LEGAL STATUS AND U.S. FARM WAGES

Transcription

1 LEGAL STATUS AND U.S. FARM WAGES Nobuyuk Iwa Internatonal Agrcultural Trade and Polcy Center Food and Resource Economcs Department PO Box 1124 Unversty of Florda Ganesvlle, FL Robert D. Emerson Internatonal Agrcultural Trade and Polcy Center Food and Resource Economcs Department PO Box 1124 Unversty of Florda Ganesvlle, FL Lurleen M. Walters Internatonal Agrcultural Trade and Polcy Center Food and Resource Economcs Department PO Box 1124 Unversty of Florda Ganesvlle, FL Abstract Usng Natonal Agrcultural Workers Survey data, we estmate U.S. farm worker wage dfferentals by legal status. In order to adequately correct sample selecton bas, we develop a Heckman-type twostage method wth an ordered probt model n the frst stage and a wage equaton model n the second stage. Keywords: Legal status, wage rates, sample selecton bas. JEL Code: J43 Selected Paper prepared for presentaton at the Southern Agrcultural Economcs Assocaton Annual Meetng, Orlando, Florda, February 5-8, 26. Copyrght 25 by Nobuyuk Iwa, Robert D. Emerson and Lurleen M. Walters. All rghts reserved. Readers may make verbatm copes of ths document for non-commercal purposes by any means, provded that ths copyrght notce appears on all such copes.

2 LEGAL STATUS AND U.S. FARM WAGES Introducton The U.S. agrcultural labor market s heavly dependent on foregn-born workers. Accordng to the Natonal Agrcultural Workers Survey (NAWS) data, 79% of agrcultural workers were foregn born n 22. Ths fgure, although ncreasng slghtly, has been rather stable durng the 199s. However, the composton of legal status of farm workers has vared dramatcally n the same perod. For the years only 16% of farm workers are unauthorzed. The porton of unauthorzed workers rose to 36% for the years , and 5% for the years and 48% for the years Ths dramatc change n legal status composton of farm workers mght have had a sgnfcant mpact on the cost structure for U.S. agrculture over the above perod. The purpose of ths study s to nvestgate whether, holdng worker characterstcs constant, there are wage dfferentals by legal status for farm workers n the U.S. For that purpose, we estmate wage equatons for each legal status worker and mplement smulatons to forecast how the wage of current unauthorzed farm workers wll change f they are gven a legal status. Lmted emprcal work has been done on the relatonshp between legal status and farm worker wage (Taylor 1992, Ise and Perloff 1995, Morett 2). In general, these studes conclude that estmated wages for authorzed, n contrast to unauthorzed, workers are sgnfcantly hgher. A problem to be dealt wth when studyng the relatonshp between legal status and farm worker wages s sample selecton bas. The wage for a worker wth a partcular legal status s observed only f the worker s n that legal status; the worker s earnngs n an The authors are grateful to Susan Gabbard, Trsh Hernandez, Alberto Sandoval and ther assocates at Agurre Internatonal for assstance wth the NAWS data, and to Danel Carroll at the U.S. Department of Labor for grantng access and authorzaton to use the NAWS data. Ths research has been supported through a partnershp agreement wth the Rsk Management Agency, U.S. Department of Agrculture; by the Center for Internatonal Busness Educaton and Research at the Unversty of Florda; and by the Florda Agrcultural Experment Staton. The authors alone are responsble for any vews expressed n the paper. 1

3 alternatve legal status are not observed. Each foregn-born worker chooses hs/her legal status, consderng condtons such as hs/her ndvdual demographc characterstcs, cost of applcaton, and beneft of the status. Ths selecton of legal status may be related to the wage of the worker. If ths s the case, the wage equaton wthout correctng for ths selecton process wll yeld based and nconsstent estmates. Correcton of selecton bas s partcularly mportant for the estmaton of the change n mean wage, should she/he attan an alternatve legal status. 1 Ths smulaton study s essental to forecast the effects of mmgraton polcy change whch could result n a change n status from unauthorzed to authorzed for a large number of workers. Ise and Perloff (1995) correct the selecton bas by usng Lee s extenson of Heckman s two-stage sample selecton method (Lee 1983, Heckman 1979). 2 In the frst stage, the multnomal logt model s run to estmate the legal status equaton assumng the error term has a Gumbel dstrbuton. However, the second-stage wage equaton wth the correcton term, whch s calculated from the frst stage result, does not generally yeld consstent estmates wth the normal dstrbuton assumpton of error term n the wage equaton (Schmertmann 1992, Bourgugnon et al. 24). We develop an alternatve Heckman-type two-stage method wth the ordered probt model n the frst stage. We use the ordered probt model n the frst stage for two reasons: (1) ths method, wth the approprate correcton term, yelds consstent estmates n the second stage wage equaton, and (2) t mantans the ordnal nature of legal status whch the multnomal logt does not. Consderng the advantages n the labor market, the alternatves can be ordered as unauthorzed, authorzed, permanent resdent, and ctzen workers. 1 Maddala (1983) emphaszes mportance of ths by showng followng examples: the evaluaton of the benefts of socal programs, and professon choce problems. 2 Taylor (1992) corrects selecton bas by the standard Heckman two-stage method usng selecton of prmary (sklled) or secondary (unsklled) farm obs nstead of legal statuses. But, the correcton terms are not statstcally sgnfcant n ether wage equaton. 2

