The Pennsylvania State University. The Graduate School. College of Agricultural Sciences ESSAYS ON WELFARE USE, THE WAGE GAP AND UNEMPLOYMENT

Transcription

1 The Pennsylvana State Unversty The Graduate School College of Agrcultural Scences ESSAYS ON WELFARE USE, THE WAGE GAP AND UNEMPLOYMENT TRANSITIONS IN THE UNITED STATES A Thess n Agrcultural, Envronmental and Regonal Economcs & Demography by Quyan Wang Copyrght 2004 Quyan Wang Submtted n Partal Fulfllment of the Requrements for the Degree of Doctor of Phlosophy May 2004

2 The thess of Quyan Wang was revewed and approved* by the followng: Jll L. Fndes Professor of Agrcultural, Envronmental and Regonal Economcs, and Demography Thess Advsor Char of Commttee Lef I. Jensen Professor of Rural Socology, Socology, and Demography James S. Shortle Dstngushed Professor of Agrcultural and Envronmental Economcs Mark D. Hayward Professor of Socology and Demography Stephen M. Smth Professor of Agrcultural and Regonal Economcs Commttee Member and Head of the Department of Agrcultural Economcs and Rural Socology *Sgnatures are on fle n the Graduate School

3 ABSTRACT Ths dssertaton offers three self-contaned emprcal analyses of dfferent demographc groups n the U.S. labor market to fulfll the requrements for the degree of Doctor of Phlosophy n the dual-ttle program n Agrcultural, Envronmental and Regonal Economcs & Demography. Specfcally, the dssertaton addresses (1) the rural-urban wage gap among women n the U.S., (2) unemployment transtons of lowncome persons n the nonmetropoltan U.S., and (3) the nfluence of the employment rate n the explanaton of the relatve declne n welfare recept between mmgrants and U.S. natve-born ndvduals after welfare reform. The frst study utlzes Panel Study of Income Dynamcs (PSID) data to evaluate possble reasons for the persstent rural-urban wage gap for women over the perod n the Unted States. The paper adopts the general equlbrum model of nteractng regons as a startng pont, then uses two-step panel data estmatons that allow consderaton of explanatons ncludng rural-urban dfferences n observed ndvdual characterstcs and work-related characterstcs, and at the same tme control for both unobserved dfferences n natural amentes and productvtes and for the potental effects of endogenety and/or sample selecton bas n hours worked. The paper fnds that sgnfcant rural-urban female wage dfferentals exst for many groups even when controllng for cost-of-lvng dfferentals, ndcatng that the functonng of rural and urban labor markets are dfferent n the Unted States.

4 The second study analyzes the labor market outcomes of unemployed low-ncome persons across metro and nonmetro U.S. labor markets. Usng matched data from the March Current Populaton Survey Annual Demographc Fles from 1988 to 2002, ths study examnes the transton probabltes of unemployed persons lvng n poor/near poor famles; n poor, sngle-parent famles; and n famles recevng welfare. The employment transtons are estmated for movement nto dfferent labor force states n a dscrete choce model framework usng multnomal probt models. Two measures of employment status monthly labor force status and yearly labor force status are used to capture the concept of unemployment wth dfferent tme frames. To supplement the transton analyss of employment outcomes, the paper also examnes the dstrbuton of earnngs of those unemployed low-ncome people who found employment, by rural and urban locaton. The study fnds that there are dfferences n the lkelhood of leavng unemployment for rural low-ncome persons compared to urban ones, by both monthly and yearly labor force status measures. Based on the monthly labor force status measure, rural low-ncome unemployed resdents are generally more lkely to fnd part-tme jobs than ther urban counterparts n transtons out of unemployment. However, gven the yearly labor force status measure, the paper fnds that greater persstence exsts n partyear or seasonal unemployment for rural low-ncome unemployed persons than urban ones. Rural low-ncome unemployed workers are generally more lkely to reman unemployed over tme. Further, for rural workers who found employment the next year, ther hourly wages and earnngs are generally lower than ther urban counterparts even after they have gotten better jobs. v

5 The thrd study attempts to explan the apparent dfferences n use of welfare between mmgrants and U.S. natve-born ndvduals after welfare reform. The Personal Responsblty and Work Opportunty Reconclaton Act of 1996 (PRWORA), enacted August 22, 1996, sgnfcantly lmts federal means-tested benefts to legal mmgrants who have not become U.S. ctzens. Ths research addresses the observed dfferences n the declnes n welfare use rates by mmgrants as compared to U.S. natves by examnng the relatonshp between welfare use and employment by natvty dfferences and by gender from 1993 to 2002 usng smultaneous-equaton technques. In part, dfferences n the rates of declne n welfare use snce PRWORA result not only from changes n the rules for qualfyng for welfare, but also from dfferences n changes n employment rates. The paper tests the hypothess that the observed post-welfare reform decrease n welfare partcpaton rates among mmgrants and natves s condtonal on changes n ther employment rates. The study fnds that the decsons relatve to welfare partcpaton and employment are negatvely related and jontly determned for most welfare programs. Moreover, the study shows that the employment rate can explan around one-thrd of the greater relatve declnes among mmgrant men compared to natve men n welfare recept n the post-prwora perod. However, the employment rate can only explan a small part of the relatve declnes n welfare use rates among mmgrant women compared to natve women n the post-reform perod. Fnally, the study fnds that natvty dfferences n welfare partcpaton rates among mmgrants and natves are senstve to the unt of analyss (e.g., at the ndvdual as compared to the household level). Ths analyss s based on data from the March Current Populaton v

6 Survey Annual Socal and Economc Supplements (formerly Annual Demographc Survey) v

7 TABLE OF CONTENTS LIST OF FIGURES x LIST OF TABLES.x ACKNOWLEDGEMENTS xv 1. RURAL-URBAN WAGE GAP OF WOMEN IN THE UNITED STATES 1.1 Abstract Introducton Lterature Revew Theoretcal Model Econometrc Model and Estmaton Strategy..14 Testng for Rural-Urban Dfferences Data Emprcal Results...24 Descrptve Analyss..24 Hours Worked by Women...26 Wages Earned by Women..29 Comparsons Across Wage Equatons...31 Statstcal Tests n the Wage Equatons...32 Labor Moblty and the Rural-Urban Wage Gap...34 Dfferences n Returns Further Robustness Test...41 v

