FOREIGN WORKERS IN SOUTHERN AGRICULTURE * Robert D. Emerson and Orachos Napasntuwong Unversty of Florda The queston addressed n ths paper s the length of tme farm workers wth dfferent characterstcs are expected to contnue dong farm work. Legal status of the worker s the characterstc of utmost relevance; dfferences between workers n the southeast and other areas of the country are also examned. To our knowledge, ths s the frst tme that anyone has examned the queston of duraton of farm work. An earler study by Perloff examned the choce of farm work over non-farm work usng Current Populaton Survey data. Isé and Perloff examned the related queston of legal status and farm worker earnngs. However, ths does not address the queston of how long a farm worker s expected to contnue dong farm work on the bass of legal status or other farm worker characterstcs. Matta addressed the queston of the combnaton of farm and non-farm work by farm workers. The queston of how long ndvduals may be expected to contnue workng n farm work remans an open queston, however. Workers from other countres are an ncreasng porton of the agrcultural work force for the Southeast and for the Unted States as a whole. The Natonal Agrcultural Worker Survey (NAWS) data for the years 1993-1998 revealed that 56 percent of the agrcultural workers n the Southeast were categorzed as unauthorzed workers. Moreover, the porton has been ncreasng from an average of 44 percent for the years 1993-1995 to 68 percent for 1996-1998. The summer of 2001 wtnessed a convergence of support to address concerns about unauthorzed workers n the Unted States, partcularly n agrculture, as worker advocacy groups and * Selected paper for presentaton at the annual meetng of the Southern Agrcultural Economcs Assocaton n Orlando, Florda, February 4-6, 2002.
2 employer groups both supported legslatve acton to address the ssue. Mexco s Presdent Fox made the ssue a major focus of hs new admnstraton, and Presdent Bush lkewse gave the mmgraton problem a hgh profle n the agenda of hs new admnstraton. Legslatve acton addressng the problem appeared the most lkely t has n the last several years when Presdent Fox addressed the Unted States Congress after meetng wth Presdent Bush n the week pror to September 11. The September 11 tragedy has agan moved the ssue off the agenda. The only queston s when t wll return. Although employers may not be aware of whch of ther employees are unauthorzed workers, most are aware of the problem n the aggregate. Wth such a large porton of the workers beng unauthorzed, there s an ncreased rsk to producers that a sudden tghtenng of enforcement wll abruptly restrct labor supply. The most severely mpacted n the short run would be producers of pershable, labor-ntensve crops. In the longer term, producton technques would be expected to adjust as wage rates rose, although short-term losses could be crtcally mportant. Indeed, the Florda ctrus ndustry has reactvated ts mechancal harvestng program n the last few years n response to concerns about the nstablty of harvest labor supply. The response followng September 11 hghlghted concerns about labor avalablty wth more thorough border enforcement. The worker perspectve s that those who are unauthorzed face lvng n an underground world, often separated from famles, and few places to turn for support n the Unted States. Unauthorzed workers tend to be concentrated n ndustres and occupatons where they are not hghly vsble and where communcaton n Englsh s not essental. Unsklled agrcultural employment s an obvous choce; other common employment areas are hotel and restaurant servces and garment work. Unauthorzed workers have ncreasngly been recognzed as a
3 sgnfcant segment of the work force for poultry and meat processng plants. Notwthstandng the relatvely mnscule number apparently nvolved n terrorst actvtes, although of crtcal mport, the vast majorty are merely seekng a better lfe for ther famles. A recurrng theme n addressng the unauthorzed worker problem s a proposed amnesty that would legalze alen workers under some set of terms typcally ncludng verfcaton of date of entry and proof of employment n the Unted States for a specfed perod of tme. A major legslatve ssue to agrcultural employer groups has been the addtonal tme that unauthorzed agrcultural workers would be requred to work n agrculture to be elgble for amnesty. Under most recent proposals, unauthorzed workers seekng amnesty would be requred to mantan a major porton of ther employment tme n agrculture for up to three years n order to be elgble for amnesty. After ths tme there would be no restrctons on where they could seek employment. The paper frst descrbes the agrcultural labor market n the southeast. The major laborntensve commodty groups are dentfed and the work force s characterzed wth avalable data on ther employment characterstcs and legal status. Wth ths background, an econometrc model s specfed addressng the explanaton of varaton n years of farm work n the U.S. The prmary focus of the estmaton s on the worker legal status and dfferences between the southeast and the remander of the U.S. Labor n Southern Agrculture The regon examned s the southeastern regon of the U.S. as defned by the NAWS data: Alabama, Arkansas, Florda, Georga, Lousana, Msssspp, and South Carolna. The 1997 Census of Agrculture reported $2.2 bllon of labor expendtures by crop farms n ths regon, representng 24 percent of producton expendtures by crop farms n the regon, and representng
