Directorate F: Social and Information Society Statistics
SASU manual
I.INTRODUCTION
1.Purpose of thedocument
II.SURVEY GUIDELINES
1.Ethics
2.Age limits
3.Interview training
4.Translation
5.Legal act/s
6.Precision requirements and sample size
7.Fieldwork
III.INTERVIEW GUIDELINES
1.Selection of respondents
2.List of variables (including therationale, definition and guidance for theinterview)
3.Priorities within victimisations
4.SASU model questionnaire
IV.CODING INFORMATION
1.Code book
2.Classifications used
3.Checking rules (applied to thedataset delivered to Eurostat)
3.1datalevel checking rules
3.2record level checking rules
3.3file level checking rules
3.4update history
V.ANNEXES:
1.Results of some cognitive tests and pilots
2.Content of quality report
3.Pre-defined tables
VI.BIBLIOGRAPHY
1
I.INTRODUCTION
1.Purpose of thedocument
This manual is intended to provide areference document for all issues relating to theimplementation of theEU-Safety Survey (SASU) in theMember States of theEuropean Union. It draws heavily on theUnited Nations publication Manual on Victimization Surveys (Geneva, 2010).
II.SURVEY GUIDELINES
1.Ethics
Theimplementation of SASU has anumber of ethical implications, some of which are indicated below.
Dataconfidentiality
National authorities should establish and adhere to procedures ensuring tight security procedures for theelectronic storage of survey data. This is especially important as regards access to, and sharing of files in which there are micro datathat can be traced to individuals.
Interviewer methods
Standard procedures should exist to ensure ethical conduct on thepart of interviewers in relation to theconduct of fieldwork, disclosure of information, etc. Two issues are of particular importance for theSASU:
- Interviewers should not disclose information pertaining to respondents, and should be made aware of theconsequences of doing so.
- Interviewers should adhere closely to theinstructions they are given as to assuring respondents that their answers will be anonymous and treated confidentially.
Interviewers may have to deal with some difficult interviews, which might be stressful for them. Stressful interviews will not necessarily be confined to those in which sexual and other violent incidents are described by respondents; serious burglaries can also cause trauma. Thenational authorities should take responsibility for setting procedures in place to support interviewers. For instance:
- There could be formal ‘debriefing’ session with interviewers after they have completed, say, 100 interviews.
- Management support should be available to deal with especially difficult ‘one-off’ interviews, with interviewers being clear as how to access this support.
Attitude towards respondents
Interviewers (and those in charge of them) have aresponsibility towards respondents. Some of theprocedures which are particularly important for theSASU are that:
- Respondents should not feel pressurised into taking part in thesurvey, but feel that have given ‘informed consent’. In practice, this can be difficult given thethin dividing line between interviewer behaviour which is needed to maintain response rates, and accepting refusals.
- Respondents should be treated respectfully (however ‘awkward’ they might be).
- To help respondents access further support and advice, interviewers should be given at least one contact telephone number to provide to respondents who seem to require some support. Countries will need to choose from themost appropriate agencies.
It might be considered whether respondents, at theend of theinterview, should be offered theopportunity of receiving asummary of results from theSASU. This is away of acknowledging their contribution, and thanking them for it. It may also improve aparticipant’s future response to requests to be surveyed.
2.Age limits
Thesurvey covers persons aged 16 and over. There is no upper age limit. In fact, thevictimisation of theelderly seems important to consider – even if only to show that risks are generally low.
Thesample could be taken either from anational registry of persons, or from arandom sample of households from which one member aged 16 or more is randomly selected. Themethod used for selecting an individual from aselected household is left to thenational authorities according to their usual practice.
3.Interview training
Professionally trained and experienced interviewers should be used. They also need to be specifically trained about thenature of thesurvey.
All elements of standard training should be maintained as regards conducting interviews efficiently, accurately, and with due regard to therespondent. But elements of training will need to focus on theSASU specifically – particularly with regard to questions on sexual and other violence victimisation and theconditions under which questions are asked about this.