8 1.8 Conclusons Endnotes References Appendxes 57 Appendx A. Rural-Urban Contnuum Code (Beale Code) n the PSID...57 Appendx B. Real Average Hourly Earnngs by State for Women Reportng Postve Hours of Work Appendx C. Treatment of Intal Condtons EMPLOYMENT PROSPECTS OF UNEMPLOYED LOW-INCOME PERSONS IN THE NONMETROPOLITAN U.S. 2.1 Abstract Introducton Background and Prevous Studes Data and Measurements Econometrc Model and Estmaton Strategy Results Descrptve Profles of Poverty Level and Labor Force Status 103 Descrptve Analyss of Unemployment Dynamcs.107 Models of Unemployment Transtons 110 Independent Varables n the Transton Models.111 Descrptve Statstcs for the Sample of Transton Models 114 v

9 Transtons Out of Unemployed/Dscouraged Workers..116 Transtons for Unemployed Low-Income Persons.122 Transtons Out of Part-Year Searchng Employment.127 Transtons for Part-Year Low-Income Workers.129 Earnngs for Low-Income Persons Conclusons Endnotes References WELFARE USE AND EMPLOYMENT: THE IMPACT OF WELFARE REFORM ON IMMIGRANTS 3.1 Abstract Introducton Background and Lterature Revew Data and Descrptve Analyss Trends n Welfare Partcpaton Rates.193 Descrptve Analyss of Employment and Selected Characterstcs of Natves and Immgrants Econometrc Model and Estmaton Strategy Results of Smultaneous Equaton Models Descrptve Statstcs of the Sample 209 Smultaneous Equaton Estmates of Employment Rates 211 x

10 Smultaneous Equaton Estmates of Welfare Partcpaton Rates..214 Natves Versus Foregn-born Welfare Partcpaton n Specfc Programs.221 Ctzens Versus Nonctzens Unt of Analyss Summary and Conclusons 227 Observatons Based on the Descrptve Results Emprcal Analyss Results Endnotes References Appendx Naturalzaton Rates (%) of Workng-age Foregn-born Persons by Selected States 242 x

11 LIST OF FIGURES 1. RURAL-URBAN WAGE GAP OF WOMEN IN THE UNITED STATES Fgure 1.1 Natural Amentes Scales for the 48 Contguous States of the U.S x

12 LIST OF TABLES 1. RURAL-URBAN WAGE GAP OF WOMEN IN THE UNITED STATES Table 1.1 Varable Defntons..64 Table 1.2 Nomnal and Real Average Hourly Earnngs by Year and Resdence..67 Table 1.3 Total Annual Work Hours by Year and Resdence...68 Table 1.4 Selected Varable Means by Resdence...69 Table 1.5 Dstrbuton of Industry and Occupaton Varables and Mean Average Hourly Earnngs for Women Reportng Postve Hours of Work Table 1.6 Maxmum Lkelhood Estmaton Results for Hours of Work Equaton...71 Table 1.7 Wage Equaton Estmates for All Sampled Women Reportng Postve Hours of Work.. 72 Table 1.8 Re-estmated Wage Equaton for All Sampled Women Reportng Postve Hours of Work...75 Table 1.9 Coeffcent Estmates for Rural Varables by Selected Characterstcs n the Wage Equatons for Women Reportng Postve Hours of Work Table 1.10 Wage Equaton for Rural-Urban Dfferences..79 Table 1.11 Wald Test of Equalty of Rural-Urban Coeffcents n Wage Equaton..80 Table 1.12 Coeffcent Estmates for Rural Varables for Dfferent Specfcatons..81 x

13 2. EMPLOYMENT PROSPECTS OF UNEMPLOYED LOW-INCOME PERSONS IN THE NONMETROPOLITAN U.S. Table 2.1 Dstrbuton of Famly Income Level by Resdence for Non-elderly Persons Table 2.2 Monthly Labor Force Status of Men and Women by Resdence.154 Table 2.3 Yearly Labor Force Status of Men and Women by Resdence Table 2.4 Detaled Yearly Labor Force Status of Men and Women by Resdence.156 Table 2.5 Dstrbuton of Poverty Level by Labor Force Status..157 Table 2.6 Outflow Transtons for Unemployed/Dscouraged Workers and Part-Year Searchng Workers Table 2.7 Inflow Transtons for Unemployed/Dscouraged Workers and Part-Year Searchng Workers Table 2.8 Descrptve Statstcs for Unemployed/Dscouraged Men Table 2.9 Descrptve Statstcs for Unemployed/Dscouraged Women.161 Table 2.10 Descrptve Statstcs for Part-Year Searchng Men Table 2.11 Descrptve Statstcs for Part-Year Searchng Women Table 2.12 MNP Estmates of Transtons for Unemployed/Dscouraged Men..164 Table 2.13 MNP Estmates of Transtons for Unemployed/Dscouraged Women.165 Table 2.14 MNP Estmates of Transtons for Unemployed/Dscouraged Poor/Near Poor Men Table 2.15 MNP Estmates of Transtons for Unemployed/Dscouraged Poor/Near Poor Women x

14 Table 2.16 MNP Estmates of Transtons for Unemployed/Dscouraged Persons Lvng n Poor Sngle-Parent Famles Table 2.17 MNP Estmates of Transtons for Unemployed/Dscouraged Persons Lvng n Famles Recevng Welfare Table 2.18 MNP Estmates of Transtons for Part-Year Searchng Men Table 2.19 MNP Estmates of Transtons for Part-Year Searchng Women Table 2.20 MNP Estmates of Transtons for Part-Year Searchng Poor/Near Poor Men Table 2.21 MNP Estmates of Transtons for Part-Year Searchng Poor/Near Poor Women Table 2.22 MNP Estmates of Transtons for Part-Year Searchng Workers Lvng n Poor Sngle-Parent Famles Table 2.23 MNP Estmates of Transtons for Part-Year Searchng Workers Lvng n Famles Recevng Welfare Table 2.24 Dstrbuton of Wage and Income for Selected Workers Lvng n Poor/Near Poor Famles Table 2.25 Dstrbuton of Wage and Income for Selected Workers Lvng n Poor Sngle-Parent Famles and Famles Recevng Welfare Table 2.26 Dstrbuton of Wage and Income for Selected Workers Lvng n Poor/Near Poor Marred-Couple Famles. 178 xv

