Printer Friendly



Over the last decade, the duration of benefit receipt has been viewed with increasing concern. Yet until recently the information available on the length of time people stay on benefit has been less than ideal. Analysts have been forced to rely on cross-sectional measures of either the duration of benefits that are current at a point in time or the duration of benefits that are cancelled in a given period. Both measures suffer from inherent bias.

* The duration of current benefits is biased upwards because those who stay longer on benefit are more likely to be in receipt on a given date than those who stay for only a short period, and biased downwards because duration is calculated only part-way through the benefit spell.

* The duration of cancelled benefits is biased downwards in that those with long spells are less likely to cancel within a given period.

Because of these biases, and because they focus only on a single period on benefit, the cross-sectional measures leave some key questions unanswered: How long do people generally stay on benefit? How likely are they to return to benefit? How common is long-term benefit receipt and to what extent is it associated with intermittent rather than continuous benefit use? Who is most at risk of long-term receipt? Answers to these questions should shape both our understanding of long-term benefit use and our policy response to it.

In 1995, development of an Information Analysis Platform in the Department of Social Welfare permitted a more robust picture of benefit duration to be obtained. It enabled the construction of a longitudinal data set that records the dynamics of benefit use at the individual level. With this "benefit dynamics" data set, we are able to examine the experiences of people who started to receive benefit at a common point in time, and to derive unbiased and more comprehensive measures of benefit duration. This paper uses a recently updated and enhanced version of the benefit dynamics data set(2) to offer new insights into the duration of receipt of the main working-age social welfare benefits.(3) It focuses on people granted benefit in 1993 for whom it is now possible to examine the pattern of benefit experiences over a full five-year period.(4)


When a person applies to receive a benefit, they are asked to supply information needed to assess whether they qualify for support, and to establish the type of benefit they should receive, the level of payment they are entitled to, and any conditions of entitlement that should apply. Once granted benefit, they are required to provide information on any changes in circumstances that might affect their entitlement, and regular renewals of their benefit are carried out to check that the information previously supplied remains current. Finally, when the benefit is cancelled, a reason for cancellation is recorded.

Information collected at these various points is entered on the benefit payments system (SWIFTT), together with the person's basic demographic details, and information generated in the administration of the benefit such as the date payment commenced and ceased, the district office through which it was paid and the rate of benefit paid. Using this data, and linking information recorded for the same individual over time, the benefit dynamics data set builds up a longitudinal picture of benefit experiences. The data set is anonymous and is used solely for research purposes.

Like other longitudinal data sets built from administrative data, the benefit dynamics data set has its limitations.(5)

* It is limited in its scope to the information that is collected in the process of benefit administration. Information on education and work history, for example, is not collected.(6)

* It is limited to information relating to periods of benefit receipt. Robust measures of work status, income levels and family circumstances between periods on benefit are not available.

* For measures that are collected, the proportion of people for whom information is missing can be sizeable. This is particularly problematic for ethnicity.

* Changes in status that are recorded may be the result of administrative practices rather than genuine changes in recipient circumstances. For example, if a person fails to make contact when requested their benefit may be suspended, then cancelled, but subsequently re-granted when contact is re-established. The person's circumstances (unemployment, incapacity, or sole parenthood, for example) may have remained unchanged throughout.

* The quality of the data is highly dependent on the accuracy of reporting by clients and coding by front-desk staff.

* Finally, the data set is a "cleaned" reconstruction of the original SWIFTT source data for each individual (see below). Findings will, to some extent, be sensitive to the choices and assumptions made in assembling the data.

Against these limitations, the use of administrative data offers considerable advantages. The frequency and detail of information on changes in status is much better than could be achieved through a longitudinal survey. In addition, increasing the size of the sample studied imposes no additional costs aside from those associated with storing and processing the data. The large samples that can be attained permit examination of the experiences of narrowly defined subgroups while avoiding the usual problems arising from sampling error.

In the case of the benefit dynamics data set, we are able to hold information on the entire population of people who received benefits over the period of study. At the time of writing, the data set recorded the benefit experiences of the 1.1 million different adults who received one of the main working-age social welfare benefits, either as a primary recipient or as a partner, at some time in the period spanning 1 January 1993 to 31 December 1998. To put this number into context, we would need a count of all the different people who could have received benefit at some time over the period. It is not possible to calculate this number with certainty. The Annex to this paper presents two estimates. Viewing these as broad indications of scale suggests that around four people out of every ten present in the working-age population at some time between the beginning of 1993 and the end of 1998 are included in the data set.


The remainder of this paper explores the benefit experiences of the 250,000 people who were granted a working-age benefit in 1993. Focussing on this group allows us to exploit the longest follow-up possible at the time of writing -- a full five years for every person granted benefit in that year.

It is important to note that those granted benefit in 1993 may have had prior spells on benefit. The 1993 "entry cohort" captures all people granted benefit in that year and is not confined to "first ever" entries into the benefit system. The experiences that emerge for any particular individual may, therefore, reflect the beginning, middle or end of a longer benefit history.

In addition, the patterns of benefit use observed for the 1993 entry cohort are not necessarily generalisable to other entry cohorts. The unemployment rate was higher in 1993 than in the years that followed. As a result, one might expect the 1993 entry cohort to have a greater representation than later entry cohorts of people for whom unemployment was a transitory experience. Changes in the age structure of the population, together with changes in net migration, patterns of family formation, fertility and labour force participation rates, are also likely to have altered the benefit experiences of successive cohorts.

