Questionnaire Dwelling Unit-Level and Person Pair-Level Sampling ...
Questionnaire Dwelling Unit-Level and Person Pair-Level Sampling ...
Questionnaire Dwelling Unit-Level and Person Pair-Level Sampling ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
6.2.4.3 Constraints on Hot-Deck Neighborhoods <strong>and</strong> Assignment of Imputed<br />
Values<br />
If possible, donor pairs in the hot-deck step of PMN were chosen with predicted means<br />
within delta 17 of the recipient pair's predicted mean(s), where the value(s) of delta varied<br />
depending upon the value of the predicted means. In this case, delta was defined as 5 percent of<br />
the predicted probability if the probability was less than 0.5, <strong>and</strong> 5 percent of 1 minus the<br />
predicted probability if the probability was greater than 0.5. This allowed a looser delta for<br />
predicted probabilities close to 0.5 <strong>and</strong> a tighter delta for predicted probabilities close to 0 or 1.<br />
The range of values for delta across various predicted probabilities is shown in Table 6.8. If no<br />
donor pairs were available with predicted means within delta of the recipient pair's predicted<br />
mean, the neighborhood was ab<strong>and</strong>oned, <strong>and</strong> the donor pair with the closest predicted mean was<br />
chosen.<br />
Table 6.8 Values of Delta for Various Predicted Probabilities<br />
Predicted Probability (p)<br />
Delta<br />
p < 0.5<br />
0.05*p<br />
p > 0.5<br />
0.05*(1-p)<br />
In general, the members of the neighborhoods were restricted to satisfy two types of<br />
constraints: "logical constraints" <strong>and</strong> "likeness constraints." Constraints that made the imputed<br />
values consistent with preexisting values of other variables were called logical constraints <strong>and</strong><br />
were required for the c<strong>and</strong>idate donor pair to be a member of the neighborhood. Likeness<br />
constraints were implemented to make donor pairs <strong>and</strong> recipient pairs as much alike as possible.<br />
Although logical constraints could not have been loosened, likeness constraints could have been<br />
loosened if they had forced the donor pool to be too sparse. Details of these imputation<br />
procedures are provided in Appendix N.<br />
In addition to the likeness constraint defined by delta, other likeness constraints also were<br />
included in the neighborhoods. These constraints follow:<br />
Older pair member age constraint, 26+-year-old pair members. The 26+ age group,<br />
associated with age group pairs 7 through 10, was split up into three groups: 26 to 34, 35 to 49,<br />
<strong>and</strong> 50+. This was most useful to delineate child-parent pairs.<br />
Marital status constraints. Each respondent's marital status, as entered in the core section<br />
of the questionnaire, was closely related to the relationship between the pair members. 18 This<br />
marital status variable had four levels among respondents aged 15 or older: married, widowed,<br />
separated or divorced, <strong>and</strong> never married. Marital status likeness constraints combined the<br />
information from this variable for both pair members, where the levels were collapsed in<br />
17 "Delta" refers to the value that defined the neighborhood of donor pairs that were "close" to the recipient<br />
pair. The difference between the predicted mean of the recipient pair <strong>and</strong> the predicted means of the donor pairs<br />
must have been less than delta. See Appendix N for more details.<br />
18 <strong>Pair</strong>s that included a pair member with an imputed marital status were not eligible to be donor pairs. If a<br />
recipient pair had a pair member with an imputed marital status, then donor pairs had any marital status, unless one<br />
of the pair members in the recipient pair had a nonimputed marital status indicating married, widowed, or divorced.<br />
42