| Achieved Purpose 1 | No, N = 591 | Yes, N = 10,6811 |
|---|---|---|
| Race of child | ||
| White | 41 (69%) | 7,223 (68%) |
| Black | 18 (31%) | 2,430 (23%) |
| Asian | 0 (0%) | 500 (4.7%) |
| AIAN | 0 (0%) | 372 (3.5%) |
| Other | 0 (0%) | 123 (1.2%) |
| Sex of child | ||
| Female | 33 (56%) | 5,324 (50%) |
| Male | 26 (44%) | 5,357 (50%) |
| Age of child | 5.0 (3.0, 9.5) | 9.0 (4.0, 13.0) |
| 1 n (%); Median (IQR) | ||
A Bayesian Hierarchical Approach to Predicting Fulfillment of TANF Purposes
Shelby Tisdale
Background
A Brief Overview of TANF Policy
- Personal Responsibility and Work Opportunity Reconcilitation Act of 1996 (PWRORA).
- Temporary Assistance for Needy Families (TANF) designed to fund services and provide direct cash assistance to low-income families with children in the U.S.
- $16.5 billion in federal funds annually (has not changed since 1996, losing 40 percent of its value)
- Funds provided to states as block grants
- states can use funds as they wish, as long as spending aligns with one of TANF’s four purposes
Source: Center on Budget and Policy Priorities (2022)
The Mississippi Welfare Scandal
- Up to $94 million in TANF block grant funds misappropriated by Mississippi Department of Health and Human Services.
- Highlights need for further analysis of how TANF block grant funds are being used.
Unobligated Funds
- States can reserve unused TANF funds without time limit.
- Every year, states accumulate unobligated balances: funds that have neither been spent or obligated toward a specific purpose.
Source: Administration for Children and Families
Direct Cash Assistance
Floyd and Pavetti (Center for Budget and Policy Priorities, 2022) note that:
TANF provides less cash assistance than Aid to Families with Dependent Children (AFDC).
Deep poverty in the U.S. has increased among children in single-mother families, accompanying decline in direct cash assistance.
Research by Shaefer, Edin, Fusaro and Wu (2020) has shown:
Strong relationship between TANF caseload reduction and increased student homelessness and food insecurity for single-mother families.
In-kind aid from TANF-funded programs often benefits families who are not experiencing poverty; cash assistance goes directly to low-income families.
Direct Cash Assistance
TANF Spending for Selected States (2017-2022)
The Four Purposes of TANF
- Provide assistance to needy families so that children can be cared for in their own homes or in the homes of relatives.
- End the dependence of needy parents on government benefits by promoting job preparation, work, and marriage.
- Prevent and reduce the incidence of out-of-wedlock pregnancies.
- Encourage the formation and maintenance of two-parent families.
Source: Office of Family Assistance, Administration for Children and Families, 2024
Objectives
While many studies evaluate TANF with respect to outcomes related to poverty and child well-being, few studies examine the four purposes TANF is designed to fulfill.
In this study, we aim to:
- Examine the relationship between state-level TANF spending and household-level fulfillment of the four purposes of TANF.
- Specifically examine how state-level spending on direct cash assistance and unobligated funds relate to household-level outcomes.
Data
Survey of Income and Program Participation (SIPP)
- Nationally representative, longitudinal
- Household data on employment, family characteristics, and participation in government and social programs.
- Data from panels 2018-2022 (years 2017-2021).
- Sample limited to households with children below the federal poverty line.
- SIPP data recoded to the fiscal year.
Source: U.S. Census Bureau, SIPP User’s Guide (2024)
Survey of Income and Program Participation (SIPP)
Key Terms
- Householders: ”the person in whose name the housing unit is owned or rented (maintained) or, if there is no such person, an adult member”
- Cohort years: first fiscal year within the 2017-2022 reference period that the observation, either a household or an individual, was present in the SIPP.
- Allowed for the models to incorporate year-level TANF financial data for each state without a longitudinal response variable.
