Understanding Recent Changes in Child Poverty

1
Understanding Recent Changes in Child Poverty

• Austin Nichols, Urban Institute
• ltanichols AT ui.urban.orggt
• Tenth Annual ACF/OPRE Welfare Research and
  Evaluation Conference
• June 6, 2007

2
What Happened to Child Poverty?
Black child poverty fell dramatically in the
1990s.
3
Why does child poverty change?

• Demographic Structure
• Marital status of parents, number of parents
  present
• Lerman (1996) increase in single parents, 1971
  to 1989
• Number of children per family
• Age of parents
• Education of workers in the family
• Earnings and Employment
• Unemployment rates
• Blank and Blinder (1986) Hoynes, Page, and
  Stevens (2005)
• Government Policies taxes and credits, welfare,
  etc.
• Meyer and Rosenbaum (2001) EITC spurred LFP in
  90s

4
Cycles in Child Poverty and Unemployment
5
Other Explanatory Factors
6
Other Recent Research

• Lichter and Crowley (2004), using the CPS,
  attribute half the decline in poverty among
  children living with single mothers between 1996
  and 2000 (and 69 of the decline for Black
  children) to increased maternal employment.
• Lichter, Qian, and Crowley (2005), using the
  PUMS, conclude employment explains 27 of
  decreased poverty for children of White
  ever-married single mothers, and 18 of decreased
  child poverty for children of Black ever-married
  single mothers, between 1990 and 2000.
• Both use standardization multiply partitioned
  child poverty rate at time 2 by time 1 category
  weightscounterfactual is what would poverty
  have been if sizes of populations had not changed?

7
Options for Methods to Explain Changes in Poverty
Rates

• Panel regression
• Shift-share analysis or standardization
• Categorical explanatory variables only
• Gomulka and Stern (1990)
• use data from time 1 and coefficients from time 2
  to partition change among coefficients and
  characteristics (not detailed decompositions)
  also reweight proportions in categories to base
  year (see Table 5), similar to Dinardo, Fortin,
  Lemieux (1996)
• Yun (2003)
• Blinder-Oaxaca-style decomposition generalized to
  nonlinear models
• Fairlie (2005)
• matching decomposition requires equal samples,
  which in practice requires repeatedly drawing
  samples from the larger group (more recent survey
  year in the case of year-to-year differences
  measured in the CPS)

8
Data

• I use the same data used in computing the
  official poverty rate, microdata from the March
  CPS, survey years 1981 to 2006. The March CPS for
  1981 refers to income earned in 1980, but
  measures family structure in early 1981.
• Highest prevailing minimum wages are measured at
  the state level, and deflated by the national CPI
  research series.
• State unemployment rates are from the DoL BLS
  Local Area Unemployment Statistics (LAUS)
  program.

9
Results

• Panel regressions
• Logit (probit produces same results)
• State and year fixed effects
• Only changes within a state can explain changes
  in poverty rates.
• Coefficients scaled by mean of f(Xb)100 so
  interpretation of effect is in percentage
  points
• Mean is 6 for White children, and 11 for Black
  children
• Decomposition explains proportion of changes in
  poverty rates due to changes in explanatory
  factors, and their influence on poverty
  (coefficients)

10
(No Transcript)
11
(No Transcript)
12
Year Pairs for Decomposition 1993-2000, 2000-2005
13
Explaining Change in Rates
14
Conclusions

• Both panel regression and various decompositions
  indicate that families of Black children are more
  exposed to risk of macroeconomic shocks.
• Panel regression shows parents work attachment,
  education, and family structure are strong
  predictors of child poverty, and state
  unemployment rates and real minimum wages are
  not.
• Decompositions are sensitive to assumptions and
  techniques, but state unemployment rates and
  minimum wages appear much more important.
• Decreasing state unemployment rates are
  associated with lower poverty for children in
  families with no education beyond high-school in
  the earlier period (1993-2000), and rising
  unemployment and falling real minimum wages are
  associated with higher child poverty in the
  latter (2000-2005).

15
More conclusions

• The panel regressions are really repeated
  cross-sections, so they measure predictors of
  child poverty at a point in timework and
  education of parents, and number of parents.
• Oaxaca-style decompositions measure whether
  observed changes in poverty rates can be
  explained by observed changes in other factors,
  but cannot hope to identify causes of changes in
  either poverty rates or explanatory factors.

16
Speculation

• Would like to knowdid welfare reform or the EITC
  expansion drive changes in labor force
  participation and employment rates? Timing
  appears coincident with EITC (but no causal
  connection established).
• Or was the employment boom of the 1990s driven
  entirely by exogenous shifts in labor demand?
  Labor demand for low-skill workers seems to have
  a small positive slope, or to be nearly perfectly
  elastic, possibly as an artifact of wage
  controls, so its implausible.

17
Open Questions

• What are the impacts of policy on explanatory
  factors, and poverty? Needs a better research
  design than any extant work. Regression horse
  races unconvincing on EITC vs. welfare reform.
• Are the later consequences (including
  intergenerational impacts) of poverty spells the
  same for children who are poor in high-poverty
  years, as for children who are poor in
  low-poverty years?
• Is a child always better off with working
  parents, if he would otherwise be poor?