One solution to this is to collect three different kinds of statistics.
One set is reported (to the police) crime. This is fairly concrete but is affected by people, for example, deciding not to report petty theft because they don't believe it will be investigated.
The second set is a survey of personal victimhood. The British Crime Survey, for example, gets a representative sample of the population and asks questions like "Did you get burgled last year?". This is a smaller (so less sensitive) dataset than the first and misses crime without an individual victim, but has far fewer misincentive problems.
The third is surveys perceptions of crime, either personal or geographical. This is a much weaker measure for actual crime, but captures how people are feeling, and is useful for understanding whether, for example, people are afraid of certain places because of perceived criminality.
The first two correlate reasonably well, though not perfectly. My understanding is that the third is much more weakly associated, has a fairly lengthy time lag between crime changing and perceptions changing, and is affected by changes in political and news attention. Plus, in general, humans are not good at guessing about things they don't have experience of. You can do a similar exercise looking at people's perception of average incomes and they're way out.
One set is reported (to the police) crime. This is fairly concrete but is affected by people, for example, deciding not to report petty theft because they don't believe it will be investigated.
The second set is a survey of personal victimhood. The British Crime Survey, for example, gets a representative sample of the population and asks questions like "Did you get burgled last year?". This is a smaller (so less sensitive) dataset than the first and misses crime without an individual victim, but has far fewer misincentive problems.
The third is surveys perceptions of crime, either personal or geographical. This is a much weaker measure for actual crime, but captures how people are feeling, and is useful for understanding whether, for example, people are afraid of certain places because of perceived criminality.
The first two correlate reasonably well, though not perfectly. My understanding is that the third is much more weakly associated, has a fairly lengthy time lag between crime changing and perceptions changing, and is affected by changes in political and news attention. Plus, in general, humans are not good at guessing about things they don't have experience of. You can do a similar exercise looking at people's perception of average incomes and they're way out.