15 3. WELFARE USE AND EMPLOYMENT: THE IMPACT OF WELFARE REFORM ON IMMIGRANTS Table 3.1 Descrptons of Selected Means-Tested Programs..243 Table 3.2 Welfare Partcpaton Rates (%) for Workng-age (18-64 years old) Natve and Foregn-born Persons Table 3.3 Welfare Partcpaton Rates (%) for Workng-age (18-64 years old) Ctzens and Nonctzens Table 3.4 Welfare Partcpaton Rates (%) by Selected States for Workng-age Natve and Foregn-born Persons Table 3.5 Welfare Partcpaton Rates (%) by Selected States for Workng-age Ctzens and Nonctzens Table 3.6 Welfare Partcpaton Rates (%) for Natves and Foregn-born Households Table 3.7 Welfare Partcpaton Rates (%) for Natve and Foregn-born Males.249 Table 3.8 Welfare Partcpaton Rates (%) for Natve and Foregn-born Females.250 Table 3.9 Dstrbuton of Welfare Partcpaton Rates (%) by Immgrant Cohorts.251 Table 3.10 Welfare Partcpaton Rates (%) by State Safety Net Avalablty for Immgrants Table 3.11 Dstrbuton of Labor Force Status for Workng-age Natves and Foregn-born Table 3.12 Dstrbuton of Selected Characterstcs for Workng-age Natves and Foregn-born xv

16 Table 3.13 Welfare Partcpaton Rates (%) by Labor Force Status for Workng-age Natves and Foregn-born Table 3.14 Welfare Partcpaton Rates (%) and Annual Employed Hours for Workng-age Natves and Foregn-born Table 3.15 Dstrbuton of Sample for Workng-age Natves and Foregn-born Table 3.16 Sample Selecton Two-Step Estmaton of Ftted Wage for Workng-age U.S. Men Table 3.17 Sample Selecton Two-Step Estmaton of Ftted Wage for Workng-age U.S. Women Table 3.18 Smultaneous Equaton Estmates of Employment Rates for Workng-age U.S. Men Table 3.19 Smultaneous Equaton Estmates of Employment Rates for Workng-age U.S. Women Table 3.20 Estmates of Welfare Partcpaton Rates for Workng-age U.S. Men (Foregn-born Versus Natves) Table 3.21 Estmates of Welfare Partcpaton Rates for Workng-age U.S. Women (Foregn-born Versus Natves).269 Table 3.22 Estmates of Welfare Partcpaton Rates n Specfc Programs for Workngage U.S. Men (Foregn-born Versus Natves) for Models 1 and Table 3.23 Estmates of Welfare Partcpaton Rates n Specfc Programs for Workngage U.S. Women (Foregn-born Versus Natves) for Models 1 and xv

17 Table 3.24 Estmates of Welfare Partcpaton Rates for Workng-age U.S. Men (Nonctzens Versus Ctzens) Table 3.25 Estmates of Welfare Partcpaton Rates for Workng-age U.S. Women (Nonctzens Versus Ctzens) Table 3.26 Estmates of Welfare Partcpaton Rates for U.S. Households Wth a Male Head Table 3.27 Estmates of Welfare Partcpaton Rates for U.S. Households Wth a Female Head xv

18 ACKNOWLEDGEMENTS Ths dssertaton would not have been wrtten wthout advce and support from many very specal people and nsttutons; to each and all of them, I am apprecatve and ndebted. I would lke to express my deepest apprecaton to my advsor and mentor, Dr. Jll Fndes, for her gudance, encouragement, and patence necessary for the successful completon of ths research. She has always beleved n my abltes and gven me many opportuntes to let my sklls and knowledge develop and mature. I have been blessed wth such a wonderful advsor and truly nsprng mentor. I am also so grateful to her extreme kndness and generosty throughout my entre graduate lfe. I also extend my deep thanks and grattude to my commttee member Lef Jensen, Jm Shortle, Steve Smth and Mark Hayward for ther valuable nsghts, constructve comments and great support for my dssertaton research and career development. I feel truly honored to have these great people as my commttee member. Ths research was also benefted greatly from several scholars help and nsghts. I am very grateful to Susumu Ima s techncal assstance, gudance and comments on my research. I also would lke to thank Gordon De Jong, Frank Heland, Shrley Porterfeld and Zhou Yu for ther helpful comments. In addton, I want to thank Davd Rbar for data assstance on my research. Ths research was conceved and carred out at the Penn State Unversty. Fnancal and nsttutonal support from the Department of Agrcultural Economcs & Rural Socology and Populaton Research Insttute s greatly acknowledged. I also want to thank Dr. Jll Fndes, College of Agrcultural Scences and Women n Scences and xv

19 Engneerng Insttute for ther generous travel support for my presentaton at the professonal meetngs. I would lke to thank the entre admnstratve staff and computer support group at Penn State, especally the knd support and crucal assstance of Claudo, Rose Ann and Donna, each of these wonderful people were always wllng to help to go beyond the call of duty to help solve a problem. I am also so grateful to all my frends at the Armsby buldng. They made my graduate lfe meanngful and enjoyable. Specal thanks to Hema, Flora, Mng-chn, Tanya, Yan-guo, Wang L, Bao-chang, Latka, Koel, Debb and Anuja for ther constant support and even presence n the Armsby buldng. I also want to thank my frends, Zhen-me, Yn Mo, Shuo Chen, Jng-xang, Jng-wen, Hu Lu and Yu-an, whose frendshp and encouragement helped me realze my dream of Ph.D. I wsh to express my deepest grattude to my parents and my dear frend Robn for ther endless support and confdence n me. Wthout ther love, none of ths would have been possble. Fnally, I want to thank everyone who made ths experence possble and absolve everyone, except myself, from any responsblty for errors, omssons, or faulty logc. xx

