This study estimates the proportion of rice yield increase in University of Arkansas Division of Agriculture's (UofA) released rice cultivars that are attributable to genetic improvements through the University's breeding program. Test plot data from eight UofA experiment stations were used to quantify the yield increases and potential yield growth decreases over time. In addition to quantifying the yield and yield variance evolution at the UofA, this study also calculates the economic benefits of the UofA rice breeding program. Results indicated that by releasing modern rice cultivars, the UofA rice breeding program increased average producer yield by 0.68 bu/ac annually. During the last decade, 1997-2007, the average annual economic benefits were 34.3 million (2007) dollars. When accounting for the spillover of UofA rice varieties to neighboring states the average annual economic benefit of the breeding program increases to 46.7 million (2007) dollars.
Key Words: economic impact of technological change, Just and Pope, public rice breeding
JEL Classifications: O13, O32, Q16
Public research in rice breeding has resulted in higher yields for Arkansas rice producers over the past three decades. This study measures the genetic and economic impact of the University of Arkansas Division of Agriculture's (U of A) rice breeding program over the last 24 years (1983-2007). Yield increases for rice varieties released by the UofA breeding program were quantified, holding growing conditions, grain length, climatic conditions, and other agronomic improvements in production constant. The yield differential for each rice variety included in the UofA's annual Arkansas Rice Performance Trials (ARPT) was quantified to isolate the percentage of yield enhancement attributable solely to the genetic improvement at the rice breeding program at the University of Arkansas. The main objective of this study was to determine the proportion of the increases in yield of University of Arkansas-released rice cultivars attributable to genetic improvements. Test plot data from eight University run experiment stations were used to quantify the yield increases and potential yield growth decreases over time. Furthermore, this study set out to determine whether modern rice cultivars released by the UofA have influenced yield variability during the same 24-year period. Yield variation may have decreased due to genetic enhancements through improved tolerance to rice blast and other pathogens, resulting in higher levels of yield stability
In addition to quantifying the yield and yield variance evolution at the University of Arkansas' rice breeding program, the third objective of this study was to calculate the economic benefits of the UofA's rice breeding program for rice to local producers in Arkansas and the surrounding states that sow UofA varieties. This is important for formulating future policy decisions, given the recent trend of budget cuts at public institutions.
Measurements of the Benefits of the UofA Rice Breeding Program
The methodology used to calculate the economic benefits of the Arkansas rice breeding program followed an extensive literature on the economic impacts of agricultural research, as summarized by Huffman and Evenson (2006) and Alston, Norton, and Pardey (1995). Previous evaluations of breeding programs have been exemplified in studies by Nalley, Barkley, and Chumley (2008); Nalley et al. (2008); Brennan (1984, 1989a); Byerlee and Traxler (1995); and Barkley (1997). Brennan (1989b) goes further by evaluating the breeding programs at different stages in their lives.
The first step in evaluating the economic impact of the Arkansas rice breeding program was to measure the increase in yields from the genetic improvement of rice, holding all other production parameters constant. This was accomplished by applying the methodology of Feyerherm, Paulsen, and Sebaugh (1984) to calculate the relative yields for each variety with data from the ARPT for rice varieties. Use of relative yield performance data from test plots implicitly assumes that actual producer yields are equivalent to test plot yields in the UofA rice experiments. Although a gap between experimental and actual yields exists, Brennan (1984) wrote, "The only reliable sources of relative yields are variety trials" (p. 182).1 Annual changes in relative yields are measured with performance test data, which represent ideal management and agronomic conditions, instead of actual rice yield performance.
The genetic contribution of the Arkansas rice-breeding research program was measured by quantifying the increase in yields attributable to genetic enhancements in rice for the period 1983-2007. Yield gains were measured from all varieties released by the UofA, beginning with Newbonnet in 1983. Salmon (1951) reported that tests over many location-years are necessary to detect differences in cultivar yields. Yield data were aggregated over all locations and years to develop a yield ratio for each variety. Following Feyerherm, Paulsen, and Sebaugh (1984), relative yield ratios were derived by calculating the mean yield ratio over all location-years where each rice variety was grown together with the control variety (Newbonnet). For ease in interpretation, yield differences were also calculated by subtracting the mean yield of each variety from the mean yield of the control variety.
