21.06.2014 Views

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 ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

the r<strong>and</strong>om imputation described earlier. Even though two-family households were included in<br />

the model for the child-focus parent-child counts, the resulting predicted means were not used.<br />

This was due to the fact that the parent-focus parent-child counts were in the same multivariate<br />

set as the child-focus parent-child counts, <strong>and</strong> the predicted means could not be used in the<br />

imputation of the parent-focus parent-child counts when two families were in the household.<br />

6.4.3.3.1 Parent-Child Counts<br />

Since parent-focus <strong>and</strong> child-focus counts were so closely related, a logical constraint<br />

was placed on donors such that if the parent-focus count was nonmissing <strong>and</strong> nonzero, then the<br />

child-focus count had to exceed 0. Similarly, a nonzero, nonmissing child-focus count required<br />

that the donor's parent-focus count exceed 0. If the child focus counts were missing, donors <strong>and</strong><br />

recipients had to have the same number of household members in the age range corresponding to<br />

the domain of interest. (Donors had to have complete data on all the roster age variables.) The<br />

same constraint was applied if the parent-focus counts were missing but the child-focus counts<br />

were nonmissing, with an additional requirement: It had to be possible that no parent-child pairs<br />

existed in the household. (If it was known that there were parents in the household for the<br />

appropriate domain, it was not necessary to limit donors to have the same child age composition<br />

as the recipient.) These were likeness constraints that were never loosened. In addition, if a<br />

recipient had two family units in the household, a regular hot deck imputation could not be done,<br />

as stated earlier. For all missing counts, the counts from the two pair members (in the case of pair<br />

recipients) <strong>and</strong> the household composition were used to create upper <strong>and</strong> lower bounds, provided<br />

valid roster information was available. These bounds acted as additional logical constraints.<br />

Besides delta, additional likeness constraints all involved the household size <strong>and</strong> additional<br />

constraints on the household composition, which are described in the following paragraph.<br />

An attempt was made to match donors <strong>and</strong> recipients in each of three age ranges that are<br />

commonly associated with children aged 12 to 20: 26 to 34, 35 to 49, <strong>and</strong> 50 or older. This<br />

likeness constraint was applied whether the child-focus or the parent-focus count was missing.<br />

However, its application in the case of a missing child-focus count <strong>and</strong> nonmissing parent-focus<br />

count required an additional condition: It had to be possible that no parent-child pairs existed in<br />

the household. (If it was known that there were children in the household who belonged to<br />

parents, it was not necessary to limit donors according to the parent age ranges.) A looser form<br />

of this constraint was to collapse the 26-to-34 <strong>and</strong> 35-to-49 age ranges into a single age range<br />

<strong>and</strong> drop the 50-or-older constraint. Other household composition constraints required donors<br />

<strong>and</strong> recipients to have the same number of household members younger than 12 years old <strong>and</strong><br />

between 18 <strong>and</strong> 25 (inclusive).<br />

The likeness constraints were loosened in the following order (where applicable): (1)<br />

ab<strong>and</strong>on the neighborhood <strong>and</strong> choose the donor with the closest predicted mean or means; (2)<br />

remove the requirement that donors <strong>and</strong> recipients had to have the same household size; (3)<br />

remove the requirement that donors <strong>and</strong> recipients had to have the same number of household<br />

members younger than 12, between 18 <strong>and</strong> 25 (inclusive), <strong>and</strong> 50 or older, <strong>and</strong> collapse the 26-<br />

to-34 <strong>and</strong> 35-to-49 age constraints; <strong>and</strong> (4) remove the 26-to-49 age constraint.<br />

61

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!