AgBoForum Volume 4, Number 2 2001 Pages 115-123 THE ESTIMATED PROFIT IMPACT OF RECOMBINANT BOVINE SOMATOTROPIN ON NEW YORK DAIRY FARMS FOR THE YEARS 1994 THROUGH 1997 Loren W. Tauer 1 Data from New York dary farms for the years 1994 through 1997 were used to estmate whether recombnant bovne Somatotropn (rbst) generated profts for adopters. Results show that the estmated proft mpact of rbst, although generally postve, was statstcally zero. Herd average mlk producton per cow clearly ncreased wth rbst use. Key Words: dary; proftablty; rbst; recombnant bovne Somatotropn. Bovne Somatotropn s a hormone produced by the dary cow that regulates mlk producton. The genetc materal for ths compound has been solated and s produced by recombnant botechnology. Ths recombnant-produced bovne Somatotropn can then be njected nto the dary cow to augment her naturally produced levels of ths hormone, enhancng mlk producton, but requrng addtonal feed and other nputs to ncrease mlk producton. The compound recombnant bovne Somatotropn (rbst) has been commercally avalable to Unted States (US) dary producers snce February of 1994. Monsanto s the only US suppler of recombnant bst under the regstered trade name POSILAC. Before approved for sale, rbst was subject to years of nvestgaton and testng. Gven the large producton response per cow that most of these tests reported, rbst was generally projected to be proftable for dary farmers, wth estmates often exceedng $100 per year per cow (Butler, 1992), although some projected lttle or no proft (Maron & Wlls, 1990). Now that US farmers have used rbst for a number of years, t s possble to estmate ther actual producton and proft responses. Tauer and Knoblauch (1997) used data from the same 259 New York producers n 1993 and 1994 to estmate the mpact of rbst on mlk producton per cow and return above varable cost per cow change between these two years. Recombnant bst was not avalable n 1993, but one-thrd of these farmers used rbst n 1994. The use of rbst had a postve and statstcally sgnfcant mpact on average producton per cow (a = 0.01), but the proft effect, although postve and large, was not statstcally dfferent from zero (a = 0.14). Stefandes and Tauer (1999) also analyzed the producton and proft effects usng the same data source, but ncluded data from 1995, resultng n a panel data set of 211 farms. They corrected for self-selecton bas by usng the two-step Heckman approach, and estmated a probt adopton functon for each year (Greene, 1997). They lkewse found a statstcally sgnfcant and postve 1 Loren Tauer s a Professor n the Department of Appled Economcs and Management at Cornell Unversty n Ithaca, New York. 2001 AgBoForum.
effect on mlk producton per cow from the use of rbst, but found that the mpact of rbst on profts was statstcally zero. Ott and Rendleman (2000) used 1996 Unted States Department of Agrculture (USDA) dary data, and from actual farm rbst response concluded that the optmal rbst use would be 73 percent of the herd, wth a herd mlk response of 2,983 pounds per cow. Unfortunately, that data set dd not collect cost nformaton, but usng cost budgets they found that rbst would ncrease profts by $126 per cow. Ths paper extends fndngs on New York farms by studyng four years of rbst use data (1994 through 1997) to determne the proft per cow from rbst for each of those four years. A number of models are estmated. The frst s an rbst use dummy varable regresson assumng the only dfference between the users and non-users s the use of rbst. A probt adopton functon s then used to correct for self-selecton bas. A more extensve proft functon s also estmated controllng for self-selecton bas. Models The data are from a group of farms, some of whch used rbst. The ntent s to determne whether the use of rbst ncreased profts for those farmers usng rbst. If determnants of proft are