An Empirical Analysis of Joint Decisions on Labor Supply and Welfare Participation

Transcription

1 An Empirica Anaysis of Joint Decisions on Labor Suppy and Wefare Participation Gi Kang, Sonya K. Huffman, and Heen H. Jensen Working Paper 04-WP 361 May 2004 Center for Agricuura and Rura Deveopment Iowa State University Ames, Iowa This research was conducted whie Gi Kang, an associate professor at Cheju Nationa University, China, was visiting at Iowa State University. Sonya Huffman is adjunct assistant professor in the Department of Economics and Heen Jensen is professor of economics and head of the Food and Nutrition Poicy Division in the Center for Agricuura and Rura Deveopment at Iowa State University. The authors acknowedge hepfu comments and suggestions from Peter Orazem. The Economic Research Service, United States Department of Agricuure, provided partia funding support for the research. This pubication is avaiabe onine on the CARD Web site: Permission is granted to reproduce this information with appropriate attribution to the authors and the Center for Agricuura and Rura Deveopment, Iowa State University, Ames, Iowa For questions or comments about the contents of this paper, pease contact Heen Jensen, 578E Heady Ha, Iowa State University, Ames, IA ; Ph: ; Fax: ; E-mai: The U.S. Department of Agricuure (USDA) prohibits discrimination in a its programs and activities on the basis of race, coor, nationa origin, gender, reigion, age, disabiity, poitica beiefs, sexua orientation, and marita or famiy status. (Not a prohibited bases appy to a programs.) Persons with disabiities who require aernative means for communication of program information (Braie, arge print, audiotape, etc.) shoud contact USDA s TARGET Center at (202) (voice and TDD). To fie a compaint of discrimination, write USDA, Director, Office of Civi Rights, Room 326-W, Whitten Buiding, 14th and Independence Avenue, SW, Washington, DC or ca (202) (voice and TDD). USDA is an equa opportunity provider and empoyer. Iowa State University does not discriminate on the basis of race, coor, age, reigion, nationa origin, sexua orientation, sex, marita status, disabiity, or status as a U.S. Vietnam Era Veteran. Any persons having inquiries concerning this may contact the Director of Equa Opportunity and Diversity, 1350 Beardshear Ha,

2 Abstract Economic and wefare program factors affect the we-being of ow-income famiies and their abor suppy decisions. This study uses data from the U.S. Survey of Income and Program Participation. A nested ogit mode is estimated to expain the joint decisions to participate in Temporary Assistance for Needy Famiies (TANF) and the abor market for the popuation of famiies potentiay eigibe for TANF. The empirica findings indicate that higher wages increase abor and decrease wefare program participation; an increase in nonabor income decreases both abor market and wefare participation. Keywords: abor suppy, ow income, wefare program, wefare reform.

3 AN EMPIRICAL ANALYSIS OF JOINT DECISIONS ON LABOR SUPPLY AND WELFARE PARTICIPATION Introduction Between 1965 and 1985 the United States experienced a arge increase in the caseoad of wefare programs. There are many expanations for the high caseoad eve, and these incude ong-term wefare dependency and work disincentives. In response, recent state and federa wefare reforms have been designed to increase work participation among wefare recipients. The Persona Responsibiity and Work Opportunity Reconciiation Act (PRWORA) enacted in 1996 brought major changes in the scope, structure, and impact of programs targeted to the ow-income popuation, incuding Aid to Famiies with Dependent Chidren (AFDC), the Food Stamp Program, Medicaid, Suppementa Security Income (SSI), chid wefare, and chid support. Foowing the introduction of PRWORA, the number of wefare recipients decined across the nation. Both the reforms introduced through PRWORA and America s growing economy contributed to the decining wefare ros. Linkages among socia assistance programs have significant effects on the behavior of ow-income individuas and famiies. This study examines the effects of cash transfers on abor suppy and wefare participation decisions in order to better understand factors affecting wefare program and abor market activities of the poor. Many researchers have anayzed the effects of government transfer programs on abor suppy decisions among the ow-income popuation (see Moffitt 1992). Much of the empirica work provides insights on how wefare transfers affect abor suppy decisions of ow-income famiies and has focused on either femaes or married coupes. For exampe, Keane and Moffitt (1998) use a structura mode to examine work and muipe wefare program participation decisions among famiies headed by singe adu femaes. Hagstrom (1996) examines the effect of the Food Stamp Program on intrafamiy abor suppy and program participation decisions. Married coupes simuaneousy