4 Methodology Our Heckman-type two-stage method s specfed wth the ordered probt model for the frst stage and the wage model for the second stage. The ordered probt model s used to explan the legal status of worker as a functon of the ndvdual s socoeconomc and polcy varables denoted as vector x ). A foregn-born worker s legal status (J) takes on four values: ( =unauthorzed, 1=authorzed, 2= permanent resdent (green card holder), and 3=ctzen. Wth the famlar argument of latent regresson (Greene 23), we can assume that an unobserved varable J * s censored as follows: * * J f J, J f < J, = = 1 1 * * J = 2 f 1 < J 2, J = 3 f < J, where J * = x α + ε ; x s a vector of 2 exogenous characterstcs of ndvdual ; and ε s a dsturbance term. The characterstcs nclude gender, martal status, Englsh speakng ablty, race (black, whte, and other), ethncty (Hspanc and other), age, age squared, educaton, educaton squared, US farm experence, US farm experence squared, and the year of ntervew (before 1993, after 21, and n-between). We assume that ε s normally dstrbuted wth a mean of zero and a standard devaton of σ ε whch s normalzed to be one. Then the lkelhood functon can be expressed as x α 1 xα xα Φ Φ Φ J = σ ε J = 1 σ ε σ ε L( α, σ ε, data) =, (1) 2 xα 1 xα 2 xα Φ Φ 1 Φ J = 2 σ ε σ ε J = 3 σ ε where Φ( ) ndcates the cumulatve dstrbuton for the standard normal. The wage equaton may be expressed as ln w = z β + u where u N(, σ ). ~ Further, we assume that the mean of log of wage for a worker wth legal status ( ln w ) depends on ndependent varables z (dummy varable for seasonal worker, dummy varable for worker pad by pece rate, dummy varable for sklled task, race (whte or not), gender, martal 3

5 status, age, age squared, educaton, educaton squared, US farm experence, US farm experence squared, Englsh speakng ablty, avalablty of free housng, regon (Calforna, Florda, and other), the year of the ntervew (before 1993 or after 21 or n-between). However, wage w s observed only f person has legal status. Ths s a typcal case for selecton bas. Assumng ε and u are bvarately normally dstrbuted wth correlaton coeffcent ρ, the mean of the log of the wage condtoned on the legal status of person s E [ ln w J = ; x, z ] = z β + ρ σ λ, where λ s the correcton term for the selecton bas whch s gven as φ λ = Φ ( γ ) φ( γ 1 ) ( γ ) Φ( γ ), 1 where γ x α = σ 1, γ 1 = ε σ ε xα. Note that we can use the result of the ordered probt model n the frst stage for estmates of γ and γ 1. Also note that 1 =, 3 = from the assumpton of normal dstrbuton. In the second stage we estmate the wage equaton (equaton (2)) by OLS wth the correcton term, λ, ncluded, whch s calculated from the frst stage. [ ln w J = ; x, z ] = z + ρ σ λ + v = z β + β λ + v. β λ (2) Note that where v s heteroschedastc. Its condtonal varance depends on as var 2 2 [ v J ; x, z ] = σ [ ρ δ ] =, 1 δ γ φ = Φ ( γ ) γ 1 φ( γ 1) ( γ ) Φ( γ ) 1 φ + Φ ( γ ) φ( γ 1) ( γ ) Φ( γ ) 1 2. N Then a consstent estmator of σ s gven by ˆ σ = [ vˆ + ˆ β ˆ λ δ ]/ N where N s the number = 1 4

6 of observatons for legal status and 2 vˆ s estmated from the least squares resduals from (2). We 3 also use a consstent estmator of the asymptotc covarance matrx for βˆ and βˆ. λ Data The data used n ths study are obtaned from the Natonal Agrcultural Workers Survey (NAWS) (Offce of Assstant Secretary for Polcy 25). We used the study perod from 1989, when the NAWS was frst avalable, to the most recent year avalable, 24. Ths secton wll descrbe the defntons of each varable used n the model. Legal status s a dscrete varable rangng from to 3. Status = unauthorzed workers means that the worker s undocumented (dd not apply to any legal status or applcaton was dened) and also ncludes those who had no work authorzaton even f they were documented. Status 1 = authorzed workers or documented workers; these workers must have a work authorzaton and may fall nto any of the followng status: havng border crossng card/commuter card, wth pendng status, or temporary resdents holdng a non-mmgrant vsa. Status 2 = permanent resdents or green card holders who have the rght to resde and work n the U.S., and status 3 = ctzens who are ctzens by brth or naturalzaton. Wage s average earnngs per hour for a worker regardless of the method of payment. If a worker s pad by pece rate, hs/her wage s calculated by (pece rate) (average peces per hour). We also deflated the wage by U.S. Consumer Prce Index (Bureau of Labor Statstcs, U.S. Department of Labor 25). 4 The varable Englsh measures the capablty to speak Englsh. The varable s a dscrete varable rangng from 1 to 4, where 1= not speakng Englsh at all, 2 = speak a lttle Englsh, 3 = 3 Jmenez and Kugler (1987) also use ordered probt n the frst stage to correct the selecton bas, but ther method does not deal wth heteroschedastcty for the dsturbance term nor does t use a consstent estmator for the asymptotc covarance matrx for βˆ and. 4 We use the standard CPI: monthly CPI for all tems for all U.S. urban consumers. βˆλ 5