Active training for theSASU might also be useful including role-playings, simulations, and group discussions.
Interviewers need to be able to access support in theevent of stressful interviews. Adebriefing exercise would be useful after aset number of interviews have been completed.
Respondents must not feel overly pressurised into agreeing to an interview, should be treated respectfully and have every confidence that theinformation they give will be anonymous and confidential. Procedures should be in place so that respondents can be referred onto asupport agency if this seems appropriate.
To help respondents to access further support and advice, interviewers should be given at least one contact telephone number to give to respondents who may require some support.
Much of what should be routine initial and refresher training will apply to theSASU. At aminimum, this should include procedures with regard, for instance, to:
- Encouraging respondents to take part, without being overly forceful. This is theprocess of ensuring ‘informed consent’.
- Being able – in CATI and CAPI interviews – to ‘manipulate’ keyboards accurately, without causing delay in questionnaire administration.
- Following questionnaire instructions accurately and quickly, without making up their own rules.
- Being able to query beforehand any interviewer instructions about which interviewers are unsure.
- Steering respondents through thequestionnaire in apatient way (especially when respondents challenge ‘why is that question being asked’).
- Thanking respondents – in agenuine tone – for taking part.
In relation to theSASU, elements of training that will need to be focussed on include:
- How to select eligible respondents from within acorrectly understood household unit.
- Allaying fears about theconfidentiality of responses.
- Appreciating theneed for questions about sexual incident and assaults / threats to be answered in a‘private’ conversation with theinterviewer, thus ensuring that therespondent is in aposition to answer questions without hindrance or heightened risk. This will ensure more honest answers. It will also forestall possible domestic trouble. In face-to-face surveys, it is easier, to protect, therespondent’s privacy by ensuring that theinterview is conducted away from other household members. In telephone interviews, it is much more difficult to gauge the‘real life’ conditions under which arespondent is answering questions. This should be fully addressed in training. Theneed for rescheduling an interview if conditions are not right should be stressed, applying to both face-to-face and telephone interviews.
As theSASU is centrally concerned with victimisation experience, it will also be important for interviewers to need to know about theusual contours of this. Some key training items would be that:
- Most people will engage well with thesubject of crime and victimisation.
- TheSASU is not meant to cover every crime arespondent might have experienced.
- Victimisation is not randomly distributed: some respondents will be victimised several times. This will make for some long interviews, whereas most will be fairly short.
- It is vitally important to make sure that events are located accurately in time
- Thescreeners questions are designed specifically simply to elicit ‘yes’ or ‘no’ answers, with other parts of thequestionnaire used for collecting details about what happened.
- Respondents are to report only on their own experience of personal crime.
- Very serious events are unlikely to be reported in interview often, although they could arise.
- Less serious victimisations can soon be forgotten, but are relevant to thesurvey and need prompting for using theexact question wording.
However, active training for theSASU might be useful including role-playings, simulations, and group discussions. It would also be advisable to arrange group discussions after thefirst days of fieldwork to exchange experiences.
4.Translation
Thenational authorities are responsible for producing anational questionnaire which makes it possible to provide theinformation specified in theRegulation. In doing so, they should take into account theinformation supplied in this manual, especially theList of Variables and, if wished, theModel Questionnaire.
Much of thedevelopment work on thesurvey was originally done in English, and specific problems may arise in using certain terms where usage differs between thenational language and English. As examples thefollowing may be cited:
- Theword for ‘stranger’ may in some languages be confused with ‘immigrant’).
- Terms used for different types of vehicles may differ so that it could be difficult to correspond exactly with theEnglish
- There may be problems in distinguishing between crime types such as ‘burglary’, ‘robbery’ and ‘theft’ in such away as to ensure equivalence between thelanguage versions
- Theword for ‘bribery’ is in some languages too serious for thetype of low level bribery which thequestion is meant to capture. More appropriate terms to use could be ‘backhander’, ‘un petit cadeau’, ‘pot-de-vin’, ‘smeergeld’ (Dutch), or ‘illegal commission’ (Southern Europe).