4 83 percent of labor expendtures on all farms n the regon. There s consderable varaton among ths group of states; Florda domnates the regon wth 59 percent of labor expendtures by crop farms n the seven state regon. Three commodty groups represent over 67 percent of labor expendtures on crop farms n the regon: greenhouse, nursery and florculture (27%), fruts and tree nuts (22%), and vegetables and melons (18%). The NAWS data provde extensve detal for farm worker characterstcs on these farms. 1 Among the workers ntervewed n the regon over the perod of 1993-1998, 50 percent were workng n vegetable crops, 16 percent were workng n feld crops, 15 percent were workng n hortcultural crops, 11 percent were workng n fruts and nuts, and eght percent were n other crops. Most (46%) were performng harvest tasks, 23 percent were conductng pre-harvest tasks, 12 percent were dong post-harvest tasks, 11 percent were sem-sklled workers, and the remander of the workers were n mscellaneous tasks. An ncreasng porton of the workers was employed by labor contractors rather than drectly by growers. The average over all sx years was 28 percent employed by labor contractors, although the average ncreased from 24 percent n the 1993-1995 perod to 33 percent n the 1996-1998 perod. The ntervewed workers worked an average of fve days per week n the week prevous to ntervew for an average of 36 hours. They had 28 weeks of farm work n the prevous year, four and one-half weeks of non-farm work, and had spent an average of 15 weeks at ther current job. A partcularly mportant characterstc n the context of ths paper s that they spent an average of ten weeks abroad durng the past year. Even more noteworthy, s that when grouped by ther job task, harvest workers spent an average of 14 weeks abroad durng the past year. In addton, the tme spent abroad by all workers has doubled from seven weeks for 1993-1995 to 1 All values for worker characterstcs n ths and the next two paragraphs were compled by the authors from the NAWS publc use data.
5 14 weeks for 1996-1998. Both the ncreasng number of weeks abroad and the larger number of weeks abroad by harvest workers hghlght the senstvty of labor supply to potental border restrctons. Average real hourly earnngs were $5.83 over the 1993-1998 perod. 2 They reported an annual ncome range of $5,000-7,499. Most workers were born outsde the Unted States; only 22 percent were born n the U.S. over the sample perod. Among those not born n the U.S., 52 percent had been dong farm work n the U.S for three or fewer years. Just over 32 percent of the non-u.s. born had been n the U.S. dong farm work for one year or less. There s some ndcaton that tenure n the U.S. may be fallng. Among those sampled n 1996-1998, 37 percent had been dong farm work n the U.S. for one year or less, whle among those sampled n 1993-1995, only 27 percent had been dong farm work n the U.S. for one year or less. The largest group of foregn born was born n Mexco, representng 61 percent of all workers. Despte ther proxmty, Puerto Rco and the Carbbean accounted for only fve percent and two percent, respectvely. Central Amerca was the orgn for ten percent of the workers. In contrast to the 22 percent of the NAWS workers born n the U.S., 77 percent consder the U.S. to be ther permanent resdence. Ths array of employment and worker characterstcs places many of the concerns by employers and workers about the current agrcultural labor market nto context. For example, harvest work represented the jobs at tme of ntervew for 46 percent of the workers. Among fruts and vegetables, tmng of the harvest s often crtcal to obtan the best qualty product, the best product prce, and for some commodtes, to mantan any salvageable product for market, harvest must occur wthn just a few days. Yet harvest workers spent consderably more weeks abroad durng the prevous year than dd workers n any other task. The fact that 78 percent of the workers were non-u.s. born, and that 32 per cent of these had been n the U.S. dong farm 2 Hourly earnngs for 1993-1997 were converted to 1998 dollars usng the consumer prce ndex.