20 1. RURAL-URBAN WAGE GAP OF WOMEN IN THE UNITED STATES 1.1 ABSTRACT Ths study utlzes Panel Study of Income Dynamcs (PSID) data to evaluate possble reasons for the persstent rural-urban wage gap for women over the perod n the Unted States. The paper adopts the general equlbrum model of nteractng regons as a startng pont, then uses two-step panel data estmatons that allow consderaton of explanatons ncludng rural-urban dfferences n observed ndvdual characterstcs and work-related characterstcs, and at the same tme control for both unobserved dfferences n natural amentes and productvtes and for the potental effects of endogenety and/or sample selecton bas n hours worked. The paper fnds that sgnfcant rural-urban female wage dfferentals exst for many groups even when controllng for cost-of-lvng dfferentals, ndcatng that the functonng of rural and urban labor markets are dfferent n the Unted States. 1

21 1.2 INTRODUCTION Emprcal research ndcates that urban wages exceed those n rural areas n all regons for both whtes and blacks (Krumm 1984). Workers n ctes earn 33 percent more than ther non-urban counterparts (Glaeser and Mare 2001). Average earnngs and ncome are lower n nonmetropoltan (nonmetro) than n metropoltan (metro) areas at every level of educaton n the Unted States (Jensen and McLaughln 1995; McGranahan and Ghelf 1991; McLaughln 1986). Medan household ncome n nonmetro areas was lower than n metro areas between 1979 and 1999 (McLaughln 2002). Indeed, the poverty rate among college-educated persons n nonmetro areas s well above the rate n metro areas (Lchter and Jensen 2000). Rural areas have also been characterzed by both unemployment rates and underemployment rates contnuously above those n major urban economes (Lchter and Costanzo 1987), especally for women (Fndes and Jensen 1998; Jensen et al. 1999; Wang 1999; Wang and Fndes 2001). Rural mothers face unque barrers or dsncentves to employment (Lchter and Jensen 2000). Sngle mothers are found to receve lower hourly earnngs n nonmetropoltan areas than n metropoltan areas (Mlls and Hazarka 2002). Such rural-urban dfferences n wages and employment may smply be due to dfferences n the composton of rural and urban populatons 1. For example, lower job experence levels are assocated wth lower earnngs, and average experence rates and levels of educaton tend to be lower n rural areas (Jensen and McLaughln 1995; Lchter et al. 1993; Swam and Texera 1991). Some research consders that the nomnal wage dfferentals between areas may be representatve of compensaton for cost of lvng and amenty dfferences, at least n part, rather than real wage dfferentals (Krumm 1987). 2

22 There s some evdence showng that the cost of lvng s lower on average n rural than n urban areas (Ctro and Mchael 1995; Nord 2000) and urban workers face a hgher cost of lvng and receve much hgher nomnal earnngs (Km 2002). However, prevous research has suggested that lower female earnng rates are partly due to the way that rural labor markets dsadvantage women (Davs, Connolly and Weber 1999; Shaffer 1998; Vera-Toscano, Phmster and Alfons 2002). It s argued that even when workers characterstcs and jobs are dentcal, the structure and operaton of nonmetro labor markets may provde lower returns to workers than metro labor markets (McLaughln and Perman 1991). The vew perssts that there s a lack of good rural jobs those that pay a decent or famly wage n the new economy (Flynt 1996; Gale and McGranahan 2001; Horan and Tolbert 1984). In general, women n rural areas are less lkely than men to fnd adequate employment and more lkely to become employed n low-wage or low-hour jobs (Fndes and Jensen 1998; Wang and Fndes 2001). Rural women are largely confned to lower status occupatons n whch experence counts more than formal qualfcatons. Labor force hardshp contrbutes to the economc vulnerablty of women, especally among female heads of households wth chldren (Lchter et al. 1993; Pearce 1978). Moreover, dfferent dynamc patterns of labor force partcpaton and hours worked between metro and nonmetro women may also play a role n the observed regonal wage dfferentals (Krumm 1984; Vera-Toscano et al. 2002). As shown by Krumm and Tolley (1983), even n the absence of real wage dfferentals, the hours worked decson mght be sgnfcantly affected by the locaton decson. Ths stems from the potental substtutablty or complementarty between market goods, amentes, and tme spent n household 3

23 producton actvty that affects the hours worked decson when market good prces and amenty levels dffer (Krumm 1984). The resultng mpact on earnng dfferentals may, thus, mask the underlyng dfferences between nomnal wages. Further, fxed employment costs may nfluence observed persstence n labor markets for all partcpants (Heckman and Wlls 1977; Nakamura and Nakamura 1985; Shaw 1994). Compared to urban labor markets, rural labor markets suggest greater occupatonal msmatch and job search costs n rural areas (Lchter and Costanzo 1987). Fnally, there s evdence showng that those factors accountng for rural-urban wage dfferentals cross-sectonally may not be the same as those factors that cause changes over tme (Farber and Newman 1987). In a dynamc growng economy characterzed by stochastc shocks that may operate dfferentally across regons, we would expect to observe shortrun nterregonal wage dfferentals because of ncomplete adjustments to recent changes n the structural demand for labor across regons (Farber and Newman 1987). Ths paper examnes the rural-urban wage gap among women n the Unted States to consder whether observed dfferences are attrbutable to dfferences n observed ndvdual demographc characterstcs, dfferences n hours worked, dfferences n the operaton of rural labor markets or dfferences n preferences. The years of the Panel Study of Income Dynamcs (PSID) are used. The paper contrbutes to the lterature n several sgnfcant ways. Frst, endogenety and/or sample selecton bas of the hours worked decson are controlled across rural-urban locatons n determnng the observed rural-urban female wage gap. Second, unobserved dfferences n amenty and productvty and/or preferences are controlled for n the estmaton of the rural-urban wage gap. Thrd, ths paper updates the lterature by examnng possble explanatons 4

24 determnng the female wage gap across rural-urban areas and changes over tme n the dynamc growng U.S. economy. To examne costs of moblty effects, followng a suggeston by Dcke and Gerkng (1998), ths paper examnes whether the rural-urban wage gap vares across dfferent groups as related to labor mmoblty. 1.3 LITERATURE REVIEW Interregonal wage dfferentals have been explaned hstorcally by a varety of dsparate as well as nterrelated economc forces (Mullen and Wllams 2001). There are two conventonal explanatons of regonal wage dfferentals (Blackaby and Mannng 1990). The neoclasscal framework, whch consders a compettve market n whch workers maxmze utlty rather than only real wages and frms maxmze proft whch need not be compatble wth the mnmzaton of wage costs, argues that regonal wage dfferentals are accounted for by three sources (Maer and Wess 1986). These nclude: (1) temporary dsequlbrum nduced by shfts n regonal aggregate demand (Addson 1975), (2) dfferences n the mx of human captal-related demographc characterstcs (Bellante 1979; Coelho and Ghal 1971, 1973), and (3) compensaton for regonal heterogeneous amentes and dsamentes that affect workers utlty and frms proft (Rosen 1987; Thaler and Rosen 1975). These wage dfferentals may therefore be dvded nto dsequlbrum components that follow regonal demand shocks and equlbrum dfferentals that follow from dfferences n heterogeneous labor composton and local amentes. The exstence of persstent dfferentals should only follow from heterogeneous amentes or contnually shftng aggregate demand n ths neoclasscal framework (Blackaby and Mannng 1990). 5

