Rates of Return to Public Agricultural Research in the Presence of. Research Spillovers. Jet Yee, Economic Research Service

Transcription

1 Rates of Return to Publc Agrcultural Research n the Presence of Research Spllovers by Jet Yee, Economc Research Servce and Wallace Huffman, Iowa State Unversty Selected Paper for Presentaton at the Amercan Agrcultural Economcs Assocaton Meetngs Chcago, Illnos August 5-8, 2001 Abstract Ths study uses new state data to examne the contrbutons of publc agrcultural research, extenson, and nfrastructure to agrcultural productvty. The estmated socal rates of return (whch take nto account spllover effects) are hgh and mply a need for federal or regonal nsttutons to coordnate publc agrcultural research fundng. Address for Correspondence: Jet Yee Economc Research Servce 1800 M Street, NW Washngton, DC Phone: (202) E-mal: jyee@ers.usda.gov The vews expressed are of the authors and do not necessarly represent the polces or vews of USDA.

2 The level of knowledge n any one state or ndustry depends not only on ts own research nvestment, but s also affected by the knowledge borrowed from other states or ndustres. That s, the productvty of the agrcultural sector n a state depends also on the research nvestments of other ndustres and states. For example, a new varety of corn developed n Iowa may be avalable to the farmers n Illnos. Smlarly, U.S. agrcultural productvty may beneft from spllovers of technology developed by foregn research expendtures. A few studes (e.g., Evenson, 1989) have found cross-state spllovers from agrcultural research to be sgnfcant. Pror efforts to explan the sources of state agrcultural productvty growth have been undertaken by Huffman and Evenson (1993) and Alston, Crag, and Pardey (1998). Huffman and Evenson (1993) constructed a state level productvty data set for 42 states for the perod They used publc and prvate research stocks and agrcultural extenson stocks to explan TFP. They used publc agrcultural research expendtures focused on dscoveres to enhance and mantan agrcultural productvty to create one publc agrcultural research stock varable for each state that contaned both a state s own contrbuton and spll-n contrbutons from adjacent states. The mpacts of publc agrcultural research on agrcultural productvty were generally postve. Alston, Crag, and Pardey (1998) constructed another state level productvty data set for 48 states, whch they used to examne the mpact of a sngle combned publc agrcultural research and extenson varable on TFP. They used essentally all of the publc agrcultural research expendtures of state agrcultural experment states to construct the research and extenson stock varable, rrespectve of whether the research was focused on producton agrculture. Interstate spllovers were not consdered. They found postve effects of the

3 2 combned publc agrcultural research and extenson varable on agrcultural productvty. Our problem n ths paper s to measure the spllover effects of the research performed n other geographcal areas on a state s agrcultural productvty. Usng newly-constructed data on agrcultural productvty (TFP), producton-orented publc agrcultural research expendtures, producton-orented agrcultural extenson expendtures, and other varables at the state level, we estmate the separate contrbuton of a state s own publc agrcultural research and spll-ns from agrcultural research n adjacent states to agrcultural productvty. Our productvty model also takes nto account publc agrcultural extenson, nfrastructure n hghways, and weather. In addton, we examne the nteracton effects between a state s own publc agrcultural research and agrcultural extenson. Our results add new nformaton about the relatve mportance of a state s own publc agrcultural research, research spllovers, extenson, nfrastructure, and weather and the nteracton effects between publc agrcultural research and extenson. Havng a better understandng of the sources of productvty growth and the rates of return to publc resources nvested n research, extenson, and nfrastructure s mportant to good publc polcy decson makng. The Model In prevous econometrc studes of agrcultural productvty, publc and prvate research, extenson, and farmers schoolng have been dentfed as the prmary sources of productvty growth. Our productvty model also allows for splln of research results from other states. Instead of poolng all 48 contguous states for the perod , we dvde the states nto several regons. We estmate a pooled regresson for each regon. In our model, the level of

4 productvty n a state depends on the state s own research stock as well as the stock of research n the other states n the same regon. Assume a sngle, output aggregate producton functon wth dsemboded techncal change: 3 (1) Q = AF( K, L, M )( ownrd) δ 1 δ 2 δ 3 ( splln) ( ext) ( hway ) δ 4 where Q s all types of outputs aggregated nto one ndex of output, A s a constant, F( ) s a well-behaved functon, K s physcal captal nput, L s qualty adjusted labor nput, M s qualty adjusted materals nput, ownrd s a state's own publc research stock, splln s the publc research stock from other states, ext s publc extenson stock, and hway s hghway stock. (Snce the measure of labor nput we use accounts for the changng educatonal attanment of the farm workforce over tme, we do not explctly consder educaton as a determnant of output growth.) Now defne total factor productvty (P) as Q δ1 δ 2 δ 3 δ 4 ( 2) P = A( ownrd) ( splln) ( ext) ( hway). F( K, L, M ) Takng natural logarthms of both sdes of equaton (2) and addng an nteracton between ownrd and ext, dummy varables for extreme weather condtons, and a random dsturbance term, we obtan the followng model that s lnear n the unknown parameters (δ j 's): (3) ln P = ln A + δ ln( ownrd) + δ 1 2 ln( splln) + δ + δ drought + δ 6 3 ln( ext) + δ 7 4 flood + u. ln( hway) + δ 5 ln( ownrd) ln( ext) Wth the nteracton term ncluded, the output elastcty of ownrd s now δ 1 + δ 5 ln(ext) and of ext s δ 3 + δ 5 ln(ownrd). Ths specfcaton of the research-extenson nteracton ncorporates the

