Certainly, Monsieur Bouchard. The fundamental issue is what is called “response bias”. People who respond to voluntary surveys are, by their nature, different from those who do not. Those who do not are often more time pressed. They are people who sometimes feel marginalized within society. If there are linguistic barriers, they are less likely to respond.
Those problems can be compensated for if one has a benchmark that says there are this number of indigenous peoples, for example, in an urban situation; or there are these many people who are unilingual francophones in this area. Then you can look at surveys that are done voluntarily and reweight or change the sampling designs to ensure that you get everyone. Without the benchmarks created by a compulsory format that tests for those very basic issues with simple questions about income, ethnicity, language, and so on, you won't have that benchmark anymore. You may have a voluntary survey, but you can no longer adjust it to a known population.
That means not only are your census results less applicable in small areas, but very importantly, you no longer have the ability to conduct further voluntary surveys, or surveys such as the labour force survey, with confidence that you are accurately reflecting a cross-section of Canadians.