[DDI-CDG] Re: [DDI-SRG] Longitudinal data proposal

[I-Lin Kuo]

Both Wendy and Sanda's answers have clarified things somewhat for me.
The hypothetical situation is almost what Wendy mentioned --  I'm
thinking about a household survey where you go and ask questions about
the income, education, age, and race of the head of household. If that
person is there, then they answer the question themselves. If that
person isn't there, then the question is asked of someone else in the
household. This information might be recorded in 2 or more ways:

A) 5 variables:  age, inc, educ, race, and
relationship_of_respondent_to_head_of_household.

B) 9 variables: age_according_to_hoh, inc_according_to_hoh,
educ_according_to_hoh, race_according_to_hoh, age_according_to_resp,
inc_according_to_resp, educ_according_to_resp, race_according_to_resp,
relationship_of_respondent_to_hoh.

In the first way, the "respondent" is not well-defined variable-wide
and is unsuitable as a comparison criteria, while in the second way,
the "respondent" is well-defined as either "hoh" or "proxy" for all
records. The second type of recording seems very unusual to me...

There are other longitudinal studies like Wisconsin Longitudinal Study
where in one wave, the respondent is asked to evaluate themselves and
in a separate wave, a sibling or parent is asked to evaluate the
original respondent. In these type of studies, the "respondent" is
also a well-defined comparison criteria, and perhaps this is the kind
of study Sanda had in mind.


On 9/15/05, Wendy Thomas <wlt at pop.umn.edu> wrote:
> Sanda, correct me if I'm wrong, but I believe the answer to I-Lin's
> question is that in a longitudianal study the respondent can change over
> time, particularly if the the sample is one of households where a
> different person might respond from month to month or year to year.