The Growth in Home-Based Wage and Salary Employment in the United States, : How Much and Why?

Transcription

1 The Growh in Home-Based Wage and Salary Employmen in he Unied Saes, : How Much and Why? Gerald S. Oeinger! Deparmen of Economics Universiy of Texas a Ausin oeinge@eco.exas.edu April 2006! The auhor hanks he Alfred P. Sloan Foundaion for financial suppor, Anne Golla for valuable research assisance, and Karen Conway, Paricia Reagan, and seminar paricipans a he Sociey of Labor Economiss, he Souhern Economic Associaion, and UC Sana Barbara for commens on an earlier draf.

2 Absrac This sudy quanifies he growh in home-based employmen among prime-age, civilian workers in he Unied Saes beween 1980 and 2000 and analyzes how he wage penaly on home-based wage and salary jobs evolved over hese weny years. Home-based employmen grew much more rapidly han on-sie employmen during his period, especially among wage and salary workers. A he same ime, he mean wage penaly suffered by wage and salary workers on home-based jobs fell dramaically over his period. These paerns hold no only in he aggregae bu also wihin almos all occupaion caegories, and lile of he aggregae change in he home-based employmen share or he home-based wage penaly is explained by change in he occupaional composiion of employmen. These findings sugges ha widespread reducions in employers coss of providing home-based work arrangemens were he main reason for he growh in home-based employmen in he las several decades. The paper also shows ha he relaionship beween he home-based wage penaly and he probabiliy ha a worker uses a compuer a work has changed markedly over ime, in ways ha sugges ha IT innovaions have had significan impacs on worker labor marke produciviy boh in he workplace and a home. 1

3 I. Inroducion Home-based employmen has grown a a rapid rae in he Unied Saes in recen decades. According o U.S. census daa, he number of home-based workers nearly doubled beween 1980 and 2000, growing from less han 2.2 million o nearly 4.2 million, while oal employmen grew a a much slower pace, from 96.6 million o million. 1 Several broader forces operaing over his ime period poenially can explain his dramaic growh in home-based employmen. For example, he employmen share of women grew subsanially over his period, and women may value home-based work arrangemens more highly han men given he radiional division of home producion asks wihin he family. Also, major advances in informaion and communicaions echnology ha may have reduced he coss of providing home-based work arrangemens occurred during hese years. Finally, shifs in he occupaional composiion of he U.S. labor force over his period may have favored growh in home-based work. The few exising empirical sudies of home-based employmen (Krau 1988; Presser and Bamberger 1993; Edwards and Field-Hendry 2001, 2002; Pabilonia 2005; Schroeder and Warren 2005) all have analyzed is deerminans or is wage consequences a a poin in ime. While valuable, hese sudies offer no insigh ino why home-based employmen has expanded so rapidly in recen years. The presen paper seeks o fill his gap by using daa from he Public Use Micro Samples (PUMS) of he 1980, 1990, and 2000 U.S. Censuses o analyze in deail boh he recen growh in home-based employmen among wage and salary workers and he concomian changes in he wages of home-based employees relaive o observaionally equivalen on-sie workers. 1 The 1980 numbers can be found a hp:// on he Census web sie. The 2000 numbers can be found hrough he American FacFinder search ool on he Census web sie. 2

4 I find ha he rapid growh in home-based wage and salary employmen has been accompanied by a dramaic decline in he wage penaly associaed wih home-based work, from roughly.30 log poins in 1980 o approximaely zero in More disaggregaed analyses reveal ha home-based employmen shares and home-based wage penalies have varied subsanially across skill groups bu ha only small fracions of he aggregae growh in he home-based employmen share and he aggregae decline in he home-based wage penaly can be explained by composiional shifs favoring skill groups wih high propensiies owards and low penalies from home-based work. These resuls srongly sugges ha widespread reducions in employer coss of providing home-based work arrangemens have been he predominan force behind he growh in home-based employmen since Advances in informaion echnology (IT) are a naural suspec as a source of falling employer coss of providing home-based work arrangemens. Such advances seem likely o have had a larger impac on relaive produciviy a home in some occupaions han in ohers. Thus, if IT innovaion was he primary cause of he increase in home-based employmen and if he elasiciy of supply ino home-based employmen was similar across all occupaions, one would expec changes in he home-based employmen share and he home-based relaive wage o have been posiively correlaed across occupaions. In he census daa, I find such a posiive correlaion beween 1990 and 2000 bu no in he preceding decade. I also examine he link beween relaive home-based produciviy and (one componen of) IT more direcly by analyzing he evoluion of he relaionship beween he home-based wage penaly and on-he-job compuer use. For his analysis, I augmen 3

5 he census samples wih prediced probabiliies of compuer use a work derived from esimaes based on Curren Populaion Survey (CPS) supplemen daa from 1984, 1993, and I find ha beween 1980 and 1990 he home-based wage penaly fell mos for workers wih he lowes probabiliy of using a compuer a work bu ha in he following decade he home-based wage penaly fell mos for workers wih he highes probabiliy of using a compuer a work. I argue ha his paern is quie consisen wih IT innovaion having played a significan role in he evoluion of home-based employmen. The remainder of his paper proceeds as follows. Secion 2 discusses he census daa and quanifies in more deail he growh in home-based employmen in he U.S. in recen decades. Secion 3 argues, using he heory of compensaing differenials, ha alernaive poenial explanaions for he rapid growh in he home-based employmen share have disinc predicions for how he wage penaly on home-based jobs should have changed in recen years. Secion 4 presens iniial esimaes ha illusrae he sharp decline in he wage penaly on home-based jobs beween 1980 and Secion 5 invesigaes he exen of heerogeneiy across occupaion groups in he home-based employmen share and he home-based wage penaly and he correlaion across occupaion groups beween changes in he home-based employmen share and changes in he home-based wage penaly. Secion 6 examines he relaionship beween he homebased wage penaly and on-he-job compuer use and he evoluion of his relaionship over ime. Concluding remarks follow in he final secion. II. Daa The empirical analyses use daa from he 5% Public Use Microdaa Samples (PUMS) of he U.S. Census of Populaion for 1980, 1990, and In each of hese years, he 4

