134 
Fishery Bulletin 112(2-3) 
(A) Capture and release location (represented by the star in the upper Roanoke River) of tagged Striped Bass ( Morone saxa- 
tilis) during the period of 1991-2008 and reference map for waterbodies in coastal North Carolina. (B) Geographic areas of 
recapture used in data analyses: 1) Albemarle Sound estuary (area shaded in gray), 2) Pamlico Sound estuary (area shaded 
in black), 3) North Carolina ocean waters (box 3), and 4) northern coastal waters (box 4). 
tions could be known to occur during a given year. 
This restriction allowed movements (recapture area) 
to be directly related to stock abundance, which was 
estimated on an annual basis (i.e., each calendar 
year) from 1991 to 2008, the terminal year in the as- 
sessment. In addition, restriction of returns to a rela- 
tively short time period at liberty (<9 months) mini- 
mized the opportunity for growth between tagging 
and recapture, thereby ensuring that fish lengths at 
tagging (the size variable used in our analyses) were 
representative of the size of fish when movement 
occurred. 
To reach another recapture area (outside the Albe- 
marle Sound estuary), tagged fish would have had to 
travel a considerable distance (>300 km) from their 
release site in the upper Roanoke River. Therefore, to 
reduce the likelihood of underestimation of fish move- 
ment, we excluded tag returns from the first 14 days 
at liberty, affording tagged fish a more realistic period 
of time to complete movement or migration to another 
system. Indeed, the earliest tag return from outside 
the Albemarle Sound estuary (in North Carolina ocean 
waters) occurred at 16 days after tagging, providing 
justification for our 14-day exclusion window. Finally, 
data from 1994 were excluded from analyses because 
of reduced tagging efforts in that year (only 9 fish were 
tagged and 1 returned). 
To determine which explanatory variables affected 
movements of Striped Bass and to assess their rela- 
tive importance, we used an information-theoretic ap- 
proach. A multinomial logistic regression model was 
run for each of the 5 possible combinations of explana- 
tory variables: 1) length, abundance, lengthxabundance 
(interaction model), 2) length and abundance, 3) length 
only, 4) abundance only, and 5) intercept only (no ef- 
fects model). Akaike’s information criterion (AIC) val- 
ues were obtained for each candidate model. We con- 
sidered the model with the lowest AIC value as the 
most parsimonious or “best,” but we also computed ad- 
ditional diagnostics, Akaike differences (A;) and Akaike 
weights (w{), to assess how other models performed in 
comparison to this single best model (Burnham and 
Anderson, 2002). The first of these other diagnostics 
was calculated as 
