178 Appendix 5 



value that a variable will have if included in the regression. When 

 included, the variable with the highest '"F to enter'' value will 

 contribute most to explanation (R^). This variable fits best the ob- 

 served data and therefore should be chosen as the research vari- 

 able in the regression. As the table indicates, the best knowledge 

 variable is the unweighted stock measure. This variable is 

 therefore used in the analysis. 



Experiments with an alternative weighting scheme were also 

 conducted: The ''dummy" coefficients, the estimated (3, values 

 in regression 2 of table 2.8 in chapter 2, were 



These values indicate the extent to which counts of publica- 

 tions in a certain sector or product exceeded the value "pre- 

 dicted" by the level of production in the sector and other vari- 

 ables included in the regression. There can be several reasons for 

 nonzero dummy coefficients: (1) the abstracting journals cover 

 some areas of research more extensively than others; (2) by tradi- 

 tion, papers are shorter or more duplication is tolerated in some 

 areas of work than in others; (3) some areas are more in fashion 

 than others; (4) research in some areas is more productive (per 

 unit of output) than in others and, to equate at the margin, more 

 effort is channeled in these directions. 



To test these possibilities, measures of research were 

 "corrected" 13 



where ^- is the dummy coefficient of sector /. 



