Impact assessment of the lone parent pilots

1
Impact assessment of the lone parent pilots

• Project team
• IFS Mike Brewer, James Browne, Claire Crawford
  Lorraine Dearden. PSI Genevieve Knight

2
Main findings outline of talk

• Participation
• after 12 months of pilots, 6 of eligible LPs
  received IWC
• But participation still rising in Phase 1 areas
  after 20 months, so impact may grow
• Impact
• Small impacts on moves off benefit and into work,
  particularly for lone parents recently on NDLP
• Rest of talk
• Methods data
• What might we expect?
• Results
• Concluding thoughts
• All based on DWP Research Report 415
  (http//www.dwp.gov.uk/asd/asd5/rports2007-2008/rr
  ep415.pdf).

3
What are we trying to achieve?

• Estimate impact of lone parent pilots on labour
  market outcomes of lone parents previously on IS
• Use administrative data
• benefit receipt from DWP, employment spells from
  HMRC (WPLS)
• Evaluation will tell us how much, not why

4
What are the lone parent pilots?

• In Work Credit (IWC)
• 40 weekly payment on entering work of 16 hrs a
  week, for max of 12 months, if been on IS for 12
  months
• Quarterly WFI (QWFI)
• mandatory quarterly interview for lone parents
  whose youngest child aged 12 or over. Operated in
  LEAs operating ESC.
• Work Search Premium (WSP)
• payment of 20 a week, for max of 26 weeks, in
  exchange for undertaking job search activities
• New Deal plus for Lone Parents
• From April 2005, some districts operating the
  LPPs became NDfLP as above plus extra money and
  services

5
Why use time-limited in-work benefits?

• Wage growth (Connolly Gottshalk, 2006 Walker
  and Lydon, 2005)
• One-off costs of starting work?
• Very high discount rate?
• Habits?
• Related policies
• SSP (Canada). Reduced welfare claims, increased
  employment, increased earnings, increased income
  BUT no long-run effects (Card and Hyslop, 2005
  Connolly and Gottshalk, 2002)
• UK experience Employment credit for older
  workers ERA Pathways to Work pilots

6
Work and Pensions Longitudinal Study (WPLS)

• Administrative data-set combining
• DWP all benefit claims and participation in NDs
  from June 1999
• HMRC data from all P45/P46s (filled in when
  employers start/stop paying someone)
• Matched on NINO plus name, DOB, gender, postcode
• Personal information age, gender, ethnicity and
  postcode
• but can merge other characteristics from other
  DWP databases.
• Outcomes
• Which benefits/programmes
• Whether in work
• Doesnt tell us earnings (yet), hours worked, or
  hourly wage

7
WPLS problems with work measure

• Lots of noise
• Some entries correspond receipt of taxable state
  benefit
• Multiple entries for (apparent) same job
• Multiple entries with same start date, different
  end date
• Jobs where start or end known approximately
  (year, but not day)
• Jobs where only end-date known
• Jobs at times inconsistent with benefit receipt
• Need not include jobs paying lt tax threshold
• Tax threshold 91. Min wage 16 hours 72
  (April 2004).
• Does not capture self-employment nor informal
  employment (but neither would be eligible for
  IWC)

8
What outcomes are we measuring, and for whom?

• Measure impact on all eligible for IWC
• Future work will measure impact on job retention
  for IWC recipients
• Divide eligible lone parents into stock and
  flow
• Stock eligible for IWC when pilot starts (large
  sample)
• Flow become eligible after pilot starts (more
  interesting in long-run)
• Outcomes measured in WPLS for people in WPLS (!)
• whether off IS/JSA/IB X days after first
  potentially eligible for IWC
• whether in work X days after first potentially
  eligible for IWC
• Benefit outcomes until 31/3/06, work outcomes
  until 30/9/05
• Work measure in WPLS based on employers telling
  HMRC when they start/stop paying an employee

9
Method

• Difference-in-differences
• Compare outcomes in LPP areas with other areas
  after LPPs started
• Compare outcomes in LPP areas with other areas
  before LPPs started
• Attribute any differences to LPP
• No sensible control group within the pilot areas
• People without children on JSA ?
• So use lone parents in all other parts of England
  as control areas, and estimate impact with
  difference in differences
• Dont identify matched control areas (Blundell et
  al (2005))
• Differences between pilot and control areas

10
Empirical specification

• Outcomes
• off benefit / in work X days after eligibility.
• Explanatory variables
• History of benefit receipt and work (30 months
  before eligibility)
• Whether claimed disability benefit, JSA, or been
  on NDLP in 30 months before eligibility,
  entitlement to IS (at start of claim)
• Personal characteristics (when first eligible)
• Age, number of children, age of youngest
  children, ethnicity, gender, month first
  eligible.
• Area characteristics (based on postcode when
  first eligible)
• Indicators for JC district, supply of formal
  childcare (ward, 2003/4), unemployment (TTWA,
  2002/3), deprivation quintile (SOA, 2002/3),
  qualifications of non-working lone parents (SOA,
  2001), employment rate (SOA, 2001), of lone
  parents who are owner-occupiers (SOA, 2001)
• No time trend, but indicators for month
• Linear probability model (ie OLS on binary
  outcome)
• Estimate impact
• Across all districts (flow only for stock,
  separate regression for each phase)
• For each phase
• For each district
• By individual characteristics

11
In Work Credit detail

• Gradual roll-out
• Phase 1 (Apr 2004) Bradford, N London, SE London
  
• Phase 2 (Oct 2004) Leicestershire, Dudley and
  Sandwell, W London, Lancashire W, Staffs, Leeds
  ( Cardiff Edinburgh)
• Phase 3 (Apr 2005) Brent, City E London, S
  London, Lambeth
• Phase 4 (Oct 2005) Surrey, Sussex, Essex, Kent,
  Hampshire, IoW, Berks, Bucks, Beds, Herts (not
  covered)
• Five districts also have extra spending for
  personal advisers (NDfLP)
• Affects around a third of LPs
• All of London and south-east in pilot areas. Argh!