Source: U.S. Census Bureau, SIPP User’s Guide (2024)
TANF Federal Expenditures
- Combined TANF and maintenance-of-effort (MOE) expenditures for all 50 states and Washington, D.C.
- 20 primary spending categories, along with total awarded funds, unobligated funds, and carryover funds.
- carryover funds: unobligated funds from the previous fiscal year, reserved for future use.
Source: Administration for Children and Families
Political Data
- Data from Wikipedia indicating the political party of a state’s governor for the year. (Democrat or Republican)
- Used to control for potential political influence on welfare policy.
Source: Wikipedia
Analysis
Methods
For each purpose, we:
- Used data from the SIPP to create a binary indicator for whether the purpose was “achieved” for a person or household.
- Estimated a three-level Bayesian hierarchical logistic regression model with random-intercepts using MCMC sampling via the
rstanarmpackage. - Assessed mixing and stability across parallel chains with trace plots, autocorrelation plots, effective sample size ratios, R-hat ratios.
- Performed classification analysis with testing data and assessed fit by calculating area under the ROC curve
Bayesian Hierarchical Logistic Regression
Advantages
- Incorporate prior knowledge on family-level outcomes into models.
- Account for within- and between-group variability at each level of the hierarchy.
Model
\[ \log{(\frac{\pi_{ijk}}{1-\pi_{ijk}})} = (\beta_{0} + b_{0j} + p_{0k}) + \sum_{n=1}^m\beta_{n_\text{I}}X_{ijkn} + \sum_{n = m+1}^{p}\beta_{n_\text{I}}X_{jkn} \]
\[ Y_{ijk} = \begin{cases} 1 & \text{purpose fulfilled for observation $i$ in cohort year $j$ and state $k$} \\ 0 & \text{purpose not fulfilled for observation $i$ in cohort year $j$ and state $k$} \end{cases} \]
\[Y_{ijk} | \beta_{0}, b_{0j}, p_{0k}, \beta_{1},...,\beta_{p} \sim \text{Bernoulli}(\pi_{ijk})\]
\(\pi_{ijk}\) is the probability that child or household \(i\) in cohort year \(j\) and state \(k\) fulfills the purpose.
\(\beta_{0}\) is the global intercept
\(b_{0j}\) is the year-specific adjustment to the intercept for cohort year \(j\)
\(p_{0k}\) is the state-specific adjustment to the intercept for state \(k\)
Weakly informative priors:
- \(b_{0j} \sim \text{N}(0, \sigma_b)\)
- \(p_{0k} \sim \text{N}(0, \sigma_p)\)
- \(\sigma_b, \sigma_p \sim \text{Exp}(1)\)
- \(\beta_{1}, \dots, \beta_{p} \sim \text{N}(0, 2.5^2)\)
TANF Purpose 1: Data
Provide assistance to needy families so that children can be cared for in their own homes or in the homes of relatives.
Subsample:
all children in SIPP sample (for all months where child has age < 18)
Criteria:
- Purpose 1 is “achieved” (
purpose = 1) for a child if, for every month they are present in the sample, they lived with at least one parent (biological, step, or adoptive), adult sibling, or other relative. - Otherwise, Purpose 1 is “not achieved” (
purpose1 = 0)
Global Prior:
\[\beta_{0_\text{I}} \sim N(5.81, 1.5^2)\]
Based on data from the Annie E. Casey Foundation: In 2021, children entered foster care at a rate of 3 per 1,000.
TANF Purpose 1: Model
Provide assistance to needy families so that children can be cared for in their own homes or in the homes of relatives.
- Outcome: Purpose 1 achievement
- Child-level predictors: Race, Sex
- State-year level predictors: Percent of TANF funds unobligated, Percent of TANF expenditures on fatherhood programs, foster care, basic assistance, and child welfare; political party of governor; whether year is during covid
TANF Purpose 1: Results
Provide assistance to needy families so that children can be cared for in their own homes or in the homes of relatives.