5 4 hypothess that stronger two-way flows of nformaton occur between publc agrcultural research and extenson conducted wthn a state than between spll-n research from other states and local extenson. In partcular, f farmers producton problems get transmtted to researchers through publc extenson personnel, we would expect the nteracton effect to be postve. If publc agrcultural research and extenson wthn a state functon ndependently, we would expect the coeffcent of the nteracton term to be nsgnfcant. If publc research and extenson effectvely are substtutes n affectng agrcultural productvty, we would expect δ 5 to be negatve. Snce we estmate equaton (3) for each of seven regons, we also nclude state dummy varables n equaton (3) to obtan our fnal econometrc model. Gven estmates of the parameters n equaton (3), own-state (prvate) and socal (ncludng nter-state spllovers) mpacts of publc agrcultural research on agrcultural productvty/output can be evaluated. The numbers can then be used to compute rate of return estmates to an ncrement n publc agrcultural research. For smplcty, frst assume no nterstate spllover effects of publc agrcultural research nvestment n one state on agrcultural productvty n other states occurs. The nternal rate of return (r) for an addtonal research expendture of ΔR t n perod t s the dscount rate whch results n the followng equalty: ( 4) or R t = Q m t+ = 0 (1 + r) ( 5) 1 = Q 1 Q T m m t+ t+ t+ = = 0 Rt (1 + r) = 0 Tt + Rt (1 + r) 1 where Q s a state s output and T s ts own publc research stock (ownrd). Snce the output

6 5 elastcty of ownrd s δ 1 + δ 5 ln(ext), Q t+ ( 6) = 1 δ 5 Tt + and [ δ + ln( ext )] t+ Q T t+ t+ ( 7) Tt + R t = w where the w s are the weghts used to construct the research stock n equaton (3). Hence, we can rewrte equaton (5) as m m Q 1 Q 1 [ δ + δ ln( ext) ] w = [ δ + ln( ext) ] w ( 8) 1 = δ 5 = 0 T (1 + r) T = 0 (1 + r) where we have replaced ln(ext t+ ), Q t+, and T t+ wth ther respectve sample mean values. Wth spllovers, equaton (8) must be modfed. An addtonal research expendture of ΔR,t by state n perod t has two mpacts. Frst, t affects state s own research stock wth a drect mpact on state s output. Second, t affects the research spllover stock avalable to all other states n the same regon as state wth ndrect mpacts on those other states outputs. By a smlar dervaton, we obtan m Q Q 1 [ δ + δ ln( ext) ] + ( n 1) w ( 9) 1 = 1 5 δ 2 T S = 0 (1 + r) where S s the average research stock from other states n the same regon (splln) and n s the number of states n the regon. Wth postve nterstate spllovers effects of publc agrcultural research on productvty of other states and other thngs equal, we see that the nternal rate of return, r, must ncrease from equaton (9) relatve to equaton (8). That s, the socal (accountng

7 6 for mpacts n all states) rate of return from the nvestment n publc agrcultural research by one state must be greater than the prvate rate of return, or return as any one state sees t. Furthermore n equaton (9), the nternal rate of return must ncrease as n, the number of states n the regon, ncreases, other thngs equal. Ths s just the regonal publc good attrbute playng tself out. When Q, T, and S are n real terms, the r s computed are margnal real nternal rates of return. Data Sources and Varable Constructon Consderable effort has gone nto the constructon of the new data used n ths study. We expect the level of productvty n a state to depend on the state s own research stock as well as the stock of research n some subset of other states,.e., agrcultural research dscoveres are mpure publc goods (Huffman and Just 1999; Huffman and Evenson 1993). Much of agrcultural producton, especally crop producton, s strongly affected by geoclmatc condtons. The geoclmate determnes day length, precptaton amounts and patterns, and sol types. The movement of plant dseases and nsects also has a strong spatal component. These are some reasons why research spllovers mght be expected to have a strong spatal component. Much of non-grazng lvestock producton s, however, less affected by geoclmate, and frut and vegetable producton tend to occur n rather unque local envronments whch mght be located long dstances apart. These are some reasons why research spllovers mght have a spece component that s transferred over long dstances. Several groupngs of states were consdered n our work. Frst, the old Economc Research Servce (ERS) farm producton regons date back to the 1950s. It conssts of the 10

