.. Not tci be cited without prior reference to the authar. C.M. 1993/D:47 ICES 1993 Ret. G Statistics Cttee. Ref. Demersal Fish Ctteie. RESÄMPLING TECHNIQUES USEb iN CALCULATING ESTIMATORS AND CONFIDENCE INTERVALS OF GASTRIC EVACUATION RATES.by Peter R. Garrahäri Graduate Schaol ofOceanography, University of Rhode Island. South Ferry Road, Narragansett, RI 02882 USA. ABSTRACT • A nonlinear model was fitted by least-squares methods to wet weight and ash-free dry weight data from gastric evacuation experiments conducted on Atlantic cod (Gadus morhua L.). The regression model contained parameters for estimating the siope and shape of the digestion cuNe. Stomach contents sampies were coilected using a modified gastric lavage method that determined the amount of particulate material that passed , ~ I ' through the collectlon screens. Computer-intensive, bootstrap and jackknife techniques were used to resample the data to esÜmate variance in evacuatlon rates. Joint 95% confidence Intervals were constructed for the model parameters in order to compare experimental effects such as prey type, meal size, and sampie consiituents. Results from different studies could also be compared by using these techniques. • INTRODUCTION There has beeil a considerable amount of discussion what is the exact model thai describes food evacuation in fishes. Several functions, ranging from linear to cUrViliriear, have been routinely fitted to gastric evacuation,data: linear (Jones, Ü)74; Bagge, 1977; Bromley, 1991), exponential (Tyler, 1970; Elliott arid Persson, 1978; Eliiott, ,1991), and square root (Jobling, 1981). Temming arid Andersen (1992) suggested using a primary . model that deflnes the depletion rate in stomach contents as deperident upon the amount of food present in the stomaeh. dS . . (1) Cit=-Re SB Where S = stomach conterits. t = time. -R = rate of depletion (slope of curve). B = degree of curvilinearity (shape of cuNe). \ I I .. .. .. \ . .. Terrüning arid Aridersen's model incorpori:ltes all of these models by not pre-defining the shape of the curve but retalning it as a parameter: B B B B = 0.00: linear model. = 0.50: square,;.root model. = 0.67: volume dependent model. = 1.00: exponentiai model. Integratirig equation (1) over a specific time gives th~ following: , ' I St = [So - R· ( 1 - B ) • t ] _'1_ 1• B (2) So = original meal size at time (0). St = storriach contents at time (t). Where t = ~ost-praridialtime. .' \' Recently computer-intensive statistical techriiques for estimating variance, namely the " . ' . "" ".1· ," . bootstrap (Efron, 1982, 1987) and jackknife (Miller, 1974; Duncan; 1978; Fox, et al., 1980), have become mor~ preval.ent in ~iO~Ogi~al\studie~., Jac.kknife a~d bootstrap methods make no assumptlons about the dlstnbutlons of vanables and in problems where . I , .. it is difficult to measure vadance directly, these techniques have been helpful. Meyer, et . '-' ' . . . ,.,. .' , . , . , .' '. . "" I , . ,.. : .... '" " al. (1986) have used resamplrng methods to compare growth rates in cladoceran . . . . . ,I. . • '. . populations by determining the vadance and reducing the bias in their estimates. ' . . I In this stLJdy, jackknife and bootstrap techniques were used in conjunction with the primary model to estimate rates and variances of 'gastric ievacuation in Atlantic cod. The " ,. , .. .,.'.' ,., I, ,.,' . .., experimental factors investigated were prey type, meal size, and stomach content sampie 'n " . \.. " I :;:I~:~S. DATA .,.; ....., "". , "I. .. ,',,'. . '" , ,., Gastric evacuatlon data were collected from experiments conducted on Atlantic cod at the ,. , , ' , ' , . '.' '.,'.' ", I. '. '." . ' ' .. '. .'