a trend toward physical isolation and greater protection, especially in the large urban areas (where both the perceived and recorded incidence of crime are higher), and among residents at higher socioeconomic levels (who can pay for isolation by building high walls and railings, and employing private security personnel, notably guachimanes). Even 20 years ago, access to wealthier households could be difficult, usually because one was reduced to speaking with what seemed like congenitally suspicious service personnel (maids, in particular) through a two-way speaker system that kept the researcher on the sidewalk and avoided any face-to-face contact. As concern about crime has increased over the years, so have the restrictions on access to households of the purposes of conducting interviews, and such restrictions have spread to middle income neighborhoods, especially apartment blocks. Interviewers are now included in the lists of prohibited "visitors" that are given to security guards and guachimanes, or occasionally posted at the entry to residential buildings. This trend toward the increasing isolation of middle and upper income householders must be added to a more stable characteristic of some lower income neighborhoods - access controlled and monitored by youths and young men who create the perception, if not always the reality, of threats to interviewers abd other outsiders.

The result of this trend has been to remove many urban households from the possibility of being included in victim surveys that rely on multistage areal cluster samples and face-to-face interviews. The probability of inclusion of any randomly selected low-income household is therefore greater than that of any middle or upper income household. For example, 15 years ago it was typical to find in my home town of Mérida that of ten low-income households systematically pre-selected from a randomly assigned block perhaps one would be unavailable for interview (refusal was and is still rare among low income households); whereas only three or four to ten upper income households would be accessible and respond to the survey. The standard instruction given to interviewers to continue adding households to the list until the quota of ten had been reached would sometimes means that they exhausted the residences in one block and continued to a contiguous block in wealthy neighborhoods. Recently, the difficulties created by increasingly restricted access were made dramatically apparent in a victim survey conducted by the Venezuelan Crime Prevention Office in Caracas in 1997: with instructions to interviewers to visit a residence three times before giving up, and no provisions for replacements if an interview could not be obtained, survey coordinators found that they were up against a 77% unavailable/refusal rate-comparable to the fall-out rates typical of many mail surveys in North America. What started our as a probability sample ended up as convenience sample. Because unavailability and outright refusal to participate in surveys (e.g., exposure to strangers, perceptions of the risk of victimization), the result of such de facto convenience sampling is, in all likelihood, the more frequent inclusion of households and individuals that take fewer precautions regarding their property and person. The Caracas victim survey was declared by its sponsors to be unusable as a portrait of victimization in the nation's capital.

A second and related problem-which has been quite constant over the years - concerns access to individuals within each household. If interviewers are instructed to take the first available adult (either as a single respondent or as a proxy respondent for the whole household), between 60 and 70% of respondents will be female. The typical organization of domestic and daily activities means that males are both less likely to be at home and less likely to answer the door. Most males therefore have a lower probability than most females of inclusion in a sample based on the quasi-random selection of housing units. The cumulative result of both problems so far mentioned is that samples generated by multistage areal clustered selection of housing units, with face-to-face interviews conducted during the day, are likely to give greatest representation to female household heads in low-income areas. Males, and upper-income males in particular, have a much lower probability of inclusion.

These problems obviously affect all survey research in large Latin American cities, but the subject matter of criminology means that we should be particularly attentive to the difficulties of obtaining representative population samples. Three years ago, my colleague Luis Galbadón and I were considering measuring community attitude as part of a project on use of force by the police. We gave up because we could not think of a data collection strategy that we could devise with available resources that would avoid the problems I have just outlined. (Of course, more money might have helped, but money only facilitates data collection, it does not set the strategy.) And that is the problem for researchers: how to generate representative samples of large urban populations.

There seems to be no currently viable alternative to selecting samples by residence (selection by institutions, e.g., schools, workplace, or by location, e.g., parks, plazas, sports events, all have their attendant problems of representative sampling). Multistage areal clustered sampling appears to be the best strategy. There are no comprehensive lists of households or individuals that are available. There may even be no specific street addresses in the lists that exist (unconsolidated squatter settlements may not have numbered their streets and dwellings; in Maracaibo, some streets have more than one number, and residents often identify their addresses by prominent local landmarks). Phone ownership rates are comparatively low and concentrated mainly in middle and upper income neighborhoods.