Some Definitions and Assumptions

Of prime concern is the duration and frequency of spells of benefit receipt, where a spell is defined as a period of continuous receipt of the same benefit. In constructing the benefit dynamics data set, a number of assumptions are made:

* If a person is observed transferring from one benefit to another, the first spell is considered to have ceased and a new spell is considered to have commenced.(7)

* If a person is observed shifting their claimant status from being the primary recipient of benefit to the partner of a primary recipient or vice versa, a new spell is considered to have commenced.

* Where we observe two spells on the same benefit separated by less than fifteen days, they are amalgamated and treated as a single, uninterrupted spell.(8)

* Where we observe a single spell interrupted by a period in which benefit payment is suspended for a period of more than fourteen days, it is treated as two separate spells.

* Where a person transfers from a two-weekly to a weekly version of the same benefit,(9) or from a pre-October 1998 benefit to its post-October 1998 Community Wage equivalent, this is treated as a continuation of the original spell on benefit. Similarly, where a person transfers from Emergency Sickness Benefit to Sickness Benefit or from Emergency Maintenance Allowance to Domestic Purposes Benefit, this too is treated as a continuation of the original spell.

The analysis is broken down by the type of benefit each individual was granted as they entered the cohort. The following schema shows the benefits that are included. Domestic Purposes and Unemployment Benefits encompass a number of benefit payment categories: those grouped together under the same bullet point are treated as the "same benefit" in applying the spell definition assumptions described above; those itemised under separate bullet points are treated as distinct benefits in applying these assumptions.(10)

If a person was granted more than one benefit in 1993, their benefit at entry is the first benefit they were granted in that year. This has two implications that are important to bear in mind.

* The size of the 1993 entry cohort for a particular benefit type is somewhat smaller than the total number of grants of that benefit that occurred in 1993.

* The 1993 entry cohort for a particular benefit type may not be representative of all those people who were granted that benefit in 1993. This is especially marked for DPB, as those who come onto this benefit having received Sickness or Emergency Sickness Benefit for the latter part of their pregnancy (a group that is younger and more likely to have a new-born at grant than the average DPB entrant) are under-represented in the DPB entry cohort as they tend to be included in the SB entry cohort.(11)

Finally, if a person's only 1993 grant was one that was created in the construction or cleaning of the benefit dynamics data set, then this person is not included in the 1993 entry cohort. For example, a grant might be created from the assumptions applied where a spell on benefit was interrupted by a lengthy period of suspension (see above). The aim of excluding such grants is to approximate the group considered to be new recipients from an administrative viewpoint.

Table 1 shows the size of the groups first granted each benefit type within our 1993 entry cohort.
Table 1 1993 Entry Cohort by Benefit at Entry

Benefit at entry Number %

Invalids Benefit 4,220 1.7
Sickness Benefit 23,142 9.3
Unemployment Benefit 172,405 69.2
Training Benefit 19,992 8.0
Domestic Purposes Benefit 27,451 11.0
Widows Benefit 1,765 0.7

All 248,975 100.0

Duration of First Observed Spells

Two measures are commonly used to summarise the duration of completed spells on benefit:

* hazard rates, which measure the probability that a spell will end at time conditional on it being at least t weeks in duration; and

* the survivor function, which measures the probability that a spell continues to at least weeks in duration.(12)

Figure 1 shows survivor functions for the 1993 entry cohort by benefit type. Of those granted IB in 1993, more than half spent all of the following five years receiving that benefit compared with a third of those granted WB, one in five of those granted DPB, and less than one in 20 of those granted SB. Just one per cent of those granted UB in 1993 remained on that initial spell for all of the following five years. Few of those granted TB in 1993 had an initial spell lasting more than a year.


Because most were granted UB, the pattern for all 1993 entrants resembles the experiences of that group. Overall:

* most entrants (54%) stayed on that benefit for less than 20 weeks;

* one in five (21%) had a spell on benefit that lasted for a year or more;

* fewer than one in ten (7%) had a spell on benefit that lasted for three years or more; and

* one in 20 (4%) had a spell on benefit that lasted for all five years of the follow-up period.

With the exception of IB, the shape of the survivor functions is concave. The probability of cohort members remaining on benefit for another week increases as length of time on that benefit increases. Figure 2 shows the corresponding hazard rates. These can be interpreted as the probability a person who has been on benefit continuously up to a given week long interval will stop receiving benefit in that week, or the "rate of exit" during that week. They confirm that, after an initial rise, the probability of a spell on benefit ending falls as benefit duration increases.(13)


There are two possible explanations for this negative duration dependence:

* It may be true duration dependence. In the context of unemployment-related receipt, a long period on benefit may cause the likelihood of an individual exiting from benefit to fall, either by eroding their work skills, the intensity of their search for work or their motivation to work, or by reducing their attractiveness to potential employers -- a scarring effect. Alternatively, long duration on benefit may cause family health problems that reduce the likelihood of exit,(14) or cause financial reserves to run down to the extent that recipients become more averse to the uncertainty in income flows that might be associated with a move off benefit.(15)

* It may instead reflect population heterogeneity. Those who have good employment or partnering prospects, have low wage demands, and are highly motivated, in good health, and able to withstand income uncertainty, leave before others. Consequently as time from entry increases those with the poorest chance of moving off benefit make up a greater and greater proportion of the group that remains.

The initial analysis of spell duration by demographic and other characteristics set out in the final section of this paper suggests that population heterogeneity explains some part of the duration dependence observed. More sophisticated analytical techniques are required to estimate the effect of unobserved heterogeneity, that is, heterogeneity that is not measured by the variables available from the data set (motivation and aspects of health status, for example), and to arrive at an estimate of the degree of true duration dependence.

