For more information on the main features and limitations of the statistical methods mentioned in the following sections, see Statistical Exploration in the Methods section.

Number of Species (in Hexagons)
A statistical analysis using stepwise regression indicated the following variables, of 66 examined, are significantly predictive of avian richness in Oregon at the hexagon scale. Together they explained 70% of the variance in species richness among hexagons:

Because the first and fourth of these variables are not independent, the first was removed and the regression rerun. The resulting model explained 61% of the variance. When the role of observer effort was discounted by also removing the "time spent atlasing" variable from the model, the resulting model accounted for 50% of the variance and included only the following variables:

Our classification tree analysis indicated the following variables are significantly predictive of avian richness in some groups of hexagons, explaining about 78% of the variance overall (variables not identified by the regression analysis are shown in bold):

When the "effort" variables (variables 2 � 4) and "number of species as predicted by types of land cover present" variable were removed, the resulting classification tree (variance explained = 67%) identified the following variables as significantly explanatory in some groups of hexagons:

An earlier study by Cablk (1997) similarly analyzed correlates of avian richness in hexagons across Oregon using a classification tree method that coincidentally also explained 67% of the variance, but (a) examined a different set of potentially explanatory variables, (b) used predictions of avian richness, not actual data as the dependent variable, and (c) used a slightly different classification tree method (CART). From her data, Cablk identified the following variables as most predictive of avian richness at the hexagon scale:

Unfortunately we were not allowed access to data for these variables for use in independent testing.

Number of Species (in Squares)
The regression analysis indicated the following variables, of 19 examined, are significantly predictive of avian richness in Oregon at the square (25 sq. km) scale. Together they explained 54% of the variance in species richness among squares:

The climate variables listed above accounted for about 12% of the variance. After dropping the observer variable (total hours), the resulting model accounts for 24% of the among-square variance in the number of species found.

The following variables, found to be significantly predictive of avian richness in the hexagons, were not analyzed in the squares due to lack of appropriate data:

Of 19 variables examined, the classification tree analysis identified only one variable � number of years a square was atlased � as explaining the number of species found per square. If squares were covered for 4 or 5 years, significantly more species were found than if squares were visited during 3 or fewer years.

Because the first variable is not truly independent, it was removed and the resulting model explained 36% of the among-hexagon variance in richness.

Our classification tree analysis indicated the following variables are significantly predictive of the "percent completion" in some groups of hexagons, explaining about 67% of the variance overall (variables not identified by the regression analysis are shown in bold):

Road density was especially predictive of the percent completion in hexagons with low predicted number of species, presumably because roads facilitated access and thus survey efficiency.

Percent of Predicted Species Not Found (in Squares)
Each square had a target list of species that observers attempted to find. The percent of these species they failed to find was compared with 19 potentially explanatory variables, and the following were found to explain 83% of the variance in this "commission error" among squares:

Because the monthly temperatures listed above are not independent, the ones for May and June were dropped and the regression was re-run, resulting in a model that was nearly as good (81% of the variance explained).

Classification tree analysis indicated the following variables are significantly predictive of the "percent of predicted species not found" in some groups of hexagons, explaining 88% of the variance overall (variables not identified by the regression analysis are shown in bold):

Number of Species Unexpectedly Found (in Hexagons)
A statistical analysis of the number of species unexpectedly found per hexagon � that is, species not predicted based on land cover and geography, also called "omission errors" -- identified the following variables as significantly predictive but together explain only 37% of the variance:

After dropping the observer variable (total hours) and the first variable (which is probably not independent), the resulting model accounts for a mere 11% of the variance.

Our classification tree analysis indicated the following variables are significantly predictive of the "number of species unexpectedly found" in some groups of hexagons, explaining about 53% of the variance overall (variables not identified by the regression analysis are shown in bold):

Number of Species Unexpectedly Found (in Squares)
Regression analysis identified no variables that are statistically related to this attribute.

Classification tree analysis indicated the following variables are significantly predictive in some groups of squares, but explain only 21% of the variance overall:

Percent of Found Species That Were Unexpected (in Hexagons)
When the above was expressed as a percent of the total species found per hexagon, the regression analysis identified the following variables as significantly predictive, although accounting for only 27% of the variance:

Dropping the observer variable (total hours), the resulting model accounts for only 23% of the variance.

Our classification tree analysis indicated the following variables are significantly predictive of the "percent of found species that were unexpected" in some groups of hexagons, explaining about 43% of the variance overall (variables not identified by the regression analysis are shown in bold):

Percent of Found Species That Were Unexpected (in Squares)
The percent of all of a square�s species that were unexpected was compared with 21 potentially explanatory variables. The following were identified as statistically significant, accounting for 32% of the variance:

Classification tree analysis indicated the following variables are significantly predictive of the "percent of found species that were unexpected" in some groups of hexagons, explaining about 51% of the variance overall (variables not identified by the regression analysis are shown in bold):

Percent of Species Found in Hexagon and Also Found in Square
Although all observations from the squares were included in the associated hexagon data, not all species observed in the hexagons were also found in the squares. The "percent overlap" was calculated for each hexagon. Regression analysis identified the variables most predictive of this percent overlap as being:

Together, these accounted for 84% of the variance in the "percent overlap" variable. Classification tree analysis also indicated that the number of species found in both the hexagon and square was predicted mainly by the number of species found in the square, but especially in the range of 36-42 overlapping species, was predicted as well by the number of species found in the hexagon.