6 work for one year or less (wth the latter percentage rsng), suggests that the avalablty of labor for crtcal tasks s susceptble to abrupt changes n the flow of labor across the border. The queston we address s the determnaton of farm work duraton n the U.S. by ndvdual workers wth dfferng characterstcs. Econometrc Model An ntal specfcaton mght be a standard regresson equaton wth the number of years of farm work explaned by varous characterstcs of the worker. Obvous explanatory varables would be the age of the worker, whether or not the worker was authorzed for work n the U.S., perhaps educaton level to reflect alternatve opportuntes outsde the agrcultural sector, and of course, whether or not the worker was employed n the southeastern regon, as s our nterest for ths paper. However, the NAWS data set s a sample of people who were dong farm work at the tme of ntervew. Consequently, t s only known that the ndvduals have been dong farm work for the number of years reported up to the tme of ntervew; t s unknown how much longer each of the workers wll contnue to do farm work nto the future. In the nomenclature of lmted dependent varables, the varable of nterest, years of farm work, s censored for every observaton. Consequently, ordnary least squares wll generate based and nconsstent estmates of the parameters. Farm workers n the NAWS data set are observed at the date of ntervew, and they report, among other thngs, the years of farm work they have done. Ths establshes a startng date for farm work, and we know the ntervew date, but we do not observe the endng date. The years of farm work varable s therefore censored from the rght. Unlke the standard censorng problem where a number of observatons accumulate at a partcular value such as wth tobt analyss, the censorng problem for the duraton varable results from the uncompleted duraton
7 of farm work. Snce workers are ntervewed at a pont n tme, and are known to be workng at that tme, the samplng s known as stock samplng (Lancaster). A partcular problem assocated wth stock samplng s that ndvduals wth dfferent years of farm work have dfferng probabltes of beng sampled n the populaton. A worker wth a long true duraton has a hgher probablty of beng observed at the samplng date than does a worker wth a short duraton of farm work. The long duraton worker has a greater chance of straddlng the ntervew date wth hs work duraton than does the short duraton worker. Analogously, there are many short duraton workers who have already completed ther terms of farm work and are thus mssed n the samplng procedure. One way of statng the samplng problem s that the farm worker s observed condtonally upon the person workng at the tme of ntervew. Moreover, the person wth s years of farm work s observed condtonally upon havng entered farm work s years ago, and contnung to perform farm work through the ntervew date. Ths s one of a group of econometrc problems known as duraton analyss (Kefer; Lancaster; Nckell). Nckell s paper dealt wth a very smlar queston of analyzng when an ndvdual mght leave a spell of unemployment for a job, gven that the ndvdual was unemployed at the tme of ntervew. Nckell s contrbuton was to recognze that one could stll glean nformaton about duraton from the data although the spells were all uncompleted by the nature of the data set,.e. all ndvduals n the data set were unemployed. Hs procedure specfed a densty functon for the startng date n addton to the condtonal probablty that the ndvdual was stll unemployed at the ntervew tme havng become unemployed at the startng date. Nckell s approach s adopted for the problem n ths paper. Drawng from Nckell, the startng pont s the specfcaton of the probablty that the th worker wll stll be a farm worker at the ntervew date, t, condtonal upon havng started farm
8 work s years pror. The logstc functon s a convenent specfcaton for the probablty gven that no duraton can be negatve: where: 1 p( x ( t, s), s) = 1+ exp( y ( s)) y = β + β South + β Age + β Edcn + β Unauth + β s + β s 2 0 1 2 3 4 5 6 where South s a dummy varable equal to unty f the worker was located n the southeast regon and zero otherwse, Edcn s hghest year of school completed, Unauth s one f the worker reported as beng unauthorzed for work n the U.S., and zero otherwse, s s the years of farm work completed and Age s chronologcal age reported n years. Nckell s specfcaton correspondng to the probablty of ndvdual beng a farm worker at perod t condtonal upon havng started farm work at perod t-v s: v 1 (, ) = f t v = τ 0 (1 p( x ( t, τ ), τ )) for v 1 Lettng the probablty that ndvdual entered farm work at perod τ be u (τ), then the probablty of the worker contnung to be a farm worker at perod t (the ntervew tme) s U ( t) = ( age 8) 1 u τ = 0 ( t τ ) f ( t, τ ) The upper lmt of the sum lmts the possble years of farm work to be fewer than the person s age less eght years,.e. the person s not assumed to do farm work for pay pror to age eght. Fnally, the densty functon of uncompleted farm work duraton (v) for the th worker s: u ( t v) f U ( t) ( t, v)
