The media has been all over the decision made by Statistics Canada to eliminate the long form census. There are ample political considerations that can be debated elsewhere, but there are some basic statistical concerns with the decision.
-
Making a survey voluntary
As much as possible, individuals from whom data is collected should be selected at random from within a population. If they are selected in any other way, there is danger that the individuals completing the survey will have more to do with each other than they do with the population at large. Individuals who volunteer to complete the survey will likely have a few things in common. For example, they may have the time to complete the survey – a luxury not afforded to all. They also might also be more likely to have higher levels of education and understand the value of the survey, or hold certain beliefs about the authority of government and therefore feel compelled to complete the survey. In terms of the data, it means that people with similar values, ideas, impressions will complete the survey and therefore the data will represent the ideas of a few rather than the whole population. In the media, the focus has been on response rate. This is short sighted. Even if a large volume of people respond, they might not be representative of you or me.
-
Longitudinal data
The long from census has been used in similar forms repeatedly. If questions are altered significantly or the way in which the data is collected changes significantly, the data can no longer be compared to data that was collected previously. At present, we can ask questions such as ‘how has the population in the Waterloo region changed in the last 15 years – demographically, economically etc.?’ Once we change our methods, important comparisons are no longer possible.
In statistical terms, the decision to scarp the long from census is problematic. If we care about the information that it provided and found it useful in developing programs, advertising, services communities, and planning for schools, the new data will no longer allow us to do that with the same certainly or accuracy.