The yield ratio and yield differential provide comparisons of variety performance (Table 1). While the Feyerherm, Paulsen, and Sebaugh (1984) method allows for estimation of relative yield differences, it does not account for differences in breeding objectives; that is, if a variety is bred for blast resistance, lodging resistance, or bred for not only maximum yield but yield stability as well. To incorporate the objective of yield stability the Just and Pope (1979) method is implemented.
The Just-Pope Model
An advantage of the Just-Pope production function is its correction for multiplicative heteroscedasticity, which is important for varietal traits because of the variations in both the species and breeding goals across cultivars. Since cultivars are intended to be grown statewide and are specifically bred for resistance to different pathogens and adaptation to various agronomic conditions, the error terms across cultivars may be heteroscedastic in nature.2 The Just-Pope production function accounts for these variations and corrects the associated multiplicative heteroscedasticity in the error terms.
Data were collected from the ARPT test plots throughout the Delta of Arkansas from 19862007. The ARPT data consisted of six universityrun experiment stations: Pine Tree (St. Francis County), Stuttgart (Arkansas County), Rowher (Desha County), Keiser (Mississippi County), Cotton Branch (Lee County), an experiment station in Missouri, and two test plots conducted on farmer's fields in Jackson (Ruteldge Farm) and Clay (Ahrent Farm) counties. Cultural practices varied somewhat across the ARPT locations, but overall the rice variety trials were conducted under conditions for high yield. Nitrogen was applied to ARPT tests located on experiment stations in a two-way split application of 100 lb Nitrogen at preflood followed by a single mid-season application of 30-60 lb Nitrogen (depending on the experiment station location). Phosphorus and potassium fertilizers were applied before seeding at the Stuttgart, Jackson County, and Clay County locations. A total of 17 varieties were tested from 1986-2007; however some of the varieties were released before 1986 but were still sown in farmer's fields and thus included in the test plots. That is, even though the test period for this data set was 1986-2007, cultivars released prior to 1986 were also included. The oldest variety in the study was Newbonnet released in 1983, and the most recent was Taggart released in 2009.3
Long grain rice grown in the southern United States has cooking qualities that can be described as typical (Moldenhauer, Gibbons, and McKenzie, 2004 and Slaton, 2001) and produces rice that is dry and fluffy (non-sticky) when cooked. Long grain rice varieties grown in Arkansas are typically milled, parboiled, quick cooked, or used in processed rice products. Medium grain rice varieties grown in Arkansas produce moist sticky rice when cooked and are used in breakfast cereals, soups, baby food, and in alcoholic brewing processes. Approximately 93% of the observations in the data set were of the long grain variety with the remaining 7% being medium grain.
Semidwarf rice varieties are those who have been bred with a dwarfing gene that allows the plant to mitigate some lodging issues associated with high yielding varieties. In other words, a shorter variety often has a reduced chance of falling over under the weight of its own grain. The traditional "standard stature" varieties produce more biomass and have the potential for an increased market as a bioenergy source. Approximately 12% of the observations in the data set are semidwarf varieties with the remaining 88% being traditional varieties.
The mean and variance of yield are specified as a function of the release year (RLYR) of each cultivar, which can be interpreted as the "vintage" of the rice breeding technology (Arrow, 1962; Traxler et al., 1995). RLYR captures the progression of rice breeding technology across time, and is the main variable for measurement and analysis of the impact of the UofA's rice breeding program on rice yield. That is, the coefficient of RLYR represents the increases in yield due to genetic enhancement attributable to the UofA's rice breeding program. Following previous work (Traxler et al., 1995), a distinction should be emphasized between release year (RLYR), which varies from 1989-2009, and the trial date, which varies from 19832007. Each cultivar has a single RLYR, the date that the cultivar is made available for planting and the embodied breeding technology for that specific year. In the multiple regression model estimated here, the coefficient on RLYR captures only the effect of rice seed technology at the year of release. Thus, each cultivar is represented by only one RLYR. A typical newly released cultivar has relatively higher yields than previouslyreleased cultivars in the early years of adoption, followed by replacement by higher-yielding releases later. Release year is not a time trend variable, but is modeled similar to the way that Arrow's (1962) growth model denoted embodied technology (Traxler et al., 1995). Arrow (1962) assigned "serial numbers" of ordinal magnitude to the embodied technology in capital. In this model, the variable RLYR represents the embodied technology for a given year of release by the UofA breeding program. An RLYR squared term allows the model to capture curvature in the estimated relationship between yield and RLYR. Moreover, curvature provides breeders, administrators, and policy makers the ability to monitor not only yield increase, but also whether it is increasing at an increasing or decreasing rate.