AUC: 0.674
| Term | Estimated Odds Ratio | 80% Credible Interval |
|---|---|---|
| Global Intercept | 375.160 | (96.315, 1785.13) |
| Race of child: Black1 | 0.813 | (0.512, 1.33) |
| Child is male1 | 1.240 | (0.829, 1.86) |
| Age of child | 1.051 | (1.011, 1.09) |
| Percent of TANF funds unobligated2 | 1.010 | (0.992, 1.03) |
| Percent of TANF expenditures on basic assistance2 | 1.005 | (0.974, 1.04) |
| Percent of TANF expenditures on fatherhood programs2 | 1.181 | (1.021, 1.44) |
| Percent of TANF expenditures on foster care services and maintenance payments2 | 1.025 | (0.978, 1.08) |
| Percent of TANF expenditures on child welfare programs2 | 1.019 | (0.985, 1.05) |
| Governor of state in year is Republican1 | 0.204 | (0.077, 0.47) |
| Year is 2020 or later1 | 0.757 | (0.293, 2.10) |
| 1 categorical variable (yes = 1, no = 0) | ||
| 2 for state and cohort year | ||
TANF Purpose 2: Data
End the dependence of needy parents on government benefits by promoting job preparation, work, and marriage.
Subsample:
All households that were in the SIPP sample for more than one year and received any government benefits (TANF, GA, SNAP, WIC) in their earliest year in the sample.
Criteria:
- Purpose 2 is “achieved” (
purpose2 = 1) for a household that did not receive benefits during its final year in the sample. - Otherwise, Purpose 2 “not achieved” (
purpose2 = 0)
Global Prior:
\[\beta_{0_\text{II}} \sim N(-1.05, 1.5^2)\]
2023 analysis of social safety net program participation by Macartney and Ghertner: among individuals with income below 200 percent of the federal poverty line, 74 percent received some government benefits.
Source: Office of the Assistant Secretary for Planning and Evaluation
| Achieved Purpose 2 | No, N = 1,0591 | Yes, N = 3651 |
|---|---|---|
| Householder is married | 480 (45%) | 203 (56%) |
| Race | ||
| White | 694 (66%) | 265 (73%) |
| Black | 282 (27%) | 66 (18%) |
| Asian | 38 (3.6%) | 19 (5.2%) |
| AIAN | 33 (3.1%) | 12 (3.3%) |
| Other | 10 (0.9%) | 2 (0.5%) |
| Householder is employed | 777 (73%) | 311 (85%) |
| Householder received job training | 206 (19%) | 67 (18%) |
| 1 n (%) | ||
TANF Purpose 2: Model
End the dependence of needy parents on government benefits by promoting job preparation, work, and marriage.
Outcome: Purpose 2 achievement
Household-level predictors: Householder job training; Householder employment status; Household receipt of TANF cash assistance; number of years in sample; Race of householder
State-year level predictors: Percent unobligated TANF funds; percent TANF expenditures on basic assistance; percent TANF expenditures on work education; party of governor; whether year is during covid
Considerations: Interaction between job training and employment
TANF Purpose 2: Results
End the dependence of needy parents on government benefits by promoting job preparation, work, and marriage.
AUC: 0.650
| Term | Estimated Odds Ratio | 80% Credible Interval |
|---|---|---|
| Global Intercept | 0.076 | (0.034, 0.160) |
| Householder received job training1 | 0.145 | (0.026, 0.474) |
| Householder employed1 | 1.587 | (1.228, 2.067) |
| Household received TANF cash assistance1 | 0.378 | (0.261, 0.535) |
| Number of years in sample | 1.294 | (1.159, 1.443) |
| Householder married | 1.328 | (1.087, 1.620) |
| Race of householder: Black1 | 0.831 | (0.650, 1.059) |
| Race of householder: Asian1 | 1.277 | (0.806, 1.994) |
| Race of householder: AIAN | 1.396 | (0.811, 2.339) |
| Race of householder: Other race1 | 0.420 | (0.059, 1.728) |
| Interaction between job training and employment1 | 5.712 | (1.701, 31.13) |
| Percent of TANF funds unobligated2 | 1.003 | (0.997, 1.007) |
| Percent of TANF expenditures on basic assistance2 | 0.991 | (0.982, 0.999) |
| Percent of TANF expenditures on work education programs2 | 1.019 | (1.007, 1.030) |
| Governor of state in cohort year is Republican1 | 1.125 | (0.894, 1.419) |
| Cohort year is 2020 or later1 | 0.615 | (0.204, 1.626) |
| 1 categorical variable (yes = 1, no = 0) | ||
| 2 for state and cohort year | ||
TANF Purpose 3: Data
Prevent and reduce the incidence of out-of-wedlock pregnancies.