4 2 / Kang, Huffman, and Jensen choose the abor suppy of the husband and wife and whether to participate in the Food Stamp Program. Hagstrom estimates a nested muinomia ogit mode and uses data from the Survey of Income and Program Participation (SIPP). He found smaer abor suppy effects for married coupes to changes in the food stamp benefit compared with those of singe parents. He aso found program participation by married coupes to be responsive to changes in food stamp benefits. The purpose of this paper is to examine the effect of the 1996 wefare reforms on the abor suppy decisions of ow-income famiies. There are three types of famiies: two-parent famiies, mae-headed famiies, and femae-headed famiies. The core of the transfer system for the ow-income popuation of the nonedery and nondisabed incudes AFDC (now Temporary Assistance to Needy Famiies [TANF]), the Food Stamp Program, Medicaid, and pubic housing. Among these, the TANF program is the most widey known cash transfer program for the poor. This study uses data from the SIPP to anayze abor market and TANF participation decisions among a ow-weah famiies in the United States. A static mode of famiy behavior is deveoped in which work and program participation is chosen to maximize a famiy utiity function given resource constraints. The mode is used to expain the joint decision to participate in TANF and the abor market for the popuation of famiies eigibe for TANF. The paper provides two approaches to expaining joint decisions of famiies. First, we estimate a bivariate probit mode of participation in the abor force and TANF program. Second, we estimate a nested ogit mode that incorporates simuaneous decisions on abor market and TANF participation. TANF Program The PRWORA gives the states a fundamenta roe in assisting poor famiies. Under TANF, the eigibiity rues and benefits differ across states. To be eigibe for TANF, an appicant famiy must pass both nonfinancia tests based on the demographic characteristics of the famiy and its members and financia tests based on the famiy s income and asset hodings. At the most basic eve of nonfinancia tests, the famiy must incude a chid or, in some states, a pregnant woman. If the head of the famiy is a

5 An Empirica Anaysis of Joint Decisions on Labor Suppy and Wefare Participation / 3 teenager, she may or may not be eigibe to receive a benefit on her own. In most states, she is eigibe ony if she is iving with her parents. The financia tests require that an appicant famiy must have sufficienty ow income and asset eves. The asset imits that states have adopted under TANF differ greaty by state. Thirty-nine states have increased the asset imit for recipients above the $1,000 imit aowed under AFDC. Twenty-two states aow recipients to accumuate additiona savings in a restricted savings account set aside for a specific purpose aowed by the state. If the famiy s tota assets exceed the amounts determined by the state, the famiy is ineigibe for TANF. Once the famiy has passed the state s asset tests, its avaiabe income is computed for eigibiity purposes. States use the tota gross income cacuated from the unit s earned and unearned income as a starting point for income eigibiity tests (Rowe 2000). Many states now impose just one income test on appicants; however, others use a combination of a gross income test, gross earnings test, and/or net income test. Net income tests require that net famiy income not exceed a maximum benefit eve that varies by famiy size and state of residence. Net income is cacuated by subtracting the state s earned income disregards from the unit s gross earned income and then adding to this amount the unit s unearned income. The net income is then compared to an income standard determined by the state. If the net income is ess than the standard, then a benefit is cacuated. Ahough states use many different formuas to determine net income, there are genera rues that most states appy. A but two states aow recipients to disregard a portion of their earned income before benefit computation and vary the units benefits by income. In more straightforward cacuations, net income is subtracted from a statedetermined standard, the so-caed payment standard, which varies by famiy size. The benefit paid is the difference. Theoretica and Empirica Mode Theoretica Mode The famiy head is assumed to choose abor suppy and wefare (here, TANF) participation simuaneousy to maximize the famiy s utiity subject to its budget