6 census long form conained a quesion abou he mehod of ransporaion used o ge o work on he mos days in he previous week. Responses o his quesion were obained for all individuals who were aged 16 or older and were employed in he previous week. I classify employed individuals who selec he response worked a home as home-based workers and all ohers as on-sie workers. Since only individuals who mainly work a home are couned as home-based, he frequency of home-based employmen in he census daa is a very conservaive lower bound on he fracion of workers who do any work a home. 2 For each census year, I consruc analysis samples by firs selecing all households ha conain one or more home-based workers and a random 1% sample of households ha conain zero home-based workers. 3 From his se of households I keep all individuals aged who were employed in paid civilian jobs in he previous week. In addiion, as is discussed furher below, I drop he self-employed and limi aenion o wage and salary workers in mos of he analyses. Thus, he sudy primarily focuses on prime-age, civilian, wage and salary workers. For all analyses of wages, I compue he hourly wage as wage and salary income in he previous calendar year divided by he produc of weeks worked in he previous calendar year and usual hours worked per week. In he 1980 census, individuals who repored wage and salary income above $75,000 had heir repors opcoded a $75,000; for hese opcoded observaions, I muliply he opcoded income by 1.4 before compuing 2 In fac, daa from he May 2001 Curren Populaion Survey (CPS) indicae ha 19.8 million workers do some work a home a leas once a week. This number is aken from he BLS news release Work a Home in 2001 available a hp:// 3 I adjus he census-provided sample weighs o accoun for he differenial probabiliies of sample inclusion for individuals from households wih home-based workers and individuals from households wihou home-based workers and I use hese adjused weighs in all of he empirical analyses. 5

7 he hourly wage. 4 I conver he nominal hourly wage in all census years o real 1999 dollars using he CPI for all urban consumers. Because repored wage and salary income corresponds o he previous calendar year, some or all of his income may have been earned on a differen job han he one held on he census dae (April 1), from which home-based work saus is deermined. Thus, home-based work saus (as well as oher job characerisics) may be measured wih some error wih respec o he wage calculaed from prior year earnings. This poenial misclassificaion of home-based work saus could cause bias in cross-secion esimaes, bu i will affec over ime comparisons only o he exen ha he durabiliy of home-based jobs has changed subsanially over ime. However, he available evidence for all jobs does no indicae any large secular rend in job sabiliy. 5 Table 1 documens he rapid growh in home-based employmen among paid civilian workers aged The op panel of he able presens he evidence for his group as a whole. While employmen of on-sie workers grew by 44.1% beween 1980 and 2000 (from almos 69.9 million o around million), employmen of home-based workers grew by a much larger 115.3% (from 1.58 million o 3.41 million) over he same period. 6 As a resul, he home-based employmen share grew from 2.21% in 1980 o 3.27% in The census bureau replaced repored wage and salary incomes above he opcode levels in he 1990 and 2000 censuses ($140,000 and $175,000, respecively) wih he sae median (in 1990) or sae mean (in 2000) income level among individuals wih wage and salary income in excess of he opcode. I do no adjus income for individuals who repored wage and salary income above he opcode in hese years. 5 For deails see Farber (1999) and he aricles in he special Ocober 1999 issue of he Journal of Labor Economics. 6 This growh rae of home-based employmen among paid civilian workers aged is higher han ha for all workers cied in he inroducion. Evidenly, home-based work grew faser among prime-age civilian workers han i did among he young, he elderly, and miliary employees over he period. 6

8 These overall numbers mask subsanial differences beween wage and salary employees and he self-employed in boh he level and he growh of home-based employmen. These differences are shown in he lower panels of Table 1. Among wage and salary workers, home-based employmen has been very rare hisorically bu has grown a an exremely rapid rae in recen decades. In 1980, fewer han 1% (abou 480,000 ou of 64.1 million) of all prime-age, civilian wage and salary workers were home-based. Since hen, home-based wage and salary employmen grew by 67.6% beween 1980 and 1990 and by 67.2% beween 1990 and 2000, yielding a 180% growh rae over he enire period. This dwarfs he 44.4% growh rae in on-sie wage and salary employmen over he same weny years. In conras, among he self-employed, homebased employmen has been much more common (hough sill unusual) hisorically bu has grown less rapidly in recen years. In 1980, almos 15% (abou 1.1 million ou of 7.42 million) of all prime-age, civilian self-employed workers were home-based. Homebased self-employmen grew by 61.9% beween 1980 and 1990, similar o he rae for home-based wage and salary workers, bu hen grew by only 15.6% beween 1990 and 2000, yielding an 87.1% growh rae over he wo decades. Alhough his is larger han 42% growh in on-sie self-employmen over he same period, he difference is no nearly as dramaic as for wage and salary employmen. For several reasons, I resric aenion o he growh in home-based wage and salary employmen in he remainder of he paper. Firs, he disinc difference in he growh raes of home-based employmen among wage and salary workers versus he selfemployed, a leas beween 1990 and 2000, suggess ha he forces causing growh in home-based employmen may have differed beween he self-employed and wage and 7

