We present field performance measurements of 80 rooftop systems in the Tucson, AZ region. We describe a framework that enables identification the causes of system performance variations. We can distinguish between shading, module orientation, outages, and weather conditions from the performance data alone, i.e., without physical inspection of the system. The modules used in the systems that we study are predominantly from 2 manufacturers: Sunpower and Schott, with two types of modules from each manufacturer. We show that the variation in system performance due to these four module types is smaller than the variations due to other system-level effects. We discuss the distribution of de-ratings due to shading, and de-ratings due to clouds, as well as the significance and duration of outages. We also examine the effect of system age on annual final yields.