7 somewhat able to speak Englsh, and 4 = speakng good Englsh. Hspanc s a dummy varable for Hspanc whch ncludes Mexcan-Amercan, Mexcan, Chcano, Puerto Rcan, and other Hspanc ethnc groups. Black (or Afrcan Amercan) and Whte are also dummy varables derved from a queston regardng ther race whch may also be Amercan Indan/Alaka Natve, Indgenous, Asan, Natve Hawaan or Pacfc Islander, or others. Age was calculated from the dfference between the date of ntervew and the date of brth, except n the earler years of the survey when age was asked drectly. Educaton s the hghest grade level for educaton, and t ranges from to 2. Experence s the number of years of dong farm work n the U.S. (not ncludng farm work experence abroad). Sklled Task s a dummy for workers who engage n sem-sklled or supervsory tasks. Although the orgnal questons have over 1 task codes, tasks are grouped nto sx categores as follows: 1 = pre-harvest, 2 = harvest, 3 = post-harvest, 4 = sem-sklled, 5 = supervsor, and 6 = other. Seasonal Worker s a dummy for workers who were workng on a seasonal bass for the employer at the tme of ntervew. Pece rate s a dummy for workers who are pad by pece rate nstead of beng pad by the hour or a salary. Labor contractor s a dummy varable for workers who are employed by labor contractors rather than the grower. Free housng s a dummy varable for workers (or workers and ther famly) who receve free housng from ther current employer. It does not nclude those who own ther house or lve for free wth frends or relatves. It also excludes those who pay for housng provded by employers or by the government or charty. The dummes for Florda and Calforna are the locaton at the tme of ntervew. Before 1993 dummy varable s for all the years pror to 1993 when the maorty of IRCA legalzaton was granted, and After 21 s the years post-september 11, 21 event. Ordered Probt Model for Legal Status Here we estmate the ordered probt model for legal status for foregn-born farm workers usng NAWS data. Table 1 shows estmates for parameters and asymptotc standard errors (gven 6

8 n the parentheses) usng 3,61 observatons of foregn-born farm workers. Usng a.5 sgnfcance crteron, we fnd that all coeffcents except educaton squared are statstcally sgnfcant. The thrd column of Table 1 shows the margnal effect of each varable on the probablty of a worker becomng legal. The probablty of worker beng legal s gven by * Pr ob( J > ) = 1 Φ( x α ). Then the margnal effect of varable k evaluated at the mean x s φ ( x α) α for the contnuous varables and k Φ( x k α k ) Φ( x k α k α k ) for the dummy varables, where x k and α k are varables and coeffcents excludng k. Females, marred, workers wth hgher Englsh speakng ablty, non-black, whte, non-hspanc are statstcally sgnfcantly more lkely to have more advantageous legal status all else beng the same. We also fnd that both age and US farm experence have a sgnfcant nonlnear effect on legal status. US farm experence has a postve effect on legal status up to 38 years. Age has a postve effect on legal status up to 78 years. Educaton has a sgnfcantly postve lnear effect on legal status. We fnd that the greatest postve margnal effect s from the female dummy followed by Englsh speakng ablty and the before 1993 dummy. The greatest negatve margnal effect s from the Hspanc dummy followed by the after 21 dummy and the Black dummy. Note that, holdng all other characterstcs the same, the workers ntervewed before 1993 are 12% more lkely and those ntervewed after 21 are 13% less lkely to be legal compared to those ntervewed between these perods. Fnally, Table 2 shows the actual-predcted legal status table. A worker s predcted to be status (unauthorzed) f x ˆ ˆ α <, and s predcted to be status 1 (authorzed) worker f ˆ ˆ ˆ < x α < 1 and so on. Table 2 shows that 84% of unauthorzed workers are correctly predcted to be unauthorzed. In the same way, 19% of authorzed workers, 67% of permanent 7