8 7 regons where states wthn a regon are contguous. The regons are: Northeast, Lake States, Corn Belt, Northern Plans, Appalachan, Southeast, Delta, Southern Plans, Mountan, and Pacfc. Second, a Natonal Research Councl (NRC) study on colleges of agrculture at the land grant unverstes (Commttee on the Future of the Colleges of Agrculture n the Land Grant Unversty System 1995) conducted a cluster analyss to classfy state agrcultural experment statons (SAES s) expendtures nto 9 commodty research clusters. 1 In ths groupng, some clusters consst of contguous states and others do not. Thrd, McCunn and Huffman (2000) group the 48 states nto seven regons, each contanng contguous states. Ther groupng bulds on earler work by Khanna, Huffman, and Sandler (1994). The McCunn and Huffman (MH) regonal groupng worked well for a study of convergence n state agrcultural TFP growth rates, The earler Khanna, Huffman, and Sandler paper examned state government decsons on fundng state agrcultural experment statons n a model of mpure publc good provson. The NRC regons dffer from the ERS and MH regons n not havng states whch are contguous. For example, Calforna, Oregon, and Washngton are n the same ERS regon (Pacfc), whle Calforna and Florda are n the same NRC regon. The ERS and MH groupng are more smlar to each other than to the NRC groupng. In fact, the Pacfc, Mountan, Northern Plans, and Northeast regons are the same for both the ERS and MH groupngs. 2 Based upon some prelmnary expermentaton, the MH groupng of states nto regons gave the best emprcal results, and t s the one used for creatng research spllovers n ths paper. Data on total factor productvty (TFP) by state are new ERS estmates prepared by Ball and Nehrng (ERS Staff Paper Number 9804). 3 The TFP numbers for each state are spatally adjusted so that they are comparable across states.

9 8 Publc agrcultural research expendtures were compled by Huffman, McCunn, and Xu (2001), after makng some mprovements n the earler Huffman and Evenson (1993; 1994) approach. The Huffman and Evenson research varables stopped n 1982, and the new seres s extended to 1995, uses refned methods, ncorporates new hstorcal data on non-usda federal expendture on SAES research , and bulds on actual annual SAES expendture/recepts over the perod. Publc agrcultural research sector n each state conssts of the state nsttutons--state agrcultural experment statons and schools/colleges of veternary medcne--and the USDA agences of the Agrcultural Research Servce and Economc Research Servce. 4 Usng the USDA s Current Research System (CRIS; see U.S. Dept. Agr. 1993) classfcaton of research, agrcultural research expendtures to enhance and mantan agrcultural productvty are taken to be a subset of all CRIS research commodty categores (see USDA 1993). They are research expendtures on all farm commodtes (e.g., corn, wheat, fruts, vegetables, beef, swne, dary) n CRIS research problem areas (RPA s) that have a focus on enhancng or mantanng productvty (see Huffman and Evenson 1994, p ), plus all expendtures on the research commodty categores of structures and nputs, farm management, nsects, and general scence. Although there s undoubtedly some reportng errors by publc scentsts n ther CRIS reports, we do not consder these errors to be serous n major aggregates. 5 The annual nomnal agrcultural research expendtures by state are converted to real (1984 = 1.00) expendtures usng Huffman and Evenson s agrcultural research prce ndex (Huffman and Evenson 1993, Table 4.1 and p ). 6 Research expendtures n a gven year are expected to have an mpact on productvty for

10 9 many years. We can nclude as separate varables a fnte number of lagged research expendtures n a producton functon such as equaton (1). However, the large number of lagged research expendtures uses up a large number of degrees of freedom. Also, the lagged values of the research expendtures tend to be hghly correlated. Grlches (1979; 1998) emphaszes that near-multcollnearty problems make t mpossble to estmate research tmng weghts wthout some type of pror restrctons. What s usually done s to construct a research stock usng current and past research expendtures: (10) T t = n = 0 w R t. The scence of research stock varables constructon remans n ts nfancy (Grlches 1998, Ch. 12). The man reason s that the true research stock s unknown, so we need a good proxy or nstrumental varable for t (Greene 1997, pp ). In studes of the mpact of prvate research n manufacturng, the stock of research captal s frequently created from research expendtures usng the perpetual nventory method. Grlches (1998) concludes the usual declnng balance or geometrc deprecaton doesn t ft very well the lkely gestaton, blossomng, and eventual obsolescence of knowledge. In agrculture, where none of the publc agrcultural research s conducted on farms or under the drect control of farmers, the perpetual nventory approach s mplausble (see Evenson 2000). Huffman and Evenson (1993; 1994) use a trapezod-shaped tmng weghts after a short gestaton perod to approxmate an agrcultural research stock. There are 7 years of rsng weghts, followed by 6 years of constant weghts, and then 20 years of declnng weghts; and the summaton of the weghts s one. Here we adopt the Huffman and Evenson trapezodal tmng weght pattern to derve the proxy varable for the