Mean and variance of yield were also modeled as a function of rice grain length: medium and long grain. A distinction between the two needed to be made given the fact they are used/ bred for different purposes. While there is no a priori expectation of a yield difference between grain lengths, given the fact they are bred for different end use purposes, grain length needed to be held constant. Long grain was selected as the default because it was the most common grain length planted during the study. The University of Arkansas has released both tradition (standard statured varieties) and semidwarf varieties. Although given a normal growing year, standard and semidwarf varieties in Arkansas yield approximately the same, ceteris paribus, given the higher probability of lodging and thus yield loss in standard varieties during a wet/windy growing season, a distinction needs to be made between the two. The standard saturated varieties were chosen as the default since the majority of the varieties released by the UofA are of standard stature.
Yield is not the only selection criteria in the rice breeding program at the University of Arkansas. Therefore, a complete analysis of returns on investment should include yield growth and other traits such as yield stability, disease resistance, and end-use quality. While the most obvious tangible improvements of a breeding program are increased yields, the substantial economic benefits achieved as a result of yield losses avoided through disease (biotic) resistance (blast, sheath blight, smut, and stem rot) should be taken into consideration. Other economic benefits of a breeding program can be in the form of abiotic improvements such as decreased water needs (number of days on flood), improved heat stress, and cold tolerance (for early season rice). Improvements in biotic resistance/tolerance and abiotic stress can lead to higher yields and reduced yield variability. Or, it could leave both mean yields and variability constant, but it still has value as rice yields might otherwise have deteriorated as pathogens evolve to overcome resistance genes. Traits other than yield growth are difficult to measure and assess, and are therefore beyond the scope of the present study.
A dummy variable was included to represent each growing year from 1983-2007 with 2007 being the base year. The year variable was included to hold growing season anomalies constant. That is, if early rains delay planting, hurricane winds cause excessive lodging (2007), abnormally dry or wet growing seasons result in yield reductions, etc. By holding these growing season anomalies constant with the year dummies, these yield reductions can be accounted for and statistically correct yield estimates can be obtained for each growing year.
A dummy variable was also included for each ARPT experiment station. The station variable is the cross-sectional component of the panel data, and plays a pivotal role in holding growing conditions constant across the growing regions. Growing conditions, climatic and agronomic, vary by location throughout the state. Rainfall and other growing conditions in Northeast Arkansas diverge from the Southeast Arkansas experiment stations. Another spatial difference would be the presence of blast, smut, sheath blight, or any other diseases, which may vary across locations. By including the station variable the model can hold these spatial variances constant across experiment stations. Where the year variable can determine if a given growing season is abnormally wet or dry for the state as whole, the station variable can determine within a year if a specific location is abnormally wet or dry and can isolate spatially specific disease outbreaks. These differences in growing conditions across experiment field locations, or stations, are accounted for by inclusion of the station variable in the regression model.
Table 1 illustrates the average yield and release year for each variety in the study, and the results from the Just-Pope model. Both the effects on yield and on yield variance and the ordinary least squares (OLS) estimates are shown in Table 2 and Table 3 presents the fixed effects regression coefficients. Using the estimated RLYR and RLYR2, while holding constant all other variables at their means, partial derivatives of the function with respect to release year were calculated and illustrated on the bottom of Table 2. More than 39 percent of the variation in rice yields was explained by the regression model for the 1983-2007 period. The medium grain varieties released by the UofA yielded 9.71 bushels per acre more than the long grain varieties, holding all else constant. Semidwarf varieties were shown to yield approximately 15.98 bushels per acre less than the traditional (standard) varieties released by UofA. This was expected since the majority of varieties released by the UofA breeding program were of standard height (93%), inferring that the majority of its time, effort, and money were spent on standard statured varieties.
