| What role does geography play? a look at vancouver's secondary schools |
 |
Home | abstract | methodology | results | discussion and error sources | links |  |
I began with base layers for GVRD Census Tract Areas, Excel Documents for GVRDctAgeSex and GVRDctIndivInc for both 2006 and 2001. I was also able to obtain a secondary school catchment area base layer from the City of Vancouver website. From the BC Ministry of Education I obtained data regarding the number of students in each school for the 2005-2006 year, however I unfortunately was not able to obtain the data from them regarding the number of students in their schools from back in 2001. I had to go back this far because the Canadian Government only collects census data every 5 years, and since 2011 will be the next data collection year, this information is unfortunately quite outdated. In order to obtain the data, I had to email the BC Ministry of Education and put in a data request, which takes quite a long time to process. Three weeks later I finally obtained the dataset that I required for my project, but it was well worth the wait. The last data that I needed to collect were the Frazer Institute Scores from the Annual Reports (for 2001 and 2006 respectively), which fortunately do go quite a ways back.
Now that I had collected all the data that I required, I began my analysis. The Procedures are as follows below:
Step 1: Join the GVRDct06 Shapefile to the GVRDct06AgeSex table based on CTUID. Create a field which I titled “06_1019” and in this field I added up the F# columns containing the counts per CT of 10-14 and 15-19 year olds (as these would be the high school aged children), and did this for both males and females. Since the titles of these categories are all encoded, I had to look up what the headings meant to ensure that I was adding together the correct columns. I then displayed the summed data as totals in each CT. Since the CTs between 2001 and 2006 changed slightly, I was required to complete the entire procedure over again with the data from 2001, using the GVRDct01 shapefile and GVRDct01AgeSex table.
Step 2: Adding the secondary school shapefile that I had obtained from the City of Vancouver website, I realized that I was going to have to find a way to calculate the number of students in each catchment area, not CT. This provided quite a challenge, as the CT boundaries and school catchment boundaries did not match up. What I elected to do then was to recalculate the population based on the population density and area of each CT contained within each school catchment boundary. To accomplish this, I first calculated the population density in a separate field in the GVRDct06 attribute table, and calculating the geometry to obtain the spatial area for each CT in kilometers squared. I then used the “Split” tool in ArcToolbox, which split the CT boundaries map into 18 different layers, one for each school catchment area (there are 18 secondary schools in Vancouver). With this I subsequently was able to recalculate the area in each of the CT’s (since area is retained when performing spatial functions using a geodatabase file, which I was), and then converted it into square kilometers for each layer. I then manually added a field to each of the 18 layers and, using the Calculate Field function (which at this point had become my favourite tool), I multiplied the population density of each CT by the area in km2 of that CT contained within the school boundary. The results are obviously quite approximate, but since I was trying to display a trend and not actual statistics, I was not overly concerned by this.
Step 3: Do all of Step 2 again for the 2001 data. Bang head on desk.
Source: DMTI. NAD83, Statistics Canada, City of Vancouver
Step 4: At this point I could now go and add up the numbers for each catchment area and add them as fields in my original base layer in the original data frame. Or so I thought. Then I discovered that during the splits, sliver polygons had been created, and there for there were sometimes up to 8 polygons/layer or catchment area of about 0.002-0.000001 km2 each. Painstakingly, I had to go back through all (now 36) layers from the 2006 and 2001 data and select by attributes for those layers which were not sliver polygons, recalculate the population of only these polygons, and then used the “summarize” function to obtain the sum from each category. It is true that some numbers and data were lost here because I was re-calculating population using area in the first place, and without the area the numbers would now be off. However, I was less concerned about this for two reasons. First, I was still only looking for the trend in movement between 2001 and 2006, and since the sliver polygons were subject to both layers, they nullified each other. Second, since the census data had only collected for 10-14 and 14-19 year olds, and since most 10, 11, and 19 year olds aren’t in high school, the numbers could afford to be reduced a little bit.
Step 5: Now with actual recorded sums for each catchment, I created two fields in the original secondary schools boundary layer and manually inputted each total (I originally tried joining an excel sheet, however it saved me no time and since I can’t spell very well it never joined up correctly, so I eventually got frustrated and gave up). With the totals of students living in the catchment areas in both 2001 and 2006, I displayed them on the map and then created another field and calculated the difference between them.
Source: DMTI. NAD83, Statistics Canada, City of Vancouver
Step 6: Using the totals that I had obtained online for each school’s student population in the 2001 and 2006 school years, I created two more fields in the secondary schools layers and displayed this data as well. Then I likewise took the difference between the 2001 and 2006 school populations, and displayed this data as well.
Source: DMTI. NAD83, Statistics Canada, City of Vancouver, BC Ministry of Education
Step 7: Using the Spatial Autocorrelation (Moran’s I) tool in ArcToolbox, I calculated the Moran’s I scores for the six different maps that I had created. As I had anticipated in my Abstract, the Moran’s I values showed correlation within 1% for the difference in student body population, but no correlation for the difference in actual catchment area population. This was also quite obvious when displayed on the map, as the colours were randomly patterned for the area population, while in the student body map a clear deficit could be seen on the east side with gains on the west side. Success. I then called my mom to tell her that something on this project had finally gone right.
Step 8: Now I had to analyze the two factors that I thought could possibly be contributing to this geographical anomaly. For the Average Income, I took much the same approach as I did with the population totals, joining the GVRDct06IndivInc and GVRDct01IndivInc tables to their respective shapefiles. As with the population data, the headings are encoded, so I have to look up the codes and manually name the field that I wanted (which was the average household income in each CT). I then mapped this data. Creating another Data Frame, I again used the “Split” tool to create another 18 layers (one for each catchment area again), and used the statistics tool to manually obtain the mean value (or average) household income for each catchment area. I finally had learned how to spell “Britannia Secondary” so I was then able to join the table to another secondary school boundaries layer to an excel document with this information. I then displayed the average household income per. catchment area, and used the Spatial Autocorrelation Tool again to determine the Moran’s I value, which was determined to be 5% likelihood that the pattern was not spatially correlated.
Source: DMTI. NAD83, Statistics Canada, City of Vancouver
Step 9: The final step was to use the Fraser Institute Scores and see if they auto correlated in the same way with the student population difference and household income. Using the values that I had obtained online as part of the attribute table for the secondary schools catchment layer, I was able to more easily display the 2001 and 2006 data. I also created another map showing the differences between the FI scores for the Vancouver schools in 2001 and 2006. I then took the Moran’s I values in the same way as mentioned above for the three resulting maps. While there was a 95% likelihood of autocorrelation for the FI values from both years, the map of the differences between them held no autocorrelation.
Source: DMTI. NAD83, Statistics Canada, City of Vancouver, The Fraser Istitute
Step 10: Now that I had a generally pretty good idea of the results from my project, I elected to map the change in school population with the best explanatory value (average household income). To best display this, I represented the Average Household Income of each catchment area with a proportional dot symbol, and displayed the change in school population as a base layer beneath it. I then imported a schools layer from the G:drive and selected by attributes for secondary schools (this is also displayed on the split CT map above). This represents the final concluding map of my project.