dstrbuted randomly across farms that use and do not use rbst, then farms can smply be sorted nto rbst users and non-users and dummy varable regresson can be used to test whether the use of rbst has an mpact on farm profts. The regresson model s specfed as follows, w = a + d + e u (1) where, w s proft per farm, a s an ntercept, d s the rbst mpact wth u as a measure of whether rbst s used, and e s the error term wth as the ndex for farms. However, the potental exsts for self-selecton bas n ths test snce farmers themselves determned ther use or non-use of rbst. Farmers that use rbst may be more or less proftable as a group even wthout the use of rbst. The result s that the error term, e, may be correlated wth the rbst treatment effect, d, leadng to bas n the estmated mpact. There are a number of remedes to ths data lmtaton, most of whch nvolve the estmaton of a separate equaton explanng the selecton decson, and usng the predcton from that equaton to correct for self-selecton bas. T hs selecton process equaton can also be estmated separately or smultaneously wth the rbst mpact equaton by Maxmum Lkelhood Estmaton. Here probt adopton functons are estmated and used to sort farmers nto rbst users and non-users. T he treatment effect of rbst s then estmated by both the Heckman two-step procedure and Maxmum Lkelhood Estmaton of the adopton and rbst mpact varables (see secton 15.8 of Davdson and MacKnnon (1993)). It s also possble that not only do characterstcs of the farms determne whether farmers use rbst, but farm characterstcs themselves may determne farm proft. To ncorporate ths concept, an explanatory proft functon s further modeled as a functon of farm characterstcs. The proft w of farm s specfed as follows, w = a + b x + d u + e (2) where, u denotes the use of rbst on farm and d s the mpact of rbst use on proft, x represents observed exogenous varables for farm and b s the mpact of those varables on proft, a s the ntercept, and e s the error term. The rbst adopton functon s modeled agan 116
as a probt functon and estmated wth equaton (2) by Maxmum Lkelhood Estmaton. All estmaton was done usng the software LIMDEP (1998). Data The data were from the New York Dary Farm Busness Summary (DFBS) Program for the years 1994 through 1997 (Knoblauch & Putnam, 1998). Ths s a record collecton and analyss project prmarly meant to assst dary farmers n managng ther operatons. Varable specfcaton s consstent wth the annual Dary Farm Busness Summary Report. The average and standard devaton of these data for the year 1997 are reported n table 1. Table 1: Defnton of Varables. Varables Defnton Mean 1997 S.D. 1997 SRPCOW Mlk recepts mnus operatng cost per cow ($) 377.00 359.00 MILKCOW Mlk producton per cow (pounds) 18,739.00 3,359.00 COWS Average number of cows 183.00 243.00 MILKPR Mlk prce per cwt. 13.70 0.73 AGE Age of prncpal owner n years 48.00 89.00 BUS_ORG Busness organzaton, 1 f multple owner 0.39 EDUC Educaton, 1 f more than hgh school 0.54 N = 280 observatons for 1997 Ths s not a random sample. It represents a populaton of farmers that actvely partcpate n agrcultural extenson and research programs. Slghtly less than one-half of the partcpants each year partcpated n the other three years of the survey. The farms are larger on average than New York dary farms and they experence hgher levels of producton per cow. Yet, there s a sgnfcant amount of heterogenety n the data. The smallest farm has only 29 cows, whle the largest has over 2,000 cows. The average number of cows s 183. The DFBS surveys for each of the four years asked farmers to ndcate ther use of rbst n one of fve categores as follows: (0) not used at all; (1) stopped usng t durng the year; (2) used on less than 25 percent of the herd; (3) used on 25-75 percent of the herd; or (4) used on more than 75 percent of the herd. Most responses over the perod