UB entrants have peaks in the hazard of exit in the 9-12 week and 25-28 week intervals and around the anniversaries of entry. Gardiner (1995) found similar spikes in the probability of exit from the job seeker register and attributed the first spike (at nine weeks) to the automatic lapse of job seekers who have not made contact within the past eight weeks, and the second and third spikes (at 26 and 52 weeks, respectively) to the effects of work focus interviews conducted at these durations. Because registration is generally a condition of entitlement for UB, these effects would also explain the patterns shown in Figure 2. The spike shortly following two years may reflect the effects of Job Action, a package of interventions introduced from late 1994 that was targeted at job seekers reaching 104 weeks duration.(16) Annual spikes are also apparent for the SB and DPB entrants, and these are likely to reflect the effects of the annual benefit renewal processes.

Total Duration of All Spells in the Follow-up

Of the 96% of the 1993 entry cohort who completed their first observed spell in the five-year follow-up, three quarters either transferred to another benefit or returned after some time off benefit in that five-year period. As a result, the total length of time spent on benefit in the five-year follow-up was greater than examination of their first observed spell would suggest.

Figure 3 shows the proportions with total duration in the follow-up greater than or equal to given numbers of weeks, broken down by the benefit type received at entry. Included in the calculation is the duration of all spells on all benefit types, including spells spent in receipt of benefit income as a partner. Overall:

* almost two thirds (62%) of the 1993 entry cohort spent a total of at least one year out of the following five in receipt of benefit income;

* one third (33%) spent three years or more out of the following five years on benefit; and

* just under one in ten (8%) spent all of the five year follow-up in receipt of benefit.


In Figure 1, the height of the survivor function at 260 weeks gives the proportion of 1993 entrants who spent all five years of the follow-up on the benefit they were first granted. In Figure 3, the height of the curve at this point gives the proportion on benefit continuously for all of the period, either as a result of remaining on the benefit they were first granted in 1993, or as a result of moving between this and other benefits. The percentage point difference is greatest for those who entered onto SB (Table 2): whereas 3% were continuously on their first observed spell on that benefit, another 12% ceased that spell but immediately transferred to another benefit and continued to receive benefit income without interruption for all of the follow-up.

Table 2 Percentage in Continuous Receipt of Benefit for Five Years from 1993 Entry by Benefit at Entry
Benefit at entry Per cent with Per cent Percentage-
 continuous with continuous point
 receipt of receipt of difference
 first observed first observed
 benefit or other

Invalids Benefit 55 58 3
Sickness Benefit 3 15 12
 Benefit 1 3 2
Training Benefit 0 3 3
Domestic Purposes
Benefit 22 26 4
Widows Benefit 32 39 7
All 4 8 4

While TB entrants had a greater likelihood than UB entrants of having a short first spell, Figure 3 shows that they were almost twice as likely as UB entrants to spend four or more years out of the following five on benefit (26% compared with 14%). As most TB recipients have received or go on to receive UB, it is likely that this difference is indicative of the effects of greater educational and labour market disadvantage within the wider job seeker population. In 1993, most TB recipients were on Training Opportunities Programme (TOP) courses, gaining entry to which generally required that they should have no or very limited formal qualifications and more than 26 weeks duration on benefit.

Combined, the 250,000 people granted benefit in 1993 spent a total of 27.7 million weeks on benefit in the five year follow-up. While on average this amounted to 111 weeks (just over two years) per person, the distribution of total benefit weeks across the cohort was far from even. The 20% of the entry cohort with the shortest durations accounted for just 2% of the total weeks spent on benefit. At the other extreme, the third of the cohort that spent three or more years of the follow-up period on benefit accounted for just over two-thirds of the benefit weeks and the 8% of the cohort that spent all of the follow-up on benefit accounted for one fifth of the benefit weeks.

Table 3 gives a measure of the degree of inequality by benefit type at entry. If each person in the cohort spent the same number of weeks in the follow-up on benefit, 20% of the group would account for 20% of the benefit weeks. Total duration was most unequal for UB entrants. Forty per cent of 1993 UB entrants accounted for only 8% of the total weeks this group spent on benefit in the five years following.

Table 3 Percentage of Total Weeks Spent on Benefit in a Five-year Follow-up by the 1993 Entry Cohort Accounted for by the 20 and 40 per cent with the Shortest Durations by Benefit at Entry
Benefit at entry % benefit weeks % benefit
 accounted for weeks accounted
 by the 20% with for by the 40%
 the shortest with the shortest
 durations durations

Invalids Benefit 6 25
Sickness Benefit 2 11
Unemployment Benefit 2 8
Training Benefit 3 14
Domestic Purposes
 Benefit 5 19
Widows Benefit 4 18
All 2 9

Probability of Receipt Over Time

Yet another picture of the duration of benefit experiences emerges if we consider the cohort's probability of being in receipt of any benefit income with increasing time from the date of their 1993 grant. Figure 4 compares the fortunes of the 1993 entrants by the benefit type they were granted at entry.


This plot provides little information about the duration of particular individuals within the cohort -- one individual could be in receipt of benefit in the first and last week of the follow-up and have a total duration of only those two weeks for example -- but it tells us much about the rate at which the cohort as a whole moved on. It suggests that for a significant proportion of cohort members complete movement out of the income circumstances which caused recourse to the benefit system was slow. Overall:

* half (50%) of the cohort were in receipt of benefit income on the date one year from their 1993 grant;

* 37% were in receipt at three years; and

* 35% were in receipt at five years.

Of those in receipt at five years, 8% had remained continuously in receipt of benefit for all of the follow-up. Another 27% had moved on and off and were back in receipt of benefit on that date, either as a primary recipient or as a partner.