.; University of Rhode Island, Graduate School of Oceanography Aquanum facrlltles dunng , . > " ."'~.' • • • j .."..:.. ; " ,.' • -, '."" . • 4: the summer of 1992. Atlantlc cod (x = 1665 g; range == 972 - 3072 g; N == 45) were caught . . .. " ' ,,',. . 1.. ." .. . south of Cape Cod, Massachusetts, USA and were maintained at 10 ± 0.5°C in' two . • , . ' . .1 . , 7000-L tanks (3 m in diameter). During experiments, individually tagged cod were . ., .", .'. .. . ' ., . I :'. , ' . . ' ., voluntarily fed known meals of prawn (Pandalus sp.) or herring (Clupea harengus) .' ,; consisting of 1, 2, or 3 prey items (Table 1). Three to five cod were. sampled .' , " .' ' . I '., " . . ', . approximately every 2 hours for 24 hours. Stornach contents were collected on baskets " . ' I..,.· ; ' .. , ., with 500-Jlm nylon-mesh screen by a modified gastric lavage techntque (Robertson, 1945; , " , .. dos Santos, 1990). 2 " , . . , ' . ,'" '., ~"":."~" •• '.'~'" - .. ~{\;.I•• ~'':'; ..• \ ; Unlike other studies that have used gastrie lavage, the filtrate water was eolleeted to determine the amount of partieulate material that passed through the sereens. Filtrate water vohjme was measured and thre~'s~b-~amples of water were filtered through glassfiber fiiters (type AlE, Gelman, Inc.) capable JIm. ot extracting 95% 01 material greater than 1 Total particuiate material was measured from these three sub-sampies and extrapoiated to the entire water sampie. Various eomponents'of the stornach content sampläs were measured: wet weight >500 JIrTI (WET), ash-free dry weight >500 Jlm (AFD), and total ash-free dry weight with particulcite material <500 Jlrn CfAFD). STATISTICAi.. ANALYSES Experimental data were cut off after 2 data p<?ints fell below 10% of the odginal meal size (Elashoff, 1982) and the resulting sets included 21 - 33 data points for each sedes (Table 1). An IBM pe equipped wlth Statistieal Analysis Systems (SAS, version 6.03) software for the pe was used to fit the model by non-Iiriear least squares' methods that 'did not use derivatives (NLIN procedure, secant method, SAS, 1988): '.. , I • For the jackknife sampies eaeh sueeessive data point was removed arid the parameter esÜmates were recalculated, resulting in N recalcuiations, each compieted with N ~ 1 data points. 8ecause -R' and 8 were highly cerreiated and non-normally distributed, the estimates had to be transformed (Miliar, 1974)~ "fhe slope parameter, -R, was transformeef by the exponential funciion transformed; while the shape parameter, 8, was square-root transforrried. The resuitirig N number of pairs of parameter estimates were multiplied by • (N - 1) and subtracted from Ntimes the transformed original estimate from the full sampie: /\ ä j =Ne Where /\ - (N - 1)e_i (3) i = (1, 2, 3, ;.. N) data point number. Si = ith pseudo-value, transformed pair of parameter estimatesj (8, -R). /\ e = full sampie estimates, transfornisd. 8_i = estimate wlth ith data point remeved, transformed. This essentially reduces the bias (Efren, 1982) and ereatss ;'pseudo-values;' whose average and variance was then further studied~ The jackknife estimator is the average of the pseudo-values, .N /\ 1~' 9 J =NL,9 j i=1 3 arid the . 