Understanding county-level COI estimates

The Child Opportunity Index (COI) demonstrates how resources important to children and families—for example, quality schools, parks and economic opportunities—differ from one area to the next. The COI was originally constructed to measure and compare conditions across neighborhoods, defined as census tracts. However, user interest has prompted us to create versions of the index for larger geographic areas, first ZIP codes and now counties. Counties are important geographic units in many parts of the U.S. in terms of governance, policy development and resource allocation that impact children’s wellbeing.

County-level COI 3.0 estimates for 2012, 2017 and 2021 are now available on our mapping platform and for download upon request. 

While county-level COI data is suitable for a number of purposes, users of this data should be aware of its limitations, which we describe below. 

We generally recommend the use of census tract-level data for applied uses, including public health surveillance and policy and program planning and implementation. In places where neighborhood opportunity varies substantially within counties, county-level Child Opportunity Scores and Levels are poor representations of the opportunities available to many children who reside in neighborhoods that have considerably higher or lower opportunity than the county average. 

How are the county-level COI 3.0 metrics calculated?

We took population-weighted averages of census tract COI data for each county and then computed county-level metrics, i.e., County Child Opportunity Scores and Levels, using the same approach we used to create census tract-level metrics (see Technical Documentation, p. 29). Using population weighting to aggregate COI data across census tracts within a county means that census tracts with a relatively larger number of children (in a given county and year) have a greater weight in the county-level estimates. 

Counties are classified into five Opportunity Levels (very low, low, moderate, high, very high), each containing approximately 20% of children living in either the nation (nationally-normed data) or the state (state-normed data). Additionally, counties are assigned a Child Opportunity Score from 1 (lowest opportunity) to 100 (highest opportunity) by grouping neighborhoods into 100 ordered groups.

COI 3.0 county-level data are computed for each year from 2012 to 2021 (data for 2012, 2017 and 2021 are presented in our interactive map). Like the census tract and ZIP code data, all metrics are benchmarked to the 2021 opportunity distribution. This means that, for Opportunity Levels, for example, we ordered neighborhoods in 2021 from lowest to highest in terms of their COI. We computed cut points (percentiles) that divide the neighborhoods into five ordered groups containing 20% of the child population each. We then applied the same cut points to data from other years as well. Because Child Opportunity Levels were defined in relation to the 2021 distribution of children across neighborhoods, we observe exactly 20% of children at each level only in 2021 data.

Counties can encompass many different neighborhoods, towns and cities. If I am using county-level estimates of child opportunity, what do I need to know about variability in opportunity within counties? 

Counties differ widely in size, from less than 50 residents (Loving County, TX) to nearly 10 million residents (Los Angeles County, CA). In larger counties and particularly in metropolitan areas, county-level estimates may obscure variation in opportunity across neighborhoods. In these counties, the county-level COI metric is a poor representation of opportunities available to many children who live in that county but reside in census tracts with considerably higher or lower opportunity than the county average. 

For example, Cook County, IL—home to the city of Chicago—is classified as a moderate opportunity county (2021 state-normed data). However, opportunity within the county varies considerably across its neighborhoods (census tracts): 29% of children (334,000 children) live in very low-opportunity neighborhoods, and 19% (221,000 children) live in very high-opportunity neighborhoods. Only 15% (174,000 children) live in moderate opportunity neighborhoods that match the county Opportunity Level. Furthermore, the children residing in very low-opportunity neighborhoods in Cook County account for 58% of all children residing in very low-opportunity neighborhoods statewide. A state policy targeting children in very low-opportunity areas using county-level data would therefore miss more than half of the children who would have been eligible had census tract-level data been used. 

Does the Child Opportunity Index mapping platform flag this variability within counties?

Yes. If you view county-level estimates of child opportunity and hover over a county, you may see a colored flag in the pop up. This flag signals that a share of children in the county—“little” (>0%), “some” (>25%) or “high” (>40%)—reside in census tracts with an Opportunity Level that is different from the county’s Opportunity Level. 

If you then click on a county, these colored flags are repeated next to the Child Opportunity Levels and Scores in the Location Details panel. The overall index and domain-specific sections of the Location Details panel contain the exact percentage of children residing in census tracts with Opportunity Levels different from that of the county. Even when the flag signals “little variability,” there still may be a significant share of children in neighborhoods with Opportunity Levels that do not match that of