The Effect of Income on Educational Attainment: Evidence from State Earned Income Tax Credit Expansions

Transcription

1 The Effect of Income on Educational Attainment: Evidence from State Earned Income Tax Credit Expansions Katherine Michelmore Policy Analysis and Management Cornell University September 12, 2013 Katherine Michelmore EITC and Education September / 1

2 Introduction Children growing up in lowest income quartile 50 percentage points less likely to go to college than children in top income quartile (Bailey and Dynarski 2011) Why are low-income children less likely to go to college? Differences in academic preparedness (Carneiro and Heckman 2002; Cunha et al. 2005) Test score gaps emerge in early childhood (Fryer and Levitt 2004; Rouse, Brooks-Gunn, McLanahan 2005) Obstacles in accessing financial aid (Bettinger et al. 2012) Katherine Michelmore EITC and Education September / 1

3 How does the Earned Income Tax Credit affect educational attainment of children? Early research focused on labor force effects (see Hotz and Scholz 2003; Meyer 2010 for review) Some recent evidence that the EITC improves short-run outcomes of children Birth weight (Hoynes, Miller, Simon 2012) Test scores (Dahl and Lochner 2012) Test score gains tend to fade out after interventions end, but may still see long-run benefits (Deming 2009; Chetty et al 2012) No research has explored the long-run impacts of the EITC on children Katherine Michelmore EITC and Education September / 1

4 Contributions Impact of EITC on educational attainment of children through two primary mechanisms: Early childhood interventions: Income transfer to low-income households with young children Financial Aid: Conditional cash transfer for children aged who remain full-time students When do income transfers matter most? Katherine Michelmore EITC and Education September / 1

5 Current Study I use variation in state EITC implementation and generosity to examine how the EITC affects the educational outcomes of children Difference in differences strategy with variation in treatment dosage Proxy for EITC-eligibility using parents education (commonly done in EITC literature, e.g. Meyer and Rosenbaum 2001) Compare education outcomes of children before and after changes in EITC benefits relative to individuals in non- implementing states Control for secular trends in education outcomes using children from affluent households Examine how effects differ depending on length of EITC-exposure Outcomes of interest: High school graduation, years of schooling, college enrollment/completion Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

6 Preview of Results Following an increase in the EITC by $1,000, year old children of likely EITC recipients are more likely to enroll in college Intent-to-treat effects: 0.7 ppt more likely to be enrolled in college full-time (4% increase) 1 ppt more likely to have ever enrolled in college (2.5% increase) 0.3 ppt more likely to complete a bachelor s degree (10% increase) Larger effects among sample actually eligible for the EITC Effects concentrated among children exposed to state EITCs at younger ages (<12 years old) Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

7 Outline Katherine Michelmore EITC and Education September / 1

8 EITC Background Largest cash transfer program in the United States $75 billion dollars in 2011 Distributed through the tax code; higher claiming rates compared to traditional welfare programs (over 80% in most years) Fully-refundable, households with no tax liability can still claim the credit Benefit based on prior calendar year s earnings, received in February-April of following year Must have earnings to qualify, no lifetime limits Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

9 EITC background, cont Several expansions to both federal and state EITCs since its inception in 1970s 24 states and DC have EITCs, supplement federal EITC Benefits calculated as a percentage of federal benefit: federal eligibility implies state eligibility Variation in both timing of implementation and generosity of benefits Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

10 Number of states with EITCs, by calendar year Benefit Level in year of implementation (as a percent of federal benefit) Benefit Level as of 2011 tax year (as a percent of federal benefit) Year of Implementation Rhode Island Vermont Wisconsin Iowa Minnesota New York Massachusetts Oregon Kansas Maryland Colorado DC Illinois Maine New Jersey Oklahoma Indiana Nebraska Delaware Virginia New Mexico North Carolina Michigan Louisiana Connecticut Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

11 Maximum federal and state EITC by time to state EITC implementation, in thousands of dollars (2011$) Maximum EITC benefit in thousands of dollars (2011$) Maximum federal and state EITC Years since state EITC implementation Maximum federal EITC only Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

12 EITC Background, cont. Benefit calculated based on the earnings of the head and spouse; varies by number of children in the household EITC schedule 97% of EITC dollars go to households with children: Childless credit worth less than $500; two-child credit worth up to $5,111 in 2011 Number of qualifying children: Up to 2 (3 for 2009 and beyond) Qualifying children: Live with parents at least half of the year; child does not provide more than half of own support Conditional cash transfer for children years old: Child must be under 19 at end of tax year, or under 24 and full-time student Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

