Download - Heads you’re in; tails you’re out: How RCTs have evolved in DWP Jane Hall [email protected]
Background
Increasing use of RCTs
Emphasis on evidence-based policy
Limited UK experience in social policy arena
Practical lessons not theoretical debate
Overview
Chronology of RCTs in DWP Site selection and preparation Identifying the eligible population Dealing with resistance Performing the random assignment Monitoring take-up
Chronology of RCTs
Restart Various New Deals Employment Zones JRRP ERA JSA Intervention Pilots ND50+ Mandatory IAP
Site Selection and Preparation
Need commitment from the top
All parties need to buy-in
Set-up is resource intensive
Personal visits
Pilot the approach
Identifying and recruiting the eligible population
Can they be easily identified Self-selection Suitability of the population Monitor P & C Group characteristics Selling techniques Sample sizes: Sub-group analysis
Dealing with resistance
Busting the myth
Significant investment in training at all levels
Aides and FAQs
Scripts
Performing the Random Allocation
Needs to be sophisticated Not open to sabotage/gaming Block allocation: Maintain P:C ratio Different techniques
– NINO– Call Centre – On-line algorithm– Random numbers
Monitoring take-up
Keep track of P & C Group
Ensure only P Group receive the treatment
Monitor key characteristics of P & C Group
Be prepared to redesign the random allocation
Expect the Unexpected
Results may not be what you anticipate
A fair allocation of resources?
Participation rates can be disappointing
Operational ChallengesThe ERA Experience
Jenny Carrino
Overview
The ERA Policy Key Challenges
– Random Assignment (RA) Process– Customer Understanding of RA– Creation of ‘Informal’ Refusers– Jobcentre Plus Target Structure– Technical Assistance
The ERA Policy
To test interventions to improve retention and advancement– Adviser support– Funding for training– Financial Incentives
6 Jobcentre Plus districts Three customer groups
NDLP ND25+ WTC To test the effectiveness of using RA to evaluate social
policy in the UK
Random Assignment 1
Issue: The random allocation process Lessons Learnt
– Importance of transparency in the allocation process– Avoiding contamination
Outcomes– Most customers and staff viewed random assignment
as fair and justified
Random Assignment 2
Issue: The Informed Consent Process Lessons Learnt
– Standardisation - adviser scripts and leaflets Outcomes
– Not everyone fully understood what they had signed up for
– Too much information at initial interviews – conduct RA as a stand alone interview
Random Assignment 3
Issue: Creation of a group of ‘informal’ refusers
What do we mean by Informal Refusers?
ELIGIBLE POPULATION
CUSTOMERS RA’D OR ON SYSTEM
FORMAL REFUSERS
INFORMAL REFUSERS
SIGNIFICANTLY DIFFERENT
Informal Refusers
Why this happened– The decision to use RA – RA to ERA was voluntary– Influences from both advisers and customers
Outcomes– Creation of a ‘third’ group– Analysis to identify whether this group are different to
the ERA population Lesson Learnt
– If possible monitor intake closely against eligible population
Ensuring a Treatment - Targets
Issue: The Jobcentre Plus Target Structure– Some adviser behaviour negatively affected – Senior management buy-in affected
Lessons Learnt– Policies need to reflect the organisations reward
system– Need to be able to monitor and feedback to
implementation managers
Technical Assistance
Issue: Ensuring the effective delivery of RA Lessons Learnt
– US model of on-site RA assistance– Avoiding contamination– Monitoring Performance
Outcomes– Advisers felt supported during the RA period – Initial confusion over the role of TAs– Some districts deferred responsibility of ERA
implementation
Summary of Key Challenges
RA Process Informed Consent Creation of informal refusers Ensuring a Treatment Providing effective support to delivery agents
Jobseekers Allowance (JSA)Intervention Pilots
Jayne Middlemas
JSA Intervention Pilots
JSA Intervention Regime The Pilots Evaluation Random assignment The data Did Random Assignment work? Results
JSA Intervention Regime
First Contact New Jobseeker Interview (NJI) Financial Assessor Interview Fortnightly Jobsearch Review (FJR) 13 week review
The Pilots
Introduced in January 2005 108 Jobcentres in 10 Districts took part Each Jobcentre took part in a single pilot Aim to deliver resource savings on the FJR without
reducing unemployment off-flow rates.
The pilots (cont)
Five different approaches: Excusal of signing for first 13 wks of claim Excusal of signing for first 7 wks of claim Telephone signing Shortened FJR Group signing
The pilots (cont)
Some groups excluded: Part-time workers 16 and 17 year olds People with no fixed abode People known to have had a fraudulent claim in the past
Evaluation
Customers randomly allocated Work study to record resources used Comparison of off-flows Qualitative evaluation
Random Assignment
50% programme, 50% control
ORC International Call Centre
Two methods: Adviser calls immediately prior to each NJI Jobcentre calls at start of day with details of all clients
due to attend an NJI that day
ORC also provided random call-in date
Data
Data collected during random assignment JUVOS data – derived from the Jobseekers Allowance
Payment System (JSAPS) HMRC Employment Data
Did Random Assignment Work?
