|
EUICS Consortium |
»home |
| MEMBERS OF CONSORTIUM | FIELDWORK | RESEARCH PAPERS | DATA | |
|
Downloads |
|
EU ICS Technical Description The EU International Crime Survey (EU ICS) is a programme to look a householders' experience with crime, policing, crime prevention and feelings of safety in a number of European countries. This survey asks a representative sample of the population about selected offences (car theft, motor theft, burglary, robbery, assaults, drugs etc.) they have experienced over a given time, applying the so-called ICVS methodology. We are interested in incidents are whether or not reported to the police, and indeed, the reasons why people do and do not choose to notify the police. They thus provide both a more realistic count of how many people are affected by crime and - if the surveys are repeated - a measure of trends in crime, unaffected by changes in victims' reporting behaviour or administrative changes in recording crime. By collecting social and demographic information on respondents questioned, ICS also allow analysis of how risks of crime vary for different groups within the populations, in terms of age, income levels etc. The ICS/ICVS was set up to serve three main aims:
To provide an alternative to police information on levels of crime Offences recorded by the police are problematic for comparing crime in different countries for three reasons. Firstly, victims report the vast majority of incidents the police know about. Thus, any inter-country variance in victims' tendency to report crime to police undermines comparisons of the amount of crime counted by the police in different counties. Secondly, there may well be differences in the amount of reported crime which is actually recorded by the police in different countries. Thirdly, official police statistics vary because of differences in legal definitions, recording practices and rules for classifying and counting offences. These limitations are well established. For the crimes it covers, the ICS asks about incidents that by and large accord with legal definitions of offences. It generally accepts respondents' accounts of what happened - or at least the accounts they are prepared to give to interviewers. Thus, it allows for a broader definition of crime than the police, who, if incidents are reported to them, are likely to filter out those which may not be estimated to merit the attention of the criminal justice system or meet the legal or organisational demands for reasonable evidence. To harness crime survey methodology for comparative purposes Despite efforts made in a number of countries over the past 20 years to develop 'crime' or 'victim' surveys to assess national or local crime problems, these reports only allow a limited comparative interpretation. The objective of the EU ICS is to provide a fully standardised questionnaire enabling a truly comparable analysis of data. And as it has always been the intention to repeat the ICS over time, it promises additional information in trends in crime in different countries. To extend information on who is most affected by crime By collecting social and demographic information on respondents, the ICS also aims to assess how crime risks vary for different groups. Variance in age, income levels and so forth are considered. It therefore offers a major advantage to police statistics, which usually only provide limited documentation of the characteristics of victims. Moreover, with its cross-national perspective, the ICS allows us to see how far the determinants and consequences of victimisation are the same in different jurisdictions, or whether country differences are evident. What does the ICS cover Also, the ICS covers a broader spectrum of crime statistics. The survey is similar to most crime surveys of householders with respect to the crime it covers. It is confined to counting crimes against clearly identifiable individuals, excluding children. (Crime surveys cannot easily cover organisational victims or victimless crimes such as drug abuse). Respondents are asked questions affecting the household at large, and are invited to report all incidents known to them. But they are also questioned about personal crimes, where they report only on what happened to them personally. Questions are open to record crime that occurred during a lap of several years: respondents are asked first about their experience of crime over the last five years. Those who mention an incident of any particular type are asked when it occurred: in the current year, in the last year, or before that. Those who reported incidents in the last year were asked how many times it had occurred. All those who said they had been victimised over the five-year period were asked a number of follow-up questions about what happened- for instance whether the police were notified. These questions were posed in relation to the 'last incident' if there had been more than one victimization of a particular type. A few other crime-related questions are also included and asked of all respondents. They cover, for instance, concern about crime, attitudes to the police, and what respondents would recommend as a sentence for a recidivist burglar. Mode of the survey All EU ICS interviews have been carried out with CATI telephone methodology. Telephone surveys have been recently widely implemented in the ICVS rounds especially in more industrialised countries with high telephone penetration rates. Interviews were carried out via fixed telephones. The average duration of the interview was 23,2 minutes. WebCATI solution In 12 countries we used a centralised multilingual