Based on our results, there were no significant relationships found between any one or any combination of our change in socioeconomic variables with the change in crime rates. However, there are some interesting findings within the results that give a relative idea of what could be used as a predictor when analyzing changes in crime rate. Figure 2 presents the notion that crime rates have decreased across Vancouver during the 10-year span between 2006 and 2016. When taking those results and comparing them to the relationships between our variables (Figure 3) we can see that there isn’t one socioeconomic variable that caused the changes related to crime rates. For this reason, we performed an Exploratory Regression to see if there is any combination of changes in socioeconomic variables that can be correlated to changes in the crime rate. The results from the Exploratory Regression did not produce any passing models, however, there were some notable findings within the analysis. In Tables 2,3,4, and 5 two socioeconomic variables remained constant as they were included in all 12 top models. These two variables were population density and the percentage of youth in the population. Together they had the highest adjusted R² value of 0.25 (Table 2) and seem to be the largest contributors to crime rate changes. This finding does back up the results from our GLR, which calculated an R² value of 0.28 between the change in crime rates and the change in youth population percentage (Figure 3). These results fall in line with previous findings surrounding crime rate predictors (Ulmer and Steffensmeier 2014), suggesting that where there is more youth there is more crime.

Limitations

Despite this study not able to identify any significant relationships, there were a few notable findings that may be able to help better explain changes in crime rates. However, this does not go without any limitations. This study had two significant limitations that potentially impacted the results and caused the lack of any significant relationships or passing models. The first limitation is the Modifiable Aerial Unit Problem (MAUP) which stems from a lack of data. For this study, the boundaries for the 22 local planning areas were collected which divides the entire city of Vancouver into 22 polygons. This is clearly a massive problem because it aggregates all crime incidents into these polygons and significantly changes the distribution of the crime data. This problem can be addressed in future studies by using smaller areas such as dissemination areas (DAs), or by using a point density analysis to avoid clumping the data by neighborhood. The second major limitation of this study stems from the first limitation. To perform a reliable and accurate regression, more than 30 features are required to perform a regression. In the case of our Exploratory Regression, since the boundary data only provided 22 features. This impacted the results of the regression and could have caused certain inaccuracies in our results. This problem can also be solved in future research by dividing the city into smaller areas, such as DAs, to have more features to compare when doing both global and local regressions to calculate statistics.

Conclusion

In conclusion, this study followed spatial analysis methods typically used within crime analyses studies. Unfortunately, no passing models were found by our Exploratory Regression. The results from the GLR suggest there is a relatively high impact of the youth population on crime rates in an area compared to the other socioeconomic variables. However, due to the limitations acknowledged in our study, it is likely that our data is skewed based on the low feature count of the boundary file. The relationship between the inverse of change in population density and change in percentage of youth in a population with crime rates should be investigated further in future research, using other methods such as kernel density and smaller areas.