13 Mechanisms: Early childhood interventions Children from low-income families may attend poor-quality schooling, not academically prepared for college Early childhood interventions lead to later life benefits (Chetty et al. 2012; Deming 2009; Krueger and Whitmore 2001) May be more cost-effective than interventions at college-age (Cunha et al. 2005; Duncan, Ludwig, Magnuson 2007) EITC improves the test scores of children (Dahl and Lochner 2012) Also leads to increased educational attainment? Katherine Michelmore EITC and Education September / 1

14 Mechanisms: Early childhood interventions Receiving the EITC in childhood may allow families to improve quality of schooling for children throughout their lives The EITC tends to benefit families with young children (Average age of oldest child <12; Ackerman, Holtzblatt, Masken 2009) Recipients often eligible for several years (70% receive the benefit more than once; Ackerman, Holtzblatt, Masken 2009) EITC allows families to pay down debt, move to better neighborhoods, increase savings (Smeeding, Ross-Phillips, and O Connor 2002; Tach and Halpern- Meekin 2013) Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

15 Mechanisms: Financial aid for college-aged children Low-income families may face obstacles in attaining financial aid Mixed evidence regarding effectiveness of federal financial aid policies on educational attainment FAFSA complexity hinders college enrollment (Bettinger et al. 2012) Cash transfers for college-aged children increase college enrollment Dynarski (2003): College enrollment fell by 3.5 ppt per $1,000 loss in aid following Social Security Student Benefit Program elimination Katherine Michelmore EITC and Education September / 1

16 Mechanisms: Financial aid for college-aged children Remaining in school could be worth over $4,000 in EITC benefits alone (for 2011 tax year) Maximum credit for one-child household: $3,094 Maximum credit for childless household: $464 One-year potential gain in EITC benefits (assuming state EITC worth 45% of federal): $4,500 Net cost of college for incomes below $32,000: $10,000 for four-year, $6,000 for two-year public institutions (College Board, school year) Katherine Michelmore EITC and Education September / 1

17 Data Survey of Income and Program Participation Household survey with information on all individuals residing in the household or living away at school Short panel: Monthly data on household characteristics for months 81,724 observations for year olds living with at least one parent Observe education outcomes in March of each survey year Katherine Michelmore EITC and Education September / 1

18 Data High-impact sample: Individuals living in a household where neither parent has schooling beyond high school (31,000 observations) Education often used as a proxy for EITC-eligibility (Meyer and Rosenbaum 2001; Baughman and Dickert-Conlin 2009; Hoynes, Miller, Simon 2012) Results not sensitive to definition of high-impact sample; similar results found using actual eligibility, bottom income quintile Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

19 Data limitations Only observe individuals who reside with a parent at the start of the survey (70% of year olds) Restrict sample to year olds (85% of year olds) Restrict sample to those who were <19 at time of survey (90% of year olds) Concern over selection into who lives with a parent Sample Little evidence that state EITC benefits affect likelihood of living at home Robustness check: Include individuals not living at home as part of treated sample SIPP lacks individual state identifiers for some small states in early panels, state-years dropped for ME and VT before 2004 Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

20 Summary Statistics: year olds, high-impact sample Never state EITC "pre" Pre-state EITC Never state EITC "post" Post-state EITC Difference in differences Currently enrolled in college (.417) (.436) (.42) (.452) (.011) Years of schooling (1.579) (1.605) (1.456) (1.482) (.032) Has a high school degree (.482) (.459) (.478) (.453) (.009) Ever enrolled in college (.48) (.491) (.477) (.495) (.008) Has at least an associate's degree (.177) (.2) (.223) (.258) (.007) Has at least a bachelor's degree (.136) (.165) (.12) (.184) (.005) Number of Observations 16,507 7,806 17,271 5,642 Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

21 Summary Statistics: year olds, high-impact sample Never state EITC "pre" Pre-state EITC Never state EITC "post" Post-state EITC Difference in differences Black (.387) (.425) (.372) (.357) (.008) Other (.239) (.191) (.265) (.25) (.005) Female (.498) (.496) (.499) (.5) (.011) Age (.988) (1.672) (1.646) (1.663) (.037) Maximum federal and state EITC (in 1,000s of 2011$ ) (1.411) (1.362) (.85) (1.07) (16.52) Number of male siblings in the household (.988) (.827) (.941) (.892) (.019) Number of female siblings in the household (.889) (.813) (.924) (.868) (.018) Living with both parents (.475) (.478) (.488) (.485) (.01) Number of Observations 16,507 7,806 17,271 5,642 CPS Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

