The first attempt to measure fraud
The new look at fraud
What the new fraud figures are telling us
Keeping fraud in perspective

This Government has set itself some clear targets for reducing the amount of money flowing out of the social security system through fraud and errors, collectively called ‘incorrectness’. For the moment these targets are solely for income support (IS) and jobseeker’s allowance (JSA) claims and are based on a baseline period (October 1997 to September 1998) when it was estimated that 9 per cent of IS and JSA payments valued at £1.38 billion were incorrect. The first targets were announced in March 1999 by the Secretary of State for Social Security, Alastair Darling: a 10 per cent reduction in IS and JSA losses by March 2002, leading towards a 30 per cent reduction by 2007. In 2000, these targets for reducing fraud and error were revised significantly:

• as before, a 10 per cent reduction by March 2002;
  
• 25 per cent by March 2004;
  
• 50 per cent by March 2006.

Observers of debates about fraud will have become familiar with these figures, but there are other statistics which are frequently dragged into the discussion which were the product of an earlier system of measurement used by the last Conservative Government. These are a few examples:
fraud…costs taxpayers and claimants an estimated £4 billion…enough to give every family with children an extra £10 a week

• a conservative estimate of fraud is £2 billion a year; but the figure could be much higher, around £7 billion if all suspicions of fraud were well founded [footnote 1]
• previous estimates have shown that £2 billion a year has definitely been lost through fraud. A further £2 to £3 billion may have been lost in cases where fraud is strongly suspected [footnote 2]
• Labour themselves have admitted there is still £7 billion of fraud in the system… [footnote 3]

It is clear from these extracts and the latest press releases from the DSS that there is a distinct lack of clarity in public pronouncements about the level of fraud. This is a bad thing. If there is one area of social security policy that needs clarity it is fraud.

In this article I will attempt to explain the origins of all these statistics, and examine how the latest figures are derived and presented. The statistics make interesting reading, but raise some doubts and concerns about how they are produced and how they are used. First, a little history.

The first attempt to measure fraud
Those with a long memory will recall how the last Conservative Secretary of State for Social Security, Peter Lilley, invented the modern sport of scaring the population with stories of rampant fraud in the social security system. Headlines proclaiming that one in five lone parents on benefit were fraudulent were prominent in 1995. But where did Mr Lilley get his information?

Under the last Conservative Government the Benefits Agency carried out a number of ‘national benefit reviews’, designed to produce measures of the extent of fraud for individual benefits. In a Poverty article in 1996, I explained how these reviews worked and how they led to the headline figures that caught the attention of public and media alike. Benefit reviews were in one sense something of a breakthrough, being the first attempt to measure the amount of fraud scientifically. Unfortunately the methodology used was suspect and the resulting global figures on fraud unreliable. Cases examined as part of the reviews were categorised as either confirmed fraud, suspected fraud, error, or correct. By definition, no-one could be certain that suspected cases were actually fraudulent because there was insufficient evidence. Nevertheless, cases categorised as ‘strong’ suspicion were lumped together with the confirmed cases to produce a figure for ‘total fraud’. Using this method, the public were told as statements of fact that, for example, 9.7 per cent of income support cases were fraudulent in 1995 at a cost of £1.4 billion, and that this had risen to 11.1 per cent in 1997, worth nearly £1.8 billion.

Benefit reviews were also carried out for housing benefit, disability living allowance, retirement pensions, unemployment benefit, invalid care allowance and child benefit and the results aggregated to produce global estimates for the amount of fraud in the social security system. These estimates are the source of the statistics used in the Labour Government documents and by politicians quoted in the introduction to this article.

Many commentators were unhappy about this representation of the extent of fraud and argued that it could stigmatise honest and legitimate claimants, contribute to non-take-up of benefits and lead to an over-emphasis on punitive policy responses. The days of these one-off benefit reviews were clearly numbered and indeed when the Labour Government took office in May 1997 work was underway on the replacement of what had become recognised as a discredited and unreliable system of measurement.

The new look at fraud
Since October 1997, a new system called ‘area benefit reviews’ (ABRs) has been used to measure the losses to the benefit budget through fraud and claimant error. Despite the similarity in name to the old benefit reviews there are some major differences between the two approaches. The principal differences are set out in the table below.

Several of the features of the new ABRs are clear improvements on the old approach. The rolling programme of reviews taking place in all 13 area directorates of the Benefits Agency produces a continuous stream of data for analysis. As long as the same methodology is applied, and applied consistently, then we have a means of measuring changes in the levels of fraud and error over the years.