Similarly, there appears to be no viable alternative to face-to-face interviews as the method of data collection. Surveys that are left for respondents to fill out run into a number of problems: a) response rates are likely to be much lower; b) some respondents, especially older and lower-income respondents, cannot read and write; c) someone else may fill out the survey for the designated respondent; and d) the survey may be filled out by one person, but based partly on the experiences and opinions of all present. Apart from the problems of representativeness implied by using phone samples, Latin Americans are notably unwilling to share information with strangers by phone.

They challenge, therefore, is to improve access to middle and upper-income residences and to male respondents for the purpose of conducting face-to-face interviews. There are a number of possible methods that might be explored: a) paying interviewers more so that they can be persuaded to work at night or on weekends; b) contacting neighborhood and apartment associations to obtain permission for access to residences; c) hiring a resident of the neighborhood to conduct the interviews; d) offering respondents compensation for their time. All of these would add to the cost of surveys. Option b) would require a considerably greater period of time (some associations function very sporadically); while option c) might introduce more measurement error because interviewer selection, training and supervision might suffer. Criminologists are caught between getting what they can with less money, or trying to spend more time and money on improving access and thereby generating more representative samples.

The Validity of Self-Administered Surveys

Self-administered surveys refer to instruments that are filled out by respondents, often under conditions on anonymity. Since at least the mid-seventies, sociologists and criminologists have used self-administered surveys in Latin America to measure drug and alcohol use 2 and delinquent behavior, and we have recently used the same kind of survey to study police officers' dispositions to use force.3 Self-administered surveys were developed in North America and thus represent a cultural transplant in Latin America.

When self-administered surveys address sensitive matters such as drug use, criminal behavior or potential misconduct, a major concern is their degree of validity. Researchers in the U.S. have therefore devoted some attention to an assessment of the validity of what are often called "self-report" surveys. One method of assessing validity has been to compare the rates of self-reported delinquency for groups that are officially recognized as criminal or delinquent with rates for groups that are not. According to O'Brien, different studies have revealed significantly higher rates of self-reported delinquency among officially recognized delinquents.4 A second method is to compare self-reported delinquency and crime with official police and court records (although it is apparent that self-reported and officially recorded crime and delinquency may not comprise the same behaviors and acts.) Once again, blatant lying in either direction (concealing or inventing misconduct) does not seem to occur with fatal frequency. However, it is interesting to note that there appear to be cultural differences in validity: black youths conceal more officially recorded delinquency than whites.5

 

Currently, almost nothing is known about the validity of self-administered surveys incorporating sensitive information in Latin America. However, several things suggest that exploration of validity would be worthwhile:

  1. Survey research is far less common in Latin America than in North America, and knowledge of survey research is much more superficial and also less common among Latin American populations. Therefore, self-administered surveys may not be perceived as exercises in academic fact-finding but disguised activities of inspection and control. [One police officer, in an attempt to be complementary to the researcher after completing a survey on dispositions to use force said " Bonito el exam" ("A nice exam.").]
  2. Related to the perception of surveys as potentially threatening is the tendency to give socially desirable responses. Based on their responses to Eysenck's Sincerity Scales (Spanish Version) 71% of police officers in our three-city study of dispositions to use force were considered to be giving socially desirable responses to our survey.
  3. A pilot study conducted in Mérida's local prison compared interviews and self-administered surveys as methods for collecting information about drug and alcohol use while in prison. The groups of prisoners were quite small (n+30), but did not differ on key variables measuring socio-demographic characteristics and criminal history. However, the proportion of prisoners reporting drug and alcohol use while in prison was 53.3% for those interviewed and 26.6% for those who filled out the survey.

In short, the validity of self-administered surveys in Latin American settings is unknown buy may be lower than in North American settings. Research on this matter would help to assess the utility of the method.


"Innovative Methodologies & Statistical Analysis in Working with Latin American Crime Data"

Ignacio Cano

Instituto de Estudos da Religião, Rio de Janeiro

 

Some points for a research agenda in this area.

