I am with John on this.  as a census manager with very limited resources 
faced with a 40% non response rate on an area of around 20,000 
householders my options were very limited.  With a sample of say 2000 
households even if I only had 50% resources I would have been much 
better placed and the results would have clearly been more 
representative.  It is pretty obvious that non responders are not a 
random sample of the population, but it isn't clear to me how anyone 
could establish their characteristics sufficiently to correct for their 
non responses, because obviously the factors which lead them not to 
respond to one approach will apply to any other sort of approach of the 
same kind. 

Furthermore if it is clear to everyone that only a sample is being 
studied this could serve to reduce fears that people who want to hide 
are being rounded up.

Martin Rathfelder
Director
Socialist Health Association
22 Blair Road
Manchester
M16 8NS
0870 013 0065
www.sochealth.co.uk

If you do not wish to be on our mailing list please let us know and we will
remove you. 



John Whittington wrote:
> At 09:02 07/11/2007 +1100, Harry Feldman wrote:
>
>> My understanding is that a census is not subject to sampling error for
>> obvious reasons.  Since there is no sampling error, I believe that non
>> response has less impact, not just because each non response 
>> represents a
>> smaller proportion of observations,...
>
> Whilst that is obviously literally true, I'm not sure I understand the 
> relevance.  If the (attempted) sample is a random one, one would 
> surely expect the same proportion of non-responders in the sample as 
> in an (attempted)  100% census - and I would have thought that was the 
> proportion which matters.
>
>> ... but also because self selection in a sample survey distorts the
>> randomness of the sample, which is not an issue with the census.
>
> I don't really get that, except in terms of very literal use of 
> words.  The fact surely is that non-random ('self-selected') 
> non-response will bias the results of a census to exactly the same 
> extent as they would bias the results of a (random) sample survey - 
> and that's surely what matters?
>
>> Furthermore, as sampling error is not a factor (although confidentiality
>> may be), it is possible to release data on small geographical areas.  
>> Such
>> data are alleged to be of use in determining constituency boundaries,
>> planning infrastructure development and services, etc.  .....
>
> As others have said, it's obviously the case that a desire to obtain 
> estimates in relation to very small subsets of the population is the 
> one (and I would probably say ONLY) thing that might kill the idea of 
> a sample survey instead of 'the Census'.  I don't know enough to be 
> able to judge the importance of this - i.e. I don't know 'how small' 
> are the smallest units of interest.  However, if they are at least 
> 'well into 5 figures' (i.e. >>10,000) then I still believe that a 
> relatively modest sample survey (still probably ~10%) would probably 
> provide estimates of adequate precision for most purpose, at a 
> fraction of the cost of a census.
>
> Indeed, one thing that I don't think has been said is that if some of 
> that cost saving was 'ploughed back' into more vigorous attempts to 
> reduce the amount of non-response (and/or incorrect response!), then a 
> sample survey might end up with BETTER quality estimates (than a 
> census), but still at a lower cost.
>
> Kind Regards,
>
>
> John
>
> ----------------------------------------------------------------
> Dr John Whittington,       Voice:    +44 (0) 1296 730225
> Mediscience Services       Fax:      +44 (0) 1296 738893
> Twyford Manor, Twyford,    E-mail:   [log in to unmask]
> Buckingham  MK18 4EL, UK
> ----------------------------------------------------------------
>
> ******************************************************
> 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]
> Disclaimer: The messages sent to this list are the views of the sender 
> and cannot be assumed to be representative of the range of views held 
> by subscribers to the Radical Statistics Group. To find out more about 
> Radical Statistics and its aims and activities and read current and 
> past issues of our newsletter you are invited to visit our web site 
> www.radstats.org.uk.
> *******************************************************
>
>
>

******************************************************
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]
Disclaimer: The messages sent to this list are the views of the sender and cannot be assumed to be representative of the range of views held by subscribers to the Radical Statistics Group. To find out more about Radical Statistics and its aims and activities and read current and past issues of our newsletter you are invited to visit our web site www.radstats.org.uk.
*******************************************************