There are aspects of the data we are using for this lab that affect the results. In the notes below I highlight some of those aspects.

Data range: All of the data (including the rescaled income data) have a minimum value of 0 and a maximum value of 100 (making the assumption that no EA would have an average income over $100,000). When working with data such as this (i.e., values limited to 0 - 100) it is important to consider how a difference of, say 10, is interpreted. For example, consider two pairs of data points, one pair with values of 20 and 30, and the other pair with values of 70 and 80. For the first pair 10 reflects a difference of 50% (relative to the 20), while for the second pair the difference is only 14% (relative to 70). As such, it should be obvious that in regions where the variables are lower valued (such as income in the east side of Vancouver relative to the west side) smaller differences can have more significant impacts on the results.

Data variance: Another aspect to consider is the variance in the variables. In areas where the variance is larger GWR is more likely to identify a larger impact (that is, the parameter values will be larger). Consider the east-west gradient in many socioeconomic variables across Vancouver. This gradient (and therefore the variance in the variables) would typically be felt strongest just east of Main Street (west of Granville the variance in many socioeconomic variables is fairly small). The figure below illustrates the variance in physical scores (based on a fixed kernel of 60 cases), using a colour gradient of blue (low variance) to yellow (medium) to red (high variance):

It is obvious that west of Granville Street there is little variance in the physical scores, but that east of Main Street the variance is high.

Reason for the variance: Another aspect to consider is the nature of the variance. For example, consider gentrification, where the typical effect will be to see the average income increase in a neighbourhood but the average family size decrease. In areas undergoing gentrification children of the original (i.e., poorer) families will not see the benefits of the 'increased' income associated with the neighbourhood and, in fact, may experience less socialization (as fewer children remain in the neighbourhood, and the parents are less able to afford 'extras' as the property taxes rise). Conversely, in an area of the city where a more 'typical' gradient of income occurs, those children living in the higher income areas would more likely benefit from the higher incomes associated with their neighbourhoods. Thus, the effects of higher income could be negative in some areas (those undergoing gentrification) and positive in others.

Statistics: Sometimes, of course, there is variation in the results that is a function of variable interactions that remain unresolved in our data (e.g., GWR assumes a linear model of regression; it could be that the relation would be better described using a nonlinear model--see the figure below). Unresolved varaible interactions and other statistical complications are likely to be happening with our data, and attempting to interpret the results of such interactions is a fruitless task. So, if you can't make sense of some results don't worry--it may be that there is, in fact, no socioeconomic factors behind the results (that is, the results are statistical artifacts and not truly meaningful).