IB entrants were the most likely to remain on or be back on benefit at the end of the follow-up (69%), followed by DPB entrants (57%) and then WB entrants (52%). Forty-seven per cent of SB and 46% of TB entrants remained on or were back on benefit at the end of the follow-up. While Figure 1 shows that entrants onto WB had a greater likelihood than DPB entrants of remaining on their first observed spell five years from grant, the latter group were more likely to be in receipt of benefit income on that date. This is likely to reflect the higher likelihood that WB entrants reached the age of eligibility for the Transitional Retirement Benefit or New Zealand Superannuation within the follow-up period.

UB entrants were the least likely to be on benefit five years after their 1993 grant, and their probability of receipt was strongly cyclical. In the first three years from grant, the proportion receiving benefit around the anniversary of the 1993 grant was two to four percentage points higher than in the preceding months. Annual peaks in rates of student unemployment and the seasonal nature of unemployment in some occupations are likely to explain this pattern. The probability of receipt flattens considerably in the fourth and fifth year from grant for UB and TB entrants. As with the concave survivor functions for these benefits, this could reflect duration dependence, population heterogeneity, or a combination of the two. Those who continue to cycle in and out of benefit receipt after three years are likely to be a group that face greater labour market disadvantage and this may be a function of a long history of repeated benefit receipt, or may be associated with factors quite independent of this history.

Repeat Spells

The sharp contrasts between the probability of being on any spell on benefit with increasing time from the 1993 grant (shown in Figure 4) and the probability of remaining on the first observed spell (shown in Figure 1) reflects the significant role that repeat spells played. Table 4 shows the proportions experiencing different numbers of spells in the follow-up, including spells spent on benefit as a partner. Just over one-quarter of 1993 entrants commenced one spell only in the following five years. Around half commenced three or more spells. IB and WB entrants were the most likely to experience a single spell, and TB and UB entrants were the most likely to experience five or more spells.

Table 4 Percentage Commencing Single and Multiple Spells in the Five Years From Entry, by Benefit at Entry
 Number of spells commenced
 in five years from entry

Benefit at entry 1 2 3 4 5+ All

Invalids Benefit 81 13 4 1 2 100
Sickness Benefit 20 30 17 12 21 100
Unemployment Benefit 25 21 17 13 25 100
Training Benefit 9 12 13 14 51 100
Domestic Purposes
 Benefit 43 20 14 8 15 100
Widows Benefit 55 35 7 2 2 100

All 27 21 16 12 25 100

Note: Rows may not sum to 100 because of rounding.

Many repeat spells began as a result of transfers between benefits. Table 5 shows the destinations of the 1993 cohort at completion of their first observed spell. Around two fifths of SB and TB entrants transferred to another working-age benefit at completion of their spell, compared with one in three WB entrants, just under one in five DPB entrants and one in six UB entrants. IB entrants were the least likely to transfer to another working age benefit on cancellation.

Table 5 Percentage with Each Destination at Completion of First Observed Spell, those Completing First Observed Spell within Five Years of 1993 Entry, by Benefit at Entry
 Benefit at entry

 Invalids Sickness Unemploy-
Destination at cancellation: Benefit Benefit ment

On Invalids Benefit 0 7 0
On Sickness Benefit 4 0 4
On Unemployment Benefit 2 20 3
On Training Benefit 0 1 6
On Domestic Purposes Benefit 2 11 1
On Widows Benefit 0 0 0
On Transitional Retirement
Benefit 5 1 1
On benefit as a partner 1 2 1
All on working age benefit 13 42 15

Employment 12 13 43
Re/partner 3 2 1
Overseas 8 3 4
Full-time study 0 1 7
Prison 5 1 1
Reached NZ Super age 10 1 0
Deceased 29 2 0
Other/unknown 20 35 29
All not on working age benefit 87 58 85

All 100 100 100

 Training Domestic Widows
Destination at cancellation: Benefit Purposes Benefit

On Invalids Benefit 0 1 5
On Sickness Benefit 3 2 0
On Unemployment Benefit 35 10 0
On Training Benefit 0 0 0
On Domestic Purposes Benefit 0 1 1
On Widows Benefit 0 0 0
On Transitional Retirement
Benefit 0 0 26
On benefit as a partner 0 4 1
All on working age benefit 38 18 33

Employment 8 22 24
Re/partner 0 32 6
Overseas 1 5 12
Full-time study 1 0 0
Prison 0 1 0
Reached NZ Super age 0 0 9
Deceased 0 0 2
Other/unknown 51 21 14
All not on working age benefit 62 82 67

All 100 100 100

For SB entrants who transferred to another benefit, the main destinations were UB (20% of cancellations) and DPB (11% of cancellations). The latter group is likely to be made up mainly of new mothers who received Sickness or Emergency Sickness Benefit for the latter part of their pregnancy. Transfers to IB were less common (7% of cancellations). Most TB entrants who transferred to another benefit moved onto UB (35% of cancellations). IB and WB entrants were the most likely to transfer to the Transitional Retirement Benefit. One in ten DPB entrants transferred to UB on cancellation. This group is likely to be largely made up of those whose DPB was cancelled upon their youngest child turning 18 years. DPB entrants were the most likely to transfer onto another benefit as a partner (4% of cancellations).

For those who did not transfer to another benefit at cancellation, Table 5 uses information on reason for cancellation and a calculation of whether they had reached the age of eligibility for New Zealand Superannuation to infer their destination. Across all benefit types, there is a large "other/unknown" group made up mainly of those for whom the reason for cancellation code entered gives no indication of destination.(17) The true proportions that cancelled benefit for some or all of the "not on benefit" destinations shown in the table will, therefore, be somewhat greater than the data available suggests.

The modal destination for those not transferring to another benefit at cancellation was employment for UB and WB entrants, partnering or re-partnering for DPB entrants, death for IB entrants, and the other/unknown category for SB and TB entrants. The likelihood of the spell on benefit ending as a result of imprisonment was highest for IB entrants, although this group accounted for only a small proportion of the total number of spells ending in this way. The codes for cancellation reasons do not distinguish cases of imprisonment in secure psychiatric institutions within this category.

