

SEA-FISHERIES LABORATORY. 149 



triennial averages are taken annually. Such three-y early- 

 averages taken yearly remove the roughness of the graphs 

 without affecting their general tendencies. 



Tables III and IV, then, are prepared in this way. The 

 averages of groups of three years are taken, thus : 1892-3-4, 

 1893-4-5, 1894-5-6, etc. The averages so found are plotted 

 opposite the centre of the segments of the horizontal axis, 

 representing the three-yearly periods ; thus the triennial 

 average for 1892-3-4 is placed above the centre of the length 

 of the horizontal axis, representing the year 1893. 



The data for the fish trawl catches for the years 1898-9 

 and 1900 are, as has been said, missing. To fill up this gap, 

 as far as possible, two yearly averages have been calculated 

 for the years 1896-7, and the result is plotted on the graph, 

 Fig. 1, at the middle of the period 1896-7. Similarly the 

 years 1901-2 have been taken, and their averages calculated, 

 and the result is plotted against the middle point of the period 

 1901-2. The dotted straight line in the graph represents the 

 results for the missing period of three years, and since it 

 appears to occur on the descending limb of a curve, it probably 

 indicates fairly what was the general tendency of the variations. 



Tables III and IV give separate averages for the year 

 1918, and 1919, and the latter year is included by itself in the 

 graphs for both fish trawl and shrimp trawl nets. 



Now we assume that the results obtained in any one year 

 may be influenced by the conditions of the year before and the 

 year after, that is the rationale of taking triennial averages 

 annually. But even then the graphs drawn from these values 

 show irregularities and if there are one or two exceptionally 

 large catches in any one year, the triennial average will be 

 affected, though to a less extent than would annual 

 It will be seen that the maxima on the graphs include those 

 years in which there are some e oeptionally large eat dies, 

 and conversely the minima are in p< nods when the eatchi are 



