Hi all,
I have aggregated count data of criminal incidents in each neighbourhood of a particular city. I also have the population of each of those particulars neighbourhoods. I want to create a ratio crime per capita, (in others words the number of crimes of each neighbourhood divided by population of that particular neighbourhood). However, for two of the neighbourhoods, this ratio is very high compared with the remaining neighbourhoods, which is making them outliers.
My objective is to conduct a linear regression analysis using several others independent variables, which by scatterplot I have seen some good correlation with crime rates, although changed by the two outliers. I know the reason for this outliers. Those neighbourhoods with very high crimes rates (per capita) are due to their functionality. For instance, there are workers zones, where few people reside there, most of the buildings are for services and business, which are inflated my crime rates.
Maybe if I include population + workers I might resolve this problem, but I don't have such data. I am thinking to change those two outliers values to the lower up crimes rate, but I do not know if is the right thing. I also discarded the possibility of removing the two outliers because I have a limited number of observations.
I am here to see if someone can give me some advice and solution to this problem.
Thank you in advance!
I wonder if "per capita" is the right unit if you have neighbourhoods of different kinds that have systematically different "resident densities". In this case, the population size in the neighbourhood is not a good standard. Adjusting the model for the "functionality" of the neighbourhood would be one possible solution (I presume), but you say you don't have the information. Maybe you can make a sensible classification into different "functionalities" and use this as a factor in your model.
Thank you Jochen Wilhelm!
Please, I would like if you could give/provide me some further advice's/suggestions about how to do this sensible classification into different functionalities.
Thank you in advance!
That's domain knowledge. I can't really help you there. You should be the expert there...
Ok, thank you! I will look forward to some solutions!
I am thinking to use the volume of built-up area to normalize my variable instead of population, I think it may solve the problem of ratio inflation.
But I need to look at the literature to see if it will make sense.
You state that you are looking at criminal incidents to create a per capita ratio
What type of incidents are you considering? There are crimes of opportunity and crimes of passion There are crimes on businesses and crimes on persons. There are victimless crimes. You should limit the types of crimes to ensure that there is no confounding by the mixture of crimes. Socio-economics affect types and frequency of crimes. This could be a further source of confounding.
You have differences in locations of the portions of permanent and transient residents. Permanent and transient residents may be involved in different kinds of crimes, particularly crimes of opportunity. There are many possible sources of confounding. Ratios can be tricky when care is not taken to normalize groups. Ratios can inflate or deflate with lack of normalized groups.
Thank you Dear Joseph L. Alvarez,
I totally agree with you, indeed I have separated type of crimes.
I just got some data about number of Jobs in each zones, I am thinking to add it together with the population in each neighborhood and then normalize the crimes ( per type), using this variable (Population + Job) in the denominator). Please, I would like to hear your suggestion about this procedure.
Anyway, I need to normalize the crimes incidents, because there are strong correlation between numbers of crimes and number of population (size of urban area as well) in zones.
Thank you in advance!
I think this would further detoriate the interpretability of the data and the usefulness of your conclusions. Just my non-expert opinon...
I think shou should makes it simpler. Pulling different measures together, using some complicated ratios etc. makes everything extremely complex and hard to interpret eventually. Sometimes different things are just not comparable and need distinct ways of analysis. Possibly focus on a better defined subgroup, for instance.
Thank you, Jochen Wilhelm!
I am starting with statistics now, maybe I might be complicating things even more. All your comments and suggestions will be taken into account and they are being very useful.
- Badges
- Science topic