The first need for researchers in this area in Latin America is simply to have more and better data. The data we have is seldom enough to answer the questions we have in front of us or to answer them to the extent and with the certainty we would like. Most of the data we use originates in public institutions. But rather than wait for better data to come, researchers can play an important role in improving and extending available data.

In the course of our work, we sometimes found out that very important data are simply not collected. For example, police in Rio de Janeiro ignored how many people they killed in armed confrontations. Public officers are to be requested to gather basic data, if anything to measure their own performance against certain criteria. The absence of a culture of institutional performance evaluation does not make this task any easier. It is not uncommon that institutions of a juridical nature, whose main task is the analysis of each individual case by juridical experts in order to produce a sentence, may not always be very amenable to data aggregation and to issues such as the reliability of a classification or the use of indexes based on a large number of not-so-well-analyzed pieces of information.

One of the main targets should be to improve the reliability and validity of the violence and crime data. The data we receive is often the result of official forms that have been poorly filled in, data files that have not been properly checked nor corrected or categories that are employed differently by different public agents with no concern for homogeneity. Unreliability and bias abound. Some of these biases may be inevitable, like the underreporting of minor crimes, but may still be improved. Others could be curtailed with proper analysis and with proposals of new ways to collect, clean and process the data. This has to be done in conjunction with the authorities. Hence, a public exposure of the problems of the data may help persuade the authorities to take action.

It is not uncommon to find police chiefs who reduce the homicide rates in their area by using other police categories (such as 'suspicious death') instead. Strange and spectacular results occur sometimes as a result of a change of criteria in the official classification of certain facts.

Traditional ways to check for the validity and reliability of data include, for instance, the monitoring of trends in time to detect any sudden anomaly, or the use of different sources to report the same phenomenon so as to expose the biases of each of them. This strategy, sometimes called triangulation, can help improve the quality of data considerably. In our research on homicides, we are comparing the two main sources: the Ministry of Health and the Police. Both types of data do not coincide very well though they tend to show a reasonable correlation with one another in time. There are reasons, linked to the way the information is collected, why these two sources should not necessarily match perfectly. In fact, the number of homicides registered by the health system should be bigger than the one collected by the police. However, we found the disturbing result that there were some periods where one would exceed the other and vice-versa.

Another source of concern is the difficulty to match a single case in the different institutions involved in what we may call the system of public security. Different police forces, forensic institutes, prosecutor's offices, juridical bodies, prisons, etc. each tends to produce and keep their registries in ways that are self-serving. Little consideration is allowed for the fact that they all should harmonize their registries so as to facilitate the follow up of a case inside the system. Any evaluation of the system as a whole and even of the individual organizations involved depends on being able to analyze cases from the beginning (violent or criminal incident) to the end (end of the sentence of the culprits). Otherwise, many basic questions will remain unanswered. Questions like, what proportion of the persons arrested by the police end up being sentenced in court? Or, what proportion of criminals with a certain profile end up in jail? Many of these problems could be solved just by keeping, for each case, a general identification number, which is common to all institutions, on top of other codes internal to each organization. Databases should also be compatible or, even better, connected, so as to allow for searches along various institutions.

Yet it is not enough to produce good data, the data must be available to all researchers and to the public at large. It is vital to insist that all data, which is not specifically classified as confidential, should automatically be made public in the quickest and easiest possible way, like via the Internet. Given the tradition of secrecy of some public organizations and the high political stakes involved in this kind of information, it is not surprising to see that access to data is often hard or time demanding, when it is not overtly dependent on the relationship (political, personal or otherwise) between each researcher and the public officers in charge. For example, police would refuse to grant data to our organization for around a year's time, after we published critical information, arguing repeatedly that our files 'got lost in the process'. It is not unusual to hear policemen say that crime data like the number of crimes per month in a certain area cannot be given out publicly for fear that "criminals may benefit from them". It is essential to separate the elements that are confidential, like the name and address of individual victims, from aggregate data that should be public. Whenever appropriate, special laws could be sought to guarantee these principles.

Last but not least, other types of information should always complement official registers. In the case of crime, regular victimization surveys are badly needed if we are ever to measure crime in a valid way. Such surveys are expensive and it is hard to think of how regular research funds can be enough to maintain them regularly. Hence, the state should be convinced to finance such surveys regularly, in the same way as population surveys, as a basic tool of public policies.

