522 
THE WILSON JOURNAL OF ORNITHOLOGY . Vol 123. No. 3. September 2011 
ica? (2) Do Northern Saw-whet Owls exhibit 
inter-annual migration-route fidelity? (3) Do 
large-scale age-differentiated movement patterns 
exist? Answering these questions at this novel 
scale will expand the Northern Saw-whet Owl 
information portfolio and illustrate the versatility 
of collective data sets. 
METHODS 
Data Source, Study Area, and Data Prepara¬ 
tion.—'Me assessed movement patterns using the 
BBL data base of 170,468 Northern Saw-whet 
Owl banding events and 2,741 reports of subse¬ 
quent encounters with banded owls (here after, 
recapture will be used for owls encountered 
post-banding, dead or alive). We examined 
information from 81,584 Northern Saw-whet 
Owls banded in 1999-2008 during fall migration 
between I September and 31 December. We 
assumed this parsing would ensure that nearly all 
records represented migrating owls. Excluding 
pre-1999 records ensured that most owls were 
banded using the audio lure mist-netting tech¬ 
nique described in Erdtnan and Brinker (1997). 
Records exist across North America, but data west 
of the Mississippi River are geographically 
disparate and small in sample size. Thus, we 
restricted the analyses (o records from eastern 
North America. 
Data Analysis. The BBL reports banding events 
as either the exact latitude and longitude of the 
banding location, or the comer coordinates of the 10- 
minute or 1-minute block that a station falls within. 
The data base does not report station or bander 
names, so it is not possible to match all banding 
event coordinates exactly to banding stations 
indicated by Project Owlnet (Huy 2010). Thus, we 
define a banding station’ us any coordinate where at 
least one Northern Saw-whet Owl was banded A 
10-minute block is <20 km wide, so the variation in 
banding coordinate precision is negligible at the 
scale of eastern North America. 
We used a geographic information system 
(GIS) to draw vectors between banding and 
recapture locations for each individual captured 
multiple times, and calculated the spherical 
lengths of each vector. Compass bearings for 
each vector were calculated using a Standard 
Mercator projection designed to represent the line 
between any two points on a sphere as a constant 
azimuth. Vectors do not necessarily follow the 
Z §r oTnZ bUt a,e SUfficien ' for understand- 
8 overall distance and direction-of-travel be¬ 
tween banding and recapture locations. All spatial 
analyses were performed using ArcView 9.3® 
(ESRI 2008). 
Migration Timing.—We subdivided eastern 
North America into lateral bars 01 latitude in 
width (Fig. 1). We aggregated all banding events 
by these bars and calculated the mean Julian 
banding day at each bar. The 01 bars were 
chosen for convenience and were sufficiently 
wide to each contain a representative number of 
banding events. We verified that the mean 
banding day at each latitude bar coincided with 
a peak in migration activity represented by a bell 
curve in frequency distribution of banding days. 
There was a unimodal Gaussian distribution at all 
except four latitude bars south of Virginia. These 
four bars were ultimately excluded from analysis 
due to small sample size. Mean banding day was 
graphed against latitude bar. and against the 
latitude of banding stations with >50 banding 
events. We used linear regression to assess the 
strength of these relationships. Similar analyses 
were conducted to assess differences between 
adult and juvenile owl movements. Mean banding 
days for each latitude bar were compared using 
Chi-square contingency tables. All analyses 
performed using 01 latitude bars were also 
performed in the same manner and over the same 
area using 01 longitude bars to simultaneously 
identify east-west movement patterns. 
We modeled migration timing in eastern North 
America by performing surface interpolations 
based on mean banding days at banding stations 
with >50 banding events. The model used 
inverse-distance weighting (ESRI 2008) of mean 
banding days at stations within a fixed-distance 
neighborhood around each predicted raster cell 
defined as an ellipse of 1.5 latitude and 05" 
longitude in radius (power = 1). The search 
neighborhood was restricted in latitude to limit 
bias of stations unevenly distributed far north or 
south of a given cell. The search neighborhood 
longitude was selected to restrict influence of 
distant stations while being sufficiently inclusive 
to interpolate the entire surface. 
Variation in banding effort among stations 
cannot be calculated with the BBL data base. 
We normalized for banding effort where possible 
by comparing proportional values among stations 
lather than using raw totals, or aggregating data 
by latitude bar instead of banding station. 
We estimated migration speed by plotting 
distance between banding and recapture over time 
