2.4. Analysis
Results from the questionnaire were analysed with IBM SPSS
Statistics for Windows 21.0. Prior to analysis, the 5 skill levels
(1 ¼ complete beginner; 2 ¼ advanced beginner; 3 ¼ moderately
experienced; 4 ¼ very experienced; 5 ¼ experts) for mountain
bikers were re-classified by merging category 1 (‘complete
beginner’) and category 2 (‘advanced beginner’) into one category
(‘beginner’) to account for their smaller sample size.
To compare frequency data, such as the socio-demographics
and rider characteristics (i.e., skill level and field vs. online), Pearson's
chi-square tests were applied. The rating-scale data were
analysed with analysis of variance (ANOVA) including skill level
and sampling mode (field vs. online). ANOVA was also applied to
continuous variables measuring self-estimated average distances
travelled to tracks in the Northern Sydney area. Open-ended
questions were analysed qualitatively by identifying the major
categories/themes emerging from participants' comments.
We used several spatial scales (‘planning areas’) in our analysis
to demonstrate the versatility of PPGIS mapping and associated GIS
analysis including inside vs. outside national parks. Tracks were
assigned to either being located ‘in park’ or ‘outside’ if most (>90%)
of the track length was located in either category, otherwise, a track
was considered located in ‘both’.
Markers collected in the field were transferred by a researcher
onto Google map and then exported together with the online
mapping data for import into a geographical information system
(ArcGIS 10.1). An attribute field was added to the database that
coded whether data had been collected in the field or online. Prior
to the data analysis, we excluded any markers placed outside the
study area (6.8%). A few mountain bikers (5%) were excluded
entirely from the analysis (both the survey and PPGIS mapping)
because they indicated in the survey they never ride in the Northern Sydney area and they placed marker(s) exclusively
outside the study area.
Descriptive tables and maps were produced to showcase the
range of possible data presentation modes. To create maps showing
the varying numbers of markers placed along the 204 individual
tracks, most markers were spatially joined in ArcGIS to the nearest
track (Fig. 3aeb). We also observed that some clusters of markers
demarcated the outlines of tracks not displayed on the map and
have provided an example in Fig. 3c.
The ‘location’ and ‘reason’ markers were typically placed at the
beginning/end of a track rather than somewhere along the track,
and we considered the attribute to apply to the entire track. In
contrast, we performed raster analysis on ‘action’ markers to
identify specific sites within 150m 150mraster cells along tracks
for the actions. Raster analysis can identify specific ‘hotspots’ in the
study area (e.g., for linkages) that require management attention
(Fig. 4).
To relate the total number of mapped attributes to rider skill
levels, we used one-way factor ANOVA. The relative frequency of
different types of markers by skill level and planning area was
examined with Chi-square statistics and standardized residuals
(difference between observed and expected cell counts). Standardized
residuals indicate which attributes were mapped more or
less often than expected (sresid.>±1.96) by skill level and planning
area categories. Chi-square tests were also applied to examine
differences in mapping results by sampling mode (field vs. online).
To examine the underlying reasons for selecting specific locations
for riding, we calculated Generalised Linear Model effects of reasons
to ride on the popularity of specific tracks, measured as the
number of reason and location markers, respectively, mapped inside
or outside of national parks in Northern Sydney.
GPS tracking data were aggregated to calculate the average
number of rides, distance covered, duration of rides, velocity,
weekly/daily activity patterns, and location of rides during the
Fig.4-weekly sampling periods. We examined how some of these trip
variables varied by rider skill level.
We compared the popularity of the 204 tracks displayed on the
PPGIS map with the results from the GPS tracking by spatially
joining the tracked rides with the displayed tracks.We then ranked
the tracks by popularity, measured by the number of markers (for
mapping) or number of tracked rides (for tracking) joined with the
displayed tracks, and calculated Pearson correlation coefficients to
determine how closely reported track usage (mapping) related to
actual track usage (tracking).
We refer to the GPS tracked routes as “rides” in contrast to PPGIS
tracks. Some of the individual rides collected by tracking were
exclusively located along the 204 PPGIS tracks. However, some
rides also contained segments that connected with starting points
at home, creating whole networks of rides.