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.
collapsed with the two-parent households. For these multiplicity counts, the fitted models were<br />
binomial logistic regression models. Only respondents who had a nonimputed pair relationship<br />
with nonmissing multiplicity counts was eligible for the model-building dataset.<br />
The other responses (parent-focus parent-child <strong>and</strong> sibling-sibling multiplicity counts)<br />
were counts, where Poisson regression models were used. However, the data were<br />
underdispersed for a Poisson distribution so that the data had to be scaled using the observed<br />
variance.<br />
Determination of predicted means. Although models were built using respondent pairs<br />
where the multiplicity was known definitively, predicted means were required for all pair<br />
domains where imputation was required. Once the models were fitted, predicted means were<br />
determined for both respondent pairs <strong>and</strong> nonrespondent pairs, using the parameter estimates<br />
from the models.<br />
6.3.3.3 Constraints on Hot-Deck Neighborhoods <strong>and</strong> Assignment of Imputed<br />
Values<br />
In the same manner as with the pair relationship imputations, donor pairs in the hot-deck<br />
step of PMN for these multiplicity domains were chosen with predicted means, if possible,<br />
within delta of the recipient pair's predicted mean. The value of delta varied depending on the<br />
value of the predicted mean. The values of delta for predicted probabilities are shown in Table<br />
6.8.<br />
Wherever necessary <strong>and</strong> feasible, logical <strong>and</strong> likeness constraints (as defined in Section<br />
6.2.4.3) were placed on the membership in the hot-deck neighborhoods. The constraints are<br />
described below. The hot-deck steps are described separately for each of the variables in turn.<br />
<strong>Pair</strong> relationship constraint. The pair relationship between the donor pair <strong>and</strong> the<br />
recipient pair had to be the same. A pair could not be a donor if its pair relationship was imputed.<br />
This constraint strengthened the requirement of matching pair relationships to include the<br />
restrictions on the ages. For example, donor pairs <strong>and</strong> recipient pairs within the domain involving<br />
12- to 17-year-olds included both 12- to 14-year-olds <strong>and</strong> 15- to 17-year-olds. This constraint<br />
was applied to all multiplicity domains.<br />
Number of parents aged 26 or older constraint. This constraint applies to the parent-child<br />
child-focus domain. Donor <strong>and</strong> recipient pairs had to have the same number of individuals in the<br />
household aged 26 or older, provided information was available for the recipient pair. Donor<br />
pairs had to have complete data on all of the household composition age count variables. If the<br />
recipient pair had only one person in the household in the age range, then the total number of<br />
parents in the household could still have been two if the other parent was younger than 26 years<br />
old. The constraint ensured that donor <strong>and</strong> recipient pairs had the same household age pattern<br />
<strong>and</strong> was never removed.<br />
Number of household members aged 26 or older constraint. Donor <strong>and</strong> recipient pairs<br />
had to have the same number of household members within the age ranges of 26 to 34, 35 to 49,<br />
<strong>and</strong> 50 or older. This constraint was applied for the parent-child, child-focus domain.<br />
49