Problems testing typological correlations with the online WALS

The ease with which WALS allows users to combine features from two maps and determine numbers of languages of the resulting types means that there is a danger of misusing the data from WALS to arrive at unsupported conclusions regarding typological correlations. I examine two instances where the overall numbers suggest a correlation and show that in only one of the two instances is there any reason to believe that there is in fact a correlation. In the case where the apparent correlation turns out to be an illusion, namely between tone and the order of object and verb, the illusion arises because most of the tone languages in WALS are in two areas which happen to be primarily VO. This illustrates the need to examine how the languages are distributed geographically. But this is information that WALS also provides, on the maps.