The "release year" variable, or the year in which each respective rice variety was released to the public, was positively associated with yield, and on average was equal to 0.68 (Table 2).5 This result would indicate that each year the UofA breeding program increased average yield for their varieties by 0.68 bushels per acre annually, equivalent to a 0.42 percent annual yield increase (0.68/162.39, where 162.39 is the average yield for all UofA varieties over the time period under investigation). During the 1983-2007 period, the UofA rice breeding program contributed 16.42 bu/ac cumulatively to rice yields (0.68 × 24). This equates to a 10.11% (16.42/162.39) increase to producer yields per acre attributed to genetic enhancement from the UofA breeding program over the entire time period.
Another important aspect of the UofA breeding program is the decrease in yield variability over the same time period. More specifically, between 1983 and 2007, the varieties released by the UofA breeding program experienced an increase in annual yield, and a decrease in yield variance. The average yearly variance decreased 0.016 (bu/acre)2 (Table 2).6 The UofA breeding program continually breeds for pathogen resistance in all of its cultivars. Improvements in blast tolerance and other pathogen resistance, so called maintenance breeding, can lead to lower yield variability and higher mean yields. If the yield variance decreases, as is the case in the results above, there is an economic value to this in the form of yield loss prevention. Without the UofA breeding program rice yields may otherwise have deteriorated as pathogens like blast and sheath blight may have lessened yield and increased variability as they overcame earlier sources of resistance through breeding.
Table 4 presents the estimates of the genetic and economic benefits of the UofA rice breeding program for rice producers in Arkansas, assuming a perfectly elastic demand for rice.7 The genetic gains on Table 4 are calculated from the results (RLYR and RLYR2) of the regression model on Table 2. An important feature of the calculation of genetic gains associated with a breeding program is to take into account the cumulative effects of the program over the entire period. That is, the yields gains attributable to the breeding program in 2007 are those observed in 2007 plus those seen in 2006. So, the genetic gains for 2007 would be the sum of the year specific genetic gain from 1983-2007. The annual and cumulative genetic gains for the UofA rice breeding program are listed in Table 4. Rice producers in Arkansas have received an average annual economic benefit of $19.5 million (2007) in United States dollars (USD) from the UofA rice breeding program from 1986-2007. These benefits are a function of several exogenous factors (acreage, rice price, and adoption rate) and the endogenous factor of genetic gains attributed to the breeding program. The average economic benefits Arkansas producers have received this decade (2000-2007) are estimated at $34.3 million (2007) USD annually (Table 4).
There are substantial "spillover" benefits outside the state of Arkansas from the UofA rice breeding program. That is, out of the total 817,240 acres sown to UofA rice varieties in 2007, 141,490 (17.31%) were sown outside of the state of Arkansas. To fully account for the "total" economic benefits of the UofA breeding program the summation of yield enhancements from the other states that sow UofA varieties must be accounted for.
Table 5 is constructed using the total rice acreage in Mississippi, Louisiana, Missouri, and Texas that were sown to UofA rice varieties, as well as the average annual long grain rice price from National Agricultural Statistics Service from 1990-2007. Missouri and Louisiana had the largest percentage of their total rice acreage sown to UofA varieties over the 1990-2007 period at 40 and 9%, respectively. Table 6 shows the total benefits, both the benefits in Arkansas and the spillover benefits in the surrounding states, of the UofA rice breeding program using the cumulative genetic gain calculated in Table 4. The average annual benefits of the UofA breeding program from 1990-2007 was 28.04 million (2007 USD). That number increases to 46.7 million (2007 USD) when you analyze the annual benefits for this decade, 2000-2007. Again, these benefits are partially by price, which is exogenous to the breeding program but driven by acreage (adoption rate) and genetic gain, which is endogenous to the breeding program.
Another way of interpreting these results is the counterfactual case. That is, what would have happened if the UofA had not invested in rice breeding from 1983-2007? The implicit counterfactual is that Arkansas and producers would have continued to grow varieties of the vintage and yield of Newbonnet and would have forfeited the benefits estimated above. However, it is more likely that Arkansas producers would have adopted rice varieties developed by other breeding programs (Louisiana State University and Texas A&M for example). In that sense these estimates would overestimate the total benefits estimated above.
