554 



Fishery Bulletin 103(3) 



assigned to a year of formation based on estimated age 

 and capture year. Chemical assay and otolith processing 

 were completed at the Stable Isotope Laboratory of the 

 University of Michigan. Each otolith was embedded in 

 epoxy resin and cut transversally with a low-speed dia- 

 mond-bladed saw. Three or four thin sections -150 /urn 

 thick were removed from the center of each otolith. The 

 thin sections were then glued with cyanoacrilate glue 

 to petrographic glass slides. Samples from multiple thin 

 sections were combined for a single assay. Each opaque 

 growth zone was sampled by using a Merchantek Micro- 

 milling system and assays were completed with a Finni- 

 gan 251 MAT mass spectrometer. All measurements 

 were reported in standard Vienna Pee Dee Belemnite 

 (VPDB) and notation as 6%c (per mil), where 



<S 18 



:((( 18 0/ 16 



sample 



/( 



'O/ I6 O) standard )-lxl000). 



A time series of 6 18 was constructed for each fish by 

 using the assay from each year-specific sample of the 

 otolith material. Assay results from the collection year 

 were not included because of differences in the season 

 of capture. Years with missing results were due to 

 micromilling or assay errors that resulted in no results 

 reported by the stable isotope laboratory. 



To gauge the accuracy of age assignments, we fitted 

 a linear model to each time series and analyzed the 

 residuals from the linear model fit. The <3 1k O value cor- 

 responding to the negative residual of greatest magni- 

 tude from that linear model would be associated with 

 the anomalously warmest oceanic conditions (El Nino). 



If the age assignments were correct, that portion of the 

 otolith corresponding to 1983 would have the negative 

 residual of greatest magnitude because the observed 

 d ls O value was much lower than the linear model pre- 

 dicted. Temporal shifts of the most anomalous negative 

 residual with respect to 1983 were interpreted as either 

 an under- or over-estimation of age. 



A randomization procedure (20,000 iterations) was 

 used to determine if the magnitude of the average re- 

 sidual in any year was more negative than expected, 

 thus identifying the signal associated with the 1983 

 El Nino. The residuals from the linear models within 

 each of the fish were randomized with respect to year. 

 Randomized residuals from all iterations were averaged 

 across all fish to produce a distribution of averages. The 

 original year-specific residual averages were compared 

 to the randomization distribution to estimate statis- 

 tical significance. We rejected the hypothesis of any 

 year with an average residual 2O if less than 5% of the 

 randomizations produced a negative average residual 

 of equal or greater magnitude than the observed year- 

 specific average residual, thus identifying anomalously 

 warm years. 



An iterative sensitivity analysis was performed by 

 retrospectively removing sequential blocks of years of 

 data and by estimating the statistical significance of 

 the originally determined anomalous years with the re- 

 duced data sets. All data taken from years more recent 

 that the cutoff year were removed, and the linear model 

 fitting and randomization procedures were recalculated. 

 The cutoff year was sequentially changed beginning 

 with 1989 to 1986. 