22 Summary Statistics: year olds, high-impact sample SIPP year olds Never state EITC "pre" Pre-state EITC Never state EITC "post" Post-state EITC Difference in differences Currently enrolled in college (.004) (.007) (.004) (.008) (.014) Years of schooling (.015) (.021) (.013) (.022) (.032) Has a high school degree (.005) (.007) (.005) (.008) (.013) Ever enrolled in college (.005) (.007) (.004) (.008) (.009) Number of Observations 10,419 4,628 10,908 3,489 Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

23 Empirical Strategy Difference-in-Differences technique using variation in the timing and generosity of state EITC benefits Treatment variable constructed as the maximum federal and state EITC benefit for a household with two children in a given state and year (in 2011$) Example: Maximum federal EITC (2011, 2 kids): $5,112 Maximum Washington, DC benefit (40% of federal): $2,045 Maximum combined federal and state EITC for DC: $7,157 Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

24 Empirical Strategy I estimate two basic specifications, the first for the high impact sample only: Y i,s,t = γx i,s,t + φv s,t + βeitc s,t + θz s + αw t + ɛ i (1) Y i,s,t X i,s,t V s,t EITC s,t β Z s W t outcome for i in state s in year t demographic controls state economic indicators maximum federal and state EITC in state s in year t Effect of a $1,000 increase in max EITC on outcome Y i,s,t state FE year FE β identified off of within-state changes to EITC benefits over time Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

25 Empirical Strategy: Triple-Difference To control for secular state-level trends in education outcomes, I then include individuals not likely to be affected by the EITC: Y i,s,t = γx i,s,t +φv s,t +β 1 EITC s,t +β 2 ELG i,s,t +β 3 EITC s,t ELG i,s,t +θz s +αw t +ɛ i (2) ELG i,s,t β 3 indicator for high-impact sample key parameter of interest in (2): interaction of eligibility with EITC β 3 identifies effect of within-state EITC benefit increase of $1,000 on schooling of EITC-eligible children compared to those not eligible Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

26 Assumptions Likelihood of being in the sample does not change as a function of state EITC benefit changes high-impact sample Changes in state EITC benefits unrelated to existing trends in education outcomes Examine trends in education outcomes prior to state EITC implementation Pre-trends Use children from highly-educated households as control for state-level trends in education outcomes Changes in state EITC benefits are not correlated with other state-level policies/trends Test whether state EITC benefit is related to other state-level indicators Test of Exogeneity Control for state-by-year indicators such as GDP, minimum wage, unemployment rate Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

27 Pre-trends for outcomes of interest by time to state EITC implementation, relative to states that never implement and low-impact sample; effects normalized to zero in the year before implementation; year olds in the SIPP CPS fig a) Likelihood of being enrolled as full-time college student b) Number of years of schooling c) Likelihood of having a high school degree d) Likelihood of ever enrolling in college Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

28 Multivariate results: Effect of the EITC on education outcomes high impact sample High-impact sample SIPP year olds No controls W/controls No controls SIPP year olds W/controls Outcome variable Currently enrolled in college (0.015) (0.012) (0.014) (0.011) Years of Schooling ** (0.075) (0.051) (0.064) (0.050) High school graduate * (0.021) (0.012) (0.022) (0.016) Ever enrolled in college ** * (0.017) (0.012) (0.018) (0.015) Demographic Controls N Y N Y Year Fixed Effects Y Y Y Y State Fixed Effects Y Y Y Y Number of Observations 31,130 31,130 19,285 19,285 *** p<.01, ** p<.05, * p<.10. Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

29 Multivariate results: Effect of the EITC on education outcomes triple difference Alternative Specifications High-impact sample SIPP year olds No controls W/controls Outcome variable Currently enrolled in college (0.015) (0.012) Years of Schooling ** (0.075) (0.051) High school graduate * (0.021) (0.012) Ever enrolled in college ** (0.017) (0.012) Demographic Controls N Y Year Fixed Effects Y Y State Fixed Effects Y Y Number of Observations 31,130 31,130 *** p<.01, ** p<.05, * p<.10. Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