Did the likelihood of a repeat spell vary depending on the reason for cancellation for those who moved completely off benefit? Table 6 explores this question by examining the probability of return to any benefit, either as a primary recipient or as a partner, within two and a half years from the date of cancellation for those who cancelled and moved completely off benefit within two and a half years of their 1993 grant.

Table 6 Percentage Returning to Any Benefit Within 2.5 Years of Cancellation, by Benefit at Entry and Destination at Cancellation
 Benefit at entry

 Invalid Sickness Unemployment
Destination at cancellation: Benefit Benefit Benefit

Employment 55 49 55
Re/partner 56 65 61
Overseas 63 63 58
Full-time study na 86 80
Prison 88 90 91
Reached NZ Super age 0 0 0
Deceased 0 15 27
Other/unknown 54 67 73

All 33 61 64

 Training Domestic Widows
Destination at cancellation: Benefit Purposes Benefit

Employment 58 51 43
Re/partner 91 61 35
Overseas 61 39 79
Full-time study 85 85 na
Prison 95 90 na
Reached NZ Super age 0 0 0
Deceased 0 0 2
Other/unknown 83 72 53

All 80 60 45

Note: Includes only those completing their first observed spell within 2.5 years and moving completely off benefit.

For all benefit types, those whose destination was in the "other/unknown" category had a greater than average chance of return. This suggests that the true proportions that returned from some or all of the destinations given will be greater than the table suggests. Those whose destination on cancellation was prison had the highest likelihood of returning to benefit (around nine in ten returning within two and a half years), closely followed by those who left benefit for full-time study. The latter group is likely to be largely made up of tertiary students making periodic use of Emergency Unemployment Benefit over the period of their study. Of those who Cancelled their benefit having reached the age of eligibility for New Zealand Superannuation, small proportions returned to benefit. This group is likely to largely comprise people who did not meet the residency criteria for New Zealand Superannuation.

Among those whose first observed spell was on DPB, chances of a sustained exit from benefit were better for those moving into work than for those moving into a new partnership or reconciling with a previous partner. Just over 50% of DPB entrants who moved off benefit and into work returned to benefit within two and a half years, compared with just over 60% of those who partnered or re-partnered. For WB, the pattern was reversed -- those who left to partner or re-partner were more likely to stay off benefit.


Who is most likely to spend a long period on benefit? Who is the most likely to have repeat spells? When someone is granted a benefit, does the information they supply at that time give an indication of their risk of spending a high proportion of the coming years in receipt of benefit income?

This section takes an initial look at these questions. The analysis does not attempt to assess the marginal contribution that different characteristics make to explaining the patterns observed. Such multivariate analyses are an area for further work.(18)

Three Measures of "High Risk"

Table 7 shows variation between those with different demographic and "as at grant" characteristics in the probability of reaching three thresholds of long-term or repeated receipt:

* having a first observed spell of at least 32 weeks in duration;

* commencing four or more spells in the five year follow-up; and

* having a total duration on benefit of at least three out of the five years of the follow-up.

Table 7 Percentage with Duration or Number of Spells Reaching Given Thresholds Counting all Spells (Including Spells as a Partner) in Five the Year Follow-up 1993 Entry Cohort by Characteristics at Entry
 Percentage with
 Percentage of First spell 32
Characteristics at entry 1993 entrants weeks +

 Male 55 30
 Female 45 35

 Under 20 23 25
 0-29 42 27
 30-39 18 40
 40-49 9 42
 50-59 6 53
 60+ 1 47

Partnership status
 Single 85 32
 Partnered 15 33

Number of children
 0 77 26
 1 11 55
 2 7 49
 3+ 5 49

Age of youngest child
 Under 6 16 53
 6-13 6 50
 14+ 2 46

All 100 32

 Number of Total duration
Characteristics at entry spells four + three years +

 Male 38 27
 Female 35 40

 Under 20 52 33
 0-29 35 27
 30-39 32 37
 40-49 29 39
 50-59 21 53
 60+ 8 24

Partnership status
 Single 38 34
 Partnered 30 29

Number of children
 0 39 29
 1 29 52
 2 27 44
 3+ 31 47

Age of youngest child
 Under 6 31 51
 6-13 26 43
 14+ 27 37

All 37 33

Roughly a third of all 1993 entrants reached each of these "high risk" thresholds. What is notable is the significant variation in the proportions reaching each threshold within different sub-groups.

The patterns vary most by age group. The probability of a long single spell was lowest for those aged under 20 at their first observed entry and increased with age before dropping for those aged 60 or over, many of whom qualified for New Zealand Superannuation within three years of their 1993 grant. The probability of having multiple spells followed the opposite pattern with half of those aged under 20 having four or more spells and the probability decreasing with age. The probability of long total duration was bimodal -- a different pattern again. Those aged under 20 at entry had the same risk of long duration as the all ages average and those aged 20-29 had a lower than average risk. The risk then rose with the age to peak at 53% in the 50-59 age group, falling to its lowest level for the group aged 60 or over.

Female entrants were slightly more likely than males to have a long initial spell, and slightly less likely to have multiple spells, but one and a half times more likely to have a total duration of three or more years out of the five (40% compared with 27%). Being single made little difference to the probability of a long initial spell, but was associated with an increased probability of multiple spells and a long total duration.

The presence and age of children appears to be an important factor in determining the risk of long-term or repeated receipt. Entrants with children were around twice as likely as those without to have a long first observed spell and a long total duration. Having a youngest child aged under six years slightly increased the probability of multiple spells, and substantially increased the probability of a long total duration.

