Hello,
I'm new in the data analysis world and working on longitudinal analysis. I have the crime rate for 100 neighbourhood crime data for the period of 5 years (ex. 2000-2005). I have done the clustering part and the neighbourhoods have been clustered in 3 groups based on the crime rate during the 5 years.
Now I want to do multinomial regression and one of the variables I'd like to test is the population and I have the population number for each neighbourhood for 2000 and population number for 2005
my question is that, when I want to include the population variable in the regression, should I put the population of the first year (in 2000) or the last year or take the average? How should I deal with this and with the change in other variables during the study period such as unemployment rate etc.
Thanks,
IS the crime data for each year? Or an average across the 5 years? If it is each year, use population from each year. If you have yearly data for both crime and population, you have small cross-sectional-time series model. If you treat it that way, your 100 neighborhoods give you 500 rather than 100 observations.
But -- the problem YOU have is that you say you have crime rates. The rate is ALREADY population adjusted, so if you predict the crime rate with population, you are predicting a variable (crime rate) with an independent variable that will be related because it contains part of the dependent variables (crime rate is crime/population x 100,000).
Michael J. Lynch Thanks for your detailed answer. I imported the full crime dataset for 5 years. Then, I grouped the data by year for each neighbourhood. After that, I used the population data of the first year to calculate the crime rate ((crime/population of 1st year x 100,000)). So I wanted to make sure that is calculating the crime rate (for all 5 years) by first year population wright or should I take the last year population into my consideration?
You should use the annual population. I assume you don't have annual data for population, but you can easily estimate it by using first and last year. So Michael Lynch is correct: you should use annual data for each variable, use the correct annual population to calculate the crime rate for each year, and use annual unemployment rates, etc. for each of the other independent variables. For those variables for which you do not have annual data, you need to interpolate -- i.e. estimate the variable based on what you do have.
- Badges
- Science topic