9 resdents and 18% of ctzens are correctly predcted n ther legal status. Our ordered probt model does a very good ob n dstngushng unauthorzed workers from legal workers, but many of authorzed workers and ctzen workers are mstakenly predcted to be permanent resdents. Wage Equaton Model wth Selecton Bas Correcton Here we estmate the wage equaton model wth selecton bas correcton usng the results from the ordered probt legal status model n the frst stage. Table 3 shows estmates for parameters and asymptotc standard errors (gven n the parentheses) for farm workers wth each legal status. Status (unauthorzed) workers have 16,195 observatons, status 1 (authorzed) workers have 2,688 observatons, status 2 (permanent resdent) workers have 9,739 observatons, and status 3 (ctzen) workers have 9,166 observatons. Based on asymptotc standard errors usng a.5 sgnfcance crteron, the coeffcents on the selectvty varables, λ, are all hghly sgnfcant except for authorzed (status 1) workers. That s, usng ordnary least squares wthout correctng for selectvty would lead to bas n all equatons except for authorzed workers. Actually, observatons for status 1 workers are much fewer than other three categores, and they are concentrated n early 199s. For the years about 26% of all workers had ths legal status (status 1), but that porton has declned to only 1.3% for the years Besdes, many workers were gven ths legal status under IRCA, whch mght weaken the explanatory power of the legal status selecton model for ths legal status category. Henceforth, the selecton bas correcton term calculated from ths result does not have as much effect on the authorzed worker wage as t does on wages for other worker categores. Many varables have a statstcally sgnfcant effect on worker wage n a common drecton for all equatons. Regardless of the legal status, workers n sklled task, non-seasonal workers, workers employed by growers, workers pad by pece rate, male workers, workers n Calforna, and workers ntervewed after 21 are statstcally sgnfcantly more lkely to have a hgher wage. Free housng has sgnfcantly negatve effect on wage for all legal statuses except 8

10 for authorzed workers for whom t does not have a sgnfcant effect. Martal status has a sgnfcantly postve effect on wage for permanent resdent and ctzen workers, but does not have a sgnfcant effect for the other two worker status groups. We also fnd that age has a sgnfcant nonlnear effect on wage for all legal statuses except for the authorzed worker for whch t does not have a sgnfcant effect. 5 All of the sgns of these coeffcents are reasonable, but an nterestng result s for the after 21 dummy. It ncreases the wage rate for unauthorzed workers by only 4% all else beng the same, whle t ncreases the wages for authorzed workers by 13%, for permanent resdents by 11%, and ctzen workers by 9%. As for the magntude of nfluence, the pece rate payment dummy and sklled task dummy domnate. The former ncreases wage more than 2% and the latter ncreases wage more than 15%, regardless of legal status. Other varables tend to have varous drectons of nfluence on farm work wage for each legal status. US farm experence has a sgnfcantly postve nonlnear effect on wage for unauthorzed and ctzen workers, but a sgnfcantly negatve nonlnear effect for permanent resdent workers. 6 Educaton has a sgnfcantly postve nonlnear effect on wage for unauthorzed workers, but a sgnfcantly negatve nonlnear effect for ctzen workers. 7 The whte dummy has a sgnfcantly postve mpact on wage for ctzen workers, but has a sgnfcantly negatve effect for authorzed workers. The before 1993 dummy has a sgnfcantly postve mpact on wage for authorzed and permanent resdent workers, but has a sgnfcantly negatve effect for ctzen workers. The Florda dummy has a sgnfcantly negatve effect on wage for unauthorzed workers, but does not have a sgnfcant effect on other legal statuses. 5 The age effect s postve up to an age of 28 years for unauthorzed, up to 25 years for permanent resdent workers, and up to 41 years for ctzen workers. 6 The US farm experence effect s postve up to an experence of 19 years for unauthorzed, and up to 34 years for ctzen workers. On the other hand, the effect s negatve through 25 years for permanent resdent workers 7 The educaton effect s postve up to an educaton of 13 years for unauthorzed, but the effect s negatve up to 4 years for ctzen workers. Consderng the mean length of educaton for each (6 years for unauthorzed, 1 years for ctzen workers), we can consder that the educaton effect s postve for both. 9

11 Next, usng estmates of each equaton, we calculate the predcted farm worker wage by legal status by averagng the predctons over all observatons for each equaton (Table 4). The results ndcate that the average predcted wage for unauthorzed workers s the lowest wth $6.85, followed by authorzed workers ($7.51) and ctzen workers ($7.78). Permanent resdent workers have the hghest average predcted wage of $8.8. That s, average predcted wages for authorzed, permanent resdent and ctzen workers are 1%, 18% and 14% hgher than for unauthorzed workers. Ths s comparable to the result from Ise and Perloff (1995) who conclude that the earnngs of legal workers n 1991 averaged 15% more than for unauthorzed workers. Smulaton Study Fnally, we mplement a set of smulatons to examne how farm work wage of a typcal unauthorzed worker would be expected to change wth a change n legal status. Ths approach solates the effect of legal status of the worker from dfferng observable characterstcs of workers by holdng the characterstcs constant across varyng legal status. In addton, we vary the tme perod (before or after 21 8 ), the locaton (Calforna or other states of the U.S. 9 ), the task (sklled or non-sklled), the type of employer (grower or labor contractor), and the type of payment (pece rate payment or others). We fx each contnuous varable at the mean of unauthorzed worker observatons, and fx each remanng dscrete varable at the category wth the maxmum number of observatons of unauthorzed workers. The profle of the typcal unauthorzed worker s llustrated n Table 5. As before, the condtonal expected wage for the unauthorzed worker wth observable characterstcs of x and z s gven by [ w ε xα; x, z ] = z β + ρ σ λ = z β + β λ, Eln λ (3) When legal status of unauthorzed worker s converted to status (=1, 2 and 3), the condtonal 8 Before 21 means years from 1993 to 21. 1