were n categores 0 and 3. Very few farms ndcated they used t on more than 75 percent of the herd. Lkewse, few farms used t on less than 25 percent of the herd. Ths rbst use codng has lmted nformatonal content. Although most of these farms are DHIA (Dary Herd Improvement Assocaton) members, that organzaton does not code rbst use on ndvdual cow records, so nether age nor producton level of ndvdually treated cows was known. Ths lack of detaled rbst management nformaton precludes analyss on rbst use tactcs, whch may be complex and unque by farm. Farmers usng rbst must beleve that t s proftable on ther farms. As such, farms were smply sorted nto rbst users f they checked categores 2, 3, or 4 and non-users f they checked categores 0 or 1. 117
Emprcal Results The dummy varable regresson wthout rbst treatment selecton correcton estmated rbst proft mpacts per cow of $46, $64, $88, and $73 for the years 1994, 1995, 1996, and 1997, respectvely (table 2). Although these are all postve values, the t-statstcs (coeffcent/standard error) are all less than 2. One must conclude then that statstcally the mpact of rbst, on average, for the farms that used rbst was zero. Table 2: Estmated Impact of RbST on Profts Per Cow, Dummy Varable Regresson (Coeffcent Estmate Dvded by Standard Error n Parentheses). Varable 1994 1995 1996 1997 Intercept 580 468 550 343 (24.75)* (18.66)* (18.12)* (11.76)* rbst Impact ($) 46 64 88 73 (1.27) (1.72) (1.99) (1.71) Adjusted R-squared 0.00 0.01 0.01-0.01 Number of Observatons 324 329 307 280 * Indcates statstcally dfferent from zero at probablty = 0.05. Bnary probt adopton functons were estmated for each year, and results show that age has a negatve mpact on adopton, and a mult-owned busness has a postve mpact on adopton (table 3). However, the standard errors were almost as large as the coeffcent estmates. It s clear, however, that educaton and farm sze have postve and statstcally sgnfcant mpacts on adopton. Largescale farmers who have more than a hgh school educaton are more lkely to adopt rbst. These results are smlar to dary farmers who adopted rbst n Wsconsn (Barham, Jackson-Smth & Moon, 2000). T he probt adopton functons were then used to adjust for selecton bas n the dummy varable regresson. Two approaches were used. T he Heckman two-step procedure used the probt results to compute the nverse Mll's rato, and that value was ncluded as a varable n the rbst dummy varable regresson. The estmated coeffcent on the nverse Mll's rato s lambda n table 4. Lambda s postve n 3 of the 4 years, whch would mply selecton bas such that more proftable farms elect to use rbst. However, snce the standard errors of the lambda estmates are large, one should conclude that there s no sample selecton bas present. Ths s the same result found by Stefandes and Tauer. T he rbst proft mpacts are nconsstent, but none are statstcally dfferent from zero. The Maxmum Lkelhood Estmaton procedure allowed lambda to be decomposed and estmated nto l = rs, where s s the standard devaton of the mpact equaton error term, and r s the correlaton between the mpact and selecton equaton error terms. A postve r mples that rbst selecton and rbst proft mpact are postvely correlated, such that more proftable farms are more apt to adopt rbst. As table 4 shows, however, the correlaton s negatve n two of the four years, and n all four years the standard errors of the estmates are large, so one must conclude that the correlaton s essentally zero statstcally. There apparently s no rbst selecton bas. 118
