Model-Based Forecast Means of Suicides and Homicides
This [unedited] guest post is by a student in my PSTAT262MC class (background post).
Please praise/critique/comment on its quality and importance to you.
![susie [thumbnails].jpg](http://jaradniemi.com/blog/includes/class/pstat262mc/susie [thumbnails].jpg)
Susie Kang says:
Since the data for both time series—annual gun-related suicides and homicides--show neither a linear trend nor seasonality, the data are best described as a local level model, or a random walk with errors.
For 2005 (1 year out), the model-based forecast mean for gun-related suicides is 0.958, with 95% confidence interval (0.231, 1.685), and the model-based forecast mean for gun-related homicides is 0.196, with 95% confidence interval (-0.120, 0.512).
For 2014 (10 years beyond the last time point of the model), the model-based forecast mean for gun-related suicides is 0.958, the same as that for 2005, but with a larger 95% confidence interval (-0.420, 2.335). Similarly, the 2014 model-based forecast mean for gun-related homicides is 0.958, the same as that for 2005, but with larger 95% confidence interval (-0.185, 0.577).
In my previous post, my predictions had unrealistically tighter bounds for 2005 and 2014: 0.8 suicides with 95% confidence interval (0.7, 0.9) and 0.15 homicides, with 95% confidence interval (0.1, 0.2).
blog comments powered by Disqus
Susie Kang says:
Since the data for both time series—annual gun-related suicides and homicides--show neither a linear trend nor seasonality, the data are best described as a local level model, or a random walk with errors.
For 2005 (1 year out), the model-based forecast mean for gun-related suicides is 0.958, with 95% confidence interval (0.231, 1.685), and the model-based forecast mean for gun-related homicides is 0.196, with 95% confidence interval (-0.120, 0.512).
For 2014 (10 years beyond the last time point of the model), the model-based forecast mean for gun-related suicides is 0.958, the same as that for 2005, but with a larger 95% confidence interval (-0.420, 2.335). Similarly, the 2014 model-based forecast mean for gun-related homicides is 0.958, the same as that for 2005, but with larger 95% confidence interval (-0.185, 0.577).
In my previous post, my predictions had unrealistically tighter bounds for 2005 and 2014: 0.8 suicides with 95% confidence interval (0.7, 0.9) and 0.15 homicides, with 95% confidence interval (0.1, 0.2).
blog comments powered by Disqus