This map colours each region by the Bayesian surprise of the rate of non-primary residence dwellings, based on how observed 2016 rates differ from our model. Intuitively we colour each region by how consistent our model is with the observed data.

The white areas show where our model is consistent with the data. The greener the area the less consistent the model is with the observations and the observations show fewer non-primary residence units than our model predicts, the pink areas have more non-primary residence buildings than the model predicts.

Overall, the rate of non-primary residence units in Metro Vancouver is consistent with our model as the rate of non-primary residence units per net new dwelling matches the Canada average, but there is considerable variation among the municipalities as well as deeper within the metro region.

Metro Toronto shows considerable inconsistency with our model by having significantly lower rate of non-primary residents units than predicted. At the municipal level Oakville is the only municipality that provides some evidence of higher than predicted rates. However, digging deeper into the Toronto region reveals many areas where observed rates provide strong evidence against our model in either direction.

Montreal is an example of a CMA that provides evidence of higher than predicted rates, with the City of Montreal being the driver of this.

As always, our maps are Canada-wide. Pan, zoom, use the search functions or simple geo-locate to whatever area interests you.

As a prior we model the rate of non-primary residence dwellings as being given by the respective 2011 rate in each area plus a rate of 10.1% for each net new unit. (10.1% is the Canadian average rate of non-primary residence buildings per net new unit, assuming the rate is unchanged for all other dwelling units.)

Being "consistent" with the model means either the observed rate is close to the predicted rate. Or there is some difference, yet not enough statistical evidence, possibly because there are very few dwellings in that region, to question the validity of our model. An area is "inconsistent" with the model assumptions if the predicted rate is different from the observed one and there is enough statistical evidence to invalidate the model.

For more details on how Bayesian Surprise Maps work please refer to our blog post or this excellent paper that this approach is based on.

Compare this to the map of modelled change in the rate of non-primary residence dwellings. That map just colours areas by the difference of our modelled rate to the actual rate of non-primary residence units. At the Census Tract level the maps look quite similar, but we can already notice that some regions have changed order. The difference becomes very clear at the dissemination block level where 2016 rates in most blocks turn out to be consistent with out model despite having changed significantly, since they have relatively small numbers of dwellings. Some blocks however still stand out. This makes it much easier to spot statistically significant changes in the rate of non-primary residence dwellings.

Similarly we can turn to the municipal level where we municipalities changing the order when compared to just considering the difference of observations and model. For example, Tsawwassen and Delta trade places. Tsawassen has an overall larger difference in observations and predictions compared to delta, but given it's very small number of dwellings the Bayesian model gives it less weight. Similarly, the Bayesian model gives the City of Vancouver a slightly higher weight than Richmond, although Richmond has a slightly larger difference of observations from the model.