Subsample:
All women in SIPP sample between ages 15 and 50, inclusive, for all years in which they were unmarried.
Criteria:
Purpose 3 is “achieved” for a woman (
purpose3 = 1) if, during her time in the sample, her number of children did not increase.Otherwise, Purpose 3 is “not achieved” (
purpose3 = 0)
Global Prior:
\[\beta_{0_\text{III}} \sim N(0.38, 2.5^2)\]
According to a 2015 report by Hymowitz (Manhattan Institute), 40.6 percent of all children born in the U.S. in 2014 were born to unmarried mothers.
| Achieved Purpose 3 | No, N = 1021 | Yes, N = 4,5981 |
|---|---|---|
| Age | 26 (23, 30) | 25 (17, 36) |
| Race | ||
| White | 60 (59%) | 3,143 (69%) |
| Black | 35 (35%) | 1,087 (24%) |
| Asian | 1 (1.0%) | 158 (3.4%) |
| AIAN | 3 (3.0%) | 148 (3.2%) |
| Other | 2 (2.0%) | 49 (1.1%) |
| 1 Median (IQR); n (%) | ||
TANF Purpose 3: Model
Prevent and reduce the incidence of out-of-wedlock pregnancies.
Outcome: Purpose 3 achievement
Person-level predictors: Number of years in sample; age of person; amount of TANF cash assistance received (in hundreds of dollars); race
State-year level predictors: Percent unobligated TANF funds; percent TANF expenditures on basic assistance; percent TANF expenditures on pregnancy prevention; party of governor; whether year is during covid
Considerations: Quadratic effect for age
TANF Purpose 3: Results
Prevent and reduce the incidence of out-of-wedlock pregnancies.
AUC: 0.835
| Term | Estimated Odds Ratio | 80% Credible Interval |
|---|---|---|
| Global Intercept | 2,411,586.715 | (372125.997, 17234748.649) |
| Number of years in sample | 0.510 | ( 0.449, 0.578) |
| Age of person | 0.513 | ( 0.454, 0.578) |
| Age of person (squared) | 1.012 | ( 1.010, 1.014) |
| Amount of TANF Cash Assistance Received1 | 1.179 | ( 1.028, 1.449) |
| Race: Black2 | 0.564 | ( 0.397, 0.810) |
| Race: Asian2 | 2.760 | ( 0.825, 16.994) |
| Race: AIAN | 0.517 | ( 0.239, 1.313) |
| Race: Other race2 | 0.436 | ( 0.159, 1.512) |
| Percent of TANF funds unobligated3 | 0.994 | ( 0.986, 1.003) |
| Percent of TANF expenditures on basic assistance3 | 1.001 | ( 0.987, 1.017) |
| Percent of TANF expenditures on pregnancy prevention programs3 | 0.972 | ( 0.923, 1.029) |
| Governor of state in cohort year is Republican2 | 0.911 | ( 0.606, 1.381) |
| Cohort year is 2020 or later2 | 130.569 | ( 13.771, 4475.929) |
| 1 in hundreds of dollars | ||
| 2 categorical variable (yes = 1, no = 0) | ||
| 3 for state and cohort year | ||
TANF Purpose 4: Data
Encourage the formation and maintenance of two-parent families.