12 expected wage would be [ w x α ; x, z ] = z β + ρ σ λ = z β + β, 1,2, 3 E ln ε λ = (4) Note that the condton n the square bracket s retaned, snce t formulates the unobservable characterstcs for legal status selecton of the worker. 1 We calculate these condtonal expected wages by usng estmates ( ˆ β, ˆ β, ˆ λ, ˆ β, ˆ β λ λ ) from prevous secton. The expected wages for ths typcal unauthorzed worker, calculated from equatons (3) and (4), are shown n Table 6. For 31 out of 32 cases, 11 unauthorzed workers workng as legal workers would have a hgher expected wage than when workng as unauthorzed workers. 12 For specfc legal statuses, 77 out of 96 smulatons have hgher expected wages than as an unauthorzed worker. The largest effects were for unauthorzed workers workng under the permanent resdent status all 32 cases were postve, varyng from 19 to 46 percent. On the other hand, for 15 out of 32 cases, unauthorzed workers workng under a ctzen status would have lower expected wages. Interestngly, these converted ctzen workers who engage n unsklled tasks would have lower expected wages n 14 out of 16 cases, whle those who engage n sklled tasks would have a hgher expected wage n 15 out of 16 cases. Smulaton results for authorzed workers are not as large n magntude as for permanent resdent workers, but the drectons are almost equally stable. Focusng on after 21, for all 16 cases, unauthorzed workers workng under authorzed worker status would have hgher expected wages, rangng from 6 to 31 percent. Even before 21, for 12 out of 16 cases, the effects are postve. Interestngly, all negatve cases happen for non-pece-rate-payment workers n non- Calforna regon. λ 9 Other states of the U.S. does not nclude Florda. 1 See p.259 n Maddala (1983) for the detaled argument. 11 Only excepton s the followng case: an unsklled worker pad by non-pece rate by labor contractor n non-calforna state before Legal worker wage s the average wage weghted by composton of three legal statuses. 11

13 We fnd a very clear tendency for three employment categores. Comparng sklled and unsklled tasks, the former has a hgher expected wage ncrease for legal status n 47 out of 48 cases. 13 Comparng workers employed by growers and those employed by labor contractors, the former have hgher expected wage ncreases for legal status n 47 out of 48 cases. 14 Comparng pece rate payment and other payment contract, the former has hgher expected wage ncrease for legal status n 47 out of 48 cases. 15 Also, the wage ncrease for legal status tends to be hgher after 21 than before. The after 21 perod has hgher expected wage ncrease for legal status than before n 47 out of 48 cases. 16 Calforna tends to have hgher expected wage ncrease for legal status than rest of U.S. Calforna has hgher expected change n wage than rest of U.S. n 32 out of 48 cases. 17 We compare our smulaton results wth those done by Ise and Peroff (1995). Here we focus on the perod before 21 because Ise and Perloff use 3,989 observatons for the years 1989 to Snce Ise and Perloff do not nclude dummy varables for employment category (dummy varables for sklled task, labor contractor employment, pece rate payment, and seasonal contract) n ndependent varables, drectly comparable cases are only the followng two: Mean of unauthorzed worker case (frst row n table 6) and Calforna case (nnth row n table 6). For the former case, Ise and Perloff predct that wages of authorzed, permanent resdent and ctzen workers are expected to be 12%, 1% and -.2% hgher than unauthorzed worker wage respectvely, whle our smulaton has -2%, 8% and -11% respectvely. For the Calforna case, 13 Only excepton s the followng case: an unsklled worker pad by non-pece rate by labor contractor n Calforna before 21 has 13 % wage ncrease from convertng to a ctzen, whle a sklled worker has.3 wage ncrease n the same stuaton. 14 Only excepton s the followng case: an unsklled worker pad by non-pece rate by labor contractor n Calforna before 21 has 13 % wage ncrease from convertng to a ctzen, whle a worker employed by grower has 9 % wage decrease n the same stuaton. 15 Only excepton s the followng case: an unsklled worker pad by non-pece rate by labor contractor n Calforna before 21 has 13 % wage ncrease from convertng to a ctzen, whle a worker pad by pece rate has 1 % wage decrease n the same stuaton. 16 Only excepton s the followng case: an unsklled worker pad by non-pece rate by labor contractor n Calforna before 21 has 13 % wage ncrease from convertng to a ctzen, whle the same knd of worker has 8 % wage decrease n the same stuaton after