6 4 / Kang, Huffman, and Jensen constraint. Assume that a famiy s utiity is a function of eisure time and disposabe income and is represented by U = U ( H, Y, δ ) (1) where H is monthy hours of work suppied by famiy, Y is monthy disposabe income, and δ represents preferences for receiving a TANF benefit. Monthy disposabe income may be written as the foowing budget constraint: Y ( H, P ) = wh + N + P [ B( H, N) C] (2) T where w is the houry wage rate, N is nonabor income, P T is defined to be an indicator equa to 1 if the famiy head participates in the TANF program and 0 otherwise, B ( ) is the TANF benefit given the famiy s abor suppy, and C is the monetary cost of participating in the TANF program. The famiy head is assumed to simuaneousy choose H and P T that maximizes his or her utiity given in equation (1) subject to the budget constraint given in equation (2). That is, the famiy head chooses the (H, P T ) combination that gives the greatest indirect utiity. Participation in the wefare program is not costess. There are costs associated with the appication process and with appication itsef. The costs incude the transportation and time costs of appying for the program and some compiance costs whie participating. These costs vary by individua famiy and by ocation. However, there are benefits with TANF participation. A famiy with no income is eigibe to receive the maximum TANF grant. For a famiy with income, the TANF benefits are cacuated as the difference between the maximum benefit and net famiy income. The TANF benefits are cacuated according to the foowing formua: T B T = min{ P, G [ N + ( wh E( H )) BRR CC]} if wh + N < 1.85L (3) T where B T is the monthy TANF benefit, P is the maximum permitted payment in the state, G is the maximum amount paid or pay standard, E ( ) is the earnings disregard, T BRR is the benefit reduction rate, CC is chid care deductions, and L is iving costs. The variabes P, G T, L, E, and BRR vary by state and famiy size.

7 An Empirica Anaysis of Joint Decisions on Labor Suppy and Wefare Participation / 5 As Figure 1 shows, the abor suppy decision depends on the TANF benefit through its effect on the budget constraint, and TANF participation depends on abor suppy through its effect on the TANF benefit. Therefore the TANF participation and abor suppy decisions are both endogenous and interdependent. Empirica Mode The choice set is simpified by assuming that each famiy chooses to work or not to work under the abor suppy decision and to participate or not to participate under the TANF participation decision. The resuing choice set has four aernatives, each of which is a combination of the abor suppy and TANF status. Each aernative provides indirect utiity V. The subscripts and t combined denote an aernative, which is a combination of the abor suppy decision and TANF participation decision. The famiy chooses the aernative such that V for a ( ). V ( )' The indirect utiity V is assumed to be a function of known, measured variabes and an unobserved stochastic component. Two types of measured variabes exist in the context of the TANF participation and abor suppy decision probem. The first type varies across aernatives and is caed a choice-specific variabe or attributes of the choices. The choice-specific variabes incude the TANF benefit for different combinations of abor suppy and wages across hours of work choices. The second type varies by famiy rather than by aernative and is caed an individua-specific variabe or FIGURE 1: Simuaneity of abor suppy and TANF participation decisions

