The first stage was to determine whether the building location interacts with sign/baseline conditions.
If it does, then this result would mean the relative effectiveness of the sign conditions is not consistent across locations.
To examine this possibility, a 9 (sign/baseline periods) x 6 (buildings) between-subjects ANOVA was performed.
Although ANOVAs are not usually the preferred method to analyze dichotomous data, Cochran (1950) concluded that ANOVA is a valid, reasonably robust test of binomial data.
In this initial stage of analysis, it is used as a way of examining the simultaneous influence of sign and building.
The ANOVA showed
that the interaction was not significant, F(40, 737) < 1.0,
indicating a consistent pattern of compliance rates for
the sign/baseline conditions across all of the buildings.
Given the lack of sign/baseline x building interaction,
the data are collapsed across buildings in subsequent
(second and third stage) analyses.