Suicide Rates in US Counties and the Relevance of Area Socioeconomic Structure

Analysis of geographic variation in suicide and other psychiatric outcomes has demonstrated the impact of latent area scales (e.g. deprivation, fragmentation, rurality). For example, recent work on suicide and attempted suicide in the UK and Ireland has demonstrated an independent effect of social fragmentation (meaning absence of community ties linked to residential turnover, many one person households, etc) after allowing for deprivation. Variables such as fragmentation and deprivation are not observed directly, but may be proxied by collections of observed indicators (e.g. census data on social and demographic structure). Using such observed data, the latent variables may be derived by conventional multivariate techniques (e.g. by principal components), by composite variable methods, or by methods which explicitly consider the spatial framework of areas, and in particular the spatial clustering of latent risks and outcomes. This talk considers a latent random variable approach to explaining geographic contrasts in suicide in the US, and develops a spatial structural equation model incorporating deprivation, social fragmentation and rurality. The approach allows for latent spatial constructs to be correlated both within and between areas. For example, whereas UK studies show some positive correlation (within areas) between deprivation and fragmentation, this may not be true of the US, where poverty tends to be higher in rural areas. Potential effects of area ethnic mix on suicide are also included, in particular the higher suicide levels among white non-Hispanics and native Americans as compared to black non-Hispanic and Hispanic groups. The model is applied to male and female suicide deaths over 2002-2006 in 3142 US counties.