Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
forecast year. The benefits in the forecast year of more size categories would probably be much greater if ARC could forecast a household size distribution rather than an average household size. Employment Employment is controlled as number of workers in a household, in four categories, with TAZ- level control in the base year. In the forecast year, the control is enforced only for the region overall, and only average number of workers per household is supplied. Therefore, the forecast- year control is really quite weak. In the base year, the controlled worker categories are extremely accurate and precise, and uncontrolled subcategories are noticeably less accurate and precise. In the back- cast, it is apparent again that inaccuracy is induced by deriving controls from averages, and the quality of the largest uncontrolled category (three workers) is not worse than the controlled categories. In addition, the use of the regional control degrades tract- level precision. There would possibly be much to gain from trying to estimate a distribution of households by number of workers rather than only an average number of workers per household, and the benefits would probably be much greater if this could be done for geographic units smaller than the region. The employment status of each person is also in the synthetic sample. In the base year, the categories of (a) employed civilian and (b) not in labor force are precise and accurate at the PUMA level, apparently because of the household- based employment controls, although they are extremely imprecise at the tract level. The back- cast validation values taken from the census are incorrect (for a yet undetermined reason), so the validity of personal employment in the back- cast cannot be determined. For employed persons, the category hours worked per week is also recorded in the synthetic population. The category working 35 or more hours per week is extremely accurate and precise, even at the tract level, in the base year and the back- cast. For the categories of 15 to 34 h per week and 1 to 14 h per week, the results are imprecise and inaccurate at the tract level but reasonably accurate and precise at the PUMA level, in both the base year and the back- cast. Age Age is controlled only in Version 316, with three cate- gories on the basis of whether a householder is over or under age 65 and, for those under 65, whether a house- holderâs own children under the age of 18 are present. For the forecast year, the control is supplied only as the regional sizes of the subpopulations age 65+ and less than 15. The forecast control categories are sized by using relationships in the base- year census PUMS data between the available values and the needed control cat- egories. In the base year, the controls noticeably improve the accuracy and precision of the corresponding house- hold categories. However, in the back- cast, Version 316 controls provide no apparent improvement in accuracy or precision over the uncontrolled Versions 52 and 128, and, for the presence of children under the age of 18, Version 316 is less accurate. Apparently, the method of transforming the population estimate into the control category induces bias in the forecast. Examination of base- year validation statistics for age categories of persons in the synthetic population shows that controlling households by age of householder in Version 316 may provide a small improvement in the accuracy and precision of person ages for the major cat- egories of 0 to 17, 18 to 64, and 65+. For detailed age subcategories, results are unusably inaccurate and impre- cise at the tract level; at the PUMA level, precision and accuracy are more acceptable. In the forecast year, Ver- sion 316 results differ from those of the other versions but with similar accuracy and precision, and for all ver- sions the quality degrades somewhat from the base year but not a lot. Family Family is controlled in Versions 128 and 316 but only in the base year. In the base year, the controls improve pre- cision and accuracy, but without the control, precision and accuracy of Version 52 are still quite good, even at the tract level. The base- year controls appear to have lit- tle carryover effect in the forecast, in which the uncon- trolled categories are no better in Versions 128 and 316 than in Version 52. Precision remains fairly high, but accuracy gets considerably worse. Housing Type and Ownership Status The accuracy and precision of these variables, which are completely uncontrolled in the base year and forecast, might nevertheless be considered good enough to be usable at the PUMA level of geography, except for the tiny category of mobile home dwellers. The tract- level results are too inaccurate and imprecise to be usable. Gender Although there are no controls related to gender, the number of males and females are fairly accurate and pre- 60 INNOVATIONS IN TRAVEL DEMAND MODELING, VOLUME 2
