Minimum Income Guarantee: medium-term social security modelling

Annex B: Methodology for estimating fiscal impact of time-limited MIG

To estimate the impact of the one-year time limit, we begin by noting that the total cost of a policy over the course of a year is a function of the number of claims active over the year, the duration of each of those claims within the year (say, in months), and the payment value of each claim (either on average across the claim or within each month of the claim). Out of these three variables, the time limit most directly affects the duration of claims. It can also affect the number of claims that fall within a given year by preventing some claims from continuing on to the following year.

For analytical purposes it is helpful to combine these two variables into ‘claim-months’ – that is, the number of claims that were active within a year multiplied by the number of months that fell within the year for each claim. All else equal, the cost of a policy will change in proportion to the number of claim-months. The time limit could also affect payment values by altering the caseload composition, i.e. if claims with higher or lower payments are more or less likely to be affected by the limit. However, this effect is more difficult to predict and is not explicitly considered here.^[5]

We estimate the impact of the time limit using cross-sectional administrative data on the duration of UC claims in Scotland between April and July 2024.^[6] We exclude those in the ‘no work requirements’ conditionality regime – i.e. those who, as a result of caring responsibilities or incapacity, are not expected to perform paid work – in line with the policy that these claimants would be exempt from the time limit.

The basic method is given by the following steps:

1. Estimate the total number of claim-months under the current UC system by summing the number of claims multiplied by their respective durations;

2. Limit claim durations to a year, except those exempt through caring responsibilities and incapacity, and re-estimate the total number of claim-months;^[7]

3. Apply the ratio between these two figures to the estimated cost of the mid-MIG, with the assumption that the distribution of claim durations in the current UC system is representative of the distribution under the mid-MIG.^[8]

However, there are a number of issues with the data which mean that we do not directly observe the number of claims nor their respective durations:

1. The data is banded: we do not know the precise duration of each claim, only the duration within fixed bands. We therefore assume that the midpoint of each band is representative.^[9]

2. The data is subject to survivor bias: longer claims are more likely to be observed than shorter claims, not because they are more frequent but simply because they are longer. If we were to observe all the claims that occurred over the course of a year, we would see that there were relatively more short claims than is indicated in the cross-sectional data.^[10] We adjust for this by weighting the claims in each band by the inverse of its duration as given by the midpoint, with the assumption that the probability of observing a claim is given by its duration.

3. The data is right-censored: we do not observe completed claims, only those that are in still in payment, and therefore do not know how long claimants actually remained on UC. We adjust for this by assuming that claimants will on average be observed halfway through a claim.^[11] This is a significant simplification; for example it assumes that the probability of finishing a UC claim does not change with duration.

The total number of claim-months can then be calculated with and without the time limit, as above, and the ratio between the two can be taken. After making a final adjustment to reflect the fact that claims longer than five years would flow back on the mid-MIG, this ratio is then applied to the estimated cost of the mid-MIG to estimate the savings that would accrue from imposing the time limit.

The method outlined above was validated against longitudinal analysis on waves 6 to 11 (2014-2015 to 2019-20) of the Understanding Society survey. As noted, this data is only collected at one point in time each year for each survey participant, without information on the interim period.^[12] We therefore assume that an individual received UC, at most, for the number of years in which they reported receiving UC, and also that claims were continuous between consecutive years in which UC was reported. The data is also censored, meaning we cannot always observe when a claim began or when it ended. This will tend to underestimate the duration of claims in the aggregate. Finally, the limitations of the sample size mean that we have had to perform this analysis at the UK level rather than Scotland specifically, and the time periods also differ.

With these caveats in mind, the data seem to show a similar picture to our analysis of the administrative data. The proportion of claims affected by the one-year limit – i.e. those that would otherwise last for more than one year and do not involve being out of work due to caring responsibilities or disability – is 19% in the Understanding Society data as compared to 16% in the Universal Credit data. The duration of affected claims and the reduction in case-months caused by the limit are lower in the Understanding Society data. However, given that this can be readily explained by the censoring inherent in the data, it seems to indicate that the assumptions we have applied to the Universal Credit data regarding case durations are in the right region.^[13]

Notes: Understanding Society data includes censored claims.

As noted, our analysis of the Universal Credit data takes into account the fact that claims longer than five years would move back onto the mid-MIG, after receiving the mid-MIG for one year and moving onto the existing UC system for four years. On the other hand, because the data is not longitudinal, we are unable to take into account the potential for a given individual to be subject to the limit by making multiple, separate claims within five years. This effect appears to be relatively insignificant, however: over a five-year period of Understanding Society data (waves 7-11), ninety-four percent of people who reported receiving UC in any year reported only one claim, with nearly all of the remaining 6% reporting two claims.^[14] It would in any case depend on the details of the policy, such as whether the one-year entitlement to the mid-MIG was available in segments or only through one continuous claim.

The exercise was repeated with an assumed time limit of six months, rather than twelve, in every five years. It was not possible to cross-validate this analysis with Understanding Society due to the latter’s annual frequency.