14 Ise and Perloff predct that wages of authorzed, permanent resdent and ctzen workers are expected to be 7%, 17% and -3% hgher than unauthorzed worker wage respectvely, whle our smulaton has 5%, 2% and -9% respectvely. In general, our result s hgher for permanent resdent workers, but lower for authorzed and ctzen workers than the result from Ise and Perloff. Taylor (1992) predcts that weekly earnngs of legal workers are expected to be 33% and 5% hgher than unauthorzed workers for prmary (sklled) obs and secondary (unsklled) obs respectvely. 18 Correspondng results from our smulaton show that wages of legal workers are expected to be 1% and 8% hgher than unauthorzed workers for sklled tasks and unsklled tasks respectvely. 19 Concluson Usng Natonal Agrcultural Workers Survey data, we estmate U.S. farm worker wage dfferentals by legal status. In order to adequately correct sample selecton bas, we develop a Heckman-type two-stage method wth an ordered probt model n the frst stage and a wage equaton model n the second stage. We also mplement smulatons to examne how farm work wage of a typcal unauthorzed worker would be expected to change wth a change n legal status. In the ordered probt legal status model, all coeffcents except educaton squared are statstcally sgnfcant, and 84% of unauthorzed workers are correctly predcted to be unauthorzed. We fnd that the greatest postve margnal effect on the probablty of a worker beng legal s from the female dummy followed by Englsh speakng ablty and the before 1993 dummy. The greatest negatve margnal effect s from the Hspanc dummy followed by the after excepton cases are as follows: wage ncrease for workers n Calforna resulted from convertng to a permanent resdent s lower than that for workers n rest of US. 18 Taylor does not separate workng hours dfferentals from weekly earnngs dfferentals. Hence, ths large earnngs dfferentals may be partally explaned by workng hours dfferentals. 19 We use the average of column legal n row 9 to 12 n table 6 for unsklled task, and the average of column legal n row 13 to 16 n table 6 for sklled task. 13

15 21 dummy and the Black dummy. The workers ntervewed before 1993 are 12% more lkely and those ntervewed after 21 are 13% less lkely to be legal compared to those ntervewed between these perods, gven all else beng the same. In the second stage wage equaton model, workers n sklled task, non-seasonal workers, workers employed by growers, workers pad by pece rate, male workers, workers n Calforna, and workers ntervewed after 21 are statstcally sgnfcantly more lkely to have a hgher wage, regardless of the legal status,. The after 21 dummy ncreases the wage for unauthorzed workers by only 4% all else beng the same, whle t ncreases the wages for authorzed workers by 13%, for permanent resdent by 11%, and for ctzen worker by 9%. As for the magntude of nfluence, the pece rate payment dummy and sklled task dummy are outstandng. The former ncreases wage more than 2% and the latter ncreases wage more than 15% for all legal status workers. Average predcted wages for authorzed, permanent resdent, and ctzen worker are 1%, 18% and 14% hgher than that for unauthorzed workers respectvely. We mplement a set of smulatons to examne how farm work wage of a typcal unauthorzed worker would be expected to change wth a change n legal status. For 77 out of 96 cases, unauthorzed workers workng as legal workers would have expected a hgher wage than workng as unauthorzed workers. The largest effects were for unauthorzed workers workng under the permanent resdent status all 32 cases were postve, varyng from 19 to 46 %. The results are smlar for the cases of unauthorzed workers workng under authorzed worker status, although the magntude of wage ncreases s smaller. On the other hand, for 15 out of 32 cases, unauthorzed workers workng under a ctzen status would have lower expected wages. In general, our result has a hgher expected wage ncrease for permanent resdent worker, but lower wage ncrease for authorzed and ctzen worker than the result from Ise and Perloff (1995). The smulaton study also shows very clear tendency for three employment categores. Sklled workers, workers employed by growers, and workers pad by pece rate have hgher expected wage ncrease for legal status than those unsklled, employed by labor contractors, and 14

16 pad by other methods respectvely. Also, legal status tends to have a hgher expected wage ncrease after 21 than before. It seems that rather clear tendences from wage equatons enable us to forecast the wage for a worker wth specfc demographcs, employment type and legal status. The mportant nformaton for farmers s how much cost wll ncrease as the legal status and demographc composton of labor force changes, gven the current technology. Suppose 5% of current employees are unauthorzed for a farmer, but all of them are converted to authorzed workers. The farmer may need to rase ther wages by about 1% on average, so that the total labor cost may ncrease by about 5%. Assumng labor cost s approxmately one thrd of total cost, the labor cost ncrease wll result n approxmately 1.7% ncrease n total cost. However, the actual ncrease n total cost may be lower than ths snce a farmer may absorb the wage ncrease by usng other factors more ntensvely. The next ssue s to combne ths wage forecast wth producton functon of farms. References Bourgugnon, F., M. Fourner, and M. Gurgand. Selecton Bas Correctons Based on the Multnomal Logt Model: Monte-Carlo Comparsons, DELTA Workng Papers 24-2, DELTA (Ecole normale supéreure) (September 24). Bureau of Labor Statstcs, U.S. Department of Labor. Consumer Prce Indexes, (August 25). Avalable at Greene, W. H. Econometrc Analyss, 5 th Edton. Prentce Hall (23). Heckman, J. Sample Selecton Bas as a Specfcaton Error, Econometrca, 47 (January 1979): Ise, S. and J. M. Perloff. Legal Status and Earnngs of Agrcultural Workers, Amercan Journal of Agrcultural Economcs, 77 (May 1995): Jmenez, E. and B. Kugler. The Earnngs Impact of Tranng Duraton n a Developng Country: An Ordered Probt Selecton Model of Columba s Servco Naconal de Aprendzae (SENA), Journal of Human Resources, 22(2) (Sprng 1987) : Lee, L. F. Generalzed Econometrc Models wth Selectvty, Econometrca, 51(2) (March 1983):

