The Rise: Region and Rurality

The Rise: Region and Rurality

Note: this is part of the Youth Suicide Rise project.

Regions

The rise in child suicide was substantially smaller in NE that in the three other regions:

While subgroups based on sex/race/ethnicity tend to have higher increases when they have lower suicide rates, regions reverse this trend, since the Northeast has had substantially lower child suicide rates than the rest of the country.

Rurality

Rural areas had a slightly higher (130%) increases 2007-2018 than non-rural areas (120%), also despite starting with already higher child suicide rates.

Demographics

The Northeast and Midwest lost some child population while the South gained some:

It turns out these changes essentially cancel each other in regards to rates, since the Midwest has high rates and the South just below average rates; thus the predicted 2018 child suicide rate -- based on changes in regional population but no changes in regional rates -- remains the same as in 2007.

Rural child population decreased by about 7% while non-rural population remained nearly the same (about 0.2% child population gain).

To summarize, regional and rural demographic shifts explain no part of the rise -- if anything, the decreasing rural population would predict lower rates in 2018.

Notes

The CDC WISQARS tool offers two 'metro' classifications: based on 2006 data or 2013 data.  Since 2013 is midpoint between 2007 and 2018, we use this classification.  Using the 2006 metro classification leads to very similar results.  Using the 2006 classification for 2007 and the 2013 classification for 2018 would lead to a greater loss of 'rural' population but once again similar results regarding child suicide rates.  Ideally, there would be a metro classification based on 2017 data, so we could use a proper 'dynamic' method of classifying rural areas -- the results would presumably be similar, except the loss of 'rural' population would be even greater.

Simpson's Paradox

According to the 2013 metro classification, the child suicide rates in 2018 were 2.29 for rural and 3.80 for non-rural areas, while according to the 2006 metro classification, the child suicide rates in 2018 were 2.27 for rural and 3.76 for non-rural areas -- thus the latter partition has lower rates for both rural and non-rural areas yet the overall rates must be necessarily the same (2.50):

M1 partition: 2.29 rural, 3.80 non-rural, 2.50 overall

M2 partition: 2.27 rural, 3.76 non-rural, 2.50 overall

This is a nice illustration of Simpson's 'paradox'.