9 The lkelhood functon for the sample s smply the product of the densty functon mmedately above over all observatons n the sample: L = n = 1 U ( t) u ( t v) f ( t, v) The functon f has been specfed parametrcally va the logstc dstrbuton for p above. The entry probablty, u (t-v), s specfed n the sprt of Nckell s approach as the lkelhood of a new entrant to farm work n perod t-v. An estmate of ths s readly avalable for the sx years of the avalable NAWS data. The new entrants n a gven year are those wth a sngle year of farm work, and can be calculated from the sample. Assumng as a denomnator, the sum of all possble farm workers, the rato provdes an estmate of the probablty of entry n a gven year. Snce u appears n both the numerator and the denomnator of the lkelhood functon, the constant denomnator of the rato cancels out, and only the number of new entrants s needed. The new entrants for the survey years of 1998 back to 1993 were 370, 353, 400, 347, 275, and 378, respectvely. For lack of other avalable data, t s assumed that the number of new entrants for each pror year was the average number of entrants for these sx years, 354. Ths clearly s an approxmaton, but snce t s multpled by the values for f, t should not greatly affect the results snce the f values become small the further away from the ntervew date for whch the calculatons are done. A fnal note s that y (s) s evaluated at y (50) for s 50. Estmaton of the parameters was done by maxmum lkelhood usng the 1998 sample of 2,066 NAWS workers wth vald data for all varables n the model. 3 The parameter estmates and ther estmated standard errors are dsplayed n table 1. All parameter estmates are hghly sgnfcantly dfferent than zero except for the constant term and the square of years of farm 3 The lkelhood functon was coded n Gauss software usng the maxlk procedure.
10 work. However, examnng the effect of years on the y varable entals the lnear combnaton of the coeffcent on years and the coeffcent on the square of years: y s = β 5 + 2β 6 The estmated effect calculated at the mean of years s 0.2053 wth an estmated standard error of 0.0270 whch s clearly hghly sgnfcantly dfferent than zero. The nterpretaton of the sgns of the estmated parameters s the drecton n whch the probablty of a farm worker leavng farm work, p(x (t,s),s), wll change wth an ncrease n the level of the assocated varable. The postve sgn on the coeffcent for the South dummy varable mples that the probablty of a farm worker n the southeast leavng farm work n the next year s hgher, ceters parbus, than for farm workers n other regons of the U.S. Smlarly, unauthorzed workers have a hgher probablty of leavng farm work n the next year than do authorzed workers. Increasng age, wth a negatve coeffcent, reduces the probablty of s Table 1. Maxmum lkelhood parameter estmates Varable Parameter Estmate Estmated Standard Error Constant 0.5441 0.3727 South 0.3891 0.1201 Age -0.1892 0.0193 Educaton 0.0448 0.0158 Unauthorzed 2.1085 0.1474 Years of farm work 0.2085 0.0328 Years of farm work 2-0.00019 0.00049
11 leavng wthn the next year. More educaton, wth a postve coeffcent, ncreases the probablty of leavng farm work wthn the next year. The educaton varable has the clearest economc nterpretaton as a reflecton of alternatve employment opportuntes; the sgn s clearly consstent wth expectatons. The effect of ncreasng years of work (as calculated n the prevous paragraph) s also postve, suggestng that the longer one works n farm work, ceters parbus, the greater the probablty of leavng farm work n the next tme perod. 4 Whle the sgns of the parameter estmates n table 1 provde a qualtatve perspectve to the estmates, the expected duratons of farm work for varous worker characterstcs provde a more drect evaluaton of the effects of the ncluded varables. Selected alternatves are presented n table 2 for comparson wth the expected duraton of farm work for the average farm worker n the sample. Frst, note that the expected duraton of farm work for the average farm worker s 7.25 years. The observed sample mean for years of farm work s 8.33 years. 5 The sample mean s an over-estmate of expected duraton due to the length-based samplng n a stock sample as dscussed earler. In ths case, the expected duraton s over-estmated by over one year, or 15 percent n comparson to the corrected estmate obtaned from the maxmum lkelhood approach. Farm workers n the southeastern regon have an expected duraton of farm work just under a year less than for workers outsde the regon. The unauthorzed worker s expected duraton s just over half that of the authorzed worker. Combnng the effect of locaton and legal status, unauthorzed southeastern farm workers have an expected duraton of 4.5 years of 4 Snce the coeffcent for the squared years s negatve, the overall effect of years of farm work could eventually turn negatve. However, at the estmated parameters, ths only occurs after 552 years of farm work, clearly outsde the relevant range of values. 5 Sample means of all varables are presented n appendx table 1.