11 10 publc agrcultural research stock. It would be useful at ths pont to make a few comments concernng technologcal spllovers. The level of knowledge n any one sector or ndustry depends not only on ts own research nvestment but s also affected by the knowledge borrowed or spllng n from other ndustres. That s, the productvty of the agrcultural sector depends also on the research nvestments of other ndustres. The problem s to measure the spllover effects of the research performed by other ndustres on agrcultural productvty. There are several ways to proceed. Let T j be the stock of techncal knowledge n ndustry j. We can treat the T j 's as separate varables n a producton functon. However ths s not feasble due to degrees of freedom and perhaps multcollnearty problems. We thus need to aggregate (smlar to the research tme lag problem). A measure of borrowed knowledge by the agrcultural sector from other ndustres s (11) T a = j w T j j where T a s the amount of knowledge borrowed by the agrcultural sector from all other sectors. The weghts (w j 's) dffer by sector and reflect the fact that not all knowledge are equally useful to the agrcultural sector. W j becomes smaller as the technologcal dstance between the agrcultural sector and sector j ncreases. Measures of technologcal dstance could be based on nter-sectoral purchases under the assumpton that borrowed research s emboded n purchased nputs or could be based on research n common scentfc felds (e.g., Jaffee 1986). In addton to technologcal spllovers from prvate research, there are also geographcal spllovers from publc agrcultural research performed by one state to other states. For example, some of the publc agrcultural research dscoveres n Iowa may spllover to some or all of the

12 11 surroundng states,.e., splls n, or Iowa benefts from some or all the publc agrcultural research conducted n surroundng states. Smlarly, U.S. agrcultural productvty may beneft from spllovers of technology developed by foregn research expendtures. The technques used to treat spllovers from prvate research can also be used to treat geographcal spllovers. The weghts (w j 's) n equaton (11) reflect the fact that applcaton of research results from another state or country s lmted by the locaton-specfc nature of much agrcultural research. We use two publc research stock varables n our work. One s a state s own stock and the second s a splln/spllover stock. Consstent wth our regonal groupng of states, we assume that dscoveres from publc agrcultural research n a gven state are an mpure publc good (Cornes and Sandler 1996; Khanna, Huffman, and Sandler 1994). In partcular, we mpose the smplfyng assumpton that benefts are regonally confned and apply a smple aggregaton technology for the publcness of agrcultural research. That s, we assume the weghts (w j 's) are the same for all states n a gven regon and equal to zero for states outsde the regon. For a gven state n a regon, the spllover (or splln) stock s defned as the total publc agrcultural research stock of all states n the regon less the state s own publc agrcultural research stock. Spllovers from prvate research s also a possblty. If an mproved product s used as an nput n the producton of another product, ts contrbuton wll show up n the productvty measure of the ndustry that purchased t. So t s possble for research performed n the agrcultural chemcal ndustry to have no mpact on ts own productvty, but have an mpact on agrcultural productvty as tradtonally measured. We, however, do not drectly consder prvate research n ths study. The omsson s due to several factors. Only natonal level data on prvate agrcultural research expendtures exst,

13 12 e.g., see Huffman and Evenson 1993, Klotz, Fugle, and Pray Although Huffman and Evenson (2000, 1993) have appled a weghtng system, whch ncorporates mpure publc good attrbutes of dscoveres, to approprate potental mpacts to the ndvdual states, ther data set has not been revsed or extended. Hence, state level data on prvate research are dffcult to obtan or construct. Fnally, prvate agrcultural research s performed because frms hope to approprate most of the benefts of ther research by ncreasng the prces of ther products. These prce ncreases should be reflected n qualty changes n qualty-adjusted prce and quantty ndexes. If the prvate frms were able to capture all the benefts of ther dscoveres and the qualty adjustments were perfect, there would be no need for a prvate research stock varable. However, prvate frms cannot expect to capture all the benefts from ther nnovatons. In the case of Bt-cotton, Falck-Zepeda, Traxler, and Nelson (2000) show that the nput supply ndustry captured a lttle less than one-half of the total socal benefts n Huffman and Evenson (1993), also, showed sgnfcant effects of prvate research on state agrcultural productvty. Thus, we are undoubtedly mssng some prvate research effects on state agrcultural productvty. There are really two types of technologcal spllovers that should be dstngushed. The frst type nvolves mproved nputs purchased from another ndustry at less than ts qualtyadjusted prce. These are really not pure spllovers, but are the result of measurement problems we just alluded to. The second type nvolves knowledge borrowed from ndustry j whch s used to enhance (make more productve) the research efforts of ndustry. Ths borrowng of knowledge may have very lttle relaton to nput purchases. For example, two ndustres may buy very lttle from each other, but are workng on smlar projects and hence can beneft from each

