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.

should be believed. This decision depended upon the "quality of the roster," where the household<br />

size was determined by the pair member with a better "roster quality." One obvious way to<br />

measure roster quality was by noting the number of cases where the ages, relationship codes, or<br />

genders were missing in the roster. Clearly, if a roster was missing one or more of these three<br />

variables for some of the roster members, the roster was of "poorer quality" than a roster with<br />

these variables nonmissing for all roster members.<br />

If only one household member was selected as a respondent, known colloquially as a<br />

"nonpair household," the rules for creating HHSIZE were the same as those that were used if two<br />

household members were selected in a pair, but only one of the pair members had a nonmissing,<br />

acceptable value for a reported household size, with one important exception. If only one<br />

household member was selected as a respondent, it was obviously permissible to have a reported<br />

household size of 1, whereas in a selected pair a reported household size of 1 was considered<br />

"bad data," necessitating the use of the screener household size as the source variable for<br />

HHSIZE.<br />

In summary, the variables used to determine HHSIZE included, for each pair member,<br />

the reported <strong>and</strong> edited household sizes, the number of cases with valid ages in the roster, the<br />

number of cases with valid ages with the count in some age categories replaced by the minimum<br />

possible in that age category, <strong>and</strong> a quality of roster count of the number of roster members with<br />

missing information. The screener household size, which was the same for each pair member,<br />

also was used. Using all of these tools, HHSIZE did not have any missing values in the 2006<br />

survey, nor did it have any in surveys from previous years. General points about the creation of<br />

the household size variable are provided in Appendix P.<br />

6.2.3.2 Household Composition Age Count Variables<br />

It would seem logical to assert that the ages of other household members would be good<br />

predictors for the domain to which a pair might belong. Such variables also would be important<br />

for imputing multiplicity <strong>and</strong> household-level domain counts. The household-consistent age<br />

counts were limited to the following age ranges: younger than 12, 12 to 14, 15 to 17, 12 to 17, 12<br />

to 20, 18 to 25, 26 to 34, 35 to 49, <strong>and</strong> 50 or older. These variables were called AGE011,<br />

AGE1214, AGE1517, AGE1217, AGE1220, AGE1825, AGE2634, AGE3549, <strong>and</strong> AGE50P,<br />

respectively.<br />

The first step in this process was to count the nonmissing ages for roster members in the<br />

household for each pair member. In some cases, it was necessary to adjust the count since the<br />

ages could not be matched exactly. For example, suppose a 38-year-old <strong>and</strong> a 17-year-old were<br />

interviewed <strong>and</strong> the 17-year-old was interviewed first. Suppose also that the 17-year-old turned<br />

18 (i.e., had his 18 th birthday) before the 38-year-old was interviewed. Hence, the 17-year-old<br />

would have had an age of 18 in the 38-year-old's roster. Because the ages for the pair domains<br />

were defined at the time of each pair member's interview, the ages of interest for pair domains<br />

would have been 17 <strong>and</strong> 38. Hence, it was necessary to account for this by creating a new roster<br />

age variable that matched the age provided in the other pair member's questionnaire. The age<br />

counts using this new roster age variable were equivalent to subtracting 1 from the previously<br />

obtained 18- to-25 count <strong>and</strong> adding 1 to the previously obtained 12-to-17 count in the 38-yearold's<br />

roster. These adjustments were made for all cases where a match was made between one<br />

36

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

Saved successfully!

Ooh no, something went wrong!