8 6 / Kang, Huffman, and Jensen characteristics of the individua. The individua-specific variabes incude information on the famiy head s age, education, marita status, number of chidren, and so forth. The stochastic error component captures the effect of unmeasured variabes and unobserved differences in preferences across famiies. Given the form of utiity function and the probabiity distribution of the stochastic term, the probabiity that the famiy wi choose aernative wt can be written Pr = Pr[ V V ], for a ( ). ( )' Mode I: Bivariate Probit Mode Severa modes are avaiabe for estimating the described random utiity mode based on different assumptions about the stochastic component of indirect utiity function, V (Green). 1 One mode for incorporating simuaneous decisions on abor market and TANF participation is a bivariate probit mode. Define P L and P T as participation in the abor market and the TANF program, respectivey. A famiies are then cassified into four mutuay excusive regimes based on the discrete choice outcome on P L and P T : R 1 : P L = P T = 1 (those who participate in the abor market and TANF) R 2 : P L =1 and P T = 0 (those who participate in the abor market but not in TANF) R 3 : P L = 0 and P T = 1 (those who participate in TANF but not in the abor market) R 4 : P L = P T = 0 (those who do not participate in the abor market or TANF) A observations have a nonzero probabiity of being assigned to one of four regimes. This probabiity can be evauated with the foowing bivariate probabiity statements: M 11 P(R 1 )=P(P L = 1,P T = 1)=P[P L * =θ L Z L +µ L >0, P T * =θ T Z T +µ T >0] (4) M 10 P(R 2 )=P(P L = 1,P T = 0)=P[P L * =θ L Z L +µ L >0, P T * =θ T Z T +µ T 0] (5) M 01 P(R 3 )=P(P L = 0,P T = 1)=P[P L * =θ L Z L +µ L 0, P T * =θ T Z T +µ T >0] (6) M 00 P(R 4 )=P(P L = 0,P T = 0)=P[P L * =θ L Z L +µ L 0, P T * =θ T Z T +µ T 0] (7) Ahough P * L and P * T are unobservabe variabes, we can observe the dummy variabes P L and P T such that P L =1 if P * L >0 and P L =0 otherwise, and P T =1 if P T >0 and P T =0 otherwise. Define Z L and Z T as vectors of exogenous variabes, θ L and θ T as parameter vectors, and µ L and µ T as disturbance terms. Maximum-ikeihood estimation

9 An Empirica Anaysis of Joint Decisions on Labor Suppy and Wefare Participation / 7 of bivariate probit regressions are used to estimate θ L and θ T. These estimates are used to cacuate the probabiities of statements (4)-(7). The empirica specification of the human-capita based wage equation is Ln( W ) β + β X + β Y + ε (8) = where X is a vector of exogenous variabes incuding education, marita status, gender, race, and metro/nonmetro ocation of the famiy head; Y is a vector of other exogenous variabes, incuding oca unempoyment rate, experience, and an interaction term between experience and education; and ε w is a norma random error. The wage equation aso needs to be corrected for potentia seection bias. w Mode II: Nested Logit Mode A second mode that is widey used in the discrete choice iterature is the muinomia ogit mode (or conditiona ogit mode), which can be easiy estimated for arge choice sets (Green 2000; Hensher and Green 2000). The muinomia ogit mode assumes that the stochastic errors are uncorreated across aernatives. When the assumption on independence of irreevant aernatives does not hod, a nested muinomia ogit mode can be used. The nested muinomia ogit mode aows the error terms to be correated across aernatives in a group but to be independent of aernatives in a different group. Figure 2 shows the nesting structure of the mode to be estimated. FIGURE 2: Tree structure for nested ogit mode

10 8 / Kang, Huffman, and Jensen The indirect utiity function V is decomposed into components in the foowing way: V = X α + Xα + Y β + Yβ + ε t (9) where X is a vector of observed attributes that vary across TANF participation aernatives, Y is a vector of observed attributes that vary across the famiy s abor suppy aernatives, and the and t take on vaues to indicate whether or not there is participation in the abor force () or in TANF (t). In addition to these choice-specific variabes, we aso incude the vectors X and Y, which vary by individua famiy rather than by aernatives. In our specification, X is a measure of the TANF benefit that varies across the famiy s abor suppy choices, and Y is a vector of wage variabes that differ across abor suppy aernatives. The individua-specific characteristics incude components of the budget constraint, variabes that capture variations in preferences. The joint probabiity of choosing aernative can be written as P (, t) = P( t ) P( ) (10) where P ( t ) is the probabiity of choosing t conditiona on the choice of, and P () is the margina probabiity of choosing for the famiy. If the ε is extreme-vaue distributed, then the probabiities in equation (10) can be written as P( t ) = exp( X α + Xα ) t' T exp( X α + Xα ) t t (11) P( ) = exp( Y β + Yβ + τ I ) ' L exp( Y β + Yβ + τ I ) (12) where I = og t' T exp( X α + Xα ). (13) t

