Each of the four output measures has strengths and weaknesses

Many have predicted that such nations must bear much of the responsibility to produce the varieties that will feed the world in the coming decades . The large number of breeding centers in China, the decentralized nature of its research system, and the great heterogeneity among its centers offer a unique research opportunity to identify the relationship between varietal production, size of institute, and mix of crops in the breeding program. Finally, the results are of interest to China’s leaders who recently have announced a new round of research reforms in agriculture . To make our analysis tractable, and because of budgetary constraints, we limited the scope of our study to the breeding institutes of two crops within northern China. We chose crop-breeding institutes because crop breeding has been central to the growth of agricultural productivity in China as well as in the world. In China, crop breeding takes the largest proportion of resources in its agricultural research system . Crop-breeding institutes were chosen also because their research outputs and their consequences can be measured relatively easily—compared with, say, less applied research, research oriented towards natural resources management, or research leading to disembodied technological change. Wheat and maize are two of the most important staple crops in China, ranking second and third respectively after rice in terms of sown area and production. Wheat and maize production occupy somewhat overlapping areas and a large share of China’s wheat and maize breeding programs are located in the same institutes and similar regions,pot raspberries which allows us to measure economies of scope.The small and highly significant coefficients of economies of scale imply a significant cost saving associated with expanding the scale of breeding institutes. Such results are robust Most of the data used in this study were collected by the authors during 12 months of field work in China that began in the summer of 2001. Enumerators assembled panel data from 46 wheat and maize breeding institutes covering the years from 1981 to 2000.

The sample institutes include 40 prefectural-level institutes and 6 provincial-run institutes, selected at random from a comprehensive list of prefectural and provincial institutes in seven major wheat and maize provinces in northern China.Thirty-two of the sample institutes produce both wheat varieties and maize varieties . Four institutes specialize in producing wheat varieties . The other ten produce only maize varieties .To collect the data, teams of enumerators visited each institute for periods of up to one week and completed a set of questionnaires filled out by accountants and by enumerators. In general, the data cover four broad categories: income, costs, research output, and data on other characteristics of the institute. Since the data were not kept by a single department in any of the institutes, a great deal of cross-checking was needed to make the data consistent among the various departments. For example, the research coordination department typically kept information on income and expenditures. Personnel departments provided the data on salaries, educational accomplishments, and other information about current and past staff. Breeders kept the best information on the varieties they produced and the methods that they used in their breeding efforts. To examine cost efficiency, information is needed on two key variables, costs and output, especially since there is an a priori reason to believe that the small scale of many institutes may be an important source of inefficiency. In using our survey data to define measures of these key variables, we have had to deal with several methodological issues. As an economic activity, crop breeding has several characteristics that make it relatively hard to measure output and match measures of output to measures of costs associated with those outputs. These characteristics include the long lags between the time when costs are incurred and the resulting output is realized , uncertainty about what is an appropriate measure of output both conceptually and in practice , and the fact that output itself is uncertain when costs are incurred .Our measure of the total variable costs of each crop’s breeding activities includes the institute’s operating expenses, such as salaries, project administration, and other direct operating expenses. For cost categories that cannot be matched directly to a breeding project , we assigned a share of the costs of each category to breeding according to the number of full-time breeding staff . We deflated total variable costs by a provincial consumer price index, putting our cost figures into real 1985 terms . We assume that the products of China’s wheat and maize variety “factories” are the varieties that the breeders produce that are adopted by farmers. To measure output, we collected information on the number of varieties that were produced by the research institutes , the area sown to the varieties, and the trial yields of each variety . With these data, we constructed four measures of research output: the number of varieties released by the institute sown in the field during a given year, the number of varieties, weighted by the trial yields of the variety , the total area sown to all of the institute’s varieties during a given year ; and the number of varieties weighted by sown area and trial yields .