The pattern and degree of variation across the three measures suggests that we cannot extrapolate from results of analyses based on one measure. It suggests that there is no reason to expect, for example, that the factors associated with a long single spell also explain the likelihood of having repeated spells, or that characteristics associated with a long single spell will have the same associations with long total duration.

Ethnicity and the Likelihood of a Long Total Duration

Ethnicity data in the benefit dynamics data set require some caution in interpretation. The proportion of cases for whom ethnicity was recorded was very low in the early part of the study period but improved significantly from 1995/96 onwards. In constructing the benefit dynamics data set, the proportion of cases for whom ethnicity is recorded has been maximised by using the most recent ethnicity information available for each individual. However, because ethnicity tends to be recorded at grant:

* members of the 1993 entry cohort who had a single spell on benefit are less likely than those who left benefit and returned at a later date to have their ethnicity recorded in the data set; and

* those who remained on their first observed spell at the end of the follow-up are less likely than those with a long duration made up of multiple spells to have their ethnicity recorded in the data set.

As a result, it is apparent that members of the 1993 entry cohort for whom ethnicity is not recorded had a much lower probability of spending three or more years of the five year follow-up on benefit than cohort members for whom ethnicity is recorded. Because the "not recorded" group accounts for one quarter of the cohort, we can have little confidence that the absolute values of the probabilities of long duration that are found for the different ethnic groups approximate their true level. The figures may, however, offer a reasonable indication of the size of the difference that being in one ethnic group rather than another makes.

Table 8 focuses on the measure that is perhaps of greatest concern from a policy point of view -- total duration on benefit -- and shows the ratio of the probability of a long total duration for each ethnic group to the probability of a long total duration for the European ethnic group.

Table 8 Relative Probability of Total Duration of Three or More Years Out of Five for Given Ethnic Groups, 1993 Entry Cohort by Characteristics at Entry
 Pacific Other ethnic
 NZ Maori: Peoples: groups:
Characteristics at entry European European European

 Male 1.8 1.4 1.1
 Female 1.8 1.4 1.0

 Under 20 1.9 1.2 0.9
 20-29 2.2 1.6 0.9
 30-39 1.5 1.2 0.9
 40-49 1.4 1.4 1.0
 50-59 1.1 1.2 1.2
 60+ 1.7 9.3 11.5

Partnership status
 Single 1.8 1.4 1.0
 Partnered 1.5 1.4 1.2

Number of children
 0 1.9 1.3 1.0
 1 1.4 1.1 0.9
 2 1.4 1.2 0.9
 3+ 1.4 1.2 1.0

Age of youngest child
 Under 6 1.4 1.1 0.8
 6-13 1.4 1.3 1.1
 14+ 1.5 1.5 1.5

All 1.8 1.4 1.0

Maori were 1.8 times more likely than European entrants to experience a long total duration. The difference was most pronounced for those aged under 30. This may reflect the existence of fairly widespread experience of benefit income for at least some time within both populations in this age group, but very different labour market experiences associated with this receipt. Members of the European ethnic group, for example, appear more likely than Maori to have received Emergency UB as tertiary students. This group had a much lower risk of a long total duration on benefit than UB recipients overall.

Some caution needs to be exercised in interpreting these and other associations that are found. There is good reason to expect that the duration experience of Maori is partly explained by their educational status and their greater tendency to be located in small towns in economic decline. Were we able to include these factors, and were we to control for them in a multivariate analysis, the strength of the association between being Maori and having long total duration would almost certainly be reduced.

Pacific Islands entrants were 1.4 times more likely than their European counterparts to have a long total duration on benefit. The difference for those with only one child and those with a youngest child aged under six was not marked however. When we examine the experiences of sole parent DPB entrants only, Pacific Islands entrants with a very young child at grant were in fact less likely than their European counterparts to have a long total duration. These differences may be partly explained by the relatively high rates of full-time employment among Pacific Islands sole mothers with younger children. Thirteen per cent of Pacific Islands sole mothers with a child aged under five were in full-time employment at the 1996 Census compared with an all ethnic groups average of 11%.(19) This may be associated with a high propensity to reside in multi-family households,(20) perhaps offering Pacific Islands sole parents greater access to informal childcare.

Overall, there was little difference in risk between members of other ethnic groups and Europeans. However, members of other ethnic groups aged 60 or over at entry, together with Pacific Islands peoples in this age group, were many times more likely than Europeans to experience a long total duration on working-age benefits. This can be explained by a higher likelihood that members of these ethnic groups did not meet the residency criteria for New Zealand Superannuation.


Long-term receipt of benefit income is more common than examination of data on single spells of benefit receipt would suggest. For most who began receiving benefit in 1993, this spell was part of a longer benefit history involving multiple spells and, in many cases, multiple benefit types. While few had a first observed spell that lasted more than a year, when all spells are counted two thirds spent a total of at least one year out of the following five on benefit, and one third spent a total of at least three years of the five on benefit.

The total elapsed time over which people continue to cycle in and out of income circumstances that bring them into contact with the benefit system appears considerable: 35% of the 1993 entrants overall and 57% of the DPB entrants were in receipt of benefit on the five-year anniversary of their 1993 grant. Few received benefit without interruption however. Most of those whose total duration exceeded three of the five years studied had periods off benefit. The issue for many appeared to be maintaining independence from the benefit system, rather than finding opportunities to gain that independence.(21)

Those who experienced a long total duration through continuous or intermittent receipt accounted for the bulk of the weeks that the 1993 entrants spent on benefit in the follow-up. The third that spent three or more years on benefit accounted for two-thirds of the total weeks. This may have implications for the design and level of investment in interventions aimed at reducing duration. If we could identify the long duration group early in their benefit history and turn that history around, the benefit savings would be great, possibly sufficient to support a greater up-front investment in assisting people to secure sufficient incomes to become and remain independent of the benefit system than has been contemplated in the past.