30 Multivariate Results: Degree completion, triple-difference No controls W/controls Outcome variable Has at least an associate's degree (0.002) (0.002) Has at least a bachelor's degree * * (0.001) (0.002) Demographic Controls N Y Year Fixed Effects Y Y State Fixed Effects Y Y Number of Observations 81,724 81,724 *** p<.01, ** p<.05, * p<.10. Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

31 Multivariate Results cont. When does money matter most? Gender, race, family composition Household size Katherine Michelmore EITC and Education September / 1

32 Effect of EITC on outcomes of interest by age at state EITC implementation (reference: never exposed to state EITC), year olds in the SIPP CPS fig 2 a) Likelihood of being enrolled as full-time college student b) Number of years of schooling c) Likelihood of having a high school degree d) Likelihood of ever enrolling in college Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

33 Multivariate Results: Differences by gender, race Differences by gender Differences by family structure Differences by race Men Women Both parents Single Parent Non-Black Black Outcome variable Currently enrolled in college * *** ** (.005) (.004) (0.003) (0.006) (0.003) (0.007) Years of Schooling (.017) (.017) (0.016) (0.028) (0.015) (0.027) High school graduate (.004) (.004) (0.004) (0.006) (0.003) (0.007) Ever enrolled in college *** *** ** *** (.005) (.004) (0.004) (0.006) (0.004) (0.009) Demographic Controls Y Y Y Y Y Y Year Fixed Effects Y Y Y Y Y Y State Fixed Effects Y Y Y Y Y Y Number of Observations 43,280 38,444 58,740 22,984 69,932 11,792 *** p<.01, ** p<.05, * p<.10. Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

34 Effect of EITC on likelihood of ever enrolling in college by number of children in the household, year olds in the SIPP Effect on likelihood of ever enrolling in college Number of children under 24 living in the household Pooled Men Women Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

35 Conclusion Results suggest significant increases in educational attainment among children growing up in low-educated households associated with $1,000 increase in EITC benefits 0.7 ppt more likely to be enrolled full-time (4% increase) 1 ppt more likely to have ever enrolled in college (2.5% increase) 0.3 ppt more likely to complete a bachelor s degree (10% increase) Plausible estimates? Approximately half of high-impact sample actually eligible for the EITC: Treatment-on-treated effects roughly twice as large Chetty, Friedman, and Rockoff (2011) estimate that effects from Dahl and Lochner (2012) should lead to 0.3 ppt increase in college enrollment Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

36 Conclusion II Limitations and future work Not clear if children complete degrees; have higher wages Does the EITC impact levels of debt taken on to finance higher education? Other implications of better schooling? Health outcomes? Teen fertility? Policy implications Several states discussed cutting back EITCs: Michigan, North Carolina, Oregon, Connecticut, Kansas Results suggest adverse impacts on schooling of low-income youth Katherine Michelmore EITC and Education September / 1

37 EITC tax schedule for 2011, single-headed households by number of children and earnings back 7000 No Kids 1 Kid 2 Kids 3+ Kids EITC benefit amount (in 2011$) Pre-tax earnings, in thousands of 2011$ Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

38 Summary Statistics: Living with parents vs. not living with parents back SIPP year olds Living with at least one parent Not living with a parent SIPP year olds Living with at least one parent Not living with a parent Currently enrolled in college 40.5% 18.3% 42.3% 21.9% 4 (.491) (.386) (.002) (.414) ( Years of Schooling (1.512) (2.003) (.006) (1.735) ( Has a high school degree 76.5% 77.0% 67.2% 64.6% 6 (.424) (.421) (.002) (.478) ( Ever enrolled in college 57.3% 45.7% 50.8% 35.2% 4 (.495) (.498) (.002) (.478) ( Has at least an Associate's degree 8.5% 12.6% 1.6% 2.8% (.278) (.332) (.001) (.164) Has at least a Bachelor's degree 5.1% 7.6% 0.2% 0.5% (.22) (.265) (.) (.067) Black 14.5% 13.6% 14.4% 16.6% 1 (.352) (.343) (.002) (.372) ( Other 6.0% 5.4% 6.2% 5.9% (.238) (.226) (.001) (.235) ( Female 45.9% 58.7% 47.3% 59.4% 4 (.498) (.492) (.002) (.491) ( Age (1.646) (1.546) (.004) (.795) ( Number of Observations 81,724 37,759 51,374 11, Livi le p Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