17 Maddala, G. S. Lmted-Dependent and Qualtatve Varables n Econometrcs. Cambrdge Unversty Press (1983). Morett, E. and J. M. Perloff. Effcency Wages, Deferred Payments, and Drect Incentves n Agrculture, Amercan Journal of Agrcultural Economcs, 84(4) (November 22): Offce of the Assstant Secretary for Polcy, U.S. Department of Labor. The Natonal Agrcultural Workers Survey: What s the Natonal Agrcultural Workers Survey (NAWS)? (May 25). Avalable at Schmertmann,C. P. Selectvty Bas Correcton Methods n Polychotomous Sample Selecton Models, Journal of Econometrcs, 6(1-2) (January-February 1994): Taylor J. E. Earnngs and Moblty of Legal and Illegal Immgrant Workers n Agrculture, Amercan Journal of Agrcultural Economcs, 74 (November 1992): Table 1. Ordered Probt Model for Legal Status for Foregn-Born Farm Workers Parameter Estmate Margnal Effect Parameter Estmate Margnal Effect Female.461* (.2).177 Educaton (.5) Marred.213*.84 Experence.164*.44 (.18) (.3) Englsh Speakng.356* (.1).141 Experence 2 -.2* (.7) Black -.22* -.88 Before *.121 (.8) (.19) Whte.146*.58 After * (.16) (.2) Hspanc -.524* (.51) * (.96) Age.36* (.4) * (.97) Age 2 -.2* (.5) * (.99) Educaton.48* (.7).16 * ndcates that the estmated coeffcent s statstcally sgnfcant at 5 percent level of sgnfcance. * The probablty of worker beng legal s gven by Pr ob( J > ) = 1 Φ( x α ). Then the margnal effect on becomng authorzed of varable k evaluated at the mean x s φ ( x α) α k for the contnuous varables and Φ( x k α k ) Φ( x k α k α k ) for the dummy varables, where x k and α k are varables and coeffcents excludng k. 16

18 Table 2. Actual-Predcted Legal Status Table Predcted Legal Status Total Actual Legal Status ,85 1,18 1, , , , ,824 1,439 6, , ,531 1,397 7,624 Total 18,241 3,856 13,63 1,461 37,161 Table 3. Wage Equaton Model for Farm Workers wth Each Legal Status Unauthorzed Authorzed Permanent Ctzen Resdent λ -.45* (.11) -.29 (.2) -.12* (.1).31* (.5) Sklled Task.165* (.57).188* (.72).26* (.28).311* (.37) Seasonal Worker -.4* (.3) -.45* (.1) -.65* (.5) -.79* (.6) Labor Contractor -.62* (.4) -.72* (.13) -.83* (.6) -.115* (.11) Pece Rate.225*.326*.261*.258* (.4) (.11) (.6) (.11) Female -.59* (.5) -.58* (.17) -.74* (.7) -.87* (.7) Marred -.1 (.4).8 (.12).13* (.6).69* (.6) Whte.9 -.7*.2.82* (.4) (.11) (.5) (.6) Age.2* (.1) -.6 (.3).5* (.1).8* (.1) Age 2 -.4* (.1) -.3 (.4) -.1* (.2) -.1* (.2) Educaton.9* (.2).1 (.5).4 (.2) -.13* (.3) Educaton 2 -.4* (.1).1 (.3) -.2 (.2).2* (.2) Farm Experence.9*.8 -.6*.8* (.1) (.4) (.2) (.8) Farm Experence 2 -.2* -.1.1* -.1* (.3) (.7) (.3) (.2) Free Housng -.38* (.5) -.17 (.14) -.13* (.7) -.7* (.8) 17

19 Table 3 (contnued). Wage Equaton Model for Farm Workers wth Each Legal Status Unauthorzed Authorzed Permanent Ctzen Resdent Calforna.27* (.4).1* (.13).24* (.5).57* (.8) Florda -.49* (.5) -.11 (.15) -.11 (.9).17 (.11) Before *.17* -.35* (.6) (.16) (.7) (.11) After 21.41* (.4).134* (.32).114* (.6).92* (.6) Constant 1.81* 2.25* 2.58* 1.68* (.17) (.78) (.4) (.28) * ndcates that the estmated coeffcent s statstcally sgnfcant at 5 percent level of sgnfcance. Table 4. Average Predcted Wage for Each Legal Status ($) Legal Status Wage Unauthorzed 6.85 Authorzed 7.52 (9.69%) Permanent Resdent 8.8 (17.85%) Ctzen 7.78 (13.51%) Values nsde the parenthess are % changes from unauthorzed worker wage. Table 5. Profle of the Typcal Unauthorzed Worker Constant 1 Female Marred Hspanc 1 Whte Black Age Englsh Speakng 1.47 Educaton 6.73 Experence 5.75 Seasonal Worker 1 Free Housng 18