950/0 joint confidence regions (Duncan, 1978) are defined by: . '.' A { S: (SJ - I, <e) ~ --E..N 7P F A T1 <e) S - (SJ - 1 - a(P. N - p} } 1 . , (5) I I . , .' . " <eJ = Jackknife estimators, transformed pair of parameter estimates. I S ==' variance-covariance matrix of transformed pseudo,;,values. A Where , , I F1 • a=. F value for N - p degrees of freedom. , • , ' , . , I , P = 2, number of parameters in model. , . ' I, ." For all data sets, bootstrap procedures consisted of randomly sampling N number of Nnumber of ti~mes. original data points (with replacement) , . .....' I Each'data point may .. ' . be' " sampled more than once or not at all. Sampled data points were tallied and weight . . . . ... , ~ , " coefficients were calculated. These steps were repeated to create 250 coefficient vectors. The 250 je N matri~ of weights was then merged with ~ach original data set to create ~50 , ," I. . . ," bootstrap sampies. Nonlinear fitting procedures generated pairs of parameter estimates , ' , , ' I,,·... ,,'. (B, -R) from the merged bootstrap sampies. These estimates were then transformed I , ' , similar to the jackknife pseudo-values. Univariate statistics were performed on the final . , ".' I . set of transformed estimates, now normally distributed. ., . , . ' I '\ The bootstrap estimator is the average of the bootstrap values, . . . SB :, 2~Ili>b \'. . (6) b=1 I Where b = \ (1, 2, 3, .;. 250) sampie number.. '" '.. I . . ~b = bootstrap value, transformed pair of parameter estimates, (B, -R). A I <eB = . Bootstrap estimator, transformed., \ Because of the larger number of resampling points, tt~et following equation (Duncan, 1978) t was used to create joint 95% confidence intervals from the Means and variances of the . , I !I bootstrap values. { S: (SB - <6»T S ·1 A Where eB S x? 1 _ a • (e I B -8) ~ ~21 _a(P}} (7) I ,I . = Bootstrap estimators, transformed pair of parameter estimates.. , l , . . = variance-covariance matrix of transformed bootstrap values. , . = chi squared value for p degrees of freedom. in m'odedI p = 2, number of parameters . I I \I 4 I, I l. I! • j:inally, bootstrap estimatörs were adjusted tor bias by using the tollowing equaiion (Meyer, et al., 1986): ," A A 8 S'(b.a.) = 29 - A es A Where 8 S (b.a.) = bias adjusted estimates. A e = tull sampie estinüites. = 800tstrap estimator. RESULTS trom e Parameter estimates the jackknife and b06tstrap procedures are summarized in Table 2. The jackknife and hootstrap values tor the estimates oi' 8 and -R were very' slmilar. The shape of herring prey evacuaiion (1Ö g) was approximately 1.0 (exponential model); while prawn evacuation varied from -0.181 to 0.815 depending upon what coristituent was measured. Siopes ranged fram -0.136 to -0.043 for the jackknife and -0.431 to -0.049 tor the bootstrap. The different herring meal sizes (10, 20, arid 30 g) represented 0.70/0,1.2%, and 1.7% wet weight 8W%, respectively (Table 1). The slope parameter was approximateiy the same for the different meals (TAFD weight basis, Table 2); while there was a trend tor decreasing shape parameter with increasing meal size. The larger meal was less curvilinear (exponentlal) and possibly more volume dependent (8 = 0.67). Equations (5) and (7) delineate ellipses in the context of transformed parameters, but the 95% confidence regions become urisymmetrical arid stl'etched sllapes when 8 and -R are untranstormed (Figures 1 and 2). 8ecause of this, a comparison of regions is more descl'iptive than a comparison based upori statistics. All confidence regions were tilted in the same direction with higher shape parameters tavored witll more positive slopes. 