39 Likelihood of living at home as a function of EITC generosity back SIPP year olds SIPP year olds Outcome variable= Living with at least one parent at start of interview Maximum federal and state EITC (in thousands of 2000 dollars) * -0.0 (0.007) (0.008) (0.01 Demographic Controls Y Y Y Year Fixed Effects Y Y Y State Fixed Effects Y Y Y CPS 1 year Number of Observations 119,483 63, ,38 Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

40 Summary Statistics: year olds CPS back Never state EITC "pre" Pre-state EITC Never state EITC "post" Post-state EITC Difference in differences Currently enrolled in college (.415) (.438) (.421) (.464) (.006) Years of schooling (1.307) (1.34) (1.279) (1.28) (.017) Has a high school degree (.499) (.494) (.498) (.489) (.006) Ever enrolled in college (.454) (.47) (.457) (.486) (.006) Black (.328) (.418) (.32) (.389) (.005) Other (.252) (.178) (.278) (.249) (.003) Female (.497) (.497) (.497) (.497) (.006) Age (.81) (.811) (.81) (.82) (.005) Maximum federal and state EITC (in 1,000s of 2011$ ) (1.246) (1.266) (.638) (.795) (.013) Number of male siblings in the household (.889) (.846) (.905) (.827) (.011) Number of female siblings in the household (.887) (.775) (.873) (.827) (.011) Living with both parents (.471) (.482) (.477) (.491) (.006) Number of Observations 19,884 6,860 19,884 8,628 Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

41 Test of changes in demographic controls with EITC High-impact sample only back year olds, SIPP year olds, SIPP year olds, CPS year olds, SIPP Outcome variable Black (.024) (.02) (.011) (.004) Other ** (.01) (.013) (.008) (.002) Female (.015) (.019) (.014) (.003) Age * (.034) (.017) (.026) (.008) Number of male siblings (.065) (.079) (.032) (.011) Number of female siblin (.041) (.042) (.036) (.011) Living with both parents *** (.018) (.018) (.018) (.004) State Fixed Effects Y Y Y Y Year Fixed Effects Y Y Y Y Number of Observations 31,130 19,285 36,063 81,724 Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

42 Likelihood of being in the high-impact sample back SIPP year olds Outcome variable= In the high-impact sample Maximum federal and state EITC (in thousands of 2000 dollars) (0.017) (0.015) ( Demographic Controls Y Y Year Fixed Effects Y Y State Fixed Effects Y Y SIPP year olds Number of Observations 81,724 51,374 9 C Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

43 Test for exogeneity of timing of state EITC changes States that ever implement EITCs Maximum Federal and State EITC in year t (in thousands of 2011 dollars) All States Maximum Federal and State EITC in year t (in thousands of 2011 Dependent Variable: Log state GDP per capita (1.217) (.66) Unemployment Rate (*100) (.041) (.023) Log real minimum wage ** * (.209) (.12) State Fixed Effects Y Y Year Fixed Effects Y Y back Number of Observations Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

44 Pre-trends for outcomes of interest by time to state EITC implementation, relative to states that never implement and low-impact sample; effects normalized to zero in the year before implementation; year olds in the CPS back a) Likelihood of being enrolled as full-time college student b) Number of years of schooling c) Likelihood of having a high school degree d) Likelihood of ever enrolling in college Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

45 Alternative specifications back Main results from table 5 Actual eligibility in first year of survey Include individuals not living with a parent in treated Outcome variable Currently enrolled in college ** *** (.003) (.003) (.002) (.004) Years of Schooling * *** *** (.014) (.011) (.014) (.013) High school graduate *** ** * (.003) (.003) (.002) (.004) Ever enrolled in college *** *** *** ** (.003) (.005) (.002) (.004) Demographic Controls Y Y Y Y Year Fixed Effects Y Y Y Y State Fixed Effects Y Y Y Y Restrict to individuals under age 19 at start of survey Number of Observations 81,724 81, ,483 42,947 *** p<.01, ** p<.05, * p<.10 Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1

46 Effect of EITC on outcomes of interest by age at state EITC implementation (reference: older than 23), year olds in the CPS back < < a) Likelihood of being enrolled as full-time college student b) Number of years of schooling < < c) Likelihood of having a high school degree d) Likelihood of ever enrolling in college Katherine Michelmore (km459@cornell.edu) EITC and Education September / 1