It is clearly not possible to make provision for all such questions of interpretation which may occur in different languages. Eurostat would be willing to be involved in discussing such issues and if necessary co-ordinating information between countries which share similar linguistic problems.
5.Legal act/s
References to theCouncil/Parliament Regulation and any others
6.Precision requirements and sample size
The types of crime covered by the SASU are: theft of a car, theft from a car, theft of a motorcycle, theft of a bicycle, burglary, robbery, theft of personal property, consumer fraud, card / on-line banking abuse, bribery (backhanders), non-partner physical violence, non-partner sexual violence, partner physical violence, partner sexual violence.
One year prevalence rates on crime on individuals were chosen as reference for establishing quality requirements.
In order to assess the values of the indicators the following ICVS indicators results were used:theft of a car, theft from a car, car vandalism, theft of a motorcycle, theft of a bicycle, burglary, attempted burglary, robbery, theft of personal property, sexual offences against women, sexual offences against men, assaults and threats.
The confidence interval becomes bigger with increase of the estimated proportion (having its maximal length for the P=50). According to the ICVS results the prevalence rate per type of crime does not exceed 10 thus the maximum value of it may be considered as the reference for establishing minimum precision requirements criteria. However, as the value of 10 was not achieved in many countries, it may not be of a high relevance there. The way of selecting another relevant prevalence rate for defining confidence interval limits would be to check the most frequently appearing maximal values.
The table below shows results of basic analysis of the maximum prevalence rates collected by the ICVS. The maximum was selected from the prevalence rates per single crime (rounded to integral number) in a country in particular year.
Maximum prevalence rate / 3 / 4 / 5 / 6 / 7 / 8 / 9 / 10Frequency of the maximum prevalence rate / 1 / 1 / 6 / 10 / 15 / 10 / 8 / 7
Cumulative freq / 1 / 2 / 8 / 18 / 33 / 43 / 51 / 58
Cumulative perc / 1.72 / 3.45 / 13.79 / 31.03 / 56.90 / 74.14 / 87.93 / 100
Assuming the ICVS results could be extrapolated to all EU countries plus Norway, Switzerland and Iceland, the prevalence rates smaller or equal to the given maximum could be distributed by countries as follows:
Maximum prevalence rate / 3 / 4 / 5 / 6 / 7 / 8 / 9 / 10Number of countries with the prevalence less or equal to the maximum / 0.52 / 1.03 / 4.14 / 9.31 / 17.07 / 22.24 / 26.38 / 30.00
Given the assumptions taken, the table above may suggest that in 17 countries the maximum prevalence rate will not be higher than 7. Thus 7 can be chosen as an alternative prevalence rate for establishing precision requirements.
Indeed types of crime in the ICVS and the SASU differ a bit but the ICVS indicators could approximate the possible order of magnitude of the SASU crimes prevalence rates.
In the ICVS maximum prevalence rate was 10 and it seemed that in majority of the countries the maximal value of prevalence was not higher than 7 (values taken for all available years).
Half length of the confidence interval was used as a reference measure of the quality at the 95% confidence level for given values of P.
8000 / 7000 / 6000 / 5000 / 3000P=5 / 0.478 / 0.510 / 0.551 / 0.603 / 0.777
P=7 / 0.559 / 0.597 / 0.645 / 0.706 / 0.910
P=10 / 0.657 / 0.703 / 0.759 / 0.830 / 1.069
(See: item 4. precision requirements and costs estimation v2.doc, on the following link:
The calculations were done with the assumption of simple random sampling. So in case of other sample schems the sample size has to be adjusted to be equivalent in terms of the quality of indicators to the proposed one.
Taking into account the available budget and burden on smaller countries (roughly appliyng also compromise allocation of the EU sample) as well as comments received during various consultations the different sample sizes of indyviduals depending from the population were defined as in the proposed Regulation text.
Thedatashall be based on nationally representative probability samples. Thesample size to be achieved, calculated on theassumption of simple random sampling shall be aminimum of:
(a)8000 persons in Member States with apopulation aged 16 and over which is higher than 10 million;
(b)7000 persons in Member States with apopulation aged 16 and over which is higher than 5 million and lower than 10 million;
(c)6000 persons in Member States with apopulation aged 16 and over which is higher than 1.5 million and lower than 5 million;
(d)5000 persons in Member States with apopulation aged 16 and over which is higher than 0.5 million and lower than 1.5 million;
(e)3000 persons in Member States with apopulation aged 16 and over which is lower than 0.5 million.
The sample size requirement could be transformed for example into the text:
Taking into account the above Eurostat proposes to require the 0.6 as a maximum half length of confidence interval of the indicator on prevalence rate by type of crime for the population aged 16+ if this indicator is lower than 7% and 0.7 as maximum half length of confidence interval for prevalence rate lower than 10%.
The half length of confidence intervals describes above may be inflated by 15% in countries with the population aged 16+ higher than 1.50 million and lower than 5 million and by 25% for the countries the population aged 16+ lower than 1.5 million. ….
It has to be noted that five types of crime investigated in the survey (bicycle, car, motorcycle theft, from a car theft and burglary) are more relevant for the whole household and the variables concern not only the respondent him/herself but also other households' members: e.g. "In the last five years have you or anyone else in your household had a car, van or pick-up truck stolen or driven away without permission?"). This way, the indicators on prevalence rates for these types of crime are more relevant if the calculation unit is a household.
To simplify, in case one person per household is interviewed number of household interviewed equals to number of individuals (assuming the non response for individuals and household questions will be the same). This way a sample size for individual and household crimes is the same.
However, given the financial capacities of the project and knowing that some countries will use the national victimisation surveys, Eurostat did not propose legally binding obligation to the Member states of interviewing only one person per household but does recommend it in the survey guidelines. Countries will decide on this issue based on the aspects of data quality, interview mode, financial capacities, sampling scheme options available but also taking into account the confidentiality of the interview and respondent comfort in answering certain questions (such as household violence).
Interviewing more than one person in a household will have an obvious consequence on the quality of indicators calculated for the households (smaller effective sample size) but also on the indicators on individuals (additional household cluster effect).
Applying an average size of the household in the MS, one can assume that effective sample of individuals will correspond to sample size of households as follows:
sample size of individuals / sample size of households8000 / 3333
7000 / 2917
6000 / 2500
5000 / 2083
3000 / 1250
That could still deliver results of relatively acceptable quality; however the quality requirements established by the regulation will not be met (by obvious reason of not being equivalent to the sample size on individuals). On the other hand, increase of the sample size required for maintaining the confidence intervals of the magnitude foreseen by the Regulation for the indicators calculated for household crimes (or imposing such a rule of inreviewing one person per household) might lead to considerables increase of the costs.
An ex-ante assessment of the lost in the quality of those indicators is required in order to guide the country decision on interviewing one or more persons in household. The work is planed to be done together with the TF.
7.Fieldwork
III.INTERVIEW GUIDELINES
1.Selection of respondents
TheSASU aims for anationally representative sample of those living in private households.[1] These are best drawn from aregister of population or households – or in thecase of CATI, from some method of contacting arandom selection of households or individuals viatelephone numbers. By and large, interviewing people who have volunteered to take part in surveys – so-called ‘intent selection’ interviewing – risks possible bias.
Victimisation surveys typically collect information on experiences of two types of crime: (i) crimes affecting thehousehold as awhole (burglaries, vandalism to thehouse, theft of family-owned vehicles); and (ii) crimes affecting individuals (e.g., ‘contact crimes’ and thefts of personal property).[2] This raises issues about how many respondents are interviewed, and what is an appropriate age range.
One respondent or two?
There is an argument for separate questionnaire to deal with household and personal crimes. For thehousehold crimes, therespondent might be thebest qualified household member (one of theadults); for personal crimes, themost appropriate choice is arandomly selected household member. Germany used this approach, and continues to propose it, although there were some problems.[3] Although not participating in thepilot studies, France also suggested in awritten comment that questions on household crime could best be put to ‘thebest qualified’ person.