WebCATI solution to collect interviews (exceptions are the UK, Ireland, and Spain). Traditional CATI requires that the PCs be linked through a network to a main PC (server) that functions as a central storage location for sample and data. WebCATI connects to the main server through the World Wide Web, and instead of terminal applications, interviewers use their browser to access to the study files and to enter the responses. WebCATI is a product that allows for Computer-Assisted Telephone Interviewing (CATI) through the Internet and/or an Intranet. Equipped with a browser interviewing interface, WebCATI offers a Windows point-and-click concept for interviewers. WebCATI uses centrally-located study files for multiple interviewing sites, minimising study management errors (don't have to send out updates, files, etc.) and simplifying process of job sharing/overflow phone hours. WebCATI provides all the same powerful features of Survent (quotas, logic, sample control and access, rotations, call-backs) as the normal CATI client. The program spec is written using a combination of Survent software and XML/HTML, though which a high-level integration of the multilingual questionnaire database, the CATI software, the sample, and the tabulation / data export component is possible. Coverage The current dataset covers 18 countries of the European Union, such as:
Besides, there was separate data collection for Northern Ireland and Scotland. Both regions are included in the United Kingdom data as well. The Northern Irish and Scottish separate studies are included under their own "country code" in the harmonised datafile. The subjects of the survey are residents of 16 years of age or older in the above countries. Sample The samples of the study were uniformly selected, along the same principles in each participating country. The samples used for the ICS were designed to provide the most complete coverage with the least bias. Therefore Random Digit Dialling (RDD) samples were used in most countries to carry out the interviews. It means that telephone numbers were not selected from a list, but were generated randomly (within so-called "working banks" - stacks of 100 telephone numbers within there is at least one positive feedback for an operating line). The RDD samples for most countries were provided by Survey Sampling International Inc. In order to avoid sampling bias, we used replicates, with a size of 100 phone numbers. New replicates were opened only if available replicates were exhausted to >90%. The samples were provided by the national field agencies in Greece, Sweden, Finland, Denmark, Hungary (in these countries sampling was based on white pages directory), Ireland, Netherlands, and the United Kingdom. The samples were proportionally allocated according to the NUTS 2 (or equivalent) regions in each participating country. There were no further clusters within the regional strata: telephone numbers were selected randomly. The eligible respondent was one with the most recent birthday among the household members at least 16 years of age, and resident of the country / capital city, for the two subsamples respectively. Technical information of Poland EU ICS Sample size The targeted number of interviews in most countries was 2000. In Luxembourg the target sample size was 800, while the. The samples in each country but Estonia, Luxembourg and Poland were divided into a larger national part (with a targeted size of 1200) and a relatively smaller capital city part (targeted N = 800). The table below illustrates the actual sample sizes in each country for both subsamples and overall: Achieved sample sizes by country
Response rate Several actions were taken to increase cooperation throughout the survey. Besides using a highly experienced field force who were specifically trained to tackle respondent reluctance even at the most sensitive issues of the survey, we applied a so-called 7+7 call design over an extended period of time. We called on each telephone number at least seven times to establish initial contact (i.e. if line was always busy, or wasn't answered) and we performed seven repeated calls to establish contact with the eligible respondent within the household. The field period has been extended to allow more flexible scheduling to reach those people as well who are only rarely at home. Achieved response rates ranged from 36,9% in Luxembourg to 56,9% in Finland, averaging 46,3% overall in the 15 countries. Achieved response rates by country
Weighting A number weighting variables are computed to compensate for over- and undersampling, deliberate or otherwise, of particular groups within the population. In the ICS, we had a sample of households and a sample of individuals from each household. Weights were developed by making the marginal distributions of the auxiliary variables in the sample conform to the population marginals and not the full joint distribution. We used raking to solve our weighting tasks. This procedure performs iterative proportional fitting in contingency table analysis. We did not trim our weights as we had artificially distorted capital city ratios that had to be weighted back to the population targets - which involves some smaller and larger weights as well. Subsample weights Subsample weights are calculated separately for the national and capital city subsamples, but they are included in the same weight variable for both subsamples. Subsamples are defined by the segm_new variable: 1-national and 2-capital city samples. Please note that such weight does not exist for Luxemburg, Estonia, and Poland because of the different