> I was trying to point out that
> totals will always be under-estimated by surveys of this sort
> for a whole
> host of reasons. To overcome this, one would have to tackle multiple
> issues of frame coverage, non-response, under-reporting,
> mis-classification, etc, simultaneously.
I don't think that is right. The point of proposing the use of admin
records as a sampling frame is to get as precise as possible data on, for
example, JSA claimants. Using admin records would also deal with many
problems of mis-classification (between claimants and non-claimants) and
under-reporting (of JSA).
It could also deal with some of the very bad features of the LFS
questionnaire design. One of the reasons the LFS undercounts claimants is
that the crucial question is not asked until late in the questionnaire.
Question 195 in the 1996 version. At that stage of the interview a question
on JSA is 'sensitive'.
The French do it better. The first question in the Enquette sur l'Emploi'
is about economic status - with registered and unregistered unemployment as
precoded options. The LFS questionnaire by contrast never uses the term
unemployment. The LFS is actually vulnerable to the charge that it is not
ethical because it obtains information without informing respondents of the
purpose of asking the question.
Another reason why the LFS undercounts claimants comes from the way benefit
units are defined. Some people register from one address but habitually
stay at another - because this enables residents of both addresses to act
independently with regard to benefit claims. Typically, another resident
at the first address will be contacted by the LFS interviewer. That
respondent will not of course tell the interviewer about the person who does
not actually live there very much. The respondent at the second address
will typically be the householder who will not give details of the
individual who is not claiming benefit at that address.
This pattern should be assessed in the light of about 25% proxy responses in
the LFS. That is information obtained from a respondent about another
member of the household.
Use of claimant records as a sampling frame would avoid all these problems.
The sampling frame would pick claimants up at the address used for the
claim. The LFS interviewer can be quite upfront in telling the respondent
that the household has been selected from a list of claimants, and it is
very unlikely that the respondent would deny that there is a claimant
registered at that address.
I don't know the FRS questionnaire, and would be interested to learn the
extent to which the problems are parallel. But since the aim of the FRS
should be to get information on family resources among recipients of
government benefits it would be very reasonable to use the admin records as
a sampling frame to ensure that the count of recipients was more or less
correct.
Don't see any reason why this should add to costs. Just about sampling.
Getting more reliable responses would enable reduction in sample size.
There would of course still be a need for a PAF stratum. But in the case
of the FRS it would clearly be sensible to take out recipient households
from the PAF sampling frame so that there are always clean comparisons
between benefit and 'no benefit' households.
Ray Thomas, Social Sciences, Open University
Tel: 01908 679081 Fax 01908 550401
Email: [log in to unmask]
35 Passmore, Milton Keynes MK6 3DY
|