Although it is the most readily measured, the obvious flaw with number of varieties is that it does not take into account any quality characteristics of each variety, either yield or its other characteristics . Yield-weighted output accounts for the relative productivity of a variety in pure output terms. However, such a measure still leaves out all other quality characteristics, which an earlier study shows may be highly valued by farmers . For this reason, our third measure, area-weighted output, should be superior to the other two measures. If farmers value the characteristics in a variety—whether high yields or some other characteristic—they demonstrate their preference by adopting the variety . The last measure, yield-area weighted output combines the second and third measures. Since the variation in trial yields is small, the correlations between the third and fourth output measures are high . Hence, we would not expect much difference to result from using one versus the other. One special feature of crop variety production is the significant time lag between the time when costs are incurred in a breeding research program and the time when the resulting research output is realized. This issue is commonly discussed in studies of the returns to agricultural R&D , especially in relation to specification of econometric models relating agricultural productivity to research expenditures. In the present setting, the lag between investment and output has some further implications, akin to those that arise more generally in agricultural production economics, associated with biological lags. In microeconomic theory texts, the firm manager first chooses an output level , and then determines the cost-minimizing combination of inputs that will produce that output at minimum cost. The crop breeding institute’s director does not have that luxury, because the output from today’s investment is uncertain and will not be known for many years . As an approximation to this problem of decision-making under uncertainty, we might suppose that the director seeks to minimize the institute’s cost based on current expectations of the output that will be produced in the future as a result of the current research expenditures. Unfortunately,plastic gardening pots we cannot observe or measure, ex post, such expectations. One option is to use the output that was actually produced from the expenditures as a proxy of those expectations, but the problem remains of matching actual outputs to particular expenditures .

To deal with this problem empirically, we defined an average research lag to represent the number of years between the time when a breeding project officially begins and the time when a variety is released for commercial extension to the fields of farmers. Using this defined lag length, we modeled the cost of variety production as a function of the research output produced after a certain lag. To find the length of lag, we designed a section of the questionnaire to ask breeders in each of the 46 institutes specifically to estimate the average lag length for each crop. Based on the data we collected, the average lag length was 5.3 years for wheat and 4.5 years for maize. In our base model, we used a 5-year lag for both wheat and maize variety production. However, we also tried different lag lengths to check the robustness of our results. China’s agricultural research system has produced a steady flow of crop varieties in the past. On average, in each year during the period 1982-1995, China’s farmers grew 200 to 300 wheat varieties and 130 to 180 maize varieties in their fields . However, the number of new varieties being produced by research institutes varied significantly over time and across institutes. Based on our survey, 141 wheat varieties and 155 maize varieties were produced by our sample institutes during the period 1985- 2000 . Nineteen percent of the wheat varieties were developed by provincial institutes. The rate of production of new wheat and maize varieties increased over time. For example, prefectural maize institutes produced 34 maize varieties during 1985-1990, 47 varieties during 1990-1995, and 74 during 1995-2000. The number of wheat varieties created and commercialized by the sample institutes rose from 31 in 1985-1990 to 55 during each subsequent period . The number of varieties, however, varies sharply among institutes. For example, the Henan provincial wheat institute produced 12 wheat varieties from 1985 to 2000. The Mianyang prefectural crop breeding institute in Sichuan produced 14 wheat varieties. In contrast, 24 out of 36 of the sample wheat institutes produced fewer than 5 varieties. In fact, three wheat institutes did not produce a single variety during the entire 15-year sample period. Maize variety production also varies greatly among the sample institutes.In the same way that output varies across time and space, so does total cost. On average, the annual real total variable costs of the breeding program per institute for our sample of wheat institutes increased from 24,000 yuan to 38,000 yuan between 1981 and 2000. Similarly, the average annual total variable breeding cost for our sample of maize institutes rose from 38,000 to 53,000 yuan.The total cost of wheat and maize breeding, however, varies greatly among institutes. For example, the average provincial institute invested five times more in wheat breeding and about six times more in maize breeding than the average prefectural institute did. When comparing prefectural breeding stations, the total cost of wheat breeding in one institute could be more than three times that of the average prefectural institute. Dandong prefectural institute in Liaoning spent five times more than the average maize-breeding institute did. The average cost of variety production also varies from institute to institute and can be seen to vary systematically with research output. To compare costs and output, we have to account for the research lag. In the analysis that follows, research output is the annual mean of five years’ total research output from one of three five-year periods, 1985-1990, 1991-1995 and 1996-2000. The average annual cost associated with this output is the annual mean of five years’ total cost, lagged by five years. Therefore, the corresponding three five-year periods of cost are, respectively, 1980-1985, 1986-1990 and 1990-1995. Unlike total costs, average costs fall as the institutes produce more varieties . For wheat the cost per variety falls from 152,000 yuan for breeding institutes that produce only one variety to 60,000 yuan for those that produce more than four varieties. Similar patterns can be seen in the data when using area-weighted output. A plot of the data reveals a distinct L-shaped relationship between average cost and the size of research output .No matter what measure of output is used, or for what crop, as research output increases, the average cost of breeding research falls. The L-shaped relationship also is robust, holding over time and over institutes . The sharp fall in average costs of breeding as an institute’s output rises suggests that China’s wheat and maize research institutes are producing in an output range with strong economies of scale, such that efficiency might be increased by expanding the scale of production of China’s wheat and maize research institutes.