Subsample:
All households in the SIPP sample with at least one child and at least one parent (biological, step, or adoptive)
Criteria:
If the household includes two-parents during its initial year in the sample and remains a two-parent household in the final year of the sample, Purpose 4 is achieved (
purpose4 = 1).If the household is a single-parent household in its first year in the sample and becomes a two-parent household, Purpose 4 is achieved (
purpose4 = 1).Otherwise, Purpose 4 is “not achieved” (
purpose4 = 0)
Global Prior:
\[\beta_{0_\text{IV}} \sim N(0, 2.5^2)\] Research from Child Trends by Wildsmith and Alivra-Hammond (2023), suggests that approximately half of all families living in poverty are single-parent families.
| Achieved Purpose 4 | No, N = 2,3761 | Yes, N = 2,4651 |
|---|---|---|
| Race of householder | ||
| White | 1,552 (65%) | 1,899 (77%) |
| Black | 676 (29%) | 266 (11%) |
| Asian | 49 (2.1%) | 207 (8.4%) |
| AIAN | 71 (3.0%) | 74 (3.0%) |
| Other | 22 (0.9%) | 17 (0.7%) |
| Sex of householder | ||
| Female | 1,875 (79%) | 1,172 (48%) |
| Male | 501 (21%) | 1,293 (52%) |
| 1 n (%) | ||
TANF Purpose 4: Model
Encourage the formation and maintenance of two-parent families.
Outcome: Purpose 4 achievement
Household-level predictors: Number of years in sample; age of person; amount of TANF cash assistance received (in hundreds of dollars); race of householder, sex of householder
State-year level predictors: Percent unobligated TANF funds; percent TANF expenditures on basic assistance; percent TANF expenditures on fatherhood programs
TANF Purpose 4: Results
Encourage the formation and maintenance of two-parent families.
AUC: 0.745
| Term | Estimated Odds Ratio | 80% Credible Interval |
|---|---|---|
| Global Intercept | 0.827 | (0.672, 1.008) |
| Amount of TANF cash assistance received1 | 0.900 | (0.867, 0.931) |
| Number of years in sample | 1.058 | (1.015, 1.102) |
| Race: Black2 | 0.378 | (0.332, 0.429) |
| Race: Asian2 | 2.358 | (1.862, 3.020) |
| Race: AIAN | 0.864 | (0.655, 1.144) |
| Race: Other race2 | 0.501 | (0.296, 0.831) |
| Householder is male2 | 3.496 | (3.174, 3.856) |
| Percent of TANF funds unobligated3 | 1.002 | (0.998, 1.004) |
| Percent of TANF expenditures on basic assistance3 | 0.997 | (0.992, 1.001) |
| Percent of TANF expenditures on fatherhood programs3 | 0.978 | (0.962, 0.993) |
| Governor of state in cohort year is Republican2 | 0.798 | (0.702, 0.908) |
| 1 in hundreds of dollars | ||
| 2 categorical variable (yes = 1, no = 0) | ||
| 3 for state and cohort year | ||
Summary
No evidence that state-level unobligated funds or state-level expenditures on direct cash assistance are associated with the household-level fulfillment of any of the four purposes of TANF.
State expenditures on programs (pregnancy prevention, child welfare, fatherhood programs, etc.) had little to no effect on achievement of the four purposes of TANF.
In Republican states, children in the SIPP sample were less likely to live with relatives (Purpose 1), households less likely to be two-parent (Purpose 4).
Increases in household-level direct cash assistance correlated with lower odds of achieving Purposes 2 and 4, higher odds of achieving Purpose 3. (Note: unclear directionality of this relationship)
Racial differences in achievement of Purposes 2 and 4.
Policy Recommendations
States should prioritize spending on direct cash assistance over spending on programs.
TANF’s four purposes should be reevaluated to better align with current research on child and family wellbeing, and align with outcomes that can be easily measured.
Evaluate the effectiveness of programs funded by the TANF block grant in supporting low-income families to achieve economic security and stability.
Limitations and Future Work
Use of survey data to assess achievement of the four purposes of TANF was not exact.
Cohort design
More research should examine the impacts of unobligated funds on other outcomes, such as poverty, food insecurity, and child homelessness.
