Areal units also have the effect of hiding the distribution of tweets inside them. In this case study, the number of tweets per area varied from 2,610 to 191,415. A lower number of tweets per area will result in less reliable topic prediction. This makes area unit selection important for two reasons. First, readers of the map need to be made aware of the discrepancy between the highest tweet count areas and the lowest. Adding background shading to indicate the relative tweets was considered for this method, but it was ultimately not used, as the resulting images became too complex. The second issue concerning tweet counts and areal units is to ensure that the correct scale is used. When the areal unit was smaller than the neighborhood level, the effect of data density wasmagnified as some areas had too little data to be adequate predictors of topics. When the area units were much larger than neighborhoods, the topic model suffered from becoming too generic, offering little information about place. With larger area units, this method might be more appropriate with limiting tweets and topics to a narrower temporal resolution.