14 13 other s research. Our agrcultural extenson varable s constructed from estmates of producton-orented agrcultural extenson expendtures n each state per farm. Over , extenson was organzed nto four program areas: agrculture and natural resources (ANR), communty resource development (CRD), 4-H youth (4-H), and home economcs (HE). The ANR program area ncludes crop producton and management, lvestock producton and management, farm busness management, agrcultural marketng and supply, and natural resources. We collected data on professonal extenson full-tme equvalents (FTE s) by state and major program areas from annual State Accomplshment Reports. To obtan an estmate of agrcultural producton related extenson expendtures for a state, we multpled the state s total expendtures on extenson by the share of ANR full-tme equvalent staff to total professonal staff n all areas. An extenson stock measure s obtaned by applyng the perpetual nventory method and a 50 percent deprecaton rate. Infrastructure refers to federal hghways n ths study. Data are avalable for on captal stock from captal outlay and captal stock from mantenance (both n 1987 dollars) from the U.S. Department of Transportaton, State Transportaton Economc Dvson. 7 The Federal Hghway Admnstraton's composte prce ndex was used to deflate the captal expendture and mantenance outlay seres to 1987 dollars. In ths data set, the standard perpetual nventory technque was used to generate the hghway captal stock from expendtures data. Followng Eberts, Park, and Dalenberg (1986) dscards are assumed to follow a truncated normal dstrbuton, wth the truncaton occurrng at one half the average lfe and one and one half tmes the average lfe. We extended the hghway captal seres to Our hghway

15 14 captal stock measure s total hghway captal stock, whch equals captal stock from captal outlays plus captal stock from mantenance. 8 Extreme weather condtons (droughts and floods) affect agrcultural productvty, and f they are controlled for, we may get better estmates of the mpacts of other sources of productvty. 9 We employed the USDA s precptaton data weghted by harvested crop acreage to create a varable (pre-plant) equal to cumulatve February to July ranfall. We then created a drought dummy varable (drought) equal to 1 f pre-plant s less than 1 standard devaton below normal (and 0 otherwse) and a flood dummy varable (flood) equal to 1 f pre-plant s more than 1 standard devaton above normal (and 0 otherwse). The Emprcal Results We dscuss n ths secton econometrc estmates of the total factor productvty (TFP) equaton for each of the 7 regons, and estmates of the margnal nternal rate of return to nvestments n publc agrcultural research. The TFP equatons Econometrc estmates of the TFP equaton (3) are estmated for each regon, wth and wthout an nteracton varable between the stock of own publc research and stock of extenson. The equatons were frst estmated by ordnary least squares wth state dummy varables ncluded. Where frst-order autocorrelaton was a sgnfcant problem (as judged by the Durbn- Watson statstc), the productvty equaton was re- ftted as a tme seres model wth an AR(1) process wth ρ allowed to dffer across states. The Southern Plans was the only regon whch dd not requre a frst-order autocorrelaton correcton. The estmaton procedure also took nto

16 15 account wthn state heteroscedastcty as well as contemporaneous correlaton of dsturbances across states n the same regon. (Tables contanng the estmated results can be obtaned by sendng an e-mal to jyee@ers.usda.gov.) For the Northeast regon, all the estmated coeffcents of the publc polcy varables are sgnfcantly dfferent from zero at the 5 percent level when the research-extenson nteracton varable s ncluded, and the estmated coeffcents of splln research and hghways are postve. When the research-extenson nteracton varable s excluded, extenson performs poorly. Thus, the TFP model wth the research-extenson nteracton varable s judged to be the best. The estmated coeffcent of the research-extenson nteracton varable s, however, negatve, suggestng own research and extenson are substtutes for mpactng agrcultural productvty n the Northeast regon. Also, as expected, drought and flood have statstcally sgnfcant negatve mpacts on agrcultural productvty. For the Southeast regon, the TFP equaton ncludng the research-extenson nteracton varable performs poorly n the sense that the coeffcents of the publc polcy varables are not sgnfcantly dfferent from zero at the 5 percent level, except for splln research. When the research-extenson nteracton term s excluded, the coeffcents of own research, splln research, and extenson are postve and each sgnfcantly dfferent from zero at the 5 percent level. We conclude that ths s the best specfcaton for the Southeast regon. However, hghways have a postve but not statstcally sgnfcant effect on agrcultural productvty n the Southeast regon. Also, drought has a statstcally sgnfcant negatve mpact on productvty here. In the Central regon, when the own-research and extenson nteracton varable s excluded from the TFP equaton, the estmated coeffcents of own research and hghways are not