In any case, research should not be limited to the data on crime and violence. A line of research that I deem particularly useful is the analysis of the perception and social representation of crime and violence in society. The impact that violence has in our societies depends not only on its degree but also on how people perceive it and, therefore, how they react to it. People who work in this area know that the relationship, for example, between crime and violence rates and people's perception of insecurity is not necessarily linear and the dynamics of this link are not always simple. If our aim is not merely to describe reality but also to help develop programs and policies to reduce violence, we need first to understand how and why people feel the way they do. The role of the media in this process cannot be overstated. High exposure to certain types of crimes or to certain types of victims and 'footnote attention' to others help shape the public's perception of these issues.

Innovative Methodologies in the Area

Often, methodologies applied are not necessarily new; it is their application to certain areas that calls our attention. I will mention a few examples.

One of the methodological strategies that are quickly expanding in Brazil is the use of geo-processing of crime data. Researchers in Rio de Janeiro (Fiocruz), Minas Gerais (Fundação João Pinheiro) and in other states like São Paulo or Rio Grande do Sul have been devoting a growing interest to this area. Crime or violent incidents are geographically coded, represented and analyzed. This not only serves to describe the spatial distribution of these phenomena, but also permits an analysis of the impact of several other variables, which are also geographically coded, on crime and violence. Thus, the unit of analysis becomes a geographical unit. Care must be taken to choose a unit with a sufficient number of occurrences to make its rates reliable. Often, a problem in this area is that the geographical units of different sources do not coincide (e.g., police districts and population census sectors), thus making it very difficult to use them simultaneously as would be desired.

Second, the development of indexes from aggregated data can help analyze phenomena for which detailed data for individual cases are not available. Take, for instance, the excessive use of force by law-enforcing officers. In Brazil, precise information of the armed incidents is very hard to obtain since witnesses are too scared to testify. Hence, in our research we used different indexes, such as the proportion of homicides committed by agents of the state and the rate of opponents killed to police officers killed, proposed by professor Paul Chevigny. The rate of opponents killed to opponents wounded, what we call the "lethality index" was used to demonstrate that the so called "bravery awards" given to policemen who took part in armed confrontations had increased not only the number of victims but also the intent to kill and the excessive use of force.

Third, techniques originating in other areas like program evaluation can successfully be applied to this area. An example is interrupted time-series analysis that can be employed to assess the impact of a new event or a new policy on a certain issue. The time series is modeled for the time prior to the intervention. Next an extrapolation of the series for the following period after the intervention has occurred is compared to the actual performance of the series to detect if the intervention had a significant impact. This was already used by Williams (in 1987) to show that refugees from El Salvador that arrived to the US in the eighties were driven by human rights violation in their country rather than by economic motivations, contrary to the Reagan administration claims. Our own research has used this technique to reveal the impact of General Cedras regime on the number of people killed by gunshot in Haiti (Doretti & Cano, in press) or to show that the awards accorded to policemen that participated in armed confrontation in Rio de Janeiro were actually increasing the number of victims and were promoting the excessive use of force, thus making these confrontations deadlier.

Fourth, many studies on this area are trying to calculate not just the intensity but also the cost of crime and violence for society at large and for particular communities. Also, cost-benefit and cost-utility analyses can be carried out to evaluate whether certain crime-preventing and violence-reducing strategies are really worth the resources they spend.


Desirable Features of Statistical Systems for Investigating Crime

James Lynch

American University

 

The papers and discussions that I have heard during this Workshop remind me very much of the discussions that took place among criminologists in the United States about a decade ago. At that time violence rates were increasing at an astounding rate and criminologists were being pressed to explain why. Then, as now, we engaged in informed speculation about causes, but when we reached for data that could inform our speculation further, they were not there. Earlier presenters have spoken of myths about crime and the exploding of myths. One of the myths that I would like to dispel is the myth that data systems pertaining to crime in the United States are that much more complete than those found in other places. While data systems in the United States may be more established than those in other nations, they are still inadequate in many ways that inhibit their use to inform the questions raised here. I would like to present some lessons about building data systems that have been learned through trial and error in the United States. Some of these may have relevance to on-going activities in Latin America.