A wider analysis than that contained in this paper is required to determine whether or not this is a sensible strategy. For a start, further work is needed to establish whether it is possible to predict total duration on benefit and to identify at entry those with the greatest risk of long-term benefit receipt. The patterns found in this paper suggest that age, partnership status, the presence and age of children, ethnicity and sex are factors that should be explored further for each benefit type. Without further work it remains unclear whether these and other variables available from the benefit dynamics data set are together sufficient to account for the benefit experiences observed. In addition, work would also be needed to establish the extent to which the benefit experiences of this group can be altered and the costs of the interventions required. It may be that those with very long durations are the hardest to move and that interventions targeted at those with shorter benefit durations are more cost effective.

Results of future work on the benefit dynamics data set may also have implications for the timing of interventions aimed at reducing the duration of benefit receipt. Currently interventions tend to be targeted on a "wait and see" basis, with access (in the case of many facilitative measures) or application (in the case of some reciprocal obligations) requiring a minimum period of continuous receipt in a single spell on benefit. The reasoning underlying this approach is that it allows the "deadweight" associated with intervention to be minimised as short-term recipients who need no assistance will move off benefit quickly. Bane and Ellwood (1994) point out that such an approach may make less sense when repeat spells are a strong feature of benefit use and targeting on total duration rather than the duration of a single spell is considered. A one- or two-year wait and see period calculated over multiple spells of receipt might exclude from eligibility only a very small proportion of the population, and fail to significantly reduce the average remaining duration of benefit use, while in the meantime eroding the potential benefit savings. If the policy goal is to reduce total duration on benefit, waiting to intervene may be less cost effective than intervening early.(22) Whether this observation holds for the different benefit types in New Zealand could be investigated further.

This paper is very much a first look at the updated benefit dynamics data set. Further work exploring the interactions between individual, economic and policy factors and the benefit experiences that are observed for successive cohorts of recipients is planned.


The Table below sets out two possible broad estimates of the number of different individuals present in the working-age population at some time in the six-year period spanning 1 January 1993 to 31 December 1998.

The first (B) is calculated by taking the population in the working-age population as at the end of the study period ((A), approximated by the provisional as at 31 December 1998 estimates of the NZ resident population -- at the time of writing these were the most recently available single year of age estimates). Added to this figure are the numbers of people in the six single year of age groups who were members of the working-age population at some point in the six year study period but had aged out of it by the end.

This estimate: assumes that all permanent arrivals during the period remained in New Zealand continuously after arrival (and are therefore captured by the end of period count) and takes no account of the impact of outward migration and mortality on the to total number of different individuals present.

The second (C) is calculated by adding to (B) the number of permanent departures from New Zealand aged 15-64. This estimate:

assumes that all permanent arrivals during the period remained in New Zealand continuously after arrival (and are therefore captured by the count A);

assumes that all permanent departures during the period had resided continuously in NZ before departure and remained absent continuously after departure;

assumes that all permanent departures are recorded as such -- no allowance is made for temporary departures that become permanent; and

takes no account of the impact of mortality on the total number of different individuals present.

Viewing these broad indications of scale suggests that the proportion of the working-age population that received a main benefit to help meet their income needs at some time over the six year period could be around four in ten.
A Estimated NZ resident population aged 15-64 2,488,030
 at end of the study period - 31 December 1998

 A Plus six single year of age groups who aged
 out of the 15-64 population in the six years
 previous (enumerated at 31 December 1998):
 66 year olds - aged out 1 year previous 26,040
 67 year olds - aged out 2 years previous 26,390
 68 year olds - aged out 3 years previous 26,560
 69 year olds - aged out 4 years previous 26,760
 70 year olds - aged out 5 years previous 25,740
 71 year olds - aged out 6 years previous 25,460

B 2,644,980

 B Plus total number of permanent departures
 from 15-64 age group in six years of the
 study period:
 1993 33297
 1994 36038
 1995 39616
 1996 43738
 1997 48835
 1998 52485

C 2,898,989

Source: Statistics New Zealand Demography Section, Estimated New Zealand Resident Population, by Sex and Single-Year-of-Age, As At 31 December 1998; INFOS, Total Permanent and Long Term Departures by Age Group, Quarterly Series.

(1) My thanks to Michael Ryan of Statistics New Zealand, Dr Garry Barrett of the Australian National University, and Ron Lovell, Stuart Irvine, Kay Goodger and Michelle Gosse of the Ministry of Social Policy for their helpful comments on earlier drafts. Any errors or omissions are my own. The views expressed are not necessarily those of the Ministry of Social Policy.

(2) The original data set was constructed as part of a joint project undertaken by the Social Policy Agency and the Treasury and covered the period 1 July 1992 to 1 June 1996. Construction and analysis of the updated data set was greatly assisted by the groundbreaking work carried out on the original data set by Sonia Chen, Stuart Irvine and Ron Lovell.

(3) Supplementary benefits and New Zealand Superannuation are not included in the data set.

(4) See De Raad (1997) for earlier findings on the dynamics of unemployment benefit spells based on a sample of 7,000 drawn from the original benefit dynamics data set.

(5) See De Raad (1997) for more full discussion of the merits and limitations of administrative records as a source of longitudinal data.

(6) The unemployment dynamics data set, developed by the Labour Market Policy Group of the Department of Labour, holds information on education and work history collected as people enter the job seeker register. Prospects for including this information in the benefit dynamics data set by linking with the unemployment dynamics data set are to be explored further.