95 % confidence regions tor wet weight sampies extended weil beyond valid vaiues of -R and 8, positive values and negative values, respectiveiy, while AFD arid TAFD confidence regioris tor both shrimp and herring prey were almost coincident. . DISCUSSION Jac~knife and bootstrap estimators ware approximateiy the same. It appears that llle jackknife and bootstrap methods worked equally weil in deflnirig gasti-ic evacuation rates. There may be an advantage to ihe jackknife, because they the same N are required for both methöds bui the jackknife is less computer intensive. 5 ,---------~ - - - -- - - - I \ \ I . . . , ' "" a wet Stomach contents weighed on , ' , ' " ,,' I" ,'. t , . . .. _,. "'" " , weight basis do not accurately estimate gastric ,,' evacuation (Hopkins and Larson, 1990). When compared to AFD weight, wet weight '. ,.,,'.;., I ' ' , '" . ' overestlmates the slope whlle underestlmatlng the shape parameter (prawn prey, flgure I ,'.. ,,' , , " 1A,2A). When shrimp prey are digested, the percentage of water in the stomach contents . '., " I , . ' increases, presumably to hydrolyse or break down chitin in the carapace and exoskeleton. " . • I The added water weight smooths a curvilinear funetion to make it appear more linear. If , I the amount of water is large enough, alag phase or delay in the begirining of evacuation . , . . _ , ,I ' .' . " may result. The differences are not as dramatic for herring (Figures 1B, 2B). , > ' , • " '.', ,> '. '. • I· . ' , On the other hand, the AFD weight of storriach contents collected on mesh screens does not accurately estimate gastric evacuation rate as co1mpared to total AFD weight contents • '" • ' I , ' inch..iding rTlElterial <500 flm (Figures 1A, 2A, and Table 2).. A considerable arriount of " ",,' " , I , , " particulate material is consistently being lost in gastric lavage filtrate water. On itie average this material represents only 5% of the original meal, but can reach up to 25%. ' .• ' , ' ', ' ",,' " " " . ' I . , '" " ,', , '" ' . .The amount of material <500 flm is not constant over, the whole digestive process. In the .' ' " . ...,". . ,"'" !' . " " .'.,,, . early stages of digestion thls fractlon does not contnbute very much but by the end can • '.. ." " '. ' , ; , I' .• ,', '." .. ,,' slgmflcantly Influence the evacuatlon curve. Thls would make a seemlngly linear process , . , ',' , : , " , "" " 1 I more curvilinear, as seen with prawn prey (figures 1A; 2A). For herring prey the effects . . ' , ' .t '.' are not as great, although -R is overestimated and B is roughly the same. '::, . . .' ; ,.; ". \, '. , , .. Experimental procedures In prevlous gastrlc evacuatlon studies have not. been , ",' . " , , . ,'", '. ',' I " ," .,. '. , . . .,,' standardized. Different components of stomach content sampies have been used to ." ' .' '. ' '. , , '," .''';.. J . . .,' ; ,, ~ ~. estlmate evacuatlon rates. Stomach contents retneved by gastric lavage are usually ," ' " ". ,'. ,I ,,_ "',, screened to remove excess water, but the mesh size of the screens has varied tram 200 , ' '" ' f , ' " ,1 ' ' .. ·.l l " " Jlm to 1 mm. The traction of material that is c6l1ec~ed on the screens can represent , . ' " . ,,' , ., . I " ' different proportions of the original meal when comparing separate studies. I CONSUMPTION RATES , \ Previous consumption estimates were calculated with shape parameters of 0.27 - 0.47. If . ' '. . ' , ' .'" ' . ,.'. " j" , .", ,.,.., .., .. ' , the gastnc evacuatlon process IS more exponentlai (B ~ 1.0), then the true consumption ,,', . . , , " . ', " " L . ,,-...' . " ',' .,. rates for cod populations are 3 - 5 times higher than those trom Temming and Andersen. I \ I References , I ", I " , , ' ,.' , Bagge, O. 1977. Meal size and digestion in cod (Gadus morhua L.) and sea scorpion (Myoxocephalus scorpius L.). , Meddr. Danm. Fisk.