Table 3: Bnary Probt Model Estmates for RbST Adopton, Maxmum Lkelhood Estmaton (Coeffcent Estmate Dvded by Standard Error n Parentheses). Varable 1994 1995 1996 1997 Intercept -3.79-3.98-3.71-5.60 (-5.63)* (-6.19)* (-5.62)* (-8.22)* Educaton 0.25 0.35 0.33 0.55 (1.61) (2.25)* (2.08)* (3.18)* Age -0.014-0.014-0.012-0.002 (-1.76) (-1.85) (-1.39) (-0.87) Busness Organzaton 0.40-0.03-0.68 0.003 (2.21)* (-0.20) (-0.39) (0.02) Log (cows) 0.84 0.93 0.86 1.11 (6.76)* (7.38)* (7.05)* (7.77)* Number of Observatons 324 329 307 280 * Indcates statstcally dfferent from zero at probablty = 0.05. Snce other varables may mpact proft and may dffer between rbst and non-rbst users, table 5 reports results where proft per cow s estmated as a functon of rbst use and other farm varables. RbST use s agan modeled as a probt adopton functon. Both functons are estmated smultaneously usng Maxmum Lkelhood Estmaton. Results are shown n table 5. (Heckman two-step estmates of the rbst mpact are also reported.) RbST mpacts for all years are postve, although the standard errors are so large that statstcally one must conclude that the mpacts are zero. The determnng factor for proftablty n any of the four years s the mlk prce receved. The estmated rho coeffcents are, agan, all statstcally zero, mplyng no exstence of selecton bas. The Heckman two-step estmates of the mpact of rbst were negatve for three of the four years, but agan none of these estmates was statstcally dfferent from zero. The ablty to measure determnants of farm proftablty s challengng snce there are numerous causaton factors, many of whch cannot be quantfed. The models specfed here faled to measure a statstcally sgnfcant mpact of rbst on farm proftablty. One reason may be an ncorrect model specfcaton. To test that possblty, the same model as reported n table 5 was estmated substtutng herd average mlk producton per cow for the prevous dependent varable of profts per cow. The results are reported n table 6. The major dfference n the results s that, where prevously the prce of mlk largely determned proft per cow, now educaton and farm sze determne producton per cow. Another major dfference s that now rbst has a postve and statstcally strong mpact on producton per cow, whereas rbst use dd not have a statstcally measured mpact on proft per cow. Ths mpact ranged from a low of 2,668 pounds per cow n 1994 to a hgher response of 3,466 pounds per 119
cow n 1996. These output responses are smlar to those found n Ott and Rendleman (2000) usng 1996 USDA dary data. Table 4: Estmated Impact of RbST on Profts Per Cow, Adjustng for Selecton Bas wth Adopton Functon (Coeffcent Estmate Dvded by Standard Error n Parentheses). Varable 1994 1995 1996 1997 -- Heckman Two-Step Procedure -- Intercept 607 433 569 378 (16.80)* (10.92)* (11.39)* (9.26)* rbst Impact ($) -17 140 47-2 (-0.23) (1.85) (0.49) (-0.03) Lambda 49-60 31 66 (0.96) (-1.16) (0.49) (1.23) -- Maxmum Lkelhood Estmates -- Intercept 544 491 514 352 (14.39)* (12.56)* (9.70)* (8.35)* rbst Impact ($) 31 67 84 75 (0.70) (1.59) (1.71) (1.59) Sgma 324 334 387 357 (25.18)* (30.15)* (26.50)* (26.16)* Rho.21-0.11 0.15-0.45 (1.06) (-0.70) (0.72) (-0.26) Adopton Estmates Smlar to Table 3 Number of Observatons 324 329 307 280 * Indcates statstcally dfferent from zero at probablty = 0.05. 120
Table 5: Estmated Impact of RbST and Other Varables on Profts Per Cow, Maxmum Lkelhood Estmaton (Coeffcent Estmate Dvded by Standard Error n Parentheses). Varable 1994 1995 1996 1997 Intercept -188-174 -996-1121 (-0.39) (-0.35) (-2.02)* (-2.74)* rbst Impact ($) 19 49 81 63 (0.38) (1.10) (1.47) (1.20) Educaton 41 18 90 59 (1.13) (0.47) (2.00)* (1.36) Age -1.66-2.27 1.98-0.24 (-0.99) (-1.16) (0.87) (-0.26) Mlk Prce 60 58 93 108 (1.68) (1.55) (2.85)* (3.70)* Number of Cows -0.04 0.05-0.04-0.10 (-0.26) (0.44) (-0.17) (-0.54) Sgma 320 331 379 347 (25.29)* (28.62)* (24.71)* (27.11)* Rho 0.18-0.92 0.01-0.11 (0.88) (-0.53) (0.43) (-0.58) Adopton Estmates Smlar to Table 3 -- Heckman Two-Step Estmate of rbst -- rbst Impact ($) -130 83-85 -112 (-1.27) (0.73) (-0.59) (-1.02) Lambda 120-26 116 121 (1.68) (-0.37) (1.27) (1.71) Addtonal Varables Number of Observatons Smlar to above estmates 324 329 307 280 * Indcates statstcally dfferent from zero at probablty = 0.05. 121