11 An Empirica Anaysis of Joint Decisions on Labor Suppy and Wefare Participation / 9 The term I is the incusive vaue in the abor suppy equation and can be interpreted as a measure of the sum of the utiity for choice t given choice of. The nested ogit aspect of the mode arises when the coefficients of the incusive vauesτ differ from 1. To be consistent with the random utiity mode, the utiity function for the four aernatives can be specified in the foowing way: where Vyy = αyy + βbbyy + βrbrr+ ( γw, + γw, t ) W + ( γm, + γm, t ) M + ( γc, + γc, t ) C + ( γ + γ ) N + ( γ + γ ) U + ε V V V N, Nt, U, Ut, yy yn = yn + β b B yn + β r BRR + γ W, W + γ M, M + γ C, C + γ N, N + γ U, ny = ny + β b Bny + β r BRR + γ W, tw + γ M, t M + + γ C, tc + γ N, t N + γ U, t nn α U + ε α U + ε = β B + β BRR + ε (14) b nn r nn B is the pay standard, BRR is the benefit reduction rate, 2 W is the predicted wage, M is a famiy head s gender, C is the number of chidren in the famiy, N is nonabor income to the famiy, and U is the state unempoyment rate. yn ny Data and Estimation Resus The effectiveness of wefare programs has been debated for a ong time. The difficuy of evauating these programs comes from the ack of appropriate data. The avaiabiity of the SIPP data now provides evidence for evauating the efficacy of wefare programs. One of the major goas of SIPP is to examine interactions among transfer programs, abor force participation, and iving arrangements because it contains detaied information about the characteristics of, and actua choices made by, both participants and nonparticipants. The 1996 SIPP (wave 3) data was used in this study. Ony nonedery (under 65 years of age), nondisabed famiy heads are incuded in the sampe because both edery and disabed peope are eigibe for other transfer programs. Famiy heads are aso excuded if they are categoricay ineigibe for the TANF program, that is, if they do not have any chidren under age 18 in the famiy. In this study we assume that the famiy s assets are exogenous. Therefore famiies with assets that exceed the asset imit, described in Tabe 1, are excuded from the sampe. We do not use income eigibiity in seecting the sampe because hours of work is an endogenous variabe, that is, the famiy head s decision to earn an amount that causes the

12 10 / Kang, Huffman, and Jensen TABLE 1. TANF asset imits State Asset Limits ($) State Asset Limits ($) Aabama 2,000 Mississippi 1,000 Aaska 1,000 Missouri 5,000 Arizona 2,000 Montana 3,000 Arkansas 3,000 Nebraska 6,000 Caifornia 2,000 Nevada 2,000 Coorado 2,000 New Hampshire 2,000 Connecticut 3,000 New Jersey 2,000 Deaware 1,000 New Mexico 1,500 Dist. of Coombia 1,000 New York 2,000 Forida 2,000 North Caroina 3,000 Georgia 1,000 North Dakota 5,000 Hawaii 5,000 Ohio 1,000 Idaho 2,000 Okahoma 1,000 Iinois 3,000 Oregon 2,500 Indiana 1,500 Pennsyvania 1,000 Iowa 5,000 Rhode Isand 1,000 Kansas 2,000 South Caroina 2,500 Kentucky 2,000 South Dakota 2,000 Louisiana 2,000 Tennessee 2,000 Maine 2,000 Texas 2,000 Maryand 2,000 Washington 1,000 Massachusetts 2,500 West Virginia 2,000 Michigan 3,000 Wisconsin 2,500 Minnesota 5,000 Wyoming 2,500 Source: Gaagher et a famiy income to exceed the breakeven eve is a matter of choice. We seect whoever gets a higher wage as the famiy head in the case of married coupe famiies, as it is more ikey that the famiy member who receives a higher wage woud work first. The resuing sampe incudes 6,404 famiies with ow weah. Of these, 58 percent are married coupe famiies, and 78 percent ive in metro areas. A the dependent variabes are defined for the month of November A famiy is recorded as a TANF participant if a member reports receiving TANF support in the month. For the abor suppy, a famiy head is cassified as not working if he or she reports