20 After 21 Table 6. Smulated Changes n Farm Wage by Legal Status a Calforna Sklled Task Labor Contractor Pece Rate Payment Unauthorzed No No No No No 6.64 No No No No Yes 8.31 No No No Yes No 6.24 No No No Yes Yes 7.81 Authorze d 6.47 (-2.44) 8.97 (7.9) 6.3 (-3.37) 8.35 (6.87) Legal Status Legal b Permanent Resdent 7.96 (19.97) 1.33 (24.26) 7.32 (17.43) 9.5 (21.63) Ctzen 5.89 (-11.28) 7.62 (-8.32) 5.25 (-15.88) 6.79 (-13.7) Legal c 6.81 (2.68) 8.92 (7.3) 6.2 (-.63) 8.11 (3.84) No No Yes No No 7.83 No No Yes No Yes 9.81 No No Yes Yes No 7.36 No No Yes Yes Yes (-.15) 1.83 (1.43) 7.28 (-1.1) 1.8 (9.38) 1.32 (31.85) (36.56) 9.49 (29.5) (33.67) 8.3 (2.65) 1.4 (6.8) 7.16 (-2.67) 9.27 (.58) 8.93 (14.4) (19.17) 8.12 (1.36) 1.63 (15.33) No Yes No No No 6.82 No Yes No No Yes 8.54 No Yes No Yes No 6.41 No Yes No Yes Yes (4.9) 9.91 (16.2) 6.66 (3.9) 9.23 (14.91) 8.15 (19.56) 1.58 (23.83) 7.5 (17.2) 9.73 (21.21) 6.23 (-8.6) 8.7 (-5.55) 5.55 (13.33) 7.19 (-1.44) 7.15 (4.93) 9.37 (9.65) 6.51 (1.55) 8.52 (6.12) No Yes Yes No No 8.4 No Yes Yes No Yes 1.8 No Yes Yes Yes No 7.56 No Yes Yes Yes Yes (7.37) (18.74) 8.4 (6.34) (17.61) 1.57 (31.39) (36.9) 9.72 (28.61) (33.2) 8.51 (5.76) 11.1 (9.29) 7.58 (.28) 9.81 (3.62) 9.37 (16.54) (21.79) 8.53 (12.78) (17.86) 19

21 After 21 Table 6 (contnued). Smulated Changes n Farm Wage by Legal Status a Calforna Sklled Task Labor Contractor Pece Rate Payment Unauthorzed Yes No No No No 6.87 Yes No No No Yes 8.61 Yes No No Yes No 6.46 Yes No No Yes Yes 8.1 Authorze d 7.37 (7.28) 1.22 (18.64) 6.87 (6.25) 9.51 (17.51) Legal Status Legal b Permanent Resdent 8.82 (28.34) (32.93) 8.12 (25.61) 1.53 (3.11) Ctzen 6.47 (-5.82) 8.38 (-2.68) 5.77 (-1.71) 7.47 (-7.73) Legal c 7.56 (9.94) 9.9 (14.89) 6.87 (6.4) 9. (11.19) Yes No Yes No No 8.11 Yes No Yes No Yes 1.16 Yes No Yes Yes No 7.62 Yes No Yes Yes Yes (9.8) (21.43) 8.29 (8.75) (2.27) (41.4) (46.8) 1.52 (38.5) (42.99) 8.84 (8.97) (12.6) 7.88 (3.32) 1.2 (6.77) 9.9 (22.11) (27.6) 9.1 (18.17) (23.49) Yes Yes No No No 7.6 Yes Yes No No Yes 8.85 Yes Yes No Yes No 6.64 Yes Yes No Yes Yes (15.35) (27.57) 7.59 (14.25) 1.51 (26.35) 9.3 (27.89) (32.47) 8.31 (25.18) 1.79 (29.66) 6.85 (-2.97) 8.87 (.26) 6.11 (-8.) 7.91 (-4.93) 7.94 (12.35) 1.39 (17.41) 7.22 (8.73) 9.45 (13.63) Yes Yes Yes No No 8.33 Yes Yes Yes No Yes 1.44 Yes Yes Yes Yes No 7.83 Yes Yes Yes Yes Yes (18.6) (3.57) 9.16 (16.93) (29.32) (4.55) 15.2 (45.58) 1.77 (37.57) (42.49) 9.36 (12.27) (16.1) 8.34 (6.45) 1.79 (1.) Notes: a All other worker characterstcs are as n Table 5. b Numbers n parentheses are percentage changes from expected wage n an unauthorzed status. c The legal category combnes the three prevous categores authorzed, permanent resdent, and ctzen. The calculaton s: 1.4 (24.79) (3.4) 9.46 (2.76) (26.2) E[ln( w 1) J = 1]Pr[ J = 1] + E[ln( w 2) J = 2]Pr[ J = 2] + E[ln( w 3) J 1 Pr[ J = ] = 3]Pr[ J = 3] 2