A slightly longer email, as this issue is so topical and in the hope that it
contributes to a recognition of what is known and what remains uncertain.
For access to the ONS explanation of the new census figures, the process to
estimate the full census day population including those not on any census
form, and the reasons why ONS were content with their new estimate though it
is nearly a million less than they had estimated previously, see
www.statistics.gov.uk/census2001 with link to methodology, response rates,
and the missing men.
The problem with interpretation of these figures is big and at least
two-sided - and certainly a problem that ONS have not prepared the media and
politicians for. Martha Riche (one of the many ex-Directors of the US Census
Bureau) warned about the need to have media, public and politicians
understand statistical adjustment to census data in RadStats 69.
The process of supplementing census records based on the biggest ever census
check (the Census Coverage Survey), was discussed in detail and agreed in
principle with census users: better an unbiased estimate of the complete
population than a biased incomplete census database. However the time-scale
required to create this new 'One Number Census' meant that all difficult
decisions were taken internally and have not been subject to any discussion.
Secrecy has prevailed over evidence and understanding during this last year
so that kites would not fly. With the result that suddenly there are a lot
of ridiculous stories flying around now, some promoted from ONS (Ibiza
raves) and others invented by journalists who have not been given the
positive story about the long-awaited unbiased census database created
through imputing records for missing people. In fact the last census also
had lots of people imputed, not for all those missing but for those the
enumerators felt sure had been missed, so precedent could have been claimed.
But there is a second big side to the story, which is confusing because it
runs counter to some of the main messages in the imputation story. It is
this. After the census had been supplemented with information based on the
census check as above, ONS were still faced with 900,000 people less than
they had expected - their expectations are based on adding births, deaths
and estimated migration to the population estimated at the time of the 1991
census. This 900,000 shortfall is mainly men aged 18-40, but there are quite
a few children under 10 too. The shortfall has been characterised by
men/women ratios lower at aged 25-40 than at ages 41-60 in many parts of the
country, in spite of higher male mortality at every age.
There are two competing explanations:
1. The ONS explanation of un-monitored emigration; they came to this
conclusion after thorough examination of the evidence but behind closed
doors; that evidence is on the website in broad terms but has not yet been
reported in detail.
2. That some and maybe all of the 900,000 are residents in the UK who have
been missed over and above those estimated and imputed as above. These would
be people whose probability of being missed by the Census was not
independent from the probability of being missed by its check - they were
hard for both to find. They may have not wanted to be counted, but just as
likely they would include people who have unfixed abode, who are out a lot
or in other ways are hard to count by a census and by the good survey check
on the census. The possibility was clearly highlighted as a weakness in the
One Number Census methodology over the past few years.
In the 1991 Census very similar sex ratios in the estimate based on
census+check were seen as evidence of further undercount (ie explanation
two), and the population was supplemented by extra undercount. That was a
big embarrassment all round - the census after all aims to be the gold
standard. Danny Dorling and I explained the situation and its consequences
in RadStats 55. So it is understandable that ONS does not want to conclude
that the robust and improved methods for 2001 have also failed. And perhaps
they have not.
But until the evidence from checks of administrative records against census
counts, demographic checks and released for peer group and media
understanding, it is likely that the 2001 census results will also be
questioned from this second side to the story.
Allan is correct that better press releases are needed from ONS. They have
been working under extreme pressure to get the census job done. Now they
have to work just as hard with journalists and social statisticians to share
evidence and convince us of their interpretation, even if it means admitting
that there is some uncertainty that will remain about the balance of the two
explanations above. The last suggestion may be hard for someone of Len
Cook's reputation for making a decision and defending it to the hilt.
Ludi
******************************************************
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]
*******************************************************
|