17 16 sgnfcantly dfferent from zero at the 5 percent level. However, when the research-extenson nteracton varable s ncluded, all of the publc polcy varables have coeffcents that are sgnfcantly dfferent from zero and the estmated coeffcents for splln research and hghways are postve. Furthermore, the estmated coeffcent of the nteracton varable s postve, suggestng that own research and extenson are complements for mpactng agrcultural productvty n the Central regon. In ths regon whch has very lttle rrgaton, both drought and flood varables have statstcally sgnfcant negatve mpacts on agrcultural productvty. In the Northern Plans regon, the own research and extenson varables perform poorly when no own research-extenson nteracton varable s ncluded. The estmated coeffcents of own research and extenson are not sgnfcantly dfferent from zero. However, when the research-extenson nteracton varable s ncluded, all the estmated coeffcents of the publc polcy varables are sgnfcantly dfferent from zero at the 5 percent level. The nteracton model s the preferred model for the Northern Plans regon. Furthermore, own research and extenson are complements n the Northern Plans regon. Drought has a statstcally sgnfcant negatve mpact on productvty here. In the Southern Plans regon, all the estmated coeffcents of the publc polcy varables are postve and sgnfcantly dfferent from zero when no research-extenson nteracton varable s ncluded n the TFP equaton. When the nteracton varable s ncluded, ts estmated coeffcent s negatve but not sgnfcantly dfferent from zero at the 5 percent level. Hence, we conclude that n the Southern Plans regon, the TFP regresson wthout the research-extenson varable s best. The mpacts of droughts and floods on TFP are negatve, but drought s statstcally stronger and more frequently occurrng n ths regon.

18 17 In the Mountan regon, the estmated coeffcent of own research s not sgnfcantly dfferent from zero and that of extenson s negatve (and sgnfcant) when no research-extenson nteracton varable s ncluded. However, when the nteracton varable s ncluded, the estmated coeffcents of all research and extenson varables are sgnfcantly dfferent from zero, and the coeffcent for own research s postve. For the Mountan regon, we conclude that the TFP equaton ncludng the research-extenson nteracton varables s best. However, the estmated coeffcent for the research-extenson nteracton varable s negatve, suggestng that own research and extenson are substtutes. Here, hghways do not have a statstcally sgnfcant effect on agrcultural TFP, and drought has a statstcally sgnfcant negatve mpact on agrcultural productvty. For the Pacfc regon, nether TFP model performs well. In partcular, extenson does not have a statstcally sgnfcant mpact n ether equaton. Extenson appears to be a dfferent actvty n ths regon than n the other sx regons. Hence, among the two equatons reported, the one wthout the research-extenson nteracton varable s best. In ths equaton, the estmated coeffcents of own research and splln research are postve, but the coeffcent for own research s only sgnfcant at the 7 percent level. Hghways have a postve, but not statstcally sgnfcant mpact on agrcultural TFP n the Pacfc regon. Wth ths regon beng one of normally low precptaton and heavy relance on rrgaton, the negatve but not sgnfcant effects of the drought and flood varables s not too surprsng. Next, we summarze the TFP elastctes assocated wth the publc polcy varables (own research, splln research, extenson, and hghways) and report them n table 1. In the models wthout the research-extenson nteracton varable, the TFP (or output) elastctes are the

19 18 estmated coeffcents. In the models contanng nteracton varables, the TFP (output) elastcty of ownrd s δ 1 + δ 5 ln(ext) and of ext s δ 3 + δ 5 ln(ownrd), and these elastctes are evaluated by regon at the respectve regonal sample mean values. For the non-nteracton model, TFP elastctes for ownrd are postve for all regons, except for the Mountan regon (where the estmated coeffcent was not sgnfcant). The TFP elastcty of publc research splln s postve for all regons. Perhaps surprsng, the TFP elastcty wth respect to publc research splln s greater than for ownrd, except for the Northern Plans. The elastcty of TFP wth respect to hway s postve for all regons. Ext, however, has a negatve TFP elastcty for three regons: the Northern Plans, Mountan, and Pacfc regons. In the TFP models ncludng the research-extenson nteracton varable, the TFP elastcty of ownrd s postve for all regons, except for the Northern Plans. The TFP elastcty of splln s postve for all regons. In general, the nteracton model, when evaluated at the regonal sample means, ncreases the TFP elastctes for ownrd and decreases the TFP elastctes for splln, compared to the non-nteracton model. Splln, however, stll has a hgher TFP elastcty than ownrd. Hway has a postve TFP elastcty for all regons. Ext has a negatve TFP elastcty for two regons (Northern Plans and Pacfc). Snce the coeffcent of the nteracton term s postve (and sgnfcant) for the Northern Plans regon, ths mples that a greater level of nvestment n a state s own research and extenson efforts s needed to make the TFP elastctes postve. As we have noted, the TFP model wth the own research and extenson nteracton varable s preferred to the non-nteracton model for most of the regons. The socal rate of return to publc agrcultural research The prvate and socal real nternal rate of return s computed for each of the seven