12
Timeline and sample
Apr 02 Mar 04 Apr 04 Oct 04 Oct 04 Mar 05 Apr 05 Mar 06
Phase 1 Unaffected Affected Affected Affected
Phase 2 Unaffected Unaffected Affected Affected
Phase 3 Unaffected Unaffected Unaffected Affected
Phase 4 Ignored Ignored Ignored Ignored
Other areas Control Control Control Control
13
Sample size how many are potentially eligible?
First eligible after pilot started
Stock Flow Flow
Stock (outcomes after 6 months) (outcomes after 12 months)
Phase 1 42,374 14,480 10,316
Phase 2 77,290 16,928 8,281
Phase 3 71,705 7,152 211
Total 191,369 38,560 18,808
Eligible on day pilot started
14
Typical profile flow
Off benefit
In work
15
Typical profile stock
Off benefit
In work
16
Typical profile lessons

• Eligible population is prone to long spells on
  benefit
• After 12 months, 15-20 of flow (10-15 of stock)
  are off benefit
• Work measure looks too high, but changes are
  more plausible.
• After 12 months, 5-10 ppt more are in work
• Pilot areas have worse outcomes than control
  areas, particularly Phases 1 3 (London)

17
Take-up (1) ever received IWC as ever
potentially eligible
Corrected, 25/10/06
Numerator DWP financial data (stops Nov 2005).
Denominator WPLS
18
Take-up (2) new IWC claims as of benefit exits
and of job starts
Numerator DWP financial data. Denominator WPLS
19
Results flow
Control
Off benefit after 9 months
Phase 1
Phase 2
Phase 3
Differences
Date first (potentially) eligible to LPPs
20
Results flow
Impact of LPP (ppts) 3 months 6 months 9 months 12 months
1 Benefit Work -0.7 0.2 -0.3 0.4 -0.4 0.8 -0.3
2 Benefit Work 0.2 0.3 0.5 0.8 0.9 0.7 0.8
3 Benefit Work -0.4 -.0.1 -0.4 -0.3
All areas -0.2 0.2 0.1 0.6 0.3 0.7 0.3
Bold and italicised means statistically different
from zero
21
Results stock (phase 1)
Off benefit
Days since eligibility to LPPs
22
Results stock (phase 2)
Off benefit
Days since eligibility to programme
23
Results stock (phase 3)
Off benefit
Days since eligibility to programme
24
Results stock
Impact of LPP (ppts) 6 months 12 months 18 months 24 months
1 Benefit Work 0.0 0.2 0.5 0.2 0.8 0.9 1.2
2 Benefit Work 0.0 0.2 0.3 0.6 0.6
3 Benefit Work 0.7 1.1 1.2
Bold and italicised means statistically different
from zero
25
Impact by subgroups

• Number of children, gender
• No consistent pattern few significant
  differences
• Age of youngest child
• Weak (statistically insignificant) evidence that
  response greater where youngest child age 3 or
  more
• Recent participation in NDLP
• Evidence that impact greater for LPs who have
  recently been on NDLP, but estimate not
  consistent/stable across districts
• By district
• Considerable variation
• IWC vs NDfLP areas
• Evidence that NDfLP areas have WORSE outcomes

26
Results flow, by age of youngest child
Off benefit after 9 months
(Results similar amongst stock)
27
District-level impacts flow
Sig diff from 0
Average (not sig diff from 0)
Sig diff from 0
28
District-level impacts flow
Sig diff from 0
Average IWC (sig diff from 0)
Average NDfLP
Sig diff from 0
29
District-level impacts stock
Sig diff from 0
Average
30
District-level impacts stock
Sig diff from 0
Average IWC (sig diff from 0)
Average NDfLP
31
Impact by recent NDLP participation
Sig diff from 0
Off benefit after 273/364 days
Recent On NDLP 6-12 months before eligible
to LPP Past On NDLP 13-30 months before
eligible to LPP
32
Summary of results

• Impact
• Small impacts on flows off benefit, particularly
  for lone parents recently on NDLP, and easier to
  detect in stock sample than flow
• Participation
• After 12 months, 6 of LPs have received IWC
• But participation still rising after 20 months,
  so impact may grow
• Why impact so small?
• Either LPs dont hear about IWC, or they arent
  responding to it yet
• Is 0-2 ppts small? Level without treatment is
  15-20

33
Problems/extensions

• Common trends
• So far, estimated many unrelated regressions
• What gain would there be from estimating a
  duration model with time-varying treatment?
• Pilot and control areas are different
• Matched difference-in-differences (Blundell et al
  NDYP)
• Not yet used data on receipt of IWC
• Joint model of NDLP IWC flows off benefit
• Impact on retention
• Technical report due early 2008, future reports
  on more data late 2008 2010.

34
End