25 An alternatve to the above paradgm of convergence to equlbrum can be found n a combnaton of the theory of polarzed development and labor market segmentaton theory whch allows for dvergent development (Maer and Wess 1986). Segmentaton theory argues that uncertanty, frcton, and nformaton costs lead to cumulatve processes (Maer and Wess 1986). Ths process nduces persstent nterregonal dfferences n development and an organzatonal dependency among plants (Massey 1979) and dstnct labor market segments (Vetorsz and Harrson 1973). As a result, the cumulatve process leads to nterregonal dfferences n economc structure and growth and to structural dfferences n regonal earnngs functons (Maer and Wess 1986). Emprcal evdence, begnnng wth aggregate regonal wage studes, has proven nconclusve. Interregonal dfferences n average wages and earnngs have been observed partcularly n the North and South of the Unted States ever snce the md- 1800s (Gerkng and Werck 1983). Ths observaton has motvated several emprcal attempts to determne the source of those dfferentals, measured both n nomnal and real terms, and to explan why they have been mantaned over tme. Those studes produce dfferent results: some argue that the dfferences n regonal characterstcs are more mportant determnants of regonal wage dfferentals (Bellante 1979; Coelho and Ghal 1971, 1973; Dcke and Gerkng 1987; Gerkng and Werck 1983); some attrbute prmary mportance to returns to characterstcs (Hanushek 1981; Krumm 1984; Sahlng and Smth 1983; Tremblay 1986); stll others argue that whle the level of characterstcs may be mportant n determnng the regonal wage dfferentals at any partcular tme, t s the change n the rates of return that are mportant n explanng the change n the dfferentals over tme (Farber and Newman 1987; Krumm 1987). Blackaby and 6

26 Mannng (1990) examned regonal wage dfferentals n Brtan wthn the hedonc wage equlbrum framework. They found a negatve assocaton between earnng dfferentals and regonal unemployment rates, whch represent a reducton n regonal labor demand. They argued that the persstence of regonal wage and unemployment dfferentals questons the sgnfcance of compettve labor market theory, whch predcts ther dsappearance through actual or potental mgraton of labor and captal. Wthn the nterregonal wage dfferentals studes, there have been relatvely few studes that consder the rural-urban wage gap explctly, despte evdence that rural labor markets may operate n a dstnct way compared to urban labor markets (McLaughln and Perman 1991). Krumm (1984) examned regonal wage dfferentals, hours worked and yearly earnngs profles for whtes and blacks takng nto account regonal and rural and urban locaton status usng data from the Panel Study of Income Dynamcs (PSID). Krumm s results ndcate sgnfcant nterregonal wage dfferences both for urban and rural resdents. He also found that there are systematc spatal dfferences n the hours worked decson, whch would confound the nferences based on only one of the wage or earnngs relatonshps. He emphaszed the need for more refned emprcal analyss to brng together the structure of regonal wage dfferentals and the labor supply responses to such dfferentals. McLaughln and Perman (1991) examned the earnngs dfferences for metro and non-metro men and women n both 1977 and They found that the returns (rewards to human captal and job characterstcs) outwegh endowments (levels of human captal) n explanng the earnngs gap for both men and women, although the mportance of returns decreases over the ten-year perod. Renkow (1996) sought to quantfy the 7

27 magntude of rural-urban dfferences n the response of earnngs to key economc varables usng county-level data from North Carolna. He found that returns to schoolng are sgnfcantly lower n rural areas than n urban areas and that local economc shocks had a much more profound mpact on earnngs n rural areas than they dd n urban areas. Krumm (1987), usng panel data from the 1973 through 1978 PSID, estmated the ntertemporal structure of regonal and racal wage dfferentals. Krumm (1987) found that a trend toward an ncrease n the urban and rural wage dfferentals over ths perod, ncreasng from less than 10 percent n 1973 and 1974 to a maxmum of almost 16 percent n Besdes ths, hs results mply that nferences based on wage dfferental estmates at any pont n tme are lkely to be ncomplete. There was a systematc changng structure of absolute and relatve regonal and racal wage dfferentals over tme. Glaeser and Mare (2001) examned the urban wage premum over tme usng multple data sets. Ther emprcal study provded evdence that the urban wage premum does not seem to be the result of omtted ablty varables, whch are correlated wth urban status. They found that the urban wage premum s related to the wage growth effect, not a wage level effect. Ths evdence suggests that ctes speed the accumulaton of human captal relatve to ther rural counterparts. Dcke and Gerkng (1998) evaluated possble reasons why nterregonal wage dfferences mght persst over long perods of tme usng Canadan Labor Market Actvty Survey panel data. They developed a general equlbrum model of nteractng regons that can consder explanatons ncludng nterregonal dfferences n producton costs, changes n relocaton (mgraton) cost and dfferences n nterregonal transfer payments. Ther emprcal 8