20 19 regons. For each regon, we compute the sample mean of Q, T, S, and ln(ext) to obtan values for a representatve state n the regon and for a representatve spllover effect from other states wthn the regon. Table 2 presents the calculated rates of return for each of the 7 regons for both the non-nteracton and nteracton models. The margnal prvate real rate of return s obtaned usng equaton (8) [or equaton (9) wth δ 2 set to zero]. The margnal socal real rate of return takes nto account own-state and spllover effects and s obtaned usng equaton (9). As expected, for all regons the margnal socal real rate of return s always greater than the own-state/prvate real rate of return. The ownstate/prvate rates of return are generally szeable but less than 100 percent and wthn the range found by other researchers for the U.S. (see Evenson 2000). The socal rates of return are several tmes the prvate rates of return. Some of the socal rates of return are surprsngly hgh, e.g., for the Northeast and Central regons the real socal rate of return exceed 600 percent. Regons havng a large number of states and large farm output per state have the largest socal rate of return, whch s as expected wth a regonal publc good (.e., summaton technology for publcness wthn the regon). Concludng Remarks Ths paper uses new data to provde new evdence on the contrbuton of publc own research, splln research, extenson, and nfrastructure n hghways to agrcultural productvty change over 1960 to The results are a sgnfcant addton to the lterature because few studes have focused on agrcultural productvty at the state level, and none has consdered the effects of nfrastructure or the drect effect of publc agrcultural research spllovers. Where our results overlap wth pror studes, the results are largely as expected. Publc

21 20 agrcultural research and hghways have postve mpacts on agrcultural productvty, and the margnal real socal rate of return to publc agrcultural research s large. The results for publc agrcultural extenson are mxed, but ths s consstent wth Huffman and Evenson (1993), and extenson does not have a postve mpact on agrcultural productvty n two regons (n the nteracton model). The model contanng an own-research-and-extenson nteracton varable was preferred to the non-nteracton model for most regons. Splln research stocks also mpact agrcultural productvty postvely n almost all regons. Furthermore, the computed real rates of return to nvestments n publc agrcultural research to any one state s less than the socal rate of return to all states n ts regon. Thus, state level plannng for publc agrcultural research would be socally suboptmal. Ths outcome s consstent wth the regonal publc good nature of publc agrcultural research, and t mples a need for a regonal nsttuton to coordnate publc agrculture research fundng and provson. Furthermore, our estmates of the margnal socal rate of return s large by comparson wth other studes reported n Evenson (2000), but ths seems to be due largely to the drect ncorporaton of research splln n our model of state agrcultural productvty. In ths study, we have not ncluded prvate research varables. If good prvate research varables were avalable, t mght alter the story somewhat. Ths s left to future work.

22 21 Table 1. TFP or Output Elastctes wth respect to Publc Polcy Varables (nteracton terms evaluated at the respectve regonal means) Non-nteracton Term Model Regon Own R&D Splln R&D Extenson Hghway Northeast Southeast* Central Northern Plans Southern Plans* Mountan Pacfc* Interacton Term Model Regon Own R&D Splln R&D Extenson Hghway Northeast* Southeast Central* Northern Plans* Southern Plans Mountan* Pacfc Notes: The preferred model s denoted by (*). In the models wthout the research-extenson nteracton varable, the TFP/output elastctes are the estmated coeffcents. In the models contanng nteracton varables, the TFP/output elastcty of ownrd s δ 1 + δ 5 ln(ext) and of ext s δ 3 + δ 5 ln(ownrd) evaluated at the regonal sample means. Northeast - ct, de, ma, md, me, nh, nj, ny, pa, r, vt Southeast - al, fl, ga, ky, nc, sc, tn, va, wv Central - n, l, a, m, mo, mn, oh, w Northern Plans - ks, ne, nd, sd Southern Plans - ar, la, ms, ok, tx Mountan - az, co, d, mt, nv, nm, ut, wy Pacfc - ca, or, wa

23 22 Table 2. Own-State/Prvate and Socal Real Rate of Return (r) to Publc Agrcultural Research n a State (percent) Interacton Non-nteracton Number of States Regon Prvate Socal Prvate Socal n the regon Northeast 106 > 600* 92 > Southeast * 9 Central 89 > 600* 12 > Northern Plans * 510 > Southern Plans * 5 Mountan * Pacfc 127 > 600 > 600 > 600* 3 Notes: The preferred model s denoted by (*). See notes for Table 1. In the Northern Plans and Mountan regons, when the estmated output elastcty was negatve, the output elastcty s set to zero n the rate of return computaton.

24 23 References Alston, J., B. Crag, and P. Pardey. Dynamcs n the Creaton and Deprecaton of Knowledge, and the Returns to Research. Internatonal Food Polcy Research Insttute, EPTD Dscusson Paper No. 35, Aug Ball, Eldon and Rchard Nehrng Patterns of State Productvty Growth n the U.S. Farm Sector. Economc Research Servce Staff Paper Number Cornes, R. and T. Sandler The Theory of Externaltes, Publc Goods and Club Goods. New York, NY: Cambrdge Unversty Press. Commttee on the Future of the Colleges of Agrculture n the Land Grant Unversty System, Natonal Research Councl Colleges of Agrculture at the Land Grant Unverstes: A Profle. Washngton, D.C.: Natonal Academy Press. Eberts, Randall, Chul Soo Park, and Douglas Dalenberg Publc Infrastructure Data Development. Natonal Scence Foundaton. Evenson, Robert E Spllover Benefts of Agrcultural Research: Evdence from U.S. Experence, Amercan Journal of Agrcultural Economcs 71(2), pp Evenson, R. E Economc Impact Studes of Agrcultural Research and Extenson. Chapter n Handbook of Agrcultural Economcs, Vol. I, B.L. Gardner and G. Rausser, Eds., Amsterdam, Netherlands: Elsever Scence/North-Holland. Falck-Zepeda, J. B., Greg Traxler, and R. G. Nelson Surplus Dstrbuton from the Introducton of a Botechnology Innovaton, Amercan Journal of Agrcultural Economcs, pp Greene, W.H Econometrc Analyss. 3rd ed. Upper Saddle Rver, N.J.: Prentce Hall, Inc. Grlches, Z., "Issues n Assessng the Contrbuton of Research and Development to Productvty Growth," Bell Journal of Economcs, 10, pp Grlches, Z R&D and Productvty: The Econometrc Evdence. Chcago, IL: The Unversty of Chcago Press. Huffman, Wallace E. and Robert E. Evenson Scence for Agrculture, Iowa State Unversty Press, Ames.