Often overlooked in breeding programs is the importance of breeding for pathogen and disease resistance, so called "maintenance breeding". While this paper captures the genetic gain attributed to the UofA rice breeding program it fails to account for the significant yield losses that would unfold overtime without UofA breeding. Previous studies (Marasas, Smale, and Singh, 2003) on breeding programs have estimated that the economic impact of a breeding programs pathogen resistance breeding efforts can be as large, if not larger than the impact of increased yields. Therefore, these estimates would underestimate the total benefit associated with the UofA rice breeding program.
Implications and Conclusions
Arkansas producers sowing UofA varieties during the period from 1983-2007 experienced a 10.11% yield increase that can be attributed to the genetic advancement of cultivars through the UofA breeding program alone. These estimates result in Arkansas producers receiving an average annual economic benefit of $19.5 million (2007) USD from 1986-2007. When you account for the spillover benefits, neighboring states that sow UofA varieties, the average annual economic benefits increase. That is, in 2007, approximately 17% of all of the UofA varieties sown were sown outside the state of Arkansas. The average total, in Arkansas and surrounding states, economic benefits of the UofA breeding program from 1990-2007 was $28.02 million (2007) annually.
Using a discount rate of 4% to calculate the discounted costs and benefits, and accounting for the 1 0 year lag between the initial cross and releasing a variety to the public, the benefit cost ratio was estimated to be 17.7: 1.8 That is, for each dollar of public funds invested in the University of Arkansas rice breeding research, nearly $ 1 8 of benefits result. Using the same assumptions but with a new discount rate of 10% the estimated benefit cost ratio decreases to 8.75:1. To put these benefits in perspective, the internal rate of return (IRR), which is computed as the discount rate that results in a value of zero for the net present value, was estimated to be 30.9% for the time period.9 The benefitcost ratio and IRR provide evidence that the economic rate of return to the University of Arkansas rice breeding program is high, although assessing these measures further is difficult without comparable values for other public investments (the opportunity cost of funds).
The most tangible improvements of a breeding program are in the form of increased yields, but the substantial economic benefits should be recognized by valuing the yield losses avoided through sheath blight, blast, and other pathogen resistance, so called "maintenance breeding". This study also found that over the 1983-2007 period that UofA varieties had reduced yield variability. This study only valued yield increases and did not attempt to quantify the value of the maintenance breeding nor attempted to monetize the value of decreased yield variability, thus the benefits estimated to producers in this study are on the conservative side. That is, without the breeding program, rice yields could have deteriorated and become more unstable as pathogens such as blast and sheath blight may have drastically reduced yield and increased yield variation as they overcame earlier resistance genes.
Holding all climatic and agronomic conditions constant, the annual genetic gain attributed to the UofA rice breeding program has increased, and the returns to the UofA rice breeding program continue to play a large role for Arkansas producers through both increasing and stabilizing rice yields. Given the estimates found in this research, the benefits of the rice breeding program outweigh the costs by a large multiple, demonstrating that investments in the UofA rice breeding program have provided large and sustained economic benefits to rice producers, consumers, and millers in Arkansas.
[Received June 2010; Accepted September 2010.]
1 In 2007 the United States Department of Agriculture reported an average yield of 160 (bu/ac) for the entire state of Arkansas compared with the Arkansas Rice Performance Trials (ARPT) average of 168 (bu/ ac), a 5% difference (University of Arkansas, CES, 2007). That being said, experimental test plot yields should be viewed as a yield "ceiling" for producers to analyze varieties by. An overestimation of a breeding program can result given the size of the yield "gap" between test plots and on-farm results.
2 Breusch-Pagan test, significant at the 1% level, indicated the presence of heteroscedasticity.
3 Two varieties, Taggart and Templeton, were released to the public after 2007 but had test plot observations before 2007.
4 The Hausman test, for fixed versus random effects, indicated random effects were appropriate. That being said, the t-ratio for fixed effects were larger, in absolute value, and the coefficient differences between the two models were negligible. Thus the fixed effects model was implemented.
