The set of demographic variables used on MTCF were derived from data available in the American Community Survey (ACS). The ACS is an annual survey sample to track demographic changes in the U.S. over shorter timeframes and in more detail than the decennial census from the U.S. Census Bureau. Larger divisions are political areas such as counties, townships, or cities. However, nested within these areas are Census Bureau-specific entities known as census tracts. Census tracts are mutually exclusive and exhaustive divisions of the U.S. that are intended to contain about 3,000-5,000 residents.
For each census tract, data are aggregated across all the surveyed residents of that tract. For example, variables include the percent of residents of a census tract who self-identify as Black, the percent of residents who report that they walk to work, or the median income of all households in the Census tract. The ACS 2019 aggregate 5-year estimate files were used, which represent data from 2015-2019, for crash data years 2016-2019. For crash years 2020 and 2021, the 2020 aggregate 5-year estimates were utilized, which represent data from 2016-2020. Crashes were then located within the available Census tracts. It is important to note that Census tract characteristics tied to a crash are specific to the Census tract in which a crash occurred and not necessarily specific to that crash and the people involved in the crash.
To select relevant and informative demographic variables from the census data, meaningful variables were chosen based on their relevance to crashes. Due to the large number of variables and their potential interrelatedness, a factor analysis was conducted to examine the proposed census variables for correlation (e.g., median income is often highly correlated with percent employed or educational attainment level). Factor analysis essentially identifies groups of related variables, which are called “factors.” Highly correlated variables were not included in the same logistic regression analysis, which was used to identify variables that were significant when modeling fatalities and suspected serious injuries. From the significant variables in the models, a total of 26 demographic variables were selected for inclusion on MTCF.
In order to display these demographic variables in a useful way on MTCF, crashes were grouped in quintiles that will enable users to generate tables and compare patterns. Quintiles divide crashes in equal percentages across groupings. In some cases, census tracts have a value of 0% for more than one quintile grouping due to low percentages across all Census tracts in Michigan. This results in some of the lower quintiles being indistinguishable. Percent American Indian/Alaska Native and Percent Arabic Language were affected and adjusted across quartiles with the smallest group set at 0%. The three language variables were selected from those available in the ACS to reflect the most commonly spoken languages across Census tracts in Michigan.
Each newly generated quintile variable based on Census tracts is prefixed with the word “Demographic” for clarity and ease of use. All 26 variables are available for use in generating charts or tables when viewing results. Although at the crash level, these variables are all shaded gray to distinguish them from the current crash level (white), vehicle level (blue), and person level (pink) variables.
Rural/urban designations were assigned to crashes using the urban areas shapefiles available from the U.S. Census Bureau. These are not necessarily the same as Census tracts, but the assignments of both rural/urban and demographic variables were linked to individual crashes. Due to the strong correlation between the demographic variables and the rural or urban crash concentrations, it is strongly recommended that the Rural/Urban filter is used when generating chart or table results.
For more information about the demographic analysis and results, please see Demographic Analysis of Michigan Traffic Crashes 2016-2019, located on the Reports page.