Incentive Effects of Social Assistance:

1
Incentive Effects of Social Assistance

• A Regression Discontinuity Approach

2
Motivation

• Sharp Discontinuity in the social Assistance
  Policy
• Substantial disparity in the receipts of
  social assistance between the childless
  recipients under the age of 30 and those that
  are above 30 years old.
• The Elimination of this Discontinuity in August
  1989.

3
Why is it Motivating?

• The break in the policy natural setting for
  evaluation using a regression discontinuity.
• Estimates credible as those of randomized
  experiment.

4
Objective

• Estimate the incentive effect of social
  assistance (using a regression discontinuity
  approach ) on
• 1- Various labor market outcomes.
• 2- living arrangements outcomes ( other)

5
Locating the Paper in the Literature

• Fortin Al(2004) used this policy experiment to
  estimate the effect of social assistance on the
  duration of social spells (DD approach before and
  after the reform)
• Dooley(1999), Dooley al.(2000) no relation
  between sex of HH-head SA benefits.
• Ford(2003), Cragg(1998), Warburton(2004) strong
  effect of earnings supplement on labor supply.
• Lee (2005) argues that the regression
  discontinuity estimates are as credible as
  those from randomized experiments
• Van der Klaauw (2002) ,Black (1999) Etc
• Between 80 and 90 many reforms approach was
  experimental treatment and control group

6
Contribution to the Literature

• This paper uses a first difference estimator in
  the context of a regression discontinuity
• This paper focuses on social assistance benefits
  on labor market behavior of men without children

7
Main Conclusions

• Evidence that generous social assistance benefits
  reduces employment.
• DD estimators may perform poorly if the control
  group is not chosen adequately.

8
Social Assistance in Quebec

• Singles and non-parents are eligible.
• If agelt30 then low benefit (187) otherwise
  (507).
• August 1st new Act? the discontinuity removed

9
Data and Descriptive Statistics

• Focus on male non-parents High School drop outs
• Census Master files
• - give employment information on the week
  prior to survey.
• -no separate item for the SA (approx. 85 of
  transfers Lemieux Milligan 2004).
• Labor Force Survey (labor marker context few obs.
  around the cut off point)

10
Employment trends
11
Main Graphical Conclusions

• Employment is cyclical ?need to have control
  group.
• Two ages groups track each other (roC) ? ages
  groups comparable.
• Two ages groups roughly dont track each other
  (QC) ? SA program lt 30.

12
Empirical Approach The model

• Model I ERC
• This model represents a Sharp Regression
  Discontinuity.
• Uses Employment rate at the census week (ERC).
• d(a) is continuous but not necessarily linear?
  only differential benefits are the sources of
  discontinuity in outcomes around age 30.
• Ya is average outcome by age group (parametric
  assumption to extrapolate)

13
Empirical Approach The Model cont.

• Model II ERL
• This model represents a Fuzzy (not sharp)
  Regression Discontinuity.
• Uses Employment rate based on the fraction of
  weeks worked in previous year (ERL).
• ? Possible problem
• Under reporting Recall bias.

14
Comparison S-D F-D
15
Comparison S-D F-D

• The employment drops abruptly at the cutoffs.
• For the census week the drop happens between the
  age of 29 and 30
• For the decline in the employment measured in the
  previous year the drop happens between the age of
  30 and 31.

16
Empirical Approach The Model cont.

• If the models are well specified ß1 ß1
• Model III ERC-ERL First Difference estimator
  (FD)
• ERC-ERL
• (ß0 ß0) ß1(Ta-Ta) ?(a) (ea- ea)
• This FD estimator
• - Exploits the longitudinal nature of the
  information.
• - Captures the change in employment of the same
  group age at ages of 29 and 30.
• - Individual-specific fixed effects are
  eliminated

17
Importance of the Model

• People in control and treatment group are subject
  to the same environment (in QC)
• People around the discontinuity have the same
  characteristics (comparing what is comparable).

18
Regression Discontinuity estimates

• Method of estimation is weighted OLS.
• Many specifications for the d were adopted.
• The models estimated are S-RD, F-RD and the
  FD-RD.
• Also look at the effect of hours of work in the
  census week.

19
Regression Discontinuity Estimates
20
Regression Discontinuity Estimates

• Results are consistent with simple labor market
  supply all of the impact is on participation
  rather than the hours of work conditional on
  employment.

21
Robustness of the Results

• ? Narrowing the window
• ? Results robust to narrowing the window
• ? Falsification Test
• ? Robust especially in the case of the Fuzzy RD
  (Star-economics!!)
• ? Broadening the target Group (All men).
• ? Smaller treatment effect.

22
Robustness of the Results

• ? Selection Bias
• - due to selecting the men w/o children
• - If getting children and living with them is
  related to the SA benefits then RD not
  appropriate.
• - If there is a manipulation then there should
  be a discontinuity of the fraction of men with
  children around the discontinuity point
• ? No evidence of discontinuity

23
Comparing RD and DD in DD

• Proposition DD approach requires a control group
  placed in the same labor market as the treatment
  group.
• Proof use three different DD estimators
• 1- 29 vs 30 (Qc vs RoC) for 86
• 2- 29 vs 30 (86 vs 91)
• 3- 29 Qc (86 vs 91) vs 29 RoC (86 vs 91)

24
Comparing RD and DD in DD

• First DD approach
• - Estimator unbiased because age trend is same
  in Qc and RoC
• - DD estimator similar to Discontinuity estimator

25
Comparing RD and DD in DD

• Second DD approach
• - Estimator may be biased recession of 1991 had
  asymmetrical effects on age groups
• - DD estimator different from Discontinuity
  estimator

26
Comparing RD and DD in DD

• Third DD approach
• - Estimator may be biased recession of 1991 had
  asymmetrical effects on provinces
• - Choosing RoC as a control group for Qc is
  inadequate

27
Triple DD !!!
28
Conclusion

• Generous Social assistance benefits substantially
  reduces employment for men at age 29-30.
• Treatment effect on the treated may not be
  generalized for other groups.
• DD estimators may perform poorly if the control
  group is not chosen adequately.