Causal approaches may be useful in determining direction of relationships
Bayesian hierarchical modeling may be used more widely in social science research
Intuitive interpretation of significance for logistic regression models
Captures within-group correlation for nested data.
References
- 2022 survey of income and program participation user’s guide (tech. rep.). (2023, September). U.S. Census Bureau.
- Administration for Children and Families. (2025). State TANF Data and Reports [[Online; accessed 24- March-2025]]. https://www.acf.hhs.gov/ofa/programs/tanf/data-reports
- Category: Lists of state governors of the united states - wikipedia [[Online; accessed 24-March-2025]]. (2025). https://en.wikipedia.org/wiki/Category:Lists_of_state_governors_of_the_United_ States
- Center on Budget and Policy Priorities. (2022). Policy Basics: Temporary Assistance for Needy Families [[Online; accessed 24-March-2025]]. https://www.cbpp.org/research/family-income-support/ policy-basics-an-introduction-to-tanf
- Congdon, P. D. (2019). Bayesian hierarchical models: With applications using r, second edition (2nd). Chapman; Hall/CRC. https://doi.org/10.1201/9780429113352
- Floyd, I., & Pavetti, L. (2022, January). Improvements in tanf cash benefits needed to undo the legacy of racist welfare policy (tech. rep.) ([Online; accessed 24-March-2025]). Center on Budget and Policy Priorities. https://www.cbpp.org/sites/default/files/1-26-22tanf.pdf
- Floyd, I., Pavetti, L., Meyer, L., Safawi, A., Schott, L., Bellew, E., & Magnus, A. (2021, August). Tanf policies reflect racist legacy of cash assistance (tech. rep.) ([Online; accessed 24-March-2025]). Center on Budget and Policy Priorities. https : / / www . cbpp . org / research / income - security / tanf - policies-reflect-racist-legacy-of-cash-assistance
- Gelman, A., Carlin, J. B., Stern, H. S., Dunson, D. B., Vehtari, A., & Rubin, D. B. (2013). Bayesian data analysis (3rd). Chapman; Hall/CRC. https://doi.org/10.1201/b16018 Hardy, B. L., Samudra, R., & Davis, J. A. (2019). Cash assistance in america: The role of race, politics, and poverty. The Review of Black Political Economy, 46(4), 306–324. https://doi.org/10.1177/ 0034644619865272
- Hymowitz, K. S. (2015, September). The relationship between poverty and unwed births (tech. rep.). The Manhattan Institute. https://manhattan.institute/article/the-relationship-between-povertyand- unwed-births Johnson, A. A., Ott, M. Q., & Dogucu, M. (2021, December). Bayes rules! an introduction to applied bayesian modeling. CRC Press. https://www.bayesrulesbook.com/
- Macartney, S., & Ghertner, R. (2023, January). How many people participate in the social safety net? [[Online; accessed 25-March-2025]]. https : / / aspe . hhs . gov / reports / how - many - people - participate-social-safety-net
- Shaefer, H. L., Edin, K., Fusaro, V., & Wu, P. (2020). The decline of cash assistance and the well-being of poor households with children. Social Forces, 98(3), 1000–1025. https://doi.org/10.1093/sf/ soz020
- The Annie E. Casey Foundation. (2022, May). Child Welfare and Foster Care Statistics [[Online; updated 27-July-2024; accessed 25-March-2025]]. https://www.aecf.org/blog/child-welfare-and-fostercare- statistics
- United States Census Bureau. (2023e, September). 2020 SIPP Data [[Online; accessed 24-March-2025]]. https://www.census.gov/programs-surveys/sipp/data/datasets/2022-data/2020.html Wang, J. S.-H. (2015). Tanf coverage, state tanf requirement stringencies, and child well-being. Children and Youth Services Review, 53, 121–129. https://doi.org/10.1016/j.childyouth.2015.03.028
Questions
Appendix
Purpose 1: ROC Curve
Purpose 2: ROC Curve
Purpose 3: ROC Curve
Purpose 4: ROC Curve
