Statistics on crime were one of the topics that divided opinion when we talked about the rest of the unit in the first week: some people definitely wanted to learn about them, others definitely didn't. I hope this week's teaching included enough information about stats for the first group and didn't scare the second group off.
I kept the Highly Technical Statistical Concepts to a minimum. In fact there were only two, which I'll quickly go over again now.
HTSC 1: Administrative vs Survey-Based Statistics
Some statistics are administrative: they're produced as a kind of by-product of the administration of an institution of some sort. Every time a pupil starts at a new school, a police officer makes an arrest or a Job Seekers' Allowance claimant signs on, somebody does some paperwork (or screenwork) and updates a total: the total number of pupils/arrests/claims in that year/month/week has just gone up by one. Administrative statistics are nothing more or less than a summary of all these updates. Usually administrative statistics record people's interactions with some kind of authority.
Administrative statistics are usually a very accurate record, but it's important to keep in mind what they're a record of. For example, hospital records can tell you exactly how many people were admitted to A&E on a particular Friday night after a bad reaction to dance drugs, but they can't tell you how many people had a bad reaction to the same drug and decided not to go to hospital - let alone how many people took that drug overall (most of them may have been fine).
Survey-based statistics, as the name implies are produced as a result of carrying out a survey: you go out in the street, or send out emails, or phone people up, and ask them a set of questions. Opinion polls are one example of a survey that produces statistics, but surveys aren't just for finding out what people are thinking; they're useful for finding out all sorts of things which can't be caught by administrative statistics. Surveys can tell you how many people are out of work (not just how many have succeeded in claiming JSA), or how many people have committed offences (not just how many have been arrested for it).
Survey-based statistics can be accurate or inaccurate: it depends how well the survey has been designed and also how well the sample has been designed. (Usually* you don't want to ask everyone in the country,
so you ask a sample of people; if you've designed the sample properly
it'll be representative of everyone else.)
The British Crime Survey is a survey (clearly), and the figures it produces are now accepted as being a much better record of the level of crime than police recorded crime figures.
HTSC 2: Validity and reliability
Remember the slide with the two targets? Keep it in mind. Reliability is the 'shots close together' picture. It means that, if you take a survey and take it again a year later - or if you look at an institution's records and look at them again a year later - you'll get the same results unless the reality has changed. Reliability means that the survey - or the records - doesn't produce wide variation in results for no good reason. Of course, if we only know about the underlying reality from the statistics, we can't always tell whether there is a good reason or not - but we can often make a good guess. If, for example, a political opinion poll tells you that support for Labour is at 38% one week, 44% the next week and 38% the week after that, it's not likely that there will really have been such a large and short-lived change; the chances are that you're looking at an unreliable poll.
A reliable survey isn't necessarily accurate: it may be consistently giving the same kind of inaccurate answers. Validity is the 'shots in the right area' picture; it measures the extent to which the statistics are an accurate representation of reality. Once again, if we only know about the underlying reality through the statistics, this is hard to measure - but again, we can make a guess, usually with the help of other statistics. Imagine that you had these three opinion poll readings for successive weeks:
Poll 1: 38%, 42%, 38%, 34%, 38%
Poll 2: 38%, 38%, 38%, 39%, 38%
Poll 3: 34%, 33%, 33%, 32%, 34%
Putting the three together, we can make an educated guess that poll 2's results are both reliable and valid, poll 1's are valid but unreliable and poll 3's are reliable but invalid. We can't be certain, though - the underlying reality could be closer to poll 1 or poll 3.
We looked at some fairly controversial statistics in the seminar. I hope you didn't feel that you were being set up to give the wrong answer. The idea was to draw attention to some areas in which the official statistics are out of line with what almost all of us think. The 'true' figures were all from official government statistics, including immigration figures from the Census and asylum figures from the Home Office; they're probably pretty accurate. Of course, we don't know for certain that they are accurate - but if anyone suggests that the real figures are very different, it's worth asking them where they get their figures from.
*The Census is a survey in which you do want to talk to everyone in the country.
No comments:
Post a Comment