12 Table 2. Expected duraton of farm work n years a Worker Characterstcs Expected Duraton Average farm worker 7.25 Worker ntervewed n southeast regon 6.61 Worker ntervewed outsde the southeast 7.41 Authorzed worker 9.30 Unauthorzed worker 5.07 Authorzed southeastern worker 8.68 Unauthorzed southeastern worker 4.51 Age = 20 2.87 Age = 30 6.12 Age = 40 10.14 Age = 50 14.54 a In each row of the table, all other characterstcs of the worker are set at the mean values over the sample. farm work n comparson to 8.7 years for authorzed southeastern farm workers. The expected duraton of farm work ncreases wth age, but far less than proportonately. Clearly, there s a dfference n duraton of farm work by authorzed and unauthorzed workers after controllng for dfferent characterstcs of the workers. Concludng Remarks Much of the debate n the context of agrcultural labor regardng amnesty, regularzaton and guest worker programs concerns the extent to whch unauthorzed farm workers would be requred to contnue workng n agrculture n order to be elgble for some form of legal status n the U.S. The ndustry concern s that wthout some restrctons, agrcultural employers would
13 face an abrupt reducton n the supply of labor snce the substantal number of unauthorzed workers would seek alternatve employment wth newly opened opportuntes outsde agrculture. Legal status s undoubtedly a promnent reason that many unauthorzed workers choose agrcultural employment. However, other consderatons such as language and communcaton ablty and other necessary sklls to partcpate n other sectors of the U.S. economy also make agrcultural employment a startng pont n the absence of those sklls. It s mportant to recognze that the latter consderatons reman unaffected by changes n legal status. Nevertheless, the calculated expected duratons for authorzed n contrast to unauthorzed workers are strkngly dfferent. The nature of the data used for analyss also lmts the nterpretaton of the results. Recall that the data are censored: all observatons of farm work are uncompleted spells of farm work. The data source does not provde any nformaton about what persons who have left farm work may be dong. They ndeed may have found non-farm employment, but they may have returned to ther home country, or they may be unemployed. The nformaton s smply not avalable from the data set. The latter are questons to be addressed from alternatve approaches and sources of nformaton.
14 References Kefer, Ncholas M. Economc Duraton Data and Hazard Functons. Journal of Economc Lterature 26 (1988):646-679. Isé, Sabrna, and Jeffrey M. Perloff. Legal Status and Earnngs of Agrcultural Workers. Amercan Journal of Agrcultural Economcs 77 (1995):375-386. Lancaster, Tony. The Econometrc Analyss of Transton Data. New York: Cambrdge Unversty Press, 1990. Matta, Benjamn N., Jr. The Off-Farm Work of Hred Farm Workers. In Seasonal Agrcultural Labor Markets n the Unted States, ed. Robert D. Emerson, pp. 140-164. Ames: Iowa State Unversty Press, 1984. Natonal Agrcultural Worker Survey (NAWS). Publc use data dsk, 1993-1998. OASP, U.S. Department of Labor, Washngton, D.C. Nckell, Stephen. Estmatng the Probablty of Leavng Unemployment. Econometrca 47 (1979):1249-1266. Perloff, Jeffrey M. The Impact of Wage Dfferentals on Choosng to Work n Agrculture. Amercan Journal of Agrcultural Economcs 73 (1991):671-680. U.S. Department of Agrculture, NASS. 1997 Census of Agrculture, Vol. 1, CD-Rom Set AC97-CD-VOL1-1A,1B, Washngton, D.C., 1999.
15 Appendx table 1. Sample means of NAWS data, 1998 Varable Sample mean South (= 1 f n southeast, =0 otherwse) 0.20 Age 32.93 Educaton (hghest grade completed) 6.59 Unauthorzed (= 1 f unauthorzed for work n the U.S., = 0 otherwse) 0.47 Years of farm work 8.33