28 fndngs suggested that dfferental labor moblty among workers may be an mportant determnant of observed wage dfferences between regons and that changes n the pattern of nterregonal wage dfferences over tme may be at least partly a demographc phenomenon. Fnally, Vera-Toscano et al. (2002) used the Survey of Labor and Income Dynamcs to examne the rural-urban female wage gap explctly n Canada. They adopted a general equlbrum model and utlzed panel data estmators that can control for the endogenety and sample selecton effect of labor partcpaton and unobserved heterogenety n earnngs equatons. Ther results mpled that the rural-urban female wage gap dfferences persst even after accountng for observed ndvdual characterstcs and unobserved ones, suggestng dfferences n the functonng of rural and urban labor markets n Canada. They also found that the labor force partcpaton sample selecton effect on the observed wage gap s small and dfferences n preferences and wages across space do not appear to affect partcpaton decsons suffcently to nfluence the observed wage gap. Gven ths perspectve, the paper outlnes the theoretcal general equlbrum model as a startng pont, and then mplements the econometrc specfcaton usng panel data that can control for unobserved heterogenety responsble for the endogenety and/or sample selecton bas of the partcpaton decson and the hours worked decson across rural-urban areas. The ssue of estmatng the mpact of hours worked on the determnaton of the hourly wage rate whle accountng for the endogenety of hours worked has attracted attenton n the labor economcs lterature (Bddle and Zarkn 1989; Mofftt 1984). It s mportant to recognze that wages are lkely to dffer between full- 9

29 tme (hgh hours) and part-tme (low hours) employment due to dfferent attractveness and transacton costs assocated wth each type of work (Averett and Hotchkss 1997; Hotchkss 1991). Blank (1988) reported that female heads of households may face sgnfcant fxed costs of work, such as transportaton and chld care. There s also substantal evdence n the labor demand lterature that workers sort themselves nto dfferent markets based on ther preferences and sklls such that there would exst wage dfferentals between dfferent groups even under dentcal wage determnaton mechansms. Lundberg (1985) presented evdence for observng the offered wage to rse n general wth the number of hours worked. On ths same pont, Owen (1978:12) offered the observaton that the fewer number of hours worked, the less lkely the employer s to nvest n tranng that ndvdual, resultng n the lowest grade jobs (thus lowest wages) beng held by ndvduals who work the fewest hours. 1.4 THEORECTICAL MODEL To develop the theoretcal framework to examne the rural-urban female wage gap, the general equlbrum model of nteractng regons orgnated by Roback s semnal papers (1982, 1988) and extended by Dcke and Gerkng (1998) s adopted as a startng pont. Ths model s flexble enough to consder explanatons ncludng nterregonal dfferences n producton costs, changes n relocaton (mgraton) costs, and dfferences n nterregonal transfer payments n addton to the regonal amentes n the determnaton of regonal wage dfferences. Consder a two-regon model wth Urban (U) and Rural (R) areas makng up a closed system. Each regon s endowed wth dfferng amounts of land and a dstnct 10

30 amenty bundle and each produces a dfferent composte good ( X U and X R ) sold n compettve markets. Regons, therefore, are dentfed wth a unque mx of mmoble productve resources and produce dfferent products. Workers are mperfectly moble between urban and rural areas. In consequence, resdents of each regon may enjoy dfferent levels of utlty. On the supply sde, the producton of X U and characterzed by constant returns-to-scale and use two varable nputs, labor (H) and land (L). The constant returns-to-scale assumpton means that the model does not allow for agglomeraton economes. Markets for labor and land are compettve and frms mnmze costs at all output levels. Unt cost equatons are X R s c c U R ( w ( w U R, r, r U R ; n ; n U R, t, t U R ) = P, ) = 1, (1.1) where w denotes the wage of labor, r denotes the rent per unt of land, P denotes the prce receved by X U producers relatve to that receved by X R producers, t denotes regon-specfc factors affectng unt costs such as techncal progress, and n denotes an nterregonal dfferent bundle of amentes. In a model set up lke ths, dfferences n the composton of amenty bundles lead regons to produce dfferent outputs and motvate ncreased attenton to trade. Constant returns-to-scale n producton and competton n all markets guarantee that factors are fully employed n each regon. Thus, land s fully allocated between goods producton and resdences of workers, and workers are fully employed n goods producton: 11

31 c X +φ H c r w X = H, = L, ( = U, R) ( = U, R) where c = c / j ( j r, w) shows use of a factor per unt of goods producton, j = φ denotes per worker resdental use of land (L), and H denotes the sze of the labor force. Labor force s nelastcally suppled jontly to both regons, but can move between regons: H U + H = H dh R = 0 The demand sde of the model requres several assumptons. Frst, resdents of a regon may receve lump-sum transfer payments ( τ ) that are exogenously determned by government acton. Transfers are nternally fnanced wthn the two-regon system such that the Urban regon gans equal the Rural regon losses and vce-versa ( τ U = τ ). R Second, the model allows for relocaton costs when workers move between regons; n consequence, resdents of each regon may not enjoy the same level of utlty. Assume that the maxmum utlty dfferental s : thus, u U R u, where s an ncreasng functon of relocaton costs. Thrd, assumng the ntal utlty n the Urban regon s greater than that n the Rural regon: u U = u R +, and ths utlty dfference s at least partally reflected n wages (that s, ntally w U > wr ). Fourth, both U X and X R can be consumed n each regon; however, prces pad for X by resdents of ( j) reflect j transportaton costs. That s, when X s shpped to regon, the fracton 1 v ) s lost j ( n transt ( 0 v < 1). Assumng that goods can be shpped costless ntraregonally, the < relatve commodty prce facng consumers n Urban areas s P U = PvR ; 12

32 13 correspondngly, n Rural areas, U R v P P / =. Ffth, each worker n each regon maxmzes an dentcal homothetc utlty functon: ), ( ), ;,, ( R U n x x f u R U = = φ (1.2) n whch j x denotes per-worker consumpton of good j n regon, subject to the budget constrants ), (, R U w r x P x P R R U U = + = + + τ φ where ), / ( ), (1/, U R U R U R U U v P P v P P P = = = and 1 = R R P are unt prces of each good n each regon. Solvng for,, R U x x and φ and substtutng nto (1.2) yelds the ndrect utlty functons: ), ( ), ;,, ( R U n P r w V u = = τ Sxth, commodty market equlbrum and balance of trade between regons are guaranteed by R R R R U R U R U R U R U U U U x H v x H X v x H x H X + = + = ) / ( ) / ( (1.3) The rght hand sdes n (1.3) are nterpreted as aggregate demand facng producers n the two regons. Under these assumptons, the model can be solved for wages n ether regon (or the dfference between the two) n terms of all exogenous parameters as shown n (1.4): R U H L L v v t t n n f w R U R U R U R U R U, ),,,,,,,,,,, ( = = τ τ (1.4) Ths model suggests the man drectons for the emprcal work estmated later n ths paper.