13 An Empirica Anaysis of Joint Decisions on Labor Suppy and Wefare Participation / 11 working zero hours during the month and as working if he or she reports working at east one hour per week during the month. Variabes used in our anaysis incude a set of demographic variabes and a set of structura variabes designed to capture differences in abor market conditions and transfer programs. The demographic variabes for the famiy head incude gender, education eve, race, marita status, and experience, which is defined by age minus education minus 6. The set of individua characteristics incude a metro variabe that indicates that the famiy ives in a metro area versus nonmetro area, the state s monthy unempoyment rate, unearned income, the number of chidren under age 6, and program participation choices. The wage rate is predicted and then used in the foowing anaysis. Tabe 2 dispays the means and standard deviations of variabes. The payment standard is the maximum TANF grant per month a participant can get if he or she is eigibe for TANF. The actua TANF benefit a person can get depends on his or her participation in the abor force and income eigibiity. Tabe 3 shows the distribution of the dependent variabes: abor market and wefare program participation. About 9 percent of the asset-eigibe famiies receive a TANF grant, and 84 percent of them participate in the abor market. The workers are concentrated, at 81 percent of the sampe, in the TANF nonparticipation ce. Ten percent of those in the sampe do not work and do not participate in TANF; 6 percent do not work and participate in TANF; and 3 percent work and participate in TANF. Before the estimation of the bivariate probit mode of abor force and wefare participation, we estimate a wage equation for the famiy heads (if singe famiy) and spouse (for married famiy), and then use the predicted wage as an instrument of the actua wage. The estimates of the wage equation are reported in Tabe 4. Added experience increases the famiy head s wage through increased abor productivity, hoding other things equa. The wage equation is concave in reation to experience. One additiona year of experience has the direct effect of increasing the wage by 5.4 percent. The findings on other variabes are consistent with other studies. Being mae increases the individua s wage. Individuas iving in metro areas receive higher wage rates than those iving in nonmetro areas.

14 12 / Kang, Huffman, and Jensen TABLE 2. Definitions, mean, and standard deviations of variabes (N=6,404) Mean (Std. Variabe deviation) Definition Age (8.74) Age of famiy head Education (2.66) Years of schooing of famiy head Mae 0.45 (0.5) Dichotomous variabe equa to 1 if famiy head is mae Married 0.58 (0.49) Dichotomous variabe equa to 1 if famiy head is married White 0.75 (0.43) Dichotomous variabe equa to 1 if famiy head is white Metro 0.78 (0.41) Dichotomous variabe equa to 1 if famiy head ives in metro Kids (0.83) Number of chidren under 6 Experience (9.05) Age-Education-6 Unempoyment rate 5.25 (1.05) Loca state unempoyment rate (%) Nonabor income (344.06) Famiy nonabor income per month ($) Wage 9.46 (1.99) Predicted houry wage ($) Payment standard (213.00) Maximum TANF grant per month given participation ($) BRR 0.53 (0.18) The benefit reduction rate is the rate at which additiona doars of earned income reduce the amount transferred Labor force participation 0.84 (0.37) Dichotomous variabe equa to 1 if famiy head works TANF participation 0.09 (0.29) Dichotomous variabe equa to 1 if famiy head participates in TANF TABLE 3. Distribution of the sampe by abor suppy and wefare participation Wefare Program (TANF) Participation No participation A Work 199 3% 5,165 81% 5,364 84% Not work 397 6% % 1,040 16% A 596 9% 5,808 91% 6, %

15 An Empirica Anaysis of Joint Decisions on Labor Suppy and Wefare Participation / 13 TABLE 4. Estimates of the og wage equation Variabes Estimates Intercept (0.062) *** Education (0.003) *** Mae (0.021) *** Metro (0.012) *** Experience (0.002) *** Experience squared (0.0001) *** Lambda (0.049) *** R-square 0.22 F statistics Number of observations 5,497 Notes: * Statisticay significant at the 10% eve. ** Statisticay significant at the 5% eve. *** Statisticay significant at the 1% eve. Standard errors are in parentheses. Tabe 5 presents the fu information maximum ikeihood (FIML) estimates of the bivariate probit mode of abor force and wefare participation. Variabes, which enter directy into the famiy budget constraint, incude the nonabor income and TANF benefits. Nonabor income incudes a nonwage famiy income excuding income from any transfer programs. As expected, nonabor income has a statisticay significant negative effect on abor and TANF participation. Predicted wage has a statisticay significant negative effect on TANF participation and a positive effect on abor force participation. As expected, a higher unempoyment rate makes it difficu to find jobs and thus has a statisticay significant negative effect on abor market participation and a positive effect on wefare participation. Having younger chidren increases the probabiity of TANF participation and decreases the probabiity of abor market participation. A TANF benefit (pay standard) woud be expected to increase the probabiity of TANF participation and decrease the probabiity of abor force participation. The resu is consistent with this, and the effect of a TANF benefit on both TANF participation and abor force participation is statisticay significant. Severa variabes are aso incuded to capture the differences in taste and opportunities across famiies. The signs of these coefficients suggest that married famiies and