9 salary secors. Thus, i makes sense o analyze he wo secors separaely. Furhermore, wihin he concepual framework presened below, an analysis of changes in he relaive wages of home-based workers can shed ligh on he causes of growh in home-based work. This framework assumes ha (i) each worker faces a consan parameric wage given her skills, (ii) observed average hourly earnings are a good proxy for his (consan) marginal wage, and (iii) here exiss a marke-deermined equilibrium (implici) price for he nonwage job aribue home-basedness. These assumpions seem reasonable for wage and salary workers bu no for he self-employed. By dropping he self-employed I avoid hese difficulies, albei a he cos of ignoring a quaniaively imporan componen of oal home-based employmen. III. Theoreical Consideraions The heory of compensaing differenials provides a useful framework for disinguishing among he possible causes of he dramaic growh in home-based wage and salary employmen in recen decades. 7 Home-basedness is a nonwage job aribue for which boh workers valuaions and employers coss of provision are presumably heerogeneous. Home-basedness is a paricularly valuable aribue for individuals wih high opporuniy coss of leaving or spending ime ouside he home. A he same ime, home-basedness can be provided a lower cos on jobs ha do no require eam producion, direc supervision, or proximiy o complemenary bu physically immobile capial inpus. 7 Rosen (1986) surveys he heory of compensaing differenials and a large empirical lieraure aemps o measure compensaing differenials for various job aribues including faaliy risk (Thaler and Rosen 1975), unemploymen risk (Abowd and Ashefeler 1981, Topel 1984), shif work (Kosiuk 1990), and employer-provided healh insurance benefis (Olson 2002). 8

10 Consider a compeiive labor marke for a homogeneous skill level composed of workers wih varying valuaions of working a home and employers wih varying coss of offering home-based work arrangemens. Le W h and W o denoe, respecively, he marke wage raes for home-based work and on-sie work so ha # " W! W is he wage penaly (or premium, if posiive) for home-based work. In general, he share of workers supplying labor o home-based jobs will be an increasing funcion of! ; as he wage penaly for home-based work shrinks (i.e., as! becomes less negaive), more workers will seek home-based jobs. On he oher hand, he share of jobs ha employers are willing o make home-based is a decreasing funcion of! ; as he wage savings on home-based jobs shrinks (i.e., as! becomes less negaive), employers will offer fewer home-based jobs. In equilibrium,! mus adjus so ha he fracion of workers seeking home-based jobs equals he fracion of jobs ha employers choose o make home-based. 8 The heory of compensaing differenials predics ha he disribuions of worker valuaions of home-basedness and employer coss of providing home-basedness will joinly deermine he equilibrium value of! and ha, in equilibrium, home-based jobs will be held by workers who value home-basedness mos and will be provided by employers who can offer home-basedness a lowes cos. If providing home-based work arrangemens were cosly for all jobs, a wage penaly would exis for home-based work in equilibrium, all else equal. Wihin his framework, he dramaic growh in he home-based share of wage and salary employmen in recen decades could be explained eiher by ouward shifs in he relaive supply of labor o home-based jobs or by ouward shifs in he relaive demand h o 8 Equilibrium also requires ha he wage levels, W h and W o, adjus so ha he aggregae quaniy of labor supplied equals he aggregae quaniy of labor demanded. 9

11 for labor in home-based jobs. Changes over ime in he demographic composiion of he labor force owards groups ha value home-based work arrangemens more highly, which may have occurred wih he rise in female labor force paricipaion, would have caused such ouward supply shifs. Increasing valuaions of home-based work over ime wihin demographic groups, perhaps resuling from changes in preferences, income, or family srucure, would have had he same effec. On he oher hand, reducions in firms nonwage coss of providing home-based work arrangemens, a possible consequence of recen advances in IT, would have caused ouward shifs in he relaive demand for labor in home-based jobs. Similarly, changes in he indusrial or occupaional srucure of U.S. employmen in favor of secors ha can offer home-based jobs more cheaply would have caused ouward demand shifs. Supply-side versus demand-side explanaions for he growh in he home-based employmen share have opposie implicaions for how he wage penaly for home-based jobs should have changed over ime. In paricular, if supply-side facors were dominan, hen he wage penaly for home-based work should have increased in recen decades. On he oher hand, if demand-side facors were dominan, hen he wage penaly for homebased work should have decreased in recen decades. Thus, an empirical analysis of how he wage penaly associaed wih home-based employmen has changed in recen decades can help discriminae among alernaive explanaions for he growh in home-based wage and salary employmen. I pursue his analysis below. IV. The Wage Penaly on Home-Based Jobs, Before reporing esimaes of he wage penaly on home-based jobs, Table 2 presens descripive saisics for wages and human capial characerisics of boh on-sie and 10

12 home-based workers in each census year. The samples of wage and salary workers are he same as earlier, bu wih he added resricion ha only observaions wih real hourly wages beween $1 and $150 are included. The able reveals a number of ineresing facs. The mean log real wage was much lower among home-based workers han among on-sie workers in 1980 bu, by he year 2000, he mean log real wage acually was higher among home-based workers. This reversal is he resul of an approximaely consan mean log real wage among on-sie workers beween 1980 and 2000 coupled wih a rapidly rising mean log real wage among home-based workers over his same ime period. Wage dispersion was much higher among home-based workers han among on-sie workers in all years. 9 The remainder of he able illusraes mean human capial differences beween home-based workers and on-sie workers a a poin in ime as well as changes in he mean human capial of boh groups of workers over ime. I commen here on some of he more sriking paerns. Firs, he fracion of home-based workers ha are female has decreased over ime, despie he significan increase over ime in he fracion female among on-sie workers. Second, while home-based workers were much more likely han on-sie workers o work on a par-ime or par-year basis or o be disabled in 1980, hese differences had eiher diminished subsanially (in he case of par-ime hours) or disappeared enirely (in he cases of par-year weeks worked and disabiliy saus) by Third, home-based workers were considerably more educaed han on-sie workers by 2000, even hough he disribuion of educaional aainmen was very similar 9 Wage dispersion increased over he sample period for males and females separaely bu did no rise much for boh sexes combined because he male-female mean log wage gap declined over he same period. 10 The rise in he repored level of disabiliy in 2000 reflecs changes in he wording of he census quesion abou disabiliy saus, bu i does no aler he observaion ha he frequency of disabiliy among homebased workers relaive o he frequency among on-sie workers declined beween 1980 and