25 Huffman, Wallace E. and Robert E. Evenson The Development of U.S. Agrcultural Research and Educaton: An Economc Perspectve. Iowa State Unversty, Department of Economcs. 24 Huffman, Wallace E. and Robert E. Evenson Structural and Productvty Change n U.S. Agrculture, , Agrcultural Economcs, 24, pp Huffman, Wallace E. and Rchard E. Just Agrcultural Research: Benefts and Benefcares of Alternatve Fundng Mechansms, Revew of Agrcultural Economcs, 21, pp Huffman, W.E., A. McCunn, and J. Xu Publc Agrcultural Research wth an Agrcultural Productvty Emphass: Data for 48 States, Iowa State Unversty, Staff Paper. Jaffee, A., "Technologcal Opportunty and Spllovers of R and D: Evdence from Frms' Patents, Profts, and Market Value," Amercan Economc Revew, 76, pp Khanna, Jyot, Wallace Huffman, and Todd Sandler Agrcultural Research Expendtures n the Unted States: A Publc Goods Perspectve, Revew of Economcs and Statstcs, 76(2), pp Klotz, Cassandra, Keth Fugle, and Carl Pray Prvate-Sector Agrcultural Research Expendtures n the Unted States, , Washngton, D.C., ERS Staff Paper No McCunn, A. and W.E. Huffman Convergence n U.S. Productvty Growth for Agrculture: Implcatons of Interstate Research Spllovers for Fundng Agrcultural Research, Amercan Journal of Agrcultural Economcs, 82, pp Morrson, Catherne J. and Amy E. Schwartz State Infrastructure and Productve Performance, Amercan Economc Revew, 86(5), pp U.S. Dept. Agr., CSRS, CRIS Manual of Classfcaton of Agrcultural and Forestry Research. Revson V. Beltsvlle, MD: USDA-CSRS.

26 25 ENDNOTES 1. See Fgure 7.4 n the NRC report for the groupng of states by research cluster. 2. In addton, MH Central = ERS Corn Belt + ERS Lake States, MH Southern Plans = ERS Southern Plans + ERS Delta, and MH Southeast = ERS Southeast + ERS Appalachan. 3. The data n the publcaton are also avalable electroncally from the ERS homepage at URL: 4. The project nformaton reported to CRIS contans a locaton code for where the research s conducted. When an ARS or ERS project s conducted n Iowa for example, the funds would go nto the publc agrcultural research expendtures n Iowa. If the locaton s gven as Washngton, DC, for ERS the funds are not counted n our publc research expendtures data. Part but not all of the ARS research at Beltsvlle was ncluded n the publc research expendtures n Maryland. 5. The annual research expendture seres, , were derved from USDA-CRIS electronc data fles. For , the total publc agrcultural research expendtures seres were derved from (1) reported expendtures by the state agrcultural experment statons to the USDA and (2) a projected rato of total publc agrcultural research expendtures to total SAES expendtures by state. The projected ratos were obtaned by fttng smple tme seres models to the actual data by state, , and then projectng backward. A major adjustment was made to reflect the dramatc shft n the share of U.S. total publc agrcultural research between state nsttutons and the USDA (see Huffman and Evenson 1993, Table 4.1 and Fgure 4.1). 6. The Huffman and Evenson research prce ndex gves about 70 percent weght to faculty salares and behaves smlar to Alston and Pardey s and Fugle s agrcultural research prce ndex over The state hghway database s located at the web page: 8. Alca Munnell at Boston College also has a database on nfrastructure ncludng the components hghways, water systems, and others by state. Munnell s database has been employed n several studes of the manufacturng sector (e.g. Morrson and Schwartz). We, however, do not employ Munnell s database n our study prmarly because her database s only from 1970 to Weather data are avalable from the ERS homepage as an ERS data product: Weather n U.S. Agrculture. Monthly temperatures and precptaton data for farm producton

27 regons and States for (URL: Weather for certan months (e.g. February to July) may be more mportant than other months n explanng agrcultural productvty (Huffman and Evenson). 26