Thursday, July 22, 2010

The Census

There are clear problems with how the government has handled this issue, but there are also some clear problems with how the long form was handled at all.

The one in five households is a random selection, but that selection needs to be corrected for what sort of a representation it is off the total public.   As it stands this is not done with the census data and each long form counts as five households without.  There are anomalous results because of this.  Communities that I know that have Quakers in them are listed as zero Quakers and ones without have people tick that box and suddenly there are 5-10 Quakers in communities where there are none.  It is a statistically useless tool for anything on a small scale but the data is still published.

The long form also has a problem in that a lot more people lie when they answer the long form than the short form.   This fact is not accounted for by Stats Canada and the assumption is that the answers are truthful.   Consider the Jedi religion phenomena in censuses in the English speaking world in the last 15 years.  

I do not know what the rate of lieing is on the census, but on the short form I assume it is quite low as there are only a few very straight forward answers.   The long form has a whole host of questions, some are of a form that people may prefer to lie than answer truthfully.   What we have is 80% of households that have a very low rate of dishonest answers and 20% with an unknown, but more than zero by a longshot, number of dishonest answers.

The long form census also suffers because it asks many questions that require estimates.  The public has real trouble estimating distances or time spent on tasks, when these sort of questions are asked, I have to take the results with a grain of salt.   There are other questions which simply give you no guideline as how to answer them.   Religion, how do you measure the faith of people that take no part in their professed faith?   How can you put them on the same level as people very active in their faith?   Languages, the terms used within the census are so vague so to make it a completely unreliable measure of language use and proficiency.  

There are three possible answers to the census issue.

First, there should be a single form that all households get and it should be something that can be answered easily and quickly, no more than three of four pages and no questions that require estimates or do not have guidelines on how to answer them.   As an example, for the language one, there should be a text in the language that the person needs to be able to read and discuss with someone over the phone.  For religion, only people that regularly attend some sort of worship service should count towards the numbers for a religion.

Second, the provinces should take over much of the census role from the federal government.   Each province has different needs and interests.   A census in Newfoundland would be asking different questions than one in BC.   Other than needing to know the population for the purposes of representation, everything else would be better asked by a provincial census.

Three, maybe we are beyond the era of needing to have every individual fill out a form.   Can we not develop a good statistical model to randomly sample on an ongoing basis to find out the answers to the demographic questions we have?

We managed to live a very long time without a whole host of the questions we ask now on the census.   We have more and better tools to answer the demographic questions than we did in the past.   This is not an issue in my opinion.

No comments: