We compared breeding bird metrics from 2005 to 2006
among the five treatments (young clearcut, young two-age,
old clearcut, old two-age, and unharvested) with two-way
nested analysis of variance (ANOVA), which is robust to conditions
of non-normality (Zar, 1999) and robust to moderate
departures from homogeneity of variance (Dowdy et al.,
2004). By examining probability plots, we found that the residuals
of the raw data did not deviate excessively from normality.
The ANOVA model included stand nested within
treatment as the error term to examine effects of treatment,
year, and treatment by year interactions on the dependent
variables: species richness, Shannon diversity, relative abundance
of each species detected in >30% of stands, and each
habitat guild. We used Waller–Duncan k-ratio t-tests to find
which treatments differed when ANOVA was significant and
a sequential Bonferroni test (Rice, 1989) to control for experiment-
wide error rates by adjusting alpha levels for each test.
When there was a significant treatment by year interaction,
treatments were tested separately for each year.