As with any GIS project, there are many areas where error can/may have occurred. As far as my measurements and data are concerned, most of these were hard numbers with reliable sources, so the initial inputs are most likely not a cause for significant concern in a project at this level. However, often those numbers related to Census Tract Areas, and not to the secondary school boundaries that I wanted them to. To overcome this, as I mentioned in my methodology, I used the “Split” tool to split the CT’s into secondary school boundary catchment zones. With population totals, I then recalculated the “population” in each part of the CT by normalizing the data as a population density. The problems associated with this are obvious. Apartment buildings may be present on one side of the CT and not the other, and by performing this operation I undoubtedly created an ecological fallacy somewhere, not to mention that I clearly started with  MAUP issue by not using Dissemination Areas instead (although this would have required many more extra recalculations after the split given that the split would have split more areas, which is why I didn’t use them). Another two problems also arise where the plit is concerned. As labeled in the legend, I used the population of 10-19 year-olds in these calculations, however clearly most 10-12 year-olds and 19 year-olds do not go to high school. I had hoped to use data for just those youth aged 13-18, however Statistics Canada makes their data available only in 4-year groupings of 10-14 and 15-19. Thus, the totals undoubtedly contain students not going to high school, but since my project was on the changes in these populations, and not specifically on the hard numbers themselves, and since the bulk of the numbers were of students going to high school, I did not rate this as too much of a concern, although it should be noted in a section regarding error. Lastly, regarding the split, the borders of the secondary school boundaries and CT areas did not quite perfectly line up where they should have. This created sliver polygons, which I subsequently checked the attribute table for, and if I determined that they were large enough (over 0.5km squared, which only 1 was), I included then in the population re-calculation. However, approximately 0.05km2 was missing from the catchment areas along the ocean-front boundaries (mostly on the northernmost part of the map), which would have marginal effects on the population re-count, given that this number was based upon area.

    Regarding the analysis of the data, I had originally intended to spatially analyze the correlations using an Ordinary Least Squared, and initially did use this tool to compare the results between the schools changes and the Average Income and FI Scores values, however I quickly realized this was not an accurate representation of my data. First of all, with only 18 data points, it is very hard to spatially analyze anything for certain, given that correlations can develop in larger spatial units that are actually not present when one breaks them down and looks more carefully at them (in smaller spatial units). As such, both my spatial autocorrelation and ordinary least squares were subject to this error. In addition, the ordinary least squares, while showing a correlation between the average income and changes in school population, did not take into account that another factor may be present in this that my analysis was not taking into account. For example, a lot of parents are concerned about sending their children to schools where there are a lot of ESL students because they believe the learning will not progress as quickly in the classroom, or that their English-speaking child will not receive the attention they need. A stronger correlation might have been present if I had looked at immigration data in the catchment boundaries. As such, I decided in the end to simply model that there was a spatial correlation between east and west side schools, and leave the average income and FI Scores as a secondary part to my project that was more or less intended to suggest possible reasons as to why this could be, not hard-line say “this is why there is spatial correlation”, but to simply cause whomever is reading this project to question possibilities as to why totals are different on the west and east sides. One place where an OLS could have come into play is between the 10-19 age difference population and schools population difference, however given that one showed a very high spatial correlation and the other did not, this would have been redundant, as the Moran’s Index already showed that these two datasets were not correlated.

    Obviously, a great amount of room is left to further investigate this topic. As I suggested above, numerous other correlations could be made, and real data about populations of youth in the catchment areas does exist (the Ministry of Education has these) and would be very helpful in any further analyzing of this topic.