Add info about zero-person households

docs/application_configuration.rst

@@ -81,7 +81,7 @@ Seed sample
 
 As mentioned in previous section, the seed sample is typically obtained from the ACS PUMS. One of the main requirements for the seed sample is that it should be representative of the modeling region. In case of ACS PUMS, this can be ensured by selecting PUMAs representing the modeling region both demographically and geographically. PUMA boundaries may not perfectly line up against the modeling region boundaries and overlaps are possible. Each sub-seed geography must be assigned to a Seed geography, and each Seed geography must be assigned to a Meta geography.
 
-The seed sample must contain all of the specified control variables, as well as any variables that are needed for the travel model but not specified as controls. For population groups that use completely separate, non-overlapping controls, such as residential population and group-quarter population, separate seed samples are prepared. PopulationSim can be set up and run separately for each population segment using the same geographic system. The outputs from each run can be combined into a unified synthetic population as a post processing step.
+The seed sample must contain all of the specified control variables, as well as any variables that are needed for the travel model but not specified as controls. For population groups that use completely separate, non-overlapping controls, such as residential population and group-quarter population, separate seed samples are prepared. In the ACS PUMS datasets, it is possible to have zero-person households in the raw data table (`NP = 0`); these records must be filtered from the seed data. PopulationSim can be set up and run separately for each population segment using the same geographic system. The outputs from each run can be combined into a unified synthetic population as a post processing step.
 
 Finally, the seed sample must include an initial weight field. The PopulationSim algorithm is designed to assign weights as close to the initial weight as possible to minimize the changes in distribution of uncontrolled variables. All the fields in the seed sample should be appropriately recoded to specify controls (see more details in next section). Household-level population variables must be computed in advance (for e.g., number of workers in each household) and monetary variables must be inflation adjusted to be consistent with year of control data (e.g., Household Income). The ACS PUMS data contain 3 or 5 years of household records, where  each record's income is reported in the year in which it was collected. The ACS PUMS data includes the rolling reference factor for the year and the inflation adjustment factor, these must be used to code each household's income to a common income year.