Objective. Explain and analyze the concept of spatial autocorrelation regarding median total household income and the occurrence of different types of crime in an urban setting. Explain and show evidence of the ways in which the Modifiable Areal Unit Problem (MAUP) impacts analyses via the scale and aggregation effects, as well as the implications that these uncertainties might represent when interpreting the results of the spatial analyses. Manipulate and analyze census data and work with statistics tools for spatial data.
Statistics Canada: Data Suppression in Dissemination Areas.
There need to be more than forty people living in a dissemination (enumeration) area to keep Statistics Canada from supressing census characteristic data for that geographic area. For income data, there need to be more than two hundred fifty people to keep Statistics Canada from practicing area suppression.
Statistics Canada practices area suppression to protect the privacy of individual participants in the given statistical study.
Random rounding is an area suppression technique where values “greater than ten are randomly rounded up or down to multiple of five [and those less than ten] are randomly rounded to zero or ten” (Puderer, 2001). This decreases data precision by normalizing the data into random multiples of zero, five, and ten, which will affect the implications drawn from generalized data values.
Spatial Autocorrelation of Residuals.
The residuals are spatially auto correlated. This is supported by the Moran’s Index report which shows that the spatial distribution of crime is clustered, and not randomly dispersed for both maps: also indicated by the z-scores and p-values shown in Table 1.
Regarding the distribution of residential break and enter crimes, the results of the residual analysis show whether the predicted value of crime incident in a neighborhood is greater than, less than, or approximately equal to the observed number of neighborhood crimes. As shown in the map, “Regression Analysis of Vancouver Residential Break and Enter Crimes and Median Total Household Income Correlation Model,” the neighborhoods coloured closer to the dark red shade are associated with observed crime greater than the predicted value and those closer to the dark blue shade are associated with observed crime less than the predicted value. The neighborhoods colored in pale yellow are those with observed crime values closest to the predicted value: where the relative abundance of these areas support that the residuals are spatially auto correlated.
See Regression Analysis of Vancouver Break and Enter Crimes and Median Total Household Income Correlation Model
Residual Patterns in Census Tracts and Dissemination Areas.
Dissemination areas are census data measured in smaller geographic units compared to census tracts maps. So the residuals represent similar patterns in both maps but the census tracts map shows a more general distribution of residential break and enter crimes: class breaks are larger areas assigned with the same colour. These groups are subdivided into smaller areas with varying colours in the dissemination areas map. The dissemination areas show different crime intensities concentrated in smaller tiles within the same census tract.
The census tracts map shows the largest deviation from the average frequency of break and enter crimes generally occurs within the same areas, but the dissemination areas map shows that they are concentrated into smaller areas. Regarding medina total household income, greater deviations (whether greater or less than) from the mean represent the degree of auto correlation between median total household income and crimes. The patterns show whether our function for crime and median total household income predicts well or poorly.
The scale affects the map user’s understanding of the distribution of residential break and enter crimes relative to median total household income. The data is averaged across a larger area in census tracts than in dissemination areas, so changing the scale will also change the distribution of the data per census tile (distribution of different colours per census tile in the map). It can imply a more clustered versus dispersed distribution of the correlation.
See Regression Analysis of Vancouver Break and Enter Crimes and Median Total Household Income Correlation Model
For example, the 0049.01 census tile has a 0.5-1.5 standard deviation in the census tracts but the dissemination area shows that approximately 60% of that census tile has a concentrated standard deviation greater than 2.5
In the 0057.02 census tile, the census tracts map shows a standard deviation greater than 2.5 but the dissemination areas show this standard deviation value concentrated only on approximately 15% of that census tile.
Grouping Analysis for Census Tracts and Dissemination Areas.
The groups are labelled according to the relative variation of neighborhood crime according to median total household income. Income and crime are ranked separately as values greater than, less than, or average compared to other neighborhoods, and the variance of income and crime levels are approximately equal for both maps.
The two maps both show the occurrence of neighborhood crimes related to median total household income: neighborhood crimes considering only mischief, auto theft, and residential breaking and entering. They show that neighborhood crimes tend to occur in low to average (medium) income neighborhoods.
In general, the results of the grouping analysis do not support the claim that income and security in Vancouver are related Despite there being a potential positive relationship between median household income and neighborhood crime, as suggested by the residual analysis, the grouping analysis suggests that this might not be the case. Other than relatively high crime rates occurring in low to average income neighborhoods, the grouping analysis suggests that there could be other variable not accounted for in the analysis but may still affect the distribution of neighborhood crimes. Although it is possible that higher income neighborhoods provide a greater incentive for residential breaking and entering, other factors such as a neighborhood’s proximity to crime hot spots, could affect the actual spatial distribution of crime in Vancouver.
See Grouping Analysis of Vancouver Neighborhood Security in Relation to Median Total Household Income
See full report here Correlation and Regression Lab Report