66,600 randomly assigned 33,100 programme & 33,400 control All pilots and Districts close to 50/50 split
Was everyone assigned? Difficult to answer precisely Number randomly assigned around 90% of total new
claims. Excluded groups likely to account for 8 to 12%
Were People Wrongly Assigned?
19% had no new JSA claim during the pilot Incorrect NI numbers may mean we can’t find some
claims Jobcentres didn’t always inform us of those who failed to
attend Can’t identify excluded groups in the data
Internal Validity
Compared characteristics for programme and control groups
Very little difference was found by gender, age or ethnic origin
Concluded that the control group is well suited to providing a counterfactual for the programme group
External Validity
Pilot Jobcentres account for small proportion of all new JSA claims across the country
Gender, age & ethnicity of new claimants in pilot areas different to country as a whole
Some difference in local unemployment rates Weighted results to take account of differences
Results
13 week excusal pilot
% still claiming after x weeks
0
20
40
60
80
100
0 2 4 6 8 10 12 14 16 18 20 22 24 26
Weeks
Per
cen
tag
e
Control Programme
Length of Claim
0
0.2
0.4
0.6
0.8
1
1.2
0 28 56 84 112 140 168
Length of claim (days)
Per
cen
tag
e
Control Programme
Results
Average length of claim is 5.9 days greater in programme group than in control group
Weighting the results to be representative nationally suggests an increase in 6.1 days in average length of claim
No difference in the proportions who moved into work
Results (cont)
4 reasons for difference in length of claim: Some people take longer to find work Some people take longer to tell us they have found work Some control group customers fail to attend and have to
start a new claim Some people fail to sign off for other reasons
Results (cont)
Work study provided estimates of savings Extra benefits paid as a result of increase in average
length of claim exceed savings Qualitative evaluation suggested that the pilot was
implemented well Customers were happy not to have to attend every
fortnight
More information
DWP Research Report 300: The Qualitative Evaluation of the JSA Intervention Regime Pilots
DWP Research Report 382: Jobseeker’s Allowance Intervention Pilots Quantitative Evaluation
Available on DWP Website: http://www.dwp.gov.uk/asd/asd5/rrs-index.asp
Job Retention and Rehabilitation Pilot
Lessons learnt in running an RCT
James Holland
Structure
1. Background to JRRP
2. Results of the trial
3. Hypotheses
4. Conclusions: Importance of complementary methodologies
Design
4 - way trial To test the effectiveness of a person centred case
management approach and increased range of treatments in helping people retain work– Health care focused– Workplace focused– Combined health care and/or workplace focused– Control group
Four service providers in six parts of the country Participants were people off work sick and unlikely to
return to work without help
Routes through the trial
Project Marketing
Approach Contact Centre
Contact Centre
Explanation
Eligibility
Screening
Decline
Ineligible
Screened out
Screened in: Randomisation
Health 25%
Work 25% Combined 25%
Control 25%
Providers make contact
Written Consent
Assessment and Intervention
Return to work
Out of work
Surveys
Evaluation Design
Impact and process evaluation Cost benefit Analysis Components of the evaluation
– Survey of those screened out and the control group– Outcome survey– Panel study – Focus studies– Database of contacts and treatments– Costs exercise
Impact Measures
Primary impact measure – 13 week return to work
Secondary impact measures– Health
– Household income Costs and benefits Operation of JRRP as a RCT
Results
13 Week Return to Work
Intervention group %
Health 43.5
Workplace 45.1
Combined 44.4
All interventions 44.4
Control 44.7
Results
6 Week Return to Work
Intervention group %
Health 55.7
Workplace 56.4
Combined 56.5
All interventions 56.2
Control 53.0
Results
2 Week Return to Work
Intervention group %
Health 61.5
Workplace 61.4
Combined 62.1
All interventions 61.7
Control 59.3
Results
Positive impact among those off work because of an injury
Negative impact among those off work because of a mental health condition
Unaffected return to work rates for those with other health conditions
Positive impact on health, particularly mild depression
A RCT can work in a voluntary labour market setting
Hypotheses for findings
1. The interventions were too weak
2. The interventions were delivered in an unhelpful way
3. Too many external barriers
4. Withdrawal rate too high
5. The self-selecting participants were the ‘wrong group’
Missing evidence
Evidence of problems mostly drawn from qual research
This generates hypotheses/explanations but does not allow for quantification
Biggest gap is (quant) understanding of behaviour of control group
In retrospect, needed data on self-motivation and better understanding of participant/provider interaction
Thoughts on how to do it better
Set out possible scenarios at start Early qual research on behaviours of participants
and control group Early impact estimates so that later research can
be adapted
Conclusions
RCTs are the gold standard programme evaluation
But a number of problems, practically and methodologically
Need supplementing with a good quality process evaluation
Contact Details
James Holland
Disability and Work Division, DWP
0114 209 8280
Reports available via www.dwp.gsi.gov.uk/asd/asd5