ABRs do not cover all social security benefits, however. In consequence we have no current measure for the total amount of fraud across the whole of the benefit system. This creates something of a presentational problem for the Government and perhaps helps to explain why the old figures are still cited.

How then has the measurement of fraud and error changed? The ABRs work like this. A sample of cases is taken from each area directorate in the country. The papers are examined by ‘review officers’ and unnotified visits and interviews held with the benefit recipient. Review officers are looking for evidence of ‘incorrectness’, ie fraud or claimant error on each case. However, before a case is classified, two months is allowed for follow-up work and further investigation. Where there is an initial strong suspicion of fraud, the case is passed to the Benefits Agency’s Fraud Investigation Service (BFIS). After its investigation the case is reclassified as fraud or not fraud, or in some cases may remain classified as ‘high suspicion’ (for example, when the case is still under investigation after two months).

In deciding how to classify each case that is reviewed, the review officers use the following system: [footnote 4]

Claimant error: this includes cases where the review officer finds a discrepancy in the claim and as a result there is a change in benefit, but where there is no suspicion of fraud or fraudulent intent by the claimant. Not reporting a change in capital might be an example here.

Fraud: this includes cases where (a) the conditions for entitlement, or for the amount of benefit in payment, are not being met, (b) the claimant can reasonably be expected to know that their benefit is incorrect, and (c) benefit stops or is reduced as a result of the review.

Included in the ‘fraud’ category are those cases where (a) there is suspected, though not proven, fraud and (b) benefit changes as a result of the review. In such cases, Review Officers must show a link between their activity and the change in benefit.

Suspicion of fraud: this category is used where there is suspicion of fraud but it cannot be proved, and where benefit does not change as a result of the review. This category is sub-divided into low suspicion and high suspicion.

An example of ‘low suspicion’ is where a claimant’s apparent spending is considerably in excess of benefit payments but there is no way of establishing how such spending is financed. Cases are classified as ‘high suspicion’ where the review officers concludes that, on the balance of probabilities, a fraud has been committed but there is insufficient information to prove fraud.

Before we look at the statistics that emerge from this system of classification, two concerns need to be raised. The first is about the definition of fraud, and the second is about the absence of a classification of official error, ie by benefit officials, that has appeared in previous published reports on the extent of benefit fraud.

A definition of fraud which includes some suspected as well as proven cases of fraud appears itself to be suspect. There is no doubt that after investigation it will still be unclear in some cases whether fraud has been committed even where the benefit has changed after the review. However, to include these in the statistics for ‘fraud’ leaves the statistics open to the same criticism levelled at the old system of measurement that the true level of fraud is exaggerated. Furthermore, we have no way of knowing how much of the ‘fraud’ which appears in government reports falls into to this ‘suspected’ category.

The second concern is about the absence of a measure of ‘official error’ from the ABR results. In the first report to come out of the ABR process [footnote 5] we are told that ‘the ABR is not designed to collect information on official errors...’ though no explanation for why is given. The previous system of national benefit reviews was certainly able to do this. Let us be clear – there are figures for official error in government reports but these do not emerge from the review process, but from a separate exercise called QST checks. QST stands for Quality Support Team, an internal Benefits Agency administrative division which checks the accuracy of benefit decisions, and which produces a measure of official error. This figure is adjusted to remove ‘overlaps’ between QST and ABR results and added to the ABR figures for claimant fraud and error to produce the overall measure of ‘incorrectness from fraud and error’. (It is this process that produced the benchmark figure of 9 per cent of IS and JSA payments being the result of fraud or error for the period October 1997 to September 1998.) This method of calculation is, to say the least, confusing and difficult to understand, and is certainly less transparent than the method used under the measurement system used in the discredited national benefit reviews.

What the new fraud figures are telling us
Having set out some reservations about the new way of measuring fraud and error, let us look at what they are saying. I am drawing on two sources for this section, the first full report to emerge from the area benefit reviews [footnote 6] and a short report published in November 2000 which presents limited data from the second full report which, at the time of writing, is expected in February or March 2000. [footnote 7]

The changes in the levels of IS and JSA fraud are shown in Table 2 below.

Table 2 shows how the headline figure for IS and JSA fraud and error combined has fallen from 9.0 per cent to 8.4 percent between October 1997 and March 2000, the basis on which the DSS claims that the ‘tide is turning on benefit cheats’. However, while the overall reduction is evident there have also been interesting fluctuations up and down over the last two years rather than a steady and continuous reduction. This general pattern is mirrored in the figures for IS, but not for JSA where an initial increase in fraud and error has been followed by a steady reduction. It is also clear from the table that the levels of fraud and error are significantly higher for JSA cases than for IS.