Include Self-report and Official Records

It is essential that data systems pertaining to crime include both self-report surveys and administrative series using official records. First, as Claudio Beato noted, it is impossible to obtain the one true measure of the level and change in level of crime. It is much too subjective a phenomenon. Moreover, the statistical systems designed to capture or record these events have a distorting influence. The best we can do in this situation is to obtain multiple measures of crime that can be used in a complimentary fashion to illuminate the crime problem. If crime measures that have very different types of strengths and weaknesses tell a consistent story, then we can have greater confidence that the story is true without determining which of the measures is the "true " measure of crime.

The second reason to include data systems that include both self-report and official statistics is that victim surveys will introduce scientific principles into the collection of data on crime. Official record systems are largely appendages to service delivery systems and as such they are by-products of that system. There is generally an absence of skepticism about data in these systems and insufficient quality controls that would safeguard the data important for statistical purposes. Quality control may be diligently exercised for items relevant to service delivery, but not for statistical systems. For example, in the United States the Uniform Crime Reports which is the national system for reporting crimes known to the police relies upon the voluntary participation of police departments has virtually no audits to assure the quality of the data. In contrast, throughout its history the NCS has re-interviewed a random 10 percent of the persons interviewed to ensure that correct interviewing techniques are being followed. This kind of skepticism about data is routine in survey research because it has been the province of scientists who structured surveys according to scientific principles. Having such a data collection fosters a demand for similarly scientific data form official records. Without the example of a victim survey official record systems will remain the hand maiden of service delivery systems of the police, the courts and corrections. And, if they remain so then the public will continue to be skeptical of the data that come from these systems.

There are also reasons to include data from official records. Victim surveys cannot provide good information on crimes that are relatively rare. Obtaining reliable information on such types of crime would require sample sizes that would be prohibitively expensive. In most cases, police record systems would include a large number (albeit selective) of these rare events that could tell us a great deal about the crime. Moreover, having these data would lessen the chance that the survey would be required by political pressure to provide data on the crime even though the data would be unreliable. Similarly, official records data may be better able to produce sub-national estimates for counties or states or localities than victim surveys. Surveys would require very large samples to produce reliable estimates for sub-national units. Official record could assume this burden and free surveys from this obligation.

For these reasons data systems on crime should include both household-based victim surveys as well as official records of crimes.

 

Accommodate Both Research and Social Indicator Uses

Social indicators of crime report on levels and change in levels of crime. If the principal purpose of a data system is to produce social indicators of crime, then it must be organized in a way that permits large sample (to ensure reliability) and stability in procedures and instrumentation (to ensure that changes are real and not due to changes in data collection). Research on the other hand is designed to discover something new. Consequently, data systems designed to serve research needs will trade-off sample size for more information on a given case . Moreover, research benefits from a constantly changing information content rather than the stability that marks systems providing social indicators data. It is beneficial to accommodate both uses in a data system on crime. This is important because when resources are scarce the social indicators community can be pitted against the research community in a way that leaves us with bad research data and bad social indicators. Moreover, the passion for funding research can wax and wane with the whims of governments but social indicators (once they are institutionalize) will become a fixture--something that is awaited much like data on interest rates are in the economic arena. The institutionalization of a survey or an official record system means that this data collection vehicle will endure and provide continuous opportunity to collect data on crime. This vehicle can be used by researchers as well as those interested in social indicators.

The way to accommodate these two uses in one system is to organize the data collection into a core set of information items and a set of one-time and episodic supplements. The core items would afford the continuity and parsimony required for social indicator uses and the supplements would provide the addition information and flexible content that researchers require.

Collect Data to Allow for Multiple Units and Linkage

The data collected on crimes, victims and offenders should be collected with one of these units as the unit of analysis and the ability to link units. So, for example, information should be collected on crime events but it should be possible to link data on the event to data on victims involved in the event as well as offenders. Links should also available to the areas in which the events occurred. Collecting information on the lowest or most disaggregated unit with the potential for linking to other units can permit multi-level models. In this way crime events or court decisions can be studied in the context of the city block or the city in which it occurred. Moreover, the ability to link information on individuals across agencies or decision points in the justice system can increase the volume and variety of information on those individuals. If an individual's court records can be linked with those from the correctional system then we can expand the amount of information available on that individual. In building information systems more thought must be given to the ability to link records both horizontally and vertically (from smaller units to larger ones). This enthusiasm for linking must be tempered by an appreciation of the threats to confidentiality that it brings.