13 for boh groups in Finally, differences in he occupaional disribuion of homebased workers and on-sie workers changed noiceably beween 1980 and Relaive o he occupaional disribuion for on-sie workers, ha of home-based workers shifed disproporionaely owards managerial, scienific, and sales jobs and away from farming and service occupaions. No surprisingly, here exis pronounced differences in he occupaional disribuion of on-sie and home-based workers in all years. The picure ha emerges from Table 2 is ha home-based wage and salary workers gained subsanially on on-sie workers in heir observable skills beween 1980 and In paricular, home-based workers became relaively more educaed, more concenraed in high-wage occupaions, less likely o have a disabiliy, and less likely o be only parially employed in he labor marke. These relaive gains in he observed skills of home-based workers likely explain par of heir relaive wage gains over his period, making i imporan o conrol for observed skills in esimaing he wage penaly on home-based jobs. To obain esimaes of he home-based wage penaly a a poin in ime, I iniially esimae models of he form lnw 19 = X i % +! $ j Dij + # H i + i, = 1980, 1990, 2000 (1) i " j= 1 for he samples of year old wage and salary workers in civilian jobs. In equaion (1), i indexes individuals, indexes census years, W i is he real hourly wage, vecor of observable characerisics ha affec wages, in occupaion caegory j, disurbance erm, and X i is a D ij is an indicaor for employmen H i is an indicaor for holding a home-based job,! i is a 19!, {! j } j =1, and! are parameers o be esimaed. The 12

14 parameer of primary ineres is!, which (approximaely) measures he percenage wage penaly on a home-based job in year, holding all oher observables consan. Ordinary leas squares esimaes of! will be biased esimaes of he rue compensaing differenial for a home-based job in year if he unobserved deerminans of an individual's wage (produciviy),! i, are correlaed wih wheher an individual has a home-based job, H i. There are a leas wo reasons o expec such a correlaion. On he one hand, as Brown (1980) and ohers have noed, workers wih high unobserved produciviy have high full income (given heir observed skills) and are likely o spend par of heir greaer full income on desirable job aribues. Thus, if home-basedness is generally considered desirable, workers wih high unobserved produciviy should be more likely o selec ino home-based employmen and herefore!ˆ will be an upward biased esimae of he rue compensaing differenial (i.e., he esimaed wage penaly will be oo small). On he oher hand, a worker wih a home-based job avoids he fixed coss of on-sie work (Edwards and Field-Hendry 2001, 2002) and also may be able o perform some home producion aciviies while doing marke work from home. These nonwage benefis of a home-based job are more likely o ouweigh he wage penaly from a homebased job for workers wih low marke wage opporuniies, which suggess ha workers wih low unobserved produciviy should be more likely o selec ino home-based employmen. This ype of selecion will cause!ˆ o be a downward biased esimae of 13

15 he rue compensaing differenial. Because hese wo poenial sources of bias work in opposie direcions, he sign of any bias in OLS esimaes of! is unclear a priori. 11 Since I seek o shed ligh on he relaive imporance of supply-side versus demand-side facors in explaining he growh in he home-based employmen share in recen decades, he change in he wage penaly on home-based jobs over ime is of greaer ineres han is level a a poin in ime. Even if he esimaed wage penaly on home-based jobs a a poin in ime is a biased esimae of he rue compensaing differenial for he reasons described above, he change in he esimaed wage penaly over ime will measure he rue change in he compensaing differenial accuraely if he bias in he esimaed wage penaly is consan over ime. Of course, i is far from clear ha he bias has been consan over ime. In paricular, Table 2 suggess ha observable skills rose much more among home-based workers han among on-sie workers in recen decades. If home-based workers made similar relaive gains in unobserved skills, he change over ime in he esimaed home-based wage penaly will oversae he relaive wage gains enjoyed by a home-based worker of fixed skill, because par of he reducion in his penaly will reflec unmeasured skill improvemens among home-based workers over ime. Laer, I address his concern o some exen by esimaing models ha allow he home-based wage penaly o vary across workers wih differen observed skills. Bu, in all of he esimaes, par of he change over ime in he esimaed wage penaly may reflec change in he unobserved skill of home-based workers (relaive o on-sie workers) 11 Ideally, one would like o find a valid insrumen for holding a home-based job and hen esimae (1) by insrumenal variables mehods. However, i is difficul o hink of a variable available in he census daa ha has srong predicive power for wheher a worker has a home-based job bu ha does no direcly affec he worker's wage opporuniies. 14