Presentation of all statistics will always be selective, but it is perhaps puzzling why claimant fraud and error are presented as a single figure distinct from official error. Another way of looking at the figures is to combine both types of error and compare that with levels of fraud. Unfortunately the short report published in November 2000 does not break down the figures enough for this comparison to be made, but the report for April 1998 to March 1999 are sufficiently disaggregated. The results are presented in Table 3.

Table 3 shows the balance between fraud and error for IS and JSA. The ABRs are undoubtedly uncovering levels of fraud which are important and cannot be ignored. However, it is rarely, if ever, mentioned in policy statements, speeches or press releases that something in the region of half a billion pounds is leaking out the benefit system through errors and that out of this amount over one third of a billion pounds is due to official errors.

There is no doubt that the amount of counter fraud activity carried out by the Benefits Agency has increased since the present government came into office. It is also clear that there have been new policies and practices aimed at improving the quality of the information provided by claimants when making claims. The introduction of tougher verification procedures should be expected to result in fewer claimant errors. However, it is not so evident that official errors are being tackled successfully. In the baseline period October 1997 to September 1998 official errors were running at 1.5 per cent for IS claims and 4.2 per cent for JSA. In the latest period, April 1999 to March 2000, the comparable figures were 1.5 per cent and 4.1 per cent respectively. So, although the headline figure for fraud and error had fallen from 9 per cent to 8.4 per cent, the rates of official error remained virtually unchanged.

Keeping fraud in perspective
In many ways the problem of social security fraud is being treated in a more balanced and sensible manner by this government than the last. There is recognition that fraud and error are part of a single problem of money leaking out of the social security system in ways unintended by government policy. There are more policies designed to prevent and deter fraud even if policies aimed at detection still appear to attract more attention and investment. In the ABRs there is the potential for producing more comprehensive and more useful data and analysis than ever before.

However, the political imperative of headline figures and demand for success stories stifle the opportunity for a more imaginative examination of the fraud figures, and for putting fraud in any form of external perspective. For example, in public debates about fraud, why are the following never mentioned?

• Social security fraud is only a fraction of the size of the informal economy, estimated by Lord Grabiner in report published in 2000, to be worth between £50 billion and £80 billion.
• Much of the money ‘lost’ to the social security system through fraud and error can be recovered from claimants. The statistics on the amounts recovered are not made public however.
• The level of unclaimed income-related benefits is comparable to levels of fraud. In 1997-1998 between £1.6 billion and £4.1 billion was not claimed. [footnote 8]

Our understanding of fraud is getting better, but the standard of public debate is not keeping pace. Thankfully we are no longer subject to the histrionics of pre-1997 Conservative Party conferences, but the political and media rhetoric still concentrates on the ‘problem’ of fraud being rooted in individual claimants’ aberrant behaviour, rather than being a product also of a hugely complex and relatively ungenerous social security system. Furthermore, conflating fraud and error in the presentation of the fraud figures and mixing old and new statistics does nothing to aid understanding or generate constructive policy thinking. And the worry remains that talking up fraud will stigmatise legitimate claimants, and lead to continuing non-claiming of benefits. The tide may be turning against benefit cheats, but it is increasingly important to protect innocent and vulnerable claimants from being swept away at the same time.

Footnotes

1. A New Contract for Welfare, Green Paper, March 1998 [back to text]
2. Beating Fraud is Everyone's Business, Green Paper, July 1998 [back to text]
3. speech by Michael Portillo, Shadow Chancellor, December 2000 [back to text]
4. these definitions are taken from the DSS report The Results of the Area Benefit Review from April 1998 to March 1999 and Measurements for the Public Service Agreement, published in February 2000 by the Government Statistical Service [back to text]
5. see note 4 [back to text]
6. see note 4 [back to text]
7. DSS Analytical Services Division, Fraud and Error in Claims for Income Support and Jobseeker's Allowance from October 1998 to September 1999 and from April 1999 to March 2000, November 2000 [back to text]
8. figures quoted by Jeff Rooker, DSS Minister, in Hansard written answer, 20 December 1999 [back to text]

Roy Sainsbury is Senior Research Fellow in the Social Policy Research Unit at the University of York

Poverty 108, Winter 2001