182 



8 



or productivlcy Impacts and distribution consequences are important parts of 

 these goals. In their approach, agricultural coomodities are generally given 

 preeminence in research priority setting. The priorities when established, 

 most likely by a research team, tend to be implemented by "top down management.' 



Evenson and Kislev (1975) have developed a model of the discovery process 

 for innovations that fits into the Norton et al. priority setting process. 

 (See Huffman and Evenson 1993, p. 227-8 for a summary.) Evenson and Kislev 

 assume a particular specification to the search function where the likelihood 

 of innovations are independent across research commodities or activities, and 

 each innovation is related linearly to the real research expenditures on the 

 research commodity or activity. Given that an expected value can be placed on 

 successful innovations and that research administrators attempt to siaximlze 

 the expected value of all research discoveries subject to a budget or resource 

 constraint, the model leads to the conclusion that research funds should be 

 allocated in proportion to the expected value of their discovery. For example, 

 if the expected value of an important discovery in com research is worth twice 

 as much as an advance in soybean research, then twice as much research money 

 should be allocated to work on the discovery in com relative to the advance 

 in soybean research. This is a type of expected benefit-research expenditure 

 allocation "congruency rule." 



There are some Important qualifications to this rule. First, when this 

 rule is applied by a particular research station or agency, the decisions tend 

 to be made myopically. Administrators generally ignore research benefits that 

 may spillover to other regions or to other research enterprises and they tend 

 to ignore the existence of potentially duplicative research conducted else- 

 where. Second, we believe that advances in science are best described by a 

 building block approach within multllayered and interconnected levels of 