- og Havunders. N.S. 7: 437·446. I . : , " '.,,' , "," "I ' Bromley, P. J. 1991. Gastric evacuation in eod (Gadus morhua L.). ICES mar. Sei. Symp. 193: 93-98. ' . :.' : • . , "" ,.. • • " I ~ <, • " l : i /; dos Santos, J. 1990. Aspects of the eco-physiology of predation in Atlantic cod (Gadus morhua L.). Dr. scient. NFH, University ofTrosma, Norway, 116 pp. I , , ' , I " " ' , , Duncan, G. T. 1978., An empirical study of jackknife-constructed confidence regions in nonlinear regression. Technometries 20: 123·129. 6 Efron, B. 1982. The Jackknife, the Bootstrap and Other Resampling Plans. Society for Industrial and Applied Mathematies, Philadelphia. 92 pp. Efron, B. 1987. Better bootstrap confidenee intervals. J. Am. Stat. Assoc. 82: 171·185. Elashoff, J. D., T. J. Reedy and J. H. Meyer. 1982. Analysis of gastrie emptying data. Gastroenterology 83: 1306·1312. Elliott, J. M. 1991. Rates of gastric evacuation in piscivorous brown trout, Sa/mo trutta. Freshwat. Biol. 25: 297-305. Elliott, J. M. and L. Persson. 1978. The estimation of daily rates of food eonsumption for fish. J. Animal Ecol. 47: 977· 991. . Fox, T., D. Hinkley and K. Larntz. 1980. Jaekknifing in nonlinear regression. Teehnometrics 22: 29-33. Hopkins, T. E. and R. J. Larson. 1990. Gastric evacuation of three food types in the black and yellow rockfish Sebastes chrysome/as (Jordan and Gilbert). J. Fish Biol. 36: 673-681. Jobling, M. 1981. Mathematical models of gastrie emptying and the estimation of daily rates of food eonsumption for fish. Fish. Biol. 19: 245-257. Jones, R. 1974. The rate of elimination of food from the stomaehs of haddock Melanogrammus aeg/efinus, eod Gadus morhua and whiting Merlangius mer/angus. J. Cons. int. Explor. Mer 35: 225-243. Meyer, J. S., C. G. Ingersoll, L. L. MeDonald and M. S. Boyee. 1986. Estimating uncertainty in population growth rates: Jaekknife vs. bootstrap teehniques. Eeology 67: 1156-1166. Miller, R. G. 1974. The jaekknife -- a review. Biometrika 61: 1-15. Robertson, O. H. 1945. A method for securing stomaeh eontents of live fish. Ecol. 26: 95-96. SAS Institute Ine. 1988. SAS I STAT User's GUide, Release 6.03 Edition. SAS Institute Ine., Cary, NC. 1028 pp. Temming, A., and N.G. Andersen. 1992. Modelling gastric evaeuation in eod. ICES Committee Meeting 19921G:61. Tyler, A. V. 1970. Rates of gastrie emptying in young eod. J. Fish. Res. Board Can. 27: 1177-1189. • 7 ,--------- -- I Il • Table 1. Summary of conditions for Atlantic cod gastric evacuation experiments. I I J . Temp. Predator size. Prey type. (±CI) , Meal size, , wet wt. (:tCI) BW% N. wet wt. basis. (±CI) i Prawn 1928± 185 9 10.1 ± 0.3 9 0.56% ± 0.07% 22 0.72% ± 0.08% ,33 20.1 ± 0.8 9 1.21%±0.14% .25 30.4 ± 1.1 9 1.67% ± 0.25% 21 I I 1638 ± 164 9 Herring 10.8 ± 0.5, 9 ! 1789 ± 208 9 Herring 1992 ± 247 9 Herring I 1 I, \ \ Table 2. Summary of results. Estimated slope and shape parameters from bias adjusted bootstrap and jackknife methods. \ I I ~arameters Prey type. Meal size. Sampie constituents. Prawn 10 9 WET Prawn 10 9 AFD (0.690, -0.1\14) Prawn 10 9 TAFD (0.846 , -0.087) Herring 10 9 WET (1.073 , -0.0~3) Jackknife estimate. (B , -R) Bootstrap estimate. I . (0.004 , -0.136) (-0.181 ,-0.431) .I I (0.682 , -0.113) 1 .I (0.815, -0.088) I Herring 10 9 AFD (1.069 , -0.055) 1 (1.216 , -o.0~9) , (1.258 , -0.090) I Herring 10 9 TAFD (1.235 , -o.O~O) (1.237 , -0.071) 1 ! Herring 20 9 TAFD (1.425 , -0.045) (1.180 , -0.049) 1! I Herring 30 9 TAFD (0.698 , -0.059) I I 1 1 I I 8 (0.610 , -0.072) • :0.20 .- 0.10 -----r-r--r.,---,------r-r-'-Il LLkU-i-+-t----t---r-~ 0) a: ::'0.00 0) ',~cu ~0.10.L-l~l/f;~f;4~f~-I~ E -0.10 .. ' .