16 raher han a pure change in he equilibrium implici price of home-basedness for a worker of fixed skill. Table 3 presens OLS esimaes of (1) for 1980, 1990, and The explanaory variables include a sandard se of demographic and human capial conrols in addiion o he home-based job indicaor. As a whole, hese variables explain approximaely 30% of he sample variaion in log real wages in each census year. The esimaed coefficiens for all of he conrol variables have he expeced signs and do no require exended discussion. In addiion, he coefficiens on mos of he conrol variables are quie sable over ime. The excepions are he increase in he esimaed reurn o educaion, he decline in he esimaed male-female wage gap, and he reducion in he esimaed wage coss of children for women beween 1980 and The esimaed wage penaly on home-based jobs fell precipiously beween 1980 and 2000, mirroring he changes over ime in he mean log real wage difference beween home-based and on-sie workers shown in Table 2. In 1980, he average log real wage difference beween a home-based worker and an observaionally equivalen on-sie worker was!. 31. By 1990, his wage penaly on home-based jobs had shrunk o!. 17 and, by 2000, i had nearly disappeared. Thus, even afer conrolling for he relaive gain in he observable skills of home-based workers, i is clear ha a rise in he relaive wage of home-based workers (or, equivalenly, a decline in he wage penaly on home-based jobs) accompanied he growh in home-based employmen share. This finding suggess ha eiher reducions in employer coss of offering home-based jobs or composiional 12 The esimaed wage penaly associaed wih disabiliy also declined beween 1990 and 2000, bu his is likely an arifac of he less sringen definiion of disabiliy used in he 2000 census. 15

17 shifs in employmen favoring jobs ha can be locaed a home a low cos were he dominan force behind he rapid growh in home-based employmen. V. Heerogeneiy in he Home-Based Wage Penaly The models esimaed above impose he resricion ha he wage penaly on home-based jobs is idenical for all ypes of workers and jobs. However, i is plausible on heoreical grounds ha he marke compensaing wage differenial for home-based work varies across occupaions, boh because employers coss of offering home-based work arrangemens depend on he asks ha workers perform and because workers valuaions of working a home may vary wih full income. In addiion, in cerain occupaions (e.g., farming, clergy) some home-based workers may receive a subsanial porion of heir compensaion in he form of employer-provided or employer-subsidized housing. In such occupaions, one would expec o see a larger wage penaly for home-based employmen. If differences in he home-based wage penaly do exis across occupaions, par of he drop in he average home-based wage penaly shown in Table 3 may resul from relaive growh (decline) in home-based employmen in occupaions wih lower (higher) home-based wage penalies raher han from acual reducions in home-based wage penalies wihin occupaions. Esimaes ha allow he home-based wage penaly o vary across occupaions can be used o examine wheher such composiional shifs were an imporan par of he explanaion for he decline in he aggregae home-based wage penaly. Similarly, if here exis differences in he home-based employmen share across occupaions, hen par of he increase over ime in he average home-based employmen share may be explained by relaive growh (decline) in employmen in occupaions wih 16

18 high (low) propensiies oward home-based work raher han from increases in homebased employmen shares wihin occupaions. The hree lef-hand columns of Table 4 presen, for each census year, he homebased employmen share in each of 20 occupaion caegories. The hypohesis ha homebased employmen shares were equal across occupaion caegories is rejeced a any convenional significance level in all hree census years. Perhaps unsurprisingly, given he empirical observaion ha moivaes his paper, he home-based employmen share grew beween 1980 and 2000 in mos occupaion groups. However, he rae of growh varied subsanially across occupaions. For example, he home-based share grew by over 500% for engineers and scieniss, who rarely were home-based in 1980, while i fell in farming-relaed occupaions, which had he highes home-based share in To allow for heerogeneiy in he wage penaly on home-based jobs, I esimae models of he form 19 lnw i = X i! + #" j D ij + # $ j D ij H i + % i, = 1980, 1990, 2000 (2) j =1 20 where all variables are as defined in (1) and {! j } j =1 is a vecor of occupaion-specific 20 home-based wage penalies in year. Esimaes of {! j } j =1 for each census year are 20 j =1 presened in he hree righ-hand columns of Table To save space, sandard errors are 13 In an unrepored analysis, I insead allowed he home-based employmen share and home-based wage penaly o vary across skill groups defined by he 32 muually exclusive and exhausive caegories derived from he full se of ineracions beween wo sexes, four educaion levels (less han a high school degree, exacly a high school degree, some college, and four years of college or more), and four age groups (25-34, 35-44, 45-54, and 55-64). The qualiaive paerns ha emerged from his analysis are quie similar o hose shown in Table 4. The resuls of his analysis are available from he auhor upon reques. 17

19 no repored. 14 Likewise, esimaes of he coefficiens on he oher covariaes in he model are no repored, alhough he esimaes of! are very similar o hose repored in Table 3. As was rue for home-based employmen shares, one easily rejecs he hypohesis ha he wage penalies on home-based jobs were idenical across occupaion caegories in all census years. Comparing he esimaed coefficiens across census years reveals ha he wage penaly for home-based employmen shrank wih he passage of ime in every occupaional caegory excep for mechanics and repair workers beween 1980 and By he year 2000, home-based workers in a few occupaions acually earned higher average pay han observaionally equivalen on-sie workers. I use sandard decomposiion echniques o assess he empirical imporance of composiion effecs in explaining changes over ime in he aggregae home-based employmen share and aggregae home-based wage penaly. Le H denoe he homebased share a dae among all wage and salary workers aged and le H denoe he home-based share a dae wihin occupaion caegory j. Then H =! j = s 1 j H j, where s j is occupaion group j's share in oal wage and salary employmen a dae. The change in he average home-based employmen share beween daes and! can be decomposed as # H H! " H = ( s j! " s j ) j! + H j & # s ) j! + s j & $ % 2 ' ( + ) $ % 2 ' ( ( H j! " H j ). (3) j =1 The firs erm in (3) is he par of he change in he aggregae home-based share ha is explained by changes over ime in he disribuion of wage and salary workers across j =1 20 j 14 The sandard errors fall in he range in all years for almos all occupaions. The sandard errors are somewha larger for he occupaion caegories wih he smalles employmen shares and he lowes incidence of home-based work. 18