6 Since the variance regression included a squared term (,RLYR^sup 2^) this would indicate that 0.0162 (bu/ acre)^sup 2^ is not a constant but rather the average decrease in annual yield variation over the time period.
7 This assumes that the increased Arkansas rice production due to genetic improvement does not influence the global price of rice. This is a realistic assumption, since Arkansas produces such a small portion of the global rice supply, and the yield increase is a relatively small shift in the total world supply of rice.
9 The internal rate of return was calculated as: 0 = Σ^sub t^[(B^sub t^ - C^sub t^)/(1 + IRR)^sup t^].
Alston, J.M., G. W. Norton, and P.G. Pardey. Science under Scarcity: Principles and Practice for Agricultural Research and Evaluation and Priority Setting. Ithaca: Cornell University Press, 1995.
Arrow, K. "The Economic Implications of Learning by Doing." The Review of Economic Studies 29(1962):155-73.
Barkley, A. "Kansas Wheat Breeding: An Economic Analysis." Report of Progress 793. Kansas State University Agricultural Experiment Station and Cooperative Extension Service, October 1997.
Brennan, J.P. "Measuring the Contribution of New Varieties to Increasing Wheat Yields." Review of Marketing and Agricultural Economics 52( 1 984) : 1 75-95.
Brennan, J.P. "An Analysis of the Economic Potential of Some Innovations in a Wheat Breeding Programme." Australian Journal of Agricultural Economics 33(1989a):48-55.
_____. "An Analytical Model of a Wheat Breeding Program." Agricultural Systems 31(1989b): 349-66.
Byerlee, D., and G. Traxler. "National and International Wheat Improvement Research in the Post-Green Revolution Period: Evolution and Impacts." American Journal of Agricultural Economics 77(1995):268-78.
Feyerherm, A.M., GM. Paulsen, and J.L. Sebaugh. "Contribution of Genetic Improvement to Recent Wheat Yield Increases in the USA." Agronomy Journal 76(1984):985-990.
Huffman, W.E., and R.E. Evenson. Science for Agriculture: A Long-Term Perspective. 2nd ed. Ames, IA: Blackwell Publishing, 2006.
Just, R., and R. Pope. "Production Function Estimation and Related Risk Considerations." American Journal of Agricultural Economics 61(1979):276-84.
Marasas, C, M. Smale, and R. Singh. The Economic Impact of Productivity Maintenance "Research: Breeding for Rust Resistance in Modem Wheat." Agricultural Economics 29(2003):253-63.
Nalley, L., A. Barkley, and F. Chumley. "The Impact of the Kansas Wheat Breeding Program on Wheat Yields, 1911-2005." Journal of Agricultural and Applied Economics 40(2008):913- 25.
Nalley, L., A. Barkley, J. Crespi, and K. Sayre. "The Impact of the CIMMYT Wheat Breeding Program: An Economic Welfare Analysis." Journal of International Agricultural Trade and Development. 5(2008): 1 1-16.
Salmon, S.C. "Analysis of Variance of Longtime Variety Tests of Wheat." Agronomy Journal 43(1951):562-70.
Slaton, N. Rice Production Handbook. Little Rock: The University of Arkansas Cooperative Extension Service, 2001.
Tassey, G. "Methods for Assessing the Economic Impacts of Government R&D." NIST Planning Report 03-1. National Institute of Standards and Technology (NIST), U.S. Department of Commerce, 2003.
Traxler, G, J. Flack-Zepeda, J. Ortiz-Monasterio, and K. Sayre. "Production Risk and the Evolution of Varietal Technology." American Journal of Agricultural Economics 77(1995): 1-7.
University of Arkansas Cooperative Extension Service (CES). "Arkansas Rice Performance Trials (ARPT), 'Rice Data'". Various years, 1996-2007, Little Rock, Arkansas.
L. Lanier Nalley is an assistant professor, and Nate Lyman is a graduate student, Department of Agricultural Economics and Agribusiness, University of Arkansas, Fayetteville, AR. Karen A. Moldenhauer is a professor, University of Arkansas Division of Agriculture Rice Research and Extension Center, Stuttgart, AR.…