16 14 / Kang, Huffman, and Jensen TABLE 5. Fu information maximum ikeihood estimates of the bivariate probit mode Variabes Labor Participation TANF Participation Intercept (0.193) (0.241) Married (0.051) *** (0.079) *** Mae (0.095) *** (0.115) *** White (0.046) *** (0.055) *** Experience (0.006) - Experience squared (0.011) *** - Kids (0.028) *** (0.032) *** Educy (0.015) (0.015) Nonabor income (0.028) *** (0.007) *** Predicted wage (0.029) *** (0.031) *** Unempoyment rate (0.022) *** (0.027) ** Pay standard (0.011) *** (0.014) *** BRR (0.121) (0.150) Rho (correation coefficient) (0.022) *** Notes: * Statisticay significant at the 10% eve. ** Statisticay significant at the 5% eve. *** Statisticay significant at the 1% eve. Standard errors are in parentheses. those whose famiy head is white are ess ikey to participate in TANF. White or mae famiy heads are more ikey to participate in the abor market whie married famiy heads are ess ikey to work. Having chidren increases the probabiity of TANF participation and decreases the probabiity of abor market participation. The cross-equation correation between the abor suppy and TANF participation decisions is 0.68 and statisticay significant. This is an indirect indication that the two decisions are reated through the error term. Tabe 6 presents the estimates of margina effects in the bivariate probit mode. In the abor suppy equation, being a married famiy head has the strongest negative effect and being a mae-headed famiy has the strongest positive effect on the probabiity of abor market participation given TANF participation. The presence of one additiona

17 An Empirica Anaysis of Joint Decisions on Labor Suppy and Wefare Participation / 15 TABLE 6. Margina effects in the bivariate probit mode Direct Indirect Tota Std. Error Labor suppy Married *** Mae *** White * Unempoyment ** Experience Experience squared *** Kids *** Educy Nonabor income *** Wage * Pay standard * BRR TANF participation Married *** Mae White *** Unempoyment Experience Experience squared *** Kids *** Educy Nonabor income *** Wage *** Pay standard *** BRR Note: * Statisticay significant at the 10% eve. ** Statisticay significant at the 5% eve. *** Statisticay significant at the 1% eve.

18 16 / Kang, Huffman, and Jensen chid under age 6 woud decrease the probabiity of working by 4 percent. Nonabor income has a significant negative effect and wages have a significant positive effect on the probabiity of working; both nonabor income and wages have statisticay significant negative effects on the probabiity of wefare participation. In the TANF participation equation, most variabes seem to have reativey smaer effects on the probabiity of TANF participation given abor market participation. This impies that if famiy heads work, it makes reativey ess difference whether or not they participate in TANF. The FIML estimates of the nested ogit mode are presented in Tabe 7. The estimated coefficients are interpreted with respect to the no abor no TANF TABLE 7. Fu information maximum ikeihood estimates of nested ogit mode Coefficients Estimates α yy (10.894) α yn (10.875) α ny (0.424) *** β b (0.034) β r (0.003) γ W, (0.189) γ W,t (0.036) *** γ M, (2.651) γ M,t (0.191) *** γ C, (0.136) γ C,t (0.049) *** γ N, (0.333) γ N,t (0.017) *** γ U, (0.311) γ U,t (0.039) * IV work (2.486) IV no work (0.772) *** Log ikeihood Notes: * Statisticay significant at the 10% eve. ** Statisticay significant at the 5% eve. *** Statisticay significant at the 1% eve. Standard errors are in parentheses.