20 occupaion groups, given he average occupaion-specific propensiies for home-based work a daes and!. The second erm in (3) is he par of he change in he aggregae home-based share ha is explained by changes over ime in he propensiies for homebased work wihin occupaion groups. Turning nex o he home-based wage penalies, he empirical specificaion in (2) and well-known properies of leas squares regression imply ha he mean log wage for on-sie workers a ime can be wrien as lnw o o = X ˆ! + D o " ˆ o o # j $ Z j ˆ% 20 j =1 (4) and he mean log wage for home-based workers a ime can be wrien as 20 ( ) lnw h h = X ˆ! + D h " j ˆ + ˆ# h $ % Z j j ˆ& h. (5) j =1 Some simple algebra yields he following expression for he change in he mean log wage difference beween home-based and on-sie workers beween daes and! : ( ) lnw h o (! " lnw! ) " lnw h o ( " lnw ) = Z h o (! " Z! ) " Z h o ( " Z ) $ $ Z h h! + Z ' $ & ) " Z o o! + Z ' ' & & ) ) % % 2 ( % 2 ( ( ˆ# o! + ˆ# o $ ' & % 2 ) ( + ˆ# o! " ˆ# o ( ) + $ Z h h ˆ# (! " Z ) h! + ˆ# h $ ' & % 2 ) ( " ˆ# o! + ˆ# o $ ' ' & & % 2 ) ) + % ( ( $ Z h h! + Z ' & ) ˆ# h! " ˆ# o (! ) " ˆ# h " ˆ# o ( ( )). % 2 ( (6) The firs erm on he righ-hand side of (6) is he par of he change in he mean log wage difference explained by changes over ime in he mean gap in observed skills beween home-based and on-sie workers. The second erm on he righ-hand side of (6) is he 19

21 par of he change in he mean log wage difference explained by changes over ime in he reurns o observed skills, given he average mean gap in observed skills beween homebased and on-sie workers. The hird erm on he righ-hand side of (6) is he par of he change in he mean log wage difference explained by changes over ime in he disribuion of home-based employmen across occupaion caegories, given he average of he occupaion-specific home-based wage penalies a daes and!. Finally, he fourh erm on he righ-hand side of (6) is he par of he change in he mean log wage difference ha is explained by changes over ime in he home-based wage penalies wihin occupaion groups. 15 Table 5 presens he resuls from he saisical decomposiions in (3) and (6) for every combinaion of = 1980, 1990,! = 1990, 2000, and <!. The upper panel shows decomposiions of he changes over ime in he home-based employmen share. The home-based share of wage and salary employmen rose by over wo-enhs of a percenage poin beween 1980 and 1990, by an addiional one-half of a percenage poin beween 1990 and 2000, and hence by abou seven-enhs of a percenage poin over he full weny year period. 16 Given he rariy of home-based employmen among wage and salary workers, hese changes represen a near doubling of he home-based share beween 1980 and Changes over ime in he disribuion of wage and salary employmen across occupaion groups accoun for jus 18% of he growh in he aggregae homebased share beween 1980 and 1990 and jus 4% of he growh in he aggregae homebased share beween 1990 and Thus, he vas majoriy of he growh in he home- 15 The inerpreaions of he las wo erms follow from he fac ha he empirical specificaion in (2) resrics! o be idenical for on-sie and home-based workers. 16 These numbers differ slighly from hose implied by Table 1 because he samples now are limied o individuals wih real hourly wages beween $1 and $

22 based share is explained by increases in he frequency of home-based employmen wihin occupaion caegories. The lower panel of he able, which shows he decomposiions of he changes over ime in he mean log wage difference beween home-based and on-sie workers, ells a similar sory. Growh in he mean log wage for home-based workers exceeded ha for on-sie workers by.187 beween 1980 and 1990, by.283 beween 1990 and 2000, and hence by.47 beween 1980 and In all hree ime inervals, approximaely one-hird of hese gains can be explained by growh in he observed skills of home-based workers relaive o on-sie workers. In conras, changes in he reurns o observed skills accoun for almos none of he relaive wage gains of home-based workers. Thus, he remaining relaive wage gains of home-based workers, which represen abou wo-hirds of he oal gains, mus be explained by changes in he disribuion of home-based workers across occupaion groups and changes in he home-based wage penaly wihin occupaion groups. The able indicaes ha only 10% of hese residual relaive wage gains of homebased workers can be explained by changes in he occupaional composiion of homebased employmen, wih he oher 90% accouned for by reducions in wage penalies on home-based jobs wihin occupaions. In summary, he resuls in Tables 4 and 5 show ha he rise in he home-based employmen share and he decline in he home-based wage penaly in recen decades was observed no only in he aggregae bu also quie generally wihin more narrowly defined occupaion groups. The evidence presened so far is consisen wih he view ha falling employer coss of offering home-based jobs, raher han rising worker valuaions for home-based jobs, have been he primary source of growh in he home-based employmen share in he 21