Table 6: Estmated Impact of RbST and Other Varables on Herd Producton Per Cow, Maxmum Lkelhood Estmaton (Coeffcent Estmate Dvded by Standard Error n Parentheses). Varable 1994 1995 1996 1997 Intercept 21,320 16,110 16,634 13,560 (7.30)* (5.40)* (4.70)* (5.11)* rbst Impact 2,668 2,953 3,466 2,973 (7.89)* (8.75)* (9.33)* (7.58)* Educaton 1,009 644 1,021 966 (3.64)* (2.21)* (3.19)* (3.02)* Age -18 13 0.33 3 (-1.42) (0.89) (0.02) (1.39) Mlk Prce -261-19 -90 175 (-1.21) (-0.08) (-0.38) (0.91) Number of Cows 2.12 2.65 3.09 3.74 (3.93)* (5.19)* (6.01)* (3.67)* Sgma 2,334 2,525 2,699 2,603 (27.28)* (26.74)* (22.07)* (23.61)* Rho 0.007 0.19 0.27 0.03 (0.04) (1.08) (1.41) (0.20) Adopton Estmates Smlar to table 3 Number of Observatons * Indcates statstcally dfferent from zero at probablty = 0.05. Conclusons 324 329 307 280 Data from New York dary farms for the years 1994 through 1997 were used to determne f rbst generates profts for adopters. RbST has been commercally avalable snce 1994 and approxmately half of these farms used rbst durng that tme. On average, farms that are usng rbst are not proftng from ts use, although they are experencng an output response. Ths study supports earler results found by Tauer and Knoblauch (1997), and Stefandes and Tauer (1999) usng earler years of ths data set. Why do New York dary farmers use rbst when the proft mpact appears to be non-exstent? Although statstcally zero, the mpact estmates are generally numercally postve, and often 122
approach $100 per cow. These results are for the treatment average. It may be that there are farms proftng from the use of rbst. The mplcaton, then, s that some farmers may be losng money usng rbst. But snce the output mpact of rbst s unambguously postve, t may be dffcult for ndvdual farmers to determne f that output mpact s translatng nto a proft. Should a dary farm use rbst? Although a strongly statstcal postve proft response for rbst was not estmated, the estmated response was often numercally postve, suggestng a proft potental. Farmers must carefully assess ther ndvdual decsons. Producton and fnancal controls are necessary to ascertan the mpact of rbst and other changes and decsons they make n ther farm busnesses. References Barham, B.L., Jackson-Smth, D., and Moon, S. (2000). The adopton of rbst on Wsconsn dary farms. AgBoForum, 3(2&3), 181-187. Avalable on the World Wde Web: http://www.agboforum.org. Butler, L.J. (1992). Economc evaluaton of bst for on-farm use. In M.C. Hallberg (Ed.), Bovne Somatotropn and Emergng Issues: An Assessment, (pp. 145-171). Boulder, CO: Westvew Press. Davdson, J. and MacKnnon, J. (1993). Estmaton and nference n econometrcs. New York: Oxford Press. Econometrc Software, Inc. (1998). LIMDEP, Verson 7.0. [Econometrc computer Software]. Planvew, NY: Author. Greene, W. (1997). Econometrc analyss. Englewood Clffs, NJ: Prentce Hall. Knoblauch, W.A. and Putnam, L.D. (1998). Dary Farm Management Busness Summary, New York State, 1997 (Research Bulletn No. 98-06). Ithaca, NY: Department of Agrcultural, Manageral, and Resource Economcs, Cornell Unversty. Maron, B.W. and Wlls, R.L. (1990). A prospectve assessment of the mpacts of bovne Somatotropn: A case study of Wsconsn. Amercan Journal of Agrcultural Economcs 72, 326-336. Ott, S.L. and Rendleman, C.M. (2000). Economc mpacts assocated wth bovne Somatotropn (BST) use based on survey of US dary herds. AgBoForum, 3, 173-180. Avalable on the World Wde Web: http://www.agboforum.org. Stefandes, Z. and Tauer, L.W. (1999). The emprcal mpact of bovne Somatotropn on a group of New York dary farms. Amercan Journal of Agrcultural Economcs 81, 95-102. Tauer, L.W. and Knoblauch, W.A. (1997). The emprcal mpact of bst on New York dary farms. Journal of Dary Scence 80, 1092-1097. 123