Survey Methodology #2006-02 - Use Of Dependent Interviewing Procedures To Improve Data Quality In The Measurement Of Change - U.s. Census Bureau Page 11
ADVERTISEMENT
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25observed at the seam across the need-based programs summarized in Part 1A is 20 percentage
points, compared to 19 percentage points for the non-need-based characteristics in Part 1B; in
terms of the percent decline observed in 2004 relative to the 2001 baseline the average figures are
31 percent and 30 percent, respectively.
4. The positive effects of DI are a result of reduced change at the seam and increased change
off the seam. As noted earlier, seam bias has been shown to be the net effect of too many changes
observed at the seam and too few changes observed within the months of a single wave’s
reference period (Moore and Marquis, 1989). The DI procedures introduced in the 2004 SIPP
panel directly countered those tendencies. As shown in the “Average Month-to-Month Change
Rates (%)” columns, for every characteristic the off-seam change rate is higher in the 2004 panel
than in the earlier panel, and for 14 of the 15 characteristics the rate of change at the seam in 2004
is lower than in 2001. DI reduced spurious change reports at the seam, and reduced spurious non-
change reports across the months within a wave.
5. SIPP’s new DI procedures acted primarily to reduce spurious “yes-to-no” change at the
seam. As shown in the right-most columns of Table 1, across almost all but two of the
characteristics which were captured with the new DI procedures in 2004, the decline in “yes-to-
no” changes at the seam is greater, in percentage terms, than the decline in “no-to-yes” changes at
the seam. This general pattern makes great sense, because the new dependent procedures
employed in 2004 are “asymmetrical” (Murray, et al., 1991), in that they only apply to those who
are in a “yes” status (enrolled, covered, participating, etc.) at the end of the prior interview –
preventing false yes-to-no change at the seam is what they were designed for.
Given that focus, the question arises as to why the SIPP 2004 DI procedures had any impact at all
on no-to-yes transitions. That they did is obvious: with the sole exception of school enrollment,
where the rate of no-to-yes change at the seam stayed the same in 2004, the amount of no-to-yes
change at the seam was consistently higher in 2001, before the new DI procedures were
implemented. We suspect that the key is the careful targeting of DI in the new SIPP panel,
especially as contrasted with the indiscriminate, quasi-DI approach that it replaced. Recall that
the new DI procedures were only triggered by spells known to be in progress at the end of the
previous wave’s reference period. In contrast to the format used in the 2001 panel, no mention
was made of any spells which were known to have ended before the end of the reference period.
We suspect that the old format’s irrelevant reminders to respondents about already-ended spells
may have served to mask the fact that the respondent had already, in the prior wave, reported the
spell’s termination, thus subtly encouraging them to mis-recall a new spell as a continuation of an
old one – thus resulting in a false no-to-yes change report at the seam. (LeMaitre (1992) also
notes this flaw in the previous SIPP design, and its possible negative consequences.)
6. Despite the improvements due to DI, much seam bias still remains. Improvements in seam
bias due to the addition of DI are unmistakable; the fact that seam bias is far from having been
eradicated is equally unmistakable. With the single exception of school enrollment, every
characteristic with improved measurement of transitions in the 2004 SIPP panel still displays an
overabundance of changes at the seam. For example: in the proportion of all changes that are
-10-
ADVERTISEMENT
0 votes
Related Articles
Related forms
Survey Methodology #2012-02 - New Relationship And Marital Questions: A Reflection Of Changes To The Social And Legal Recognition Of Same-sex Couples In The U.s. - U.s. Census Bureau
Legal
Early Childhood Program Participation, From The National Household Education Surveys Program Of 2012 - U.s. Department Of Education
Legal
Form Hud-9887 And 9887-a - Document Pachagefor Applicant's/tenant's Concent To The Release Of Information - U.s. Department Of Housing And Urban Development
Legal
Related Categories
Parent category: Legal