[ -0.20 ib ~AFD .. 0 "TAFD. ,~·-0.30 o 0. ~0.40 0) 0.-0.30 ~0.5 .- 0.10 a: ::' 0.00 0) ~ cu 0. -0.20 ~ -0.30 02, -Cf) ':'0.40 0.0 I I , I I " I I I I I I I I I I I ·-0.50 . . --0.5 0.0 I ' .,-, I 0.5 1.01.5 2.0' 2.5 Shape parameter (8) 3.0 .3.5 I 11 I • •• ' 0 ) - oE c'..c: 3= ~ :;: :'0"0 I/' "'WET • ~AFD ..0· TAFD /. / fI i' / // ~>' , . . .. . . . 2.5 '3.03.5 1.01~52.0 CIS cu: c 11. .2: c."eu.<t:' c I- Cl .~ ~.~. CIS Cl.O) C. 'oc V Ify .... ; 0HERRING ~ -0.20 :0. ~ ' .: V' ..... 0 10.9 meal '0) .0. -0.30 o 'Cf) ~0.40 ---- - -'0 . . '-0.50 . . -0.5 0.0 209 meal ~0309meal . . . . . 0.5 '1.01.5 2.0 ·2.5 Shape parameter (8) ;e . . 3.0 3.5 -T __ 0':>Cl ce'" 'c .0 .. Cf)'O> -= 0 00. c ....0 ,0.'0. Cl t::: 0' E E LO,CUO·.c C ~. cu ~PRAWN ., ... ....... Cf) "0 0'> ... l...----" l--""" .- k-=- -'- . ... .. ... :.~ ... ~ ~. .......... 0,.......... ~ .. /. ::' 0.00 2. ~ -0.10 ~ .0- - 1000""" cCl 0.0) D. Herring prey, TAFD weight basis. a: . 1/ } 'J . 0.5 Shape parameter (8) .- 0.10 1,_ , . 0.20 . / -;7, '1 .. '(1) ..0> . ~ .;:; .•Q) L. C. Ci; Cl .o·~ I' I +J . ~ -0.10 >. .~~. CIS.>' c.c . -0.40 C.10 9 meal size,·WETweight basis. 0.20 0) c.:> 0 ) ' cn .2 _ .... CiS ~ [I) 0..0 .':1 ....... .Q Cf) -0.50~, / y. ......... ...:;:/.W:' ......... .- - ......... :. - _.- - ,V/ 'cu L. "'WET ' cu cu c gE-: > 0 0 /.~ ..- 0.10 0) ~~·-0.20 o:o:w ,:;:; ~' :> cu .. .... ;::> ':J .0.20 '~ O.ooJLJ-i1--+-t=t~E~~ ... Cf) c,cJ- B. Herring prey, 10,g meal size. /A.'Prawn prey,'10g meal size. 'Cf) .;:; cu .- U - Cl __ . - N ß .• 0 u :Ci) . _.c .•.."0 : 0 C Cf) >- ,0) Cf) L.: CiS Cl E 0.' ..".0 '"" - CIS .;:; ... .cu 0. .,9 E··~ c CIS 0 1::0 E c.:>. 0) (I) .:;:; _..c: .;:; Cf),,- CIS 0) -.11:0. _0) "0.,c:E :0) 0 :E .0 .:J Ü .~ 0 ' " ~ :0 ;'(ij.'- o ; : c Cl . 5. 0 •• -, ~ o .. ~ ·cu .. < ~ ,... ~ 0. ,,, .~ :E .c .....~.Cf) 0) ... CIS..c: cn.Ql u:cu-O) .... 0:3: '0') 2 0'> .• Siope parameter (R) Siope parameter (R) , .!=> o !=>.o 0 !=>.o 0 ~ cu N ..... 0 ..... N 01 o 0 0 0 0 0 0 ,0 o I I • I ,", 0.. ~ o o o 000.. ~ 1\.- < ~ ";':"';'" \" ~ ::r Pl ."0 ..... CDO @ 3 ..... 01 (j)1\) - ""'0 - N' ",- \ Öl 0 cu :J: '1J m :D Z Z o 0.. j,O o 0.. 0 ~ 0 , ;..I. CD 0 N 0 0 @ 3 '. - .~.<' \ ..... o :E ~ ce 3 :E CD ~ üS' ::r N' CD cn - ~ C" cn !i5' '-J , 0 0 ..... N 0 0 I I~ \ \ 0 , , 0 ~ 0 0 , , !=> 0..... 0 N 0 .~ 0 0 0 0 ..... 0 0 0 N 0 .m ~ ·· .. · \ \ -I\) CD 3 :~ 3 CD CD e.. e...e.. :: \ . \ \01 :\ ,:. \. I CD 0 ii3 0.. ~~ I ~\ I\) I : I. ~ , ~ "'\\: i..... 3 I I - 0 I : I: :. : \, :. : \ " . I\) , ~9+ i cu o ~ \ I:':' Öl I - ,~ . 05' I \: : ~ :': I!=> -g ·1....... , , ·· .. ~\ ~ <\.. Pl "0 , ·. .... cu I\) 0 0 0 ce ce ce -- 3 , 00 ::r , '~! \ , , I . Öl - - 000 I• 10 \ ,, \ ~ 0 ~ >:E 'Tl I 0 cu v. : t9!:!l ~, ~ .. .. 01 01 , , 01 . Siope parameter (R) !o ~, r::? ~\ \ \ ..... 01 cu \~ II !=>, ",.- 1\. I CD OJ 0 01 -I\) - 0 1 "0 Pl ,0 CD ' 0 ~ wo.. ::r Pl , 0 0 ..... 0 0 0 ,- 0 !=> o o "0 0 0 N CD ~ i>arämeter (R) Siope ,. 0 ..... .j> Ci> , 0 " '0 ~ :D :D o N cu , 0 !=> ..... 0 Pl \ ~ . cn 1\ OJ -I\) ,. 0 o "'- :\ \, CD 0 ce 3 0' 0' (:J , !=> 0 ~ 0 0 01 ..... 1\ \ "0 Pl 0',"- , I I ' : . " , . , .. , '" ... ' .. '.! ,. : --.,', ".. . ..... ,; . Figura 2. Bootstrap estimators and joint 95% confldence Intervals for gastnc evacuatlon rätes in Ätlahtic A) C6mpaHson ~f sampie c6~situents, prawn prey; B) Comparison ' I , " , " •.•• . .•.. " of sampIe cönsituents, herring prey; C) Comparison of prawn and herring prey on a WET weight basis; 0) Comparison of meal size, herring p1reyon a weight basis. cod. '. ' . 0'. ' . , " . ' , , . ' . , •• TAFO 10