Employ an Attribute-based Logic in Collection and Analysis

Administrative series based upon police, court or correction records have the tendency to collect information that asks line staff such as police officers or guards to make complex coding decisions in which a great deal of information is collapsed into a much reduced set of codes. The police, for example, must describe a complex set of interactions between victims and offenders in a charge or offense classification. This requires the police officer to compare the attributes of the event to some classification scheme and then give the case the charge code that corresponds to the appropriate combination of attributes, e.g. assault with injury or weapon threat-aggravated assault. Whenever possible it is better to avoid this complex coding process in the field. It is better because it reduces the burden on the officer or other agency staff. Asking them to describe very specific aspects of the event is much more straightforward and should be a simpler task to perform. Instead of asking whether the event is an aggravated assault simply ask if the event involved a threat or injury, how severe the injury was and whether a gun was present. If the data are collected in this manner then analysts can combine attributes to form crime classes after the fact. Moreover, while pre-coded data can only be analyzed in one way, attribute data can be combined and recombined in many different ways. While the introduction of this attribute-based logic is common place in surveys and in research more generally, it is a very new concept in criminal justice agencies where crime codes and charges have real meaning to the line staff in their daily tasks. It is even surprisingly difficult in crime surveys where legal codes and classification have so constricted our thinking about crime and crime classification.

Include Longitudinal and Cross-sectional Data

Another unfortunate vestige of the legal dominance of the study of crime is the point -in time logic of most data collection systems. Crimes and their consequences are believed to occur at a point in time. In many cases, however, crimes actually part of a process or sequence of events that occur over time. A fight at time one begets another assault at time two, but it is impossible understand why the assault occurred unless one knows about the prior fight. Similarly the victim of a car theft reports the loss of the car in a victim survey three weeks after the event. In the fourth week the car is recovered and there is no loss. A robbery at time one becomes a homicide three months later when the victim dies of his wounds. There are countless examples of how ignoring longitudinal processes distorts our picture of crime and severely limits our ability to understand why crime occurs. It is important, therefore, to collect both longitudinal and cross-sectional data on crime and criminal justice issues. The ideal victimization survey, for example, might have a longitudinal component within a larger cross-sectional design. The longitudinal component would permit the monitoring of processes that change over time.

Make the Self-report Survey Free-Standing

In several discussions that I have had with participants outside of the formal presentations, mention has been made of adding victimization questions to an on-going survey top obtain victimization data. If there is any choice at all here, I would recommend a free-standing victimization survey wherein a sufficient number of screening questions could be added to the survey to ensure more complete reporting of criminal victimization. Ten years of research and development work on the National Crime Victimization Survey has clearly shown that exhaustive reporting of crimes by victims requires a great deal of cueing because not everyone defines crimes similarly and memories of crime are stored and retrieved differently by respondents. This means that time must be spent clarifying the definition of the crimes that we want respondents to report on and a wide range of cues must be given to prompt recall by even those who store crime events under the category "Family matter." This type of cueing is especially important for events that are not "stereotypic" crimes, like spouse abuse, and therefore are not well reported to the police. These are exactly the events that victim surveys must capture because police records never will. Enough time must be spent on cueing to ensure exhaustive reporting of these events and it is unlikely that adding questions to an on-going survey will provide the amount of cueing required.

These recommendations come from the mistakes that have been made in the building of statistical systems in the United States over the past twenty-five years. Many of these mistakes come not from a lack of foresight but from the entrenched nature of pre-existing statistical systems. These systems were built in an earlier age with much different technologies and they were organized in a manner consistent with those technologies. Their influence on our perceptions of crime and crime statistics was enormous and very difficult to overcome. These perceptions limited the introduction of organizing principles more consistent with modern data collection technology. Nations with less entrenched pre-existing statistical systems may be able to avoid these pitfalls.

FIGURE 1.