(7) Note that spells on benefit can be redefined to suit the demands of a particular analysis. For example, a spell could be defined as a period of continuous receipt of any benefit income. In this case a transfer between benefits would be treated as a continuation of the original spell.

(8) This is intended to remove some of the administratively generated cessation and re-grant activity that occurs when clients fail to make contact when requested. It may have the unintended effect of removing some short spells off benefit that occur for other reasons.

(9) From November 1996 onwards, recipients of IB, WB and DPB have been able to elect to receive benefit on a weekly rather than two-weekly basis.

(10) For example, a transfer from Young Job Seekers Allowance to the Statutory Unemployment Benefit is treated as a continuation of a single spell, but a transfer from the Student category of Emergency Unemployment Benefit to the Statutory Unemployment Benefit is treated as the commencement of a new spell.

(11) The TB entry cohort is affected by a similar bias -- those who were aged under 20 at their first grant are slightly over-represented as they are less likely than other TB recipients to have come onto TB by way of transfer from UB.

(13) Note that to simplify the distributions the average hazard rate for four-weekly intervals ending at the points shown has been plotted.

(14) Ford et al. (1996) found that a members of a cohort of UK lone parents had an increasing likelihood of reporting long-standing or limiting illness or poor health over time. Children developing health problems was strongly related to the experience of hardship.

(15) See McLaughlin et al. (1989).

(16) See Fletcher (1995).

(17) This category was particularly large for TB entrants, a large proportion of whom have reason for cessation coded to "left TOP course" or "non-return of declaration/renewal", "non-renewal", or "not registered with NZES/registration lapsed" codes. The other/unknown category also accounted for a third of cancellations for Sickness Benefit entrants, a large proportion of whom have reason for cessation coded as "lack of medical coverage".

(18) See Barker and Maloney (1998) for a multivariate analysis of the duration of single spells of UB receipt based on a sample from the original benefit dynamics data set. No multivariate analysis of the likelihood of repeat benefit spells or the total duration on benefit has been carried out using the data to date.

(19) Statistics New Zealand, 1996 Census, unpublished tables.

(20) See Davey (1998).

(21) See Bane and Ellwood (1994) and Barrett and Cragg (1998) for discussion of similar patterns of repeat use found


Bane, M.J. and D.T. Ellwood (1994) Welfare Realities: From Rhetoric to Reform, Harvard University Press, Cambridge MA.

Barker, G. and T. Maloney (1998) An empirical analysis of welfare dynamics using a NZ tax welfare database, Annex 6 in Final Report on the Feasibility Study into the Costs and Benefits of Integrating Cross-sectoral Administrative Data to Produce New Social Statistics, Statistics New Zealand, October.

Barrett, G.F. and M.I. Cragg (1998) "An untold story: The characteristics of welfare use in British Columbia" Canadian Journal of Economics, 31(1):165-188.

Davey, J. (1998) "New Zealand children in households and families: An update" New Zealand Population Review, 24:103-118.

De Raad, J.P. (1997) The Duration and Repeat of Unemployment Benefit Spells: A Study A Cohort of Unemployment Beneficiaries 1992-1995, thesis submitted for the degree of Master of Arts (Honours) in Social Policy, Department of Sociology and Social Policy,

Fletcher, M. (1995) "Changes in registered unemployment: Duration and flows" Labour Market Bulletin, Issue 1.

Ford, R., A. Marsh and L. Finlayson (1996) What Happens to Lone Parents: A Cohort Study 1991-1995, Department of Social Security Research Report No. 77, The Stationary Office, London.

Gardiner, P. (1995) "An analysis of exit rates and duration dependence in registered unemployment" in Labour, Employment and Work in New Zealand, Proceedings of the Sixth Conference, 24 and 25 November 1994, Victoria University of Wellington.

Kiefer, N.M. (1988) "Economic Duration Data and Hazard Functions" Journal of Economic Literature, 26: 649-679.

McLaughlin, E., J. Millar and K. Cooke (1989) Work and Welfare Benefits, Avebury, Aldershot.

RELATED ARTICLE: Invalids Benefit (IB)

Sickness Benefit/Community Wage -- Sickness (SB) includes Sickness Benefit, Emergency Sickness Benefit, Community Wage -- Sickness and Community Wage -- Emergency Sickness

Unemployment Benefit/Community Wage -- Job Seeker (UB)

* Unemployment Benefit, which includes Statutory Unemployment Benefit, Young Job Seekers Allowance, 55+ Benefit and Community Wage -- Job Seeker

* Job Search Allowance

* Independent Youth Benefit

* Emergency Unemployment Benefit -- Students and Community Wage -- Emergency Student

* Emergency Unemployment Benefit

* Community Wage -- Emergency Job Seeker

Training Benefit/Community Wage -- Training (TB) includes Training Benefit, Community Wage -- Training and Community Wage -- Emergency Training

Domestic Purposes Benefit (DPB)

* DPB for sole parents, which includes DPB Basic and Emergency Maintenance Allowance

* DPB woman alone

* DPB caring for sick and infirm

Widows Benefit (WB)

Moira Wilson(1) Ministry of Social Policy
COPYRIGHT 1999 Ministry of Social Development
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1999, Gale Group. All rights reserved. Gale Group is a Thomson Corporation Company.

Article Details
Printer friendly Cite/link Email Feedback
Author:Wilson, Moira
Publication:Social Policy Journal of New Zealand
Geographic Code:8NEWZ
Date:Dec 1, 1999

Related Articles
The prevalence and persistence of low income among New Zealand children: indicative measures from benefit dynamics data.
Children on benefit: who stays longest?
Understanding the growth in invalid's benefit receipt in New Zealand.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters