«Edited by ANNE MASON Research Fellow, Centre for Health Economics University of York and ADRIAN TOWSE Director, Ofﬁce of Health Economics Radcliffe ...»
As with criminal victimisation, there is very little longitudinal evidence in relation to the fear of crime, so it is difﬁcult to determine its relationship (if any) with actual victimisation and its cumulative effects. What seems clear from an accumulation of cross-sectional evidence, though, is that victimisation does not, in itself, explain fear (Semmens, 2004). Indeed, victimisation may prompt reactions to crime that actually mediate fear (Winkel, 1998). Perhaps the most useful study for estimating the overall emotional losses from the fear of crime was carried out by Farrall and Gadd (2004), in which a general population sample were asked ‘In the past year have you ever felt fearful about becoming a victim of crime?’. Those who respond positively were asked ‘How frequently have you felt like this in the past year’, and also asked ‘On the last occasion, how fearful did you feel?’ (with response categories of ‘not very fearful’, ‘a little bit’, ‘quite fearful’, and ‘very fearful’). However, the time period speciﬁed will inﬂuence how respondents interpret the questions asked of them. For example, Winkielman et al. (1998) asked respondents about how much anger they had experienced over time frames of either one week or one year. People reported large amounts of anger for the week compared to what would be expected for a year if we extrapolated from the weekly reports to a full year.
However, the situation is not entirely hopeless. In their excellent review, Shapland and Hall (2004) provide a list of the effects commonly found in the criminological literature and this list, with some slight modiﬁcations, is given
1 initial shock and guilt at having become the victim of crime 2 physical injury, including permanent incapacity in a small number of cases 3 psychological effects, including fear, anger and depression, which, for some, may turn into longer-term depressive effects including sleeplessness and anxiety and, occasionally, into post-traumatic stress disorder (PTSD) 4 behavioural effects, involving changes to the victim’s lifestyle, normally to avoid the situation or context in which the offence occurred 5 a loss of autonomy and increased vulnerability 6 a loss of trust/faith in society, particularly in the local community or in relation to the social group or place where the offence occurred.
138 THE IDEAS AND INFLUENCE OF ALAN WILLIAMSThe effects of the fear of crime would appear to be similar to the effects of criminal victimisation itself. For example, fearful people may modify their behaviour, e.g. by not engaging in pleasurable activities as much as they would like (walking the dog, going to the pub etc) or by undertaking additional security measures, such as locking windows and doors, which have an opportunity cost in terms of time. Or they may change their views about society, or feel vulnerable. The psychological losses will be similar too, ranging from affective reactions to more cognitive assessments of the likelihood of victimisation (Dolan and Peasgood, 2006).
Against this background, we have developed the preliminary descriptive system shown in the Appendix to this chapter. We have recently begun asking members of the general public about the face validity of this measure, and early results are encouraging (although there have been some questions about whether ‘no constraints on behaviour’ is really such a good thing). This descriptive system generates 45 = 1024 possible outcomes. A six-dimension system could also be explored, where one of the dimensions could be monetary; this would allow for the direct estimation of monetary values, as discussed in the next section.
VALUING THOSE LOSSESMonetary valuations An ideal stated-preference study would give information about the current baseline risks of experiencing particular outcomes and encourage respondents to consider their WTP to achieve reductions in the risks of those outcomes (Loomes, 2004). This raises challenging problems for stated-preference work since it is well known that many respondents are insufﬁciently sensitive to information about small changes in small risks, and it may well prove impossible to ﬁnd any satisfactory way of communicating the risk information.
As an alternative, ‘certainty’ scenarios might be used and valued using willingness to accept (WTA) rather than the more traditionally used WTP. The principle underlying WTA is that if an individual contemplates some adverse event, the loss of welfare could be offset by the welfare generated by some increase in wealth. Clearly, this does not necessarily apply to more severe levels of harm, such as death, where no amount of money may adequately compensate for the loss (and where respondents may be offended that such a comparison is sought at all). However, for many crime outcomes, it may be possible to argue that there are ﬁnite sums that would offset the experience (or at least, to explore how far that may be the case).
Rather than elicit valuations directly, a contingent ranking exercise offers an indirect way of eliciting monetary values. The contingent ranking approach asks respondents to compare alternatives, each involving various differences along the kind of dimensions outlined in the appendix. The general idea is that the relative weights assigned to different levels along the different dimensions
THE MEASUREMENT AND VALUATION OF PUBLIC SAFETY 139
In the event that we are not conﬁdent that respondents are sufﬁciently sensitive to risk, the relativities obtained from contingent rankings may allow us to ‘chain’ the values for more serious outcomes (which we cannot plausibly obtain from WTP questions about preventing the certainty of those outcomes) to values for less serious outcomes that may be amenable to WTP (and WTA) questions using certainties rather than risks. However, the statistical procedures for inferring the various trade-offs often make somewhat restrictive assumptions about the underlying structure of people’s preferences, and a large number of choices might be required to obtain the minimum amount of information to enable the estimation procedure to be implemented.
In a recent WTP study in crime, Cohen et al. (2004) found an implied WTP per avoided crime well in excess of previous estimates based on victim costs, and suggest that the difference may be in part due to their study incorporating costs that have been overlooked in previous studies, including the anxieties people have about anticipated crime. The study ﬁnds an implied WTP for avoiding one burglary of $25 000, avoiding one rape or sexual assault of $237 000 and avoiding one murder of $9 700 000. However, these differences may well be driven to a signiﬁcant extent by the different frequencies of these crimes, since the mean willingness to pay for a 10% reduction in each of these crimes was very similar; $104, $127, and $146, respectively. The results do give weight to the notion that the fear of crime substantially reduces wellbeing, but the dominance of probabilities on the ﬁnal WTP ﬁgures may be thought to raise serious doubts about the reliability of the results.
More generally, there has been much discussion in the literature about potential sources of bias in WTP/WTA studies and we will not rehearse them in any detail here (see Carson et al., 2001). Some of the more pervasive problems, and ones that will be relevant to the context of crime, are: warm-glow effects, which occur when the respondent feels some moral obligation towards the scenario being valued (Diamond and Hausman, 1993); embedding effects, which occur when the respondent reacts to a good’s general symbolic meaning instead of to the speciﬁc levels of provision described (Mitchell and Carson, 1989); and scope effects, which occur when responses are insensitive to the quantity of the good being valued. Some of these biases have been reduced by improved research design (Arrow et al., 1993) but scope effects, for example, have proved to be particularly pervasive. Moreover, even if studies could perfectly replicate a market, there are questions about the degree to which the market itself is free of the some of these biases.
140 THE IDEAS AND INFLUENCE OF ALAN WILLIAMSNon-monetary valuations The aim here is to elicit the values of the outcome descriptions relative to one another. A subsequent task is then to ‘peg’ these to some monetary value(s).
The valuations are relative to two anchor points. In the health-state valuation literature, death and full health are typically used as the anchor points, and given the value of zero and one, respectively. The upper anchor in the context of crime valuation should be the combination of no problems across all dimensions, with no mention of crime. However, it is not clear whether the lower anchor should be death caused by crime, or death with no cause speciﬁed. The latter will match better with the upper anchor, but if death by crime is perceived as worse than death by, say illness, or by accident, then this should be taken into account.
Direct valuations for crime outcomes could be elicited using the standard gamble (SG) method or the time trade-off (TTO) method (Dolan, 2000).
The basic SG asks the respondent to choose between the certainty of an intermediate health state and an alternative treatment with two probabilistic outcomes, one of which is better than the certain outcome and one of which is worse. The probabilities are changed until the respondent is indifferent between the two scenarios. The basic TTO asks the respondent to choose between two alternatives, one is to live for a deﬁned period of time in poor health and then die, and the second is to live for a shorter period of time in full health and then die. The length of time in full health is changed until the respondent is indifferent between the two scenarios.
There are then ways to express the duration of a given outcome. One alternative is to make the duration of the outcome the same as that of the entire scenario, e.g. ‘you will experience outcome i for t years, and then die’.
Since many crime outcomes will last for much shorter durations than the life expectancy of a typical respondent, this may be difﬁcult for respondents to take at face value and engage in. Therefore, the second alternative is to make the duration of the outcome shorter than the overall duration of the scenario, e.g. ‘you will experience outcome i for t years (or m months), and then survive to your naturally expected age of death in full health’.
However, there are two issues associated with this. Firstly, suppose the duration of the outcome is set very short relative to the time horizon, such as when an outcome will last for one month during the next 10 years. If a respondent to a TTO question is willing to trade off more than one month at the end of the 10 years, then (in the absence of discounting) the outcome in question will have a negative value, implying that it is worse than being dead.
Likewise, if a respondent to a SG question is willing to take more than a 0.008 risk of death, then the outcome will also be valued as if it were worse than being dead. Secondly, the valuations will not be anchored at 0 for dead, and there needs to be an additional exercise to provide this link.
A further issue arises when respondents are faced with outcomes that involve relatively mild losses, which they acknowledge are worse than full
THE MEASUREMENT AND VALUATION OF PUBLIC SAFETY 141health but for which they are also unwilling to trade off any survival or take any risk of death. One way to address this issue is to value these mild states using a more severe outcome that is preferred to death as the bottom anchor in the TTO and SG elicitations (rather than using death, as in the basic formats described above). If this intermediate outcome is then valued against full health and death, the results of this elicitation can be used to ‘chain’ the results obtained for all the other outcomes to the standard 0 to 1 (death to full health) scale. This chaining method results in values for mild outcomes that are less than the value for full health. The problem with chaining is that it may introduce an additional source of noise and error. An alternative, which is to set an intermediate additional anchor (rather than replacing the bottom anchor), may be preferable in the case of crime, but again will need to be investigated in terms of whether it means the same to all respondents.
It is possible to elicit non-monetary valuations indirectly by asking respondents to rank a number of different outcomes (McCabe et al., 2006), or by asking them to make a series of choices between different pairs of outcomes, as in discrete choice experiments. The data from the ranking or discrete choice exercise are then analysed in ways that allow cardinal values at the group level to be generated from ordinal data at the individual level. Where the descriptive system does not include money, the results can be anchored to full health and being dead by including dead as one of the states in the discrete choice or ranking exercise. Where the system does include the monetary dimension, the results could be reported in terms of monetary value per one-level change in each dimension.
DISCUSSION In order to determine how best to allocate resources that seek to reduce the impact of crime on society, we need to describe the important consequences of crime and the fear of crime in ways that facilitate the practical collection of epidemiological data and the elicitation of robust valuation data. As well as the description and valuation issues identiﬁed in this paper, there are a number of other methodological and practical challenges we face in trying to develop a measure of public safety. The effects of crime and the fear associated with it can be felt not only by individual victims, but also by their families, friends and colleagues. There may well be similar externalities associated with ill-health, but they would seem to be more pronounced in the context of crime, particularly in relation to murder and sexual assaults. This suggests that samples need to be drawn from a wide range of sources and not just from victims of crime.