19 An Empirica Anaysis of Joint Decisions on Labor Suppy and Wefare Participation / 17 participation category. Higher wages ead to ess participation in TANF (γ W,t ) and, ahough not statisticay significant, make famiy heads work more (γ W, ). The coefficients on TANF benefits (β b ) and BRR (β r ) are not statisticay significant. Mae famiy heads tend to work more (γ M, ) and participate ess in TANF (γ M,t ). Famiies with more young chidren (under age 6) participate more in TANF(γ C,t ). Famiy heads with more unearned income participate ess in TANF (γ N,t ) and, ahough not statisticay significant, tend to work ess (γ N, ). The unempoyment rate has a positive and statisticay significant effect on the probabiity of TANF participation (γ U,t ). Easticities for the reationship between the TANF benefit and wages and the probabiity of choosing aernative (abor suppy and TANF participation) are reported in Tabe 8. A 10 percent increase in the pay standard wi increase by 1.60 percent the probabiity of the choice to participate in the abor market and in TANF (L-T), whie a 10 percent increase in the wage wi increase by 4.72 percent the probabiity of that choice. The easticity with respect to the (predicted) wage is quite arge in the choice in which the famiy has determined not to work and to participate in TANF (N-T). An increase in the wage by 10 percent decreases the probabiity of this choice (N-T) by percent. This resu suggests the importance of improved wages for moving famiies into the abor force and away from wefare program participation. TABLE 8. Easticities of abor suppy and TANF participation decisions with respect to pay standard and wages Direct Easticities Pay standard L-T L-N N-T N-N Wages L-T L-N N-T N-N n/a Notes: L-T is choice to work and to participate in TANF. L-N is choice to work and not to participate in TANF. N-T is choice not to work and to participate in TANF. N-N is choice not to work and not to participate in TANF.

20 18 / Kang, Huffman, and Jensen Concusions This study anayzes the abor force and wefare participation choices made by owweah famiies and the effect of the 1996 wefare reform on the abor suppy decisions of these famiies. Two modes were estimated: the bivariate probit and the nested ogit modes. From the bivariate probit mode we found evidence of endogeneity: the two choices (abor suppy and program participation) are reated through the error term, and the correation is negative and statisticay significant. The estimates from the nested ogit mode of the joint househod choices indicate that higher wages increase abor force participation and decrease wefare program participation; an increase in nonabor income decreases both abor and wefare participation. These resus support strategies for improving wages and other nonabor income for ow-resource famiies as a means to reduce their reiance on wefare programs.

21 Endnotes 1. Maddaa (1983) presents an extensive discussion of imited-dependent and quaitative-variabe modes in econometrics. 2. The term B is the maximum TANF grant per month; BRR is the benefit reduction rate, the rate at which additiona doars of earned income reduce the TANF benefit.

22 References Gaagher, J., M. Gaagher, K. Perese, S. Schreiber, and K. Watson One Year After Wefare Reform: A Description of State Temporary Assistance for Needy Famiies (TANF) Decisions as of October Research Report # The Urban Institute, Washington, DC. May. Green, W Econometric Anaysis, 4th ed. Engewood Ciffs, NJ: Prentice-Ha. Hagstrom, P The Food Stamp Participation and Labor Suppy of Married Coupes: An Empirica Anaysis of Joint Decisions. Journa of Human Resources 31(2, Spring): Hensher, D., and W. Green Specification and Estimation of the Nested Logit Mode: Aernative Normaisations. Mimeo. February. Keane, M., and R. Moffitt A Structura Mode of Muipe Program Participation and Labor Suppy. Internationa Economic Review 39(3): Maddaa, G.S Limited-Dependent and Quaitative Variabes in Econometrics. Cambridge: Cambridge University Press. Moffitt R Incentive Effects of the U.S. Wefare System: A Review. Journa of Economic Literature 30(March): Rowe, G State TANF Poicies as of Juy In Wefare Rues Databook. The Urban Institute, Washington, DC. November.