There are a number of different issues. One is the reliability of a sample
survey, and whether that sample survey accepts proxy interviews, versus a
self-completion census. The second is the nature of the questions asked.

The ONS release (I haven't yet read the linked article as the link from the
release didn't work when I tried it) addresses more the second question than
the first.

Some of the ILO definitions followed in the LFS are, to a naive person,
counter-intuitive.

The ILO definition of employment is one hour or more in the reference week, in
paid work or unpaid family work. Clearly a teenager babysitting for a
neighbour for an hour and getting paid counts as employment. Equally said
teenager looking after the family shop for an hour and not getting paid is
equally employment, albeit of the unpaid family variety. Not quite sure
whether teenager babysitting for a cousin and not getting paid, for an hour,
counts as being in employment.

These types of 'employment' are stretching what naive respondents would see as
employment, and therefore, in a self-completion census, even with notes,
numbers of people who would be picked up as being in employment by a routed
telephone interview will, in a self-completion census form, respond as not
being in employment.

Similarly, with unemployment. The ILO definition is based on a respondent
having looked for work, including scanning job ads in the local paper, within
the last four weeks, and being available to start in two (and not working for
even one hour a week). A routed interview is very much better at picking up
the detail of this sort of definition, because you can ask if respondent has
done any of a number of things, such as looked at job ads in the paper or the
newsagent, and then make a derived variable to define those who looked for
work as being all who did one or more of those things. In a census, it is
easier to ask if someone is looking for work, when they might not conceive
glancing at the cards in a newsagent, including some jobs, as looking for
work.

In a census with self-completion you will clearly get people saying they are
unemployed if they are claiming JSA and doing unpaid family work (or working
under the 16 hours limit). Equally you get people saying they are unemployed
while they are removed from unemployment counts by being on government
employment and training programmes (a UK interpretation of ILO definitions
which does not seem to be universally shared by national statistics offices).

I actually think the ILO definitions are the best we have, despite
difficulties, because they are reasonably close to economic behaviour.
Unemployment might be defined as someone with a probability of applying for a
job above a certtain level - the effect on earnings inflation depending on
actual applications etc. The problem is operationalising them in a
self-completion questionnaire.

Therefore different data collection methods (and different questions) will
inevitably produce different results.

The ONS feels that it has to reconcile different sources or at least expalin
the differences. This is the case with the press release Ray referred to.

Issues with confidence limits around LFS estimates are developing. ONS has now
started publishing confidence limits for local LFS estimates on NOMIS. This
might help people realise that sample surveys have inherent sampling
variation, although end-users are unlikely to want their spurious precision
exposed.

On Thursday 13 Jan 2005 22:41, Ray Thomas wrote:
> I'm told that there is no means of measuring the speed of light
> independently of the method of measurement.
>
> It is a bit surprising therefore to see the ONS claim in a Press Release
> today that the 2001 Census understimated the numbers in employment and
> overestimated the numbers in unemployment.    The source for this
> declaration is a sample survey, the Labour Force Survey, that is largely
> conducted by telephone interview, often by proxy, and whose results depend
> upon blowing up the sample size by estimates of total population based upon
> the very census statistics that are being castigated.
>
> Is not this a case of a dog chasing its own tail?    Who can say whether
> the head bites the tail or whether the tail wags itself into the dog's
> mouth? Or indeed whether the a bit actually happens?
>
> Isn't it time that the ONS started being scientific by acknowledging that
> the Labour Force Survey does not have a special alliance with Appolo, the
> Greek God of Truth, or our own good Lord who often claims a monopoly in the
> area?
>
> Surely it would be widely recognised in any truly scientific community that
> the measurement of large human populations is even more dependent upon the
> method of measurement that the measurement of phenomena of the physical
> world?    If so why in the ONS damaging the status of official statistics
> and reputation of statisticians by producing such a ludicrously worded
> Press Releases?
>
> What about the million men missing from the Census?   What about the
> longstanding failure of the LFS to count the number of claimant unemployed?
> What about the response rate of the LFS itself that is well below that of
> the Census statistics.
>
> A paragraph from the Press Release is given below.
>
> Ray Thomas
> ****************************************
> The LFS found 640,000 more people in employment than did the
>
> Census but 203,000 fewer unemployed people. This was very much in
>
> line with ONS predictions made before the publication of the Census,
>
> which expected that the Census would underestimate employment by
>
> 0.5 to 1.0 million and overestimate UK unemployment by 100,000 to
>
> 250,000. This is due to the self-completion nature of the Census which
>
> does not permit the International Labour Organisation definitions of
>
> employment and unemployment to be applied as rigorously as they are
>
> in the LFS.
>
> ************************************************************
>
> ******************************************************
> Please note that if you press the 'Reply' button your
> message will go only to the sender of this message.
> If you want to reply to the whole list, use your mailer's
> 'Reply-to-All' button to send your message automatically
> to [log in to unmask]
> *******************************************************

******************************************************
Please note that if you press the 'Reply' button your
message will go only to the sender of this message.
If you want to reply to the whole list, use your mailer's
'Reply-to-All' button to send your message automatically
to [log in to unmask]
*******************************************************