Key Neighborhood Indicators


The basic idea of the Key Neighborhood Indicators is to provide an alternate way of providing a picture of what a neighborhood is "like" by taking important data items (such as poverty rate, median household income, percent African-American, median age, percent Over 65, etc.) and placing them in data sets that have not just the actual values of these items but also related variables (3, so far) that measure how the indicator compares to other neighborhoods in the universe. These measures include the percentile rank of the indicator (i.e. is this a very low or very high value compared to other neighborhoods in the universe), the difference between the value and the universe mean, and the "Z score" indicating the number of standard deviations above or below the mean.

What Are the Neighborhoods?

We use the latest (2010) Census Tracts as our neighborhoods. These are statistical areas that are defined for use in the decennial census. They are also used for summarizing data from the American Community Survey for the decade following the census. (So we should have data for these units for at least until 2022 before needing to revise and start using 2020 tracts.)

Census tracts are contiguous and somewhat homogeneous areas, usually defined by local census tract committees (at least in urban areas; in rural areas they often wind up just being assigned by the Census Bureau and are much less likely to follow locally recognized "neighborhoods"). So even though we have census tracts in rural areas, there are significant differences between urban and rural tracts. The latter tend to be much larger spatially, but not always in terms of population, and also are typically less homogeneous than their urban counterparts. Of course, there are also significant differences in terms of the indicators data for the two types of area. Age distributions, racial mix, educational attainment, mobile homes and housing values tend to vary significantly between urban and rural areas. This is why we have designed the system to allow treating these two subcategories separately.

While census tracts are not formally classified as urban/rural (about 26% of the nation's over 73,000 tracts are partly both; about 11% are entirely rural and 62% entirely urban) we can easily determine what portion of a tract is urban (based on population in census blocks from the 2010 census). So we can (for the purposes of KNI classifications) define a tract as urban if at least 50% of its population lived in areas (census blocks) classified as urban by the Census Bureau as of the 2010 census. (Rural is then simply not urban).

What Are the Geographic Universes Used for Ranking?

Much of the data found in these data sets are measures of how the indicator raw values compare with all tracts within a ranking universe. We use the following four ranking universes:

  1. All tracts in the same state.
  2. All tracts in the same state with the same urban/rural category as the tract.
  3. All tracts in the United States.
  4. All tracts in the United States with the same urban/rural category as the tract.

How Are the Data Organized Within the Data Directory?

We have 51 x 2 = 102 state-based data sets. For each of the 51 states we have a data set with indicators ranked versus all tracts within the state, and a second that has indicators ranked versus the appropriate urban/rural categories (per the 50% rule discussed above). If you want to look at data for Pennsylvania using all tracts for comparison you want the data set pakni<yy> (where <yy> is the ACS vintage year, currently 14). The data set paurkni14 has data similar to pakni14 except that here the ranking universes used are the appropriate urban or rural tracts for the state.

We also have a pair of data sets named uskni14 and usurkni14 that contain data for all census tracts in the entire country where the ranks are based on all tracts in the entire country (50 states + DC) rather than just the same state.

What Indicators are Available?

The best source of information regarding the indicators (aka variables) is the Metdata.html file (web page) in this data directory. Note in particular the paragraph titled "Key Indicators Come in Sets of Five".

Can I Access These Data Without Using Uexplore/Dexter?

Yes -- but not in the way we hope to do it eventually. For now, we have only a very crude web-based demo at testkni.shtml that allows you to enter a census tract code (assumes you know what these look like) and have a page generated that shows the highlights for that tract (neighborhood). You can also enter just a county code and get access to highlights for all tracts in that county.

Do You Have a Better Idea?

This filetype and the tools that we hope to build to make the data accessible and useful is very much a work in progress. We would like to get feedback from users on how they would like to see this go. In particular, we would be interested in knowing what other key indicators would be of interest (as well as comments regarding current indicators that you think the world could live without.) We want to keep the number of indicators relatively small and avoid redundancy (multiple measures of essentially the same thing). Send your comments to us using the Questions/Comments link at the very bottom of this page.

Potential Applications

Here are some ideas for how we might make these data accessible. They may not all be feasible (given the limited resources of OSEDA/MCDC) but it won't hurt to dream. Let us know which ones you'd be interested in.

  1. The web application used to specify the tract(s) of interest can have various ways of allowing the user to specify the tract(s). Such as:

    1. A series of drop down menus where user selects a state (a parm can be used to specify the initial --default-- value), then a county, then 1 or more tracts within the chosen county.
    2. User enters a code for a county or a PUMA and gets a map of tracts for that PUMA (displayed with a Google map background showing street detail) as well as tract outlines and labels. User clicks within a tract to select and get detailed report. (Moving the mouse over a tract on the map displays a few key indicators for that tract - a sort of preview of what is displayed when they click on it.)

    3. Application allows user to enter a point location (lat-long) and have the app determine the tract and display data for it.

    4. Similar to previous but runs on a cell phone and uses the current GPS coordinates as input. Has an option where it automatically detects when you cross into a new tract and displays the data for new neighborhood.

    5. Application lets you enter a street address and have it assign a tract to that location and then display the key indicators for it.

  2. Each key indicator has a set of 4 US google maps associated with it (for the 4 universe types). The user selects the universe type and the indicator to display the map. E.g. a map of PctPoor using all tracts in the entire US as the universe. Or medianHouseholdIncome for all urban/rural tracts in the state. One or more states can be specified to limit display. An Urban/rural filter can also be selected to display only tracts for urban or rural areas. These maps would have a standard shading with 3 shades or red to indicate lowest values and 3 shades or green to indicate highest. Tracts in the "big middle" (25th to 75th percentile) would be unshaded. We could switch the red/green low/high for certain variables (such as PctPoor) where red would be high and green low. Some variables (e.g. Percent Asian) are skewed distributions that have only one end of the spectrum.

  3. Dynamic query app that lets user select geography (by state, county or PUMA) along with other filters based on indicator values. E.g. user could select all neighborhoods in California where the percentile rank category for PctPoor (PctPoor_pcat) was 5 or 6, and could also specify they wanted PctBlack to be >= 50. User could also specify what variables to keep. You can do this now with Dexter but we would want to have a more friendly user interface. It would also be nice to have a map display of the resulting filtered data set. Having a tract-level data set showing poor neighborhoods may not be meaningul to most users not familiar with tract geography - but a map of the area showing those tracts with ability to see street-level detail would be much useful for most users.

  4. Create new pseudo-geography categories that could be incorporated into MABLE/Geocorr. The Source and Target select lists would be supplemented with Source & Target neighborhood types. These would have names such as "Very Poor", "High Income", "Predominantly black", "Commercial" (lots of businesses), "Significant Hispanic", etc.