sample design. The household sample is "self-weighting" within the national and capital city subsamples in a sense that the selection remained totally random and proportionate to regional location. We have, however, controlled their selection probability by the number of telephone lines they used, and attributed an inverse weight to households with more than one telephone line. We have developed a household level weight to be used for estimating household level attributes, and household level crimes (e.g. burglary), called ss_hhwght. At the individual level we have controlled selection (and victimisation) probability with weighting according to household size as well3. As nonresponse rates vary by social segments and the household filter introduces selection bias towards the loners (and frequently the elderly) the sample characteristics will reflect such differences as well (i.e., there are usually less males and less young people in the samples than in the universe.) Therefore we applied a post stratification at the individual level, to control for the main demographic characteristics of the nations involved: age and gender4. In addition, within the national subsample we continued to control for the NUTS Level 2 regional distribution in the post-stratification phase (at the sampling stage there already was a built-in control for geographically proportional selection of respondents). The variable developed for individual-level weight is ss_indwght which has to be used when subsample-level estimations of individual attributes are calculated. National weights National weights were computed to combine estimation from the capital city and national subsamples. Weights were developed similarly to the subsample weights, but we imputed the capital city ratio as a post-stratification factor in the raking process. This weight exists for each country. The household sample is "self-weighting" within the national and capital city subsamples, as we described above. We have controlled their selection probability by the number of telephone lines they used, and attributed an inverse weight to households with more than one telephone line. Also, where it was applicable, we weighted the households according to the disproportionate selection within the overall N=2000 target sample - assigning less weight for households from the capital city and more weight to those elsewhere in the country. We have developed a household weight to be used for estimating household level attributes in the total sample, and household level crimes (e.g. burglary), called hhwght. At the individual level we have controlled selection (and victimisation) probability with weighting according to household size as well. Therefore we applied a post stratification at the individual level, to control for the main demographic characteristics of the nations involved: age and gender, and of course the artificially distorted capital city / non-capital city ratio5 (where it was applicable). In addition, we continued to control for the NUTS Level 2 regional distribution in the post-stratification phase. The variable developed for individual-level weight is indwght which has to be used when individual-level estimations are calculated for the total sample. Capital city weights Capital city weights were computed for each case from the capital cities, i.e. the capital city subsample and the capital city cases from the nationa subsample, combined. The 'cap' variable is to be used to sort or select the capital city cases within the total sample (1: capital city, 2: rest of the country) Individual and household level weights were developed similarly to the above described methods. The household sample is "self-weighting". We have, however, controlled their selection probability by the number of telephone lines they used, and attributed an inverse weight to households with more than one telephone line. We have developed a household level weight to be used for estimating household level attributes, and household level crimes (e.g. burglary), called cap_hhwght. At the individual level we have controlled selection (and victimisation) probability with weighting according to household size as well as age and gender6. In addition, within the national subsample we continued to control for the NUTS Level 2 regional distribution in the post-stratification phase (at the sampling stage there already was a built-in control for geographically proportional selection of respondents). Cross-national individual weights We added a weight variable that projects the indwght to the relative size of the country within the total geographical area covered. This weight is to be used for estimations based on more than one country (i.e. joiunt Benelux estimations, or EU-18 estimations). The separate Northern Irish and the Scottish datafiles are not part of this weighting system. The variable is called eu_indwght. This is an individual level weight. The countries' proportions are based on the number of 16+ population, with the following data (latest available from Eurostat / local statistical institutes):
Region size We added a variable that has the population of each NUTS 2 level region included in the sample. This variable can be used as source for estimations created for groups of regions, it gives the total number of 16+ people in the NUTS 2 region of the respondent. This is not a weight variable in its current form, but it can be used for developing inter-regional weights during the analyses. The variable is called: 'nuts2pop'
(1) Copenhagen metro area |
Copyright © 2005-2007 EUICS. All rights reserved. |
Help | Sitemap
| |