33 1.5 ECONOMETRIC MODEL AND ESTIMATION STRATEGY From equaton (1.4), t follows that the emprcal model of rural-urban wage dfferentals needs to control for the effects of amentes, producton cost, transfer payment, relocaton cost, transportaton cost and the endowments of the factor supples. One key ssue here s that the model treats labor as a homogeneous nput. However, n fact, workers dffer n ther observed sklls and human captal, demographc characterstcs and unobserved ones. Moreover, many of the productvty dfferences and local amentes are ndvdual specfc and are generally unobservable. However, t s possble to control for ndvdual observed and unobserved heterogenety across workers usng panel data n the estmaton of regonal wage dfferentals (Vera-Toscano et al. 2002). An econometrc model that conssts of a two-step estmaton of the wage equaton whch accounts for the unobserved heterogenety responsble for the endogenety/sample selecton bas of hours worked wll be used. The econometrc model was examned by Rdder (1990) and Njman and Verbeek (1992) and generalzed by Vella and Verbeek (1999). The model has the followng form: ' ln w = x, β 1 + m( hours ; β 2 ) + µ + η, (1.5) t 1 t t t hours * t ' = x θ + hours θ + α + ν, (1.6) 2, t 1, t 1 2 t hours t t = hours hours = 0, w * t t f hours * t not observed > 0, f hours * t 0, (1.7) where ndexes ndvduals ( =1,, N) and t ndexes tme (t=1,,t). The parameters of equaton (1.5) are of prmary focus whle equaton (1.6) s the reduced form for the explanatory varable of hours whch s endogenous and the bass of the selecton rule. 14

34 Equaton (1.7) reflects the censorng and selecton rules. The hourly wage rate of ndvdual at tme perod t; exogenous varables specfed n the wage equaton; n the hours equaton; counterpart * hourst w t represents the offered x 1, t represents vectors of observed x 2, t represents exogenous varables represents latent endogenous varables wth ts observed hours t, the number of hours worked; m denotes a polynomal of known ' ' ' ' ' ' length wth unknown coeffcents β 2, whle β = ( β, β ) and θ = ( θ, θ ) are parameters to be estmated. The equatons errors comprse random ndvdual effects, µ and α, and random ndvdual-specfc tme effects, ηt and v t, whch are assumed to be ndependent across ndvduals. Denote ε µ + η ; vector of u t s for ndvdual. Wrtng 1 t = t u t + vt 2 1 = α ; and u as the T x = x,..., x ]' and X = x, x ], we assume [ 1 T 2 [ 1 2 u X 2 2 ~ N. I. D.(0, σ α ll' + σ I) (1.8) v ( ε, ) = + τ u (1.9) E t X u τ1ut 2 where l s a T vector of ones; u T 1 = T u t t= 1 ; and τ1 and τ 2 are unknown constants. Equaton (1.8) mposes normalty and a strct error components structure and excludes any form of autocorrelaton n v t. Equaton (1.9) allows for heteroskedastcty and autocorrelaton n ηt but mposes the strct exogenety of x t. Ths model ncorporates a potental role for state dependence n the reduced form (equaton (1.6)). Ths ensures that the error components do not ncorrectly capture the dynamcs whch should be attrbuted to lagged dependent varables. 15

35 To estmate ths model, the two-step estmaton wth censored endogenous varables and selecton bas n the sprt of Heckman s (1979) sample selecton procedure s used (Vella and Verbeek 1999). In the frst step, t s possble to consstently estmate the parameters n equaton (1.6) by estmatng a random effect Tobt model usng maxmum lkelhood under the usual regularty condtons. In ths procedure, adjustment for the ntal condton problem s accomplshed usng a procedure suggested by Heckman (1981) and Vella and Verbeek (1999). Gven the estmates of equaton (1.6) t s possble to construct estmates of the two correcton terms, the generalzed resdual and ts average over tme to be ncluded n the prmary equaton (1.5), wth coeffcents τ1 and τ 2. These parameters can be estmated jontly wth β n the second step from condtonal moment restrctons such as ordnary least squares. Under the null hypothess of no endogenety, τ = τ 0, standard errors can be computed n the usual way. 1 2 = Consequently, the standard Wald test of the sgnfcance of the addtonal terms s an endogenety test. In general, standard errors should be adjusted for heteroskedastcty, autocorrelaton and for the estmaton of the correcton terms. The second-step wage equaton s estmated over the sub-sample of women reportng postve hours. The strength of ths model s that the ncluson of the correcton terms smultaneously accounts for the endogenety of the partcpaton decson and hours worked. As the decson to not partcpate corresponds to a zero value for hours t, the ncluson of the correcton terms also account for the selecton bas from estmatng over the sub-sample of workers. Moreover, the ncluson of a lagged dependent varable n the reduced form solates the role of dynamcs and state dependence (Vella and Verbeek 16

36 1999). The ncluson of a lagged dependent varable s partcularly mportant as t explots the panel nature of the data and enables solaton of the ndvdual effects from state dependence (Vella and Verbeek 1999). Although two-step estmates are generally neffcent (Newey 1987), ths approach has a sgnfcant computatonal advantage compared to maxmum lkelhood estmaton. In some cases, the two-step estmator s asymptotcally effcent wthn a lmted nformaton framework (LIML). Vella and Verbeek s (1999) method provdes ntal consstent estmators for a LIML approach so that asymptotcally effcent estmators can be obtaned n one teraton. Testng for Rural-Urban Dfferences The extent of the rural-urban U.S. female wage gap s consdered n several ways. Frst, by provdng a dstrbuton of real and nomnal wage rates for all women by rural-urban locaton over tme, t s possble to examne how the overall female rural-urban wage gap vares. Second, by addng rural dummy varables to varous wage specfcatons, t s also possble to examne how the overall female rural-urban wage gap vares after controllng other ndvdual, demographc and work-related varables. Thrd, to test the role of moblty costs on rural-urban female wage dfferences, ths paper follows Dcke and Gerkng s (1998) by lookng at the value of rural dummy varables across groups of female workers related to mmoblty. For example, the wage dfferentals are examned by age groups, educaton, famly characterstcs, and dsablty status, as examples. Fourth, whether rural-urban wage dfferences persst over long perods of tme s explored because the returns to labor market-valued characterstcs vary. For example, 17