COMPARISON OF RAPE INCIDENCE ESTIMATES FROM

DIFFERENT SELF-REPORT SURVEYS

NATIONAL CRIME SURVEY
130,000
 
NATIONAL CRIME VICTIMIZATION SURVEY
310,00
 
NATIONAL WOMEN'S STUDY
683,000


Session VIII: Discussion On

"Innovative Methodologies And Statistical Analysis In

Working With Latin American Crime Data"

Rapporteur Julie Reid (Sociology)

 

A comment was made regarding the organizational differences in crime data and its collection and the possibility of evaluating or reviewing differences with regression analysis. This could allow the comparison of crime rates between various organizations. A need for correct data in crime mapping was also stated.

Roberto Briceño-León addressed the problem of low response rates, which particularly afflict inadequately funded projects. In his research, the non-response rate was reduced from 25% to 15% through modification of data collection techniques, which included using pay as a motivator, reporting to different neighborhoods at select times, and making strategic use of stereotypes, such as having women interviewers in certain areas where men tend to viewed as dangerous.

Problems related to how death is categorized is highlighted in Brazil, where doctors often refuse to label a death as a homicide and favor other categories, such as "other violence." Alba Zaluar's 1993 study found this a particularly serious problem in Rio, where that 30% were labeled as resulting from "other violence" (in São Paulo the figure was 5%). They concluded about half of Rio's violent deaths where cause was doubted were in fact homicide cases. This may explain why Rio's health data closely mirrors police data.

The need for standardization in data coding was emphasized along with the usefulness of dividing questions into more specific sub-questions before the interview rather than afterwards. The popularity and usefulness of data imputation was also discussed, and the need to know the amount of estimation in various data sources in order to avoid confusion in interpretation.

Standardization and improved data collection were noted as being particularly important at this critical time, when different agencies are working together using statistical banks and shared data. Good crossmatching and cooperation would help to avoid possible problems with the coordination of different agencies.

Ignacio Cano responded that his study does take estimated death rate figures into account, and although it is preferable to deal with these issues beforehand, there are methods for dealing with stability of rates afterwards. A problem is that there needs to be temporal correlation, which isn't always the case.

The issue of rates stability is related to the situation in Brazil where security being privatized both informally and formally (i.e., police are selling their booths). In a community where such sales have dominated, the police-community relationship is not an exceptional problem.

Although expensive, marketing tactics, such as mounting a television campaign and/or posting a picture of the interviewer in the residential area prior to the interviews, can effectively increase survey interview response rates. Cold contact interviews generally produce low response rates, but sending letters to potential interviewees before making personal contact, as done by the Census Bureau, can substantially increase response rates. It was noted that researchers are being forced to spend increasing amounts of money in order to obtain access to residential settings, a difficult situation in environments where resources are scarce.


Endnotes

1 Birkbeck, Christopher y Gary LaFree (1990-1991) "Hacia el desarrollo del análisis situacional en crimonología: Un estudio de tres tipos de delito en Estados Unidos y Venezuela". Capítulo Criminológico (Maracaibo, Venezuela) 18-19:49-82. And see also LaFree, Gary; Christopher Birkbeck (1991) "The neglected situation: A cross-national study of the situational characteristics of crime." Criminology 29(1):73-98.

2 Williams, Harvey (1977) Progressive Drug Involvement: Marihuana Use Careers among Nicaraguan Private Secondary School Students. Vanderbilt University, U.S.: Ph.D. Dissertation. Also, Birkbeck, Christopher (1995) "La conducta problemática juvenil según dos encuestas de auto-revelación realizadas en las ciudad de Mérida" Revista Cenipec (Mérida, Venezuela) 16:35-67

3 Birkbeck, Christopher; Luis Gabaldón (1998) "The effect of citizens’ status and behavior on Venezuelan police officers decisions to use force." Policing and Society 8:315-338.

4 O’Brien, Robert M. (1985) Crime and Victimization Data. Beverly Hills, CA: Sage.

5 Hidelang, M.J., et al. (1981) "Variations in sex-race-age-specific incidence rates of offending." American Sociological Review 46:461-474.


LANIC | Back to "Rising Violence and Criminal Justice...
Copyright © 1998 University of Texas at Austin