The fiducial interval of a treatment nnean nnay be calcu- 

 lated as follows (Snedecor 1948s p^ 221); 



- + 2„0 |/~462_ 

 \' "?„ 14 



= + 16 



The mean for the controls is 36,2 seconds; those for the 

 experiments are included m table 2o Again, the means for only four 

 experiments, NoSo 95, 97, 99j and 100, as before, differ significant- 

 ly (by 32 seconds or more) from the mean for the controls. There 

 are no significant differences between the means for any of the seven 

 experimentSo This latter conclusion is also evident from a further 

 analysis of the variance of the combined results: 



Treatment vs. control 

 Treatment nneans 

 Individual periods 



The above analysis, based on quantitative data, has 

 demonstrated a significant response in four of the seven experiments. 

 Observation, however, showed a response m all seven, nannely, ob- 

 \'ious sensing of the material connbined with one or more of the several 

 components of the reaction^ as described previously. Only in No. 98 

 was there any doubt; even here, however, there was an obvious change 

 in pace and surfacing of the fish during the first and second test 

 periods o The data and their analyses do not always reflect a response 

 which is obvious and "significant" to the observer o Other experiments 

 could be selected in which the discrepancies between the data and the 

 observations are even more pronounced. 



The difficulties outlined above were recognized early. 

 Nevertheless tinning and counting were continued: they did measure 

 certain phases of the reaction and have been used in some quantitative 



21 



