Powered by DM Albums™

Data analyses The number of nest sites varied from year to year. During the study there were as many as 19 available sites in a given year, 7 of which were selected at least once. However, examination of the data suggested that site selection from year to year was not independent; specifically, site selection in a given year appeared to be conditional on site selection during the previous year. We assumed that site selection was only a function of the previous year (ie Markov process). We chose nest site to be the experimental unit; yearly site selections were then considered repeated measures within a unit.  No consideration was given to the exact pair of birds that were nesting. Correlation of yearly site selection The correlation from year to year was determined empirically from the data.  To determine this correlation, we examined the site selection pattern for 2002-2006.  The study began in 2001; therefore there was no potential for correlation with the previous year. This implied that there were 24 site selections during years 2002-2006.  Twenty-one of these 24 possible site selections resulted in the use of a nest site that was used the previous year, yielding a conditional annual probability of site re-use of 0.875 (21/24).

Because of the data characteristics, standard assumptions such as normality, homogeneity and independence, were largely invalid.  The exceptions were the variables distance to nearest active nest at time “t” and time “t-1″, which also varied from year to year at any given site depending on actual patterns of site use; all other site variables were considered constant during the study.

We used two-sample t-tests (Sokal and Rohlf 1981) to evaluate differences in distances to nearest active nests.  For all other variables, our solution was to conduct exact tests based on the distribution of a given test statistic.  First, we proposed a meaningful measure of the data (test statistic) and calculated its value using the actual data.  Second, we used a computer algorithm to simulate multiple replicate data sets.  Third, we calculated the proposed test statistic for each of the replicate data sets.  Finally, we determined the p-value for the actual test statistic based on the distribution of test statistics from the replicate data sets.  This type of simulation approach accounted for all of the characteristics of the data and the design of the study without having to make unwarranted assumptions about any given test statistic.

Simulation method
A Monte Carlo simulation method was used to calculate univariate tests of significance to detect relationships between site selection and nest site variables.  The method involved simulating parrot site selection preference under the assumption that all sites were equally desirable for nesting.

The simulation was started by randomly assigning the 5 nesting pairs to the available nest sites in 2001. The selection of nest sites in 2002 was conditional on which sites were selected in 2001.  If a given site was selected in 2001, then the probability that it would be selected in 2002 was 0.875.  The first step was to determine how many sites were re-selected, thus determine how many new sites would need to be selected.  If new site selection was required, the new sites for 2002 were then randomly determined from the available sites that were not selected in 2001.

The end result is that 5 sites would be used in 2002.  This process was repeated for all years.  For 2005 to 2006 the number of nesting pairs dropped from 5 to 4.  To simulate this occurrence, one nest site that was used in 2005 was randomly dropped for 2006.  The remaining 4 sites then used the same simulation process as described above.

The test statistic used was the sum of the values for any given nest site covariate for all 29 selected sites.  First, this sum was first calculated for the actual data.  Next, the sum was then determined for each of 10,000 replicate data sets.  In this manner an empirical distribution of the test statistic was created.

The p-value for the actual test statistic was determined by calculating the proportion of observations from the simulation that were more extreme than the sum from the actual data.  Because this is an exact test there are no degrees of freedom.  Due to small sample sizes and associated reduced statistical power, we considered differences significant at P ≤ 0.10 to reduce Type II error (Sokal and Rohlf 1981, Taylor and Gerrodette 1993).  We report all means with standard error.

RESULTS

Habitat and spatial data (Table 1) were obtained at 19 nest sites, of which 7 were classified as “used” and 12 as “unused” sites.  Site elevation was not a significant factor in site selection (P = 0.84), nor was distance to nearest active nest at time “t” (t = 0.19, df = 31, P = 0.85) or “t-1″ (t = 0.23, df = 29, P = 0.82).  There was no difference in height of nest tree (P = 0.94), diameter of nest tree (P = 0.18), height of nest entrance above ground (P = 0.82), or distance below canopy (P = 0.38) between selected and nonselected nest sites.

Site selection also was not a function of distance from nest entrance to nearest horizontal (P = 0.32) or vertical (P = 0.95) branch, or distance to nearest snag (P = 0.96).  However, nest entrance aspect was a factor in site selection, with the East/West aspect being significant (P = 0.04), while the North/South aspect lacked significance (P = 0.64).  Nest entrances at selected sites were oriented primarily (5/7) in a westerly direction, with those at unused sites oriented primarily (8/12) in an easterly direction.

Ground slope was not a significant predictor of site selection (P = 0.62), although nest entrance aspect relative to site slope was marginally significant (P = 0.10) at used and unused nest sites.  Nest entrances at selected sites mainly (6/7) faced downhill, whereas nest entrances at other sites tended (7/12) to face uphill.  Canopy cover also was a significant (P = 0.01) factor in site selection, as was horizontal visibility in the rear hemisphere (P = 0.01).  However, horizontal visibility in the frontal hemisphere did not differ (P = 0.30) between used and unused nest sites.

Openness relative to adjacent canopy trees also was significant (P = 0.08) in site selection.  Overall, used nest sites were characterized by greater outward visibility from the nest entrance, both horizontally and vertically, than unused sites.  There was no difference (P = 0.93) in the number of woody stems (>10 cm dbh) present within 15 m, although the number of mature sierra palms was greater (P = 0.04) at selected sites compared to other sites.  Nearly twice as many mature palms were found in proximity to used nests as unused nests (Table 1).  Finally, nest site selection did not appear to be a function of the number of canopy-emergent trees within 100 m (P = 0.77).