37 the paper explores whether returns to educaton or work experence are sgnfcantly lower n rural labor markets over a long tme. 1.6 DATA Panel Study of Income Dynamcs (PSID) for the years of Publc Release II data are used for ths study 2. The Panel Study of Income Dynamcs (PSID), begun n 1968, s a longtudnal study of a representatve sample of U.S. ndvduals (men, women, and chldren) and the famly unts n whch they resde. It emphaszes the dynamc aspects of economc and demographc behavor, but ts content s broad, ncludng socologcal and psychologcal measures. As a consequence of low attrton rates and the success n followng young adults as they form ther own famles and re-contact efforts (of those declnng an ntervew n pror years), the sample sze has grown from 4,800 famles n 1968 to more than 7,000 famles n 2001 (PSID 2003). The PSID sample, orgnatng n 1968, conssted of two ndependent samples: a cross-sectonal natonal sample and a natonal sample of low-ncome famles. The crosssectonal sample was drawn by the Survey Research Center (SRC). Commonly called the SRC sample, ths was an equal probablty sample of households from the 48 contguous states and was desgnated to yeld about 3,000 completed ntervews. The second sample came from the Survey of Economc Opportunty (SEO), conducted by the Bureau of the Census for the Offce of Economc Opportunty. In the md-1960's, the PSID selected about 2,000 low-ncome famles wth heads under the age of sxty from SEO respondents. The sample, known as the SEO sample, was confned to Standard Metropoltan Statstcal Areas (SMSA's) n the North and non-smsa's n the Southern 18

38 regon. The PSID core sample combnes the SRC and SEO samples. From 1968 to 1993, the PSID ntervewed and re-ntervewed ndvduals from famles n the core sample every year, whether or not they were lvng n the same dwellng or wth the same people. Adults have been followed as they grow older, and chldren have been observed as they advance through chldhood and nto adulthood, formng famly unts of ther own. In keepng the sample representatve because of the changng nature of mmgraton n the Unted States snce the begnnng of the study n 1968, the PSID sample added 2,000 Latno households, ncludng famles orgnally from Mexco, Puerto Rco, and Cuba n Whle ths sample represents three major groups of mmgrants, t mssed out on the full range of post-1968 mmgrants, Asans n partcular. Because of ths shortcomng, and a lack of suffcent fundng, the Latno sample was dropped after As a result, ths study wll only use the core sample SRC and SEO samples as the data source 3. The PSID contans questons relatng to county of resdence n each year and detaled measures of educaton, work experence, as well as nformaton on workplace, job characterstcs and demographc nformaton 4. Thus, a more complete specfcaton of the wage equaton s permtted and the possblty of aggregaton error s reduced. The defnton of rural and urban samples s based upon the concept of the ruralurban contnuum codes devsed by Calvn Beale and Peggy Ross of the U.S. Department of Agrculture n the PSID data. The rural-urban contnuum codes have been added to each wave of the PSID begnnng n 1985 to 1993 n Publc Release II data 5. The codes form a classfcaton scheme that dstngushes metropoltan countes by sze, and nonmetropoltan countes by degree of urbanzaton and proxmty to metro areas (U.S. 19

39 Department of Agrculture 2003). The codes comprse a 10-part county codfcaton whch ncludes 4 metro and 6 nonmetro categores (see 1.11 Appendx A). Metropoltan status was announced by the Offce of Management and Budget (OMB). Adjacency was determned by both physcal boundary adjacency and a fndng that at least 2 percent of the employed labor force n the non-metropoltan county commuted to metropoltan central countes. The rural and urban defnton n ths paper adopted the defnton of rural-urban contnuum code for the metro and nonmetro countes. Consderng that the rural countes n the U.S. are heterogeneous n populaton sze, urbanzaton and degree of accessblty to larger economes centers of nformaton, communcaton, trade, and fnance, etc, the paper further dfferentates the nonmetro (rural) countes adjacent to metro areas and nonmetro (rural) countes not adjacent to metro areas. The mportance of dfferentatng the nonmetro countes by adjacency to metro areas has been dscussed n Fsher and Weber (2002). The sample of women ncludes female heads or spouses (ncludng cohabtants) of male heads n the famly who were aged between 20 and 57 n Ths makes the oldest sample observaton 64 years old n The sample excludes women who ndcated retrement n the survey. The age restrctons and sample exclusons permt us to abstract from endogenous retrement decsons. The sample further excludes those who are full-tme students, and those who do not receve a unquely-dentfed wage,.e., those recevng proft-type ncome from ther jobs and those recevng multple wages. Self-employed workers were excluded snce ther estmated hourly earnngs and annual hours of work may be very naccurate (Farber and Newman 1987), and earnngs may 20

40 nclude returns to physcal captal (Dcke and Gerkng 1998). Multple job-holders were excluded snce they may face non-convex budget constrants (Farber and Newman 1987). Fve types of nformaton, obtaned from the sampled women, are used n ths study (see Table 1.1 for varable defntons). Frst, the endogenous varables nclude average hourly earnngs, hours and lagged hours. The varable average hourly earnngs s defned as the real wage rate snce the nomnal wage pad, n equlbrum, should reflect lvng cost dfferences between geographc areas (Coelho and Ghal 1971). Because a comprehensve cost-of-lvng ndex does not exst (U.S. General Accountng Offce 1996), the Consumer Prce Index (CPI), produced by the Bureau of Labor Statstcs (BLS), serves as an approxmaton of cost-of-lvng ndex to deflate the nomnal wage rate year by year (Fxler 1993). The wage rate s measured by the average hourly earnngs for the year pror to the survey year. The endogenous varable hours s defned as the total annual work hours for the sampled women n PSID. The values for ths varable represent the actual total annual work hours on all jobs ncludng overtme for the year pror to the survey. The lagged hours are created based on the panel nformaton n PSID. The 1986 observatons use presample (1985) PSID nformaton for lagged annual work hours. Second, the personal and famly characterstcs nclude women s age, agesquared, martal status, race/ethncty, chld status, dsablty status, women s asset ncome, and housng value. The martal status varable s created based on the headshp nformaton, couple status and martal status of the head n PSID. The concept here ndcates legal spouse or permanent cohabtant present n the famly unt. The race/ethncty varable s created based on the race and ethnc orgn nformaton n PSID. 21