23 las weny years. Wha caused his decline in coss? Advances in IT are one obvious possibiliy. Several recen sudies have provided evidence suggesing ha innovaions in IT may have played an imporan role in recen decades in he widening of educaional wage differenials (Auor, Kaz, and Krueger 1998), he rise in female employmen and decline in male-female wage differenials (Weinberg 2000), and he adopion of new producion mehods and organizaional pracices by firms (Bresnahan, Brynjolfsson, and Hi 2002). I seems likely ha hese IT advances would have raised he relaive produciviy of home-based workers, and hereby reduced employers coss of offering home-based work arrangemens, by more in some occupaions (e.g., managers and business specialiss) han in ohers (e.g., food or cleaning service). Thus, if IT advances were he main reason ha employers coss of offering home-based jobs fell and if he supply elasiciy ino home-based employmen is similar across occupaions, hen one would expec ha he occupaions ha experienced he larges increases in he home-based share of employmen also had he larges gains in he home-based relaive wage (i.e., larges declines in he home-based wage penaly). Moreover, since IT use rises wih educaion (Auor, Kaz, and Krueger 1998), one migh addiionally expec he larges gains in he home-based employmen share and home-based relaive wage o have occurred in occupaions wih high average educaion. Figure 1 illusraes he relaionship beween changes in home-based relaive wages and changes in home-based employmen shares across he 20 occupaion groups for he and periods. Home-based relaive wages and home-based employmen shares are measured in levels (logs) in he wo graphs on he lef (righ) side 22

24 of he figure. The size of each daa poin in he scaer plos is proporional o he average fracion of oal employmen accouned for by he corresponding occupaion in he wo relevan census years. Occupaions in which he percenage of college graduaes is a leas 1.5 imes as large as he percenage of college graduaes in he full sample are idenified as high educaion occupaions. Mos of he daa poins lie in he posiive quadran of each graph, indicaing ha boh he home-based employmen share and he home-based relaive wage increased beween census years in mos occupaions. However, he correlaion beween changes in he home-based employmen share and changes in he home-based relaive wage across occupaions differs beween decades. In he period, occupaions ha saw larger increases in he home-based employmen share appear o have seen smaller gains in he home-based relaive wage. 17 In conras, beween 1990 and 2000, he change in he home-based employmen share and he change in he home-based wage penaly are posiively correlaed across occupaions. 18 In addiion, above average gains in he homebased relaive wage appear o have been concenraed in high educaion occupaions only in he laer decade. Overall, he evidence in he figure would seem o sugges ha IT innovaion played a more prominen role in he growh in home-based wage and salary employmen beween 1990 and 2000 han for he similar growh ha occurred in he preceding decade. 17 The weighed regression line for level changes in he upper lef graph of Figure 1 has a slope coefficien of!8.743 wih a sandard error of The weighed regression line for he log changes in he upper righ graph of Figure 1 has a slope coefficien of!.302 wih a sandard error of.069. The weighs are equal o he produc of occupaional employmen in he wo relevan census years divided by he sum of occupaional employmen in he wo relevan census years. 18 The weighed regression line for level changes in he lower lef graph of Figure 1 has a slope coefficien of wih a sandard error of The weighed regression line for he log changes in he lower righ graph of Figure 1 has a slope coefficien of.067 wih a sandard error of.041. The weighs are he same as described in foonoe

25 VI. On-he-Job Compuer Use and he Home-Based Wage Penaly One limiaion of he evidence in Figure 1 is ha he supply elasiciy of workers ino home-based jobs may vary across occupaions, in which case changes in he home-based employmen share and changes in he home-based wage penaly need no be posiively correlaed, even if hese changes are caused by IT-induced changes in employer coss of offering home-based work arrangemens ha vary across occupaions. Thus, addiional evidence on he link beween IT use and he home-based wage penaly would be useful. Forunaely, informaion on compuer use a work has been colleced on an occasional basis in supplemens o he CPS since Probi models for on-he-job compuer use can be esimaed wih hese daa and he resuling coefficien esimaes can be used o impue prediced probabiliies of compuer use a work for observaions in he census samples. This allows an examinaion of how he home-based wage penaly has varied wih (he prediced probabiliy of) on-he-job compuer use. Table 6, which updaes Table 4 from Auor, Kaz, and Krueger (1998), repors he proporions of wage and salary workers aged who repor using a compuer a work in he Ocober 1984, Ocober 1993, and Sepember 2001 CPS. I uilize hese paricular CPS supplemens because hey roughly correspond o he en year inervals in he census daa. There exis large differences in he incidence of on-he-job compuer use across race, gender, educaion, and occupaion caegories in each year. While he proporion of workers using compuers a work has risen subsanially over ime in all demographic and skill caegories, he paern of differences in on-he-job compuer use across caegories has remained sable over ime. 24

26 Table 7 presens resuls from probi esimaes of he deerminans of compuer use a work in 1984, 1993, and The covariaes are nearly idenical o hose used in he log wage regressions in Table 3; he only variables no available in he CPS are indicaors for wheher he individual is a par-year worker, wheher he individual is a home-based worker, and (in 1984 and 1993) wheher he individual has a disabiliy. The upper porion of he able repors marginal effecs for seleced covariaes. The signs of he marginal effecs are as expeced given he raw proporions shown in Table 6 and are sable across survey years. However, he magniudes of he marginal effecs imply ha differences in on-he-job compuer use by race, educaion, and occupaion have increased wih he passage of ime, as on-he-job compuer use became more common. The probi coefficiens can be used o calculae a prediced probabiliy of compuer use a work for each individual in each CPS sample, and he lower porion of Table 7 repors summary saisics on hese prediced probabiliies. Several poins are worh noing. Firs, in each year, he prediced probabiliy of compuer use a work varies grealy wih observable individual characerisics. Second, occupaional affiliaion is an especially imporan predicor of on-he-job compuer use, accouning for a leas 80% of he variance in prediced probabiliies of compuer use a work in each year. Even so, he variaion in prediced probabiliies of on-he-job compuer use wihin occupaion groups is non-rivial. Finally, increases over ime in (he prediced probabiliy of) compuer use a work appear o have occurred hroughou he skill disribuion. Since he census samples include all of he variables ha were used as covariaes in he probi models for on-he-job compuer use, he probi esimaes for he 1984, 1993, and 2001 CPS samples also can be used o impue person-specific prediced probabiliies 25

27 of on-he-job compuer use for he observaions in he 1980, 1990, and 2000 census samples, respecively. I do no repor summary saisics for hese prediced probabiliies, as hey are virually idenical o hose repored in Table 7. Of course, hese prediced probabiliies slighly oversae he rue probabiliies of on-he-job compuer use for all observaions in each census year, since each CPS sample was obained a few years laer han he corresponding census sample. However, given he apparen sabiliy over ime in he correlaes of on-he-job compuer use, his should have lile effec on he paern of variaion in he prediced probabiliies across individuals wihin each census sample. 19 Having impued a prediced probabiliy of on-he-job compuer use for each census observaion, I esimae log wage equaions of he form 19 lnw ˆ ˆ i = X i & + '% j Dij + $ H i + # Pi + " Pi H i +! i, j= 1 (7) and lnw ˆ ˆ i = X i & + '% j Dij + ' $ j Dij H i + # Pi + " Pi H i +! i, j= 1 j= 1 (8) for each census year, where Pˆ i is he prediced probabiliy of on-he-job compuer use for individual i in year and all oher variables are as defined in (1) and (2). These equaions exend he specificaions in (1) and (2) by allowing he home-based wage penaly o vary wih an individual s prediced probabiliy of using a compuer a work. If advances in IT have had an imporan impac on he relaive produciviy of working a home, changes 19 Indeed, here is an exremely high correlaion beween he hree alernaive prediced probabiliy of compuer use a work measures derived from he 1984, 1993, and 2001 CPS esimaes, respecively. The correlaion beween he measures compued using he 1984 and 1993 esimaes ranges beween.95 and.96, depending on he census year. Similarly, he correlaion beween he measures compued using he 1993 and 2001 esimaes is.97 in all census years. Even when he prediced probabiliy measures are compued using he esimaes from 1984 and 2001, which are separaed by 17 years, he correlaion ranges beween.88 and.91, depending on he census year. 26

28 over ime in he home-based wage penaly are likely o have differed among workers who vary in heir (prediced probabiliies of) compuer use a work. The specificaion in (7) resrics he home-based wage penaly o be idenical across occupaions, excep for differences associaed wih differences across occupaions in he probabiliy of on-he-job compuer use. In conras, he specificaion in (8) allows for occupaional differences in he home-based wage penaly unrelaed o he use of compuers a work. Table 8 repors he esimaed coefficiens on he home-based work variables in (7) and (8) for he 1980, 1990, and 2000 census samples, wih he odd-numbered columns showing esimaes of! and! from (7) and he even-numbered columns showing esimaes of! from (8). Since he hypohesis ha home-based wage penalies are he same across occupaions is rejeced overwhelmingly in each census year, I focus on he esimaes in he even-numbered columns. In 1980, here was no relaionship beween he home-based wage penaly and he probabiliy of on-he-job compuer use, afer accouning for occupaional differences in he home-based wage penaly. One decade laer, workers wih higher probabiliies of using a compuer a work faced larger wihinoccupaion home-based wage penalies, oher hings equal. However, by he year 2000, he paern had reversed and he wihin-occupaion home-based wage penaly was smaller for individuals wih higher probabiliies of on-he-job compuer use. Since home-based wage penalies were falling on average hroughou he period, a concise summary of he resuls is ha home-based wage penalies decreased mos for workers wih he lowes probabiliies of on-he-job compuer use beween 1980 and 1990 bu decreased mos for workers wih he highes probabiliies of on-he-job compuer use beween 1990 and

29 The recen hisory of IT innovaion offers a plausible explanaion for boh he direcion and iming of hese changes in he relaionship beween he home-based wage penaly and he probabiliy of compuer use a work. The personal compuer (PC) era began wih he inroducion of he Apple II in 1977 and he IBM PC in VisiCalc, he firs spreadshee sofware, was released in The 1980 s brough faser microprocessors, new business sofware, dramaic declines in he real price of compuing, and vas expansion of he insalled compuer base. These innovaions likely alered job asks and increased produciviy more for workers wih skills ha were complemenary o compuers han for hose wihou such skills. 20 However, given he froniers of IT in 1990, few of he gains in on-sie worker produciviy resuling from increased compuerizaion could have been realized in a home-based work seing. Thus, any reducion in home-based wage penalies beween 1980 and 1990 likely would have been smaller among workers wih he higher probabiliies of using a compuer a work, consisen wih he empirical evidence in Table 8. In he 1990 s, progress coninued on he same margins as in he 1980 s bu, imporanly, breakhroughs occurred on new frons as well. In paricular, hyperex language was invened, and he World Wide Web was born, in The nex few years saw he arrival of he firs commercial web browsers and inerne service providers, and wih hese came a massive expansion of file sharing. 21 These laer innovaions, ogeher wih coninued qualiy improvemens and price declines for compuer hardware, allowed individuals who used compuers inensively in heir work o perform many more of heir 20 Auor, Levy, and Murnane (2003) derive his implicaion in a formal model of he effecs of compuers on he ask composiion of jobs. 21 Several sources repor ha raffic on he World Wide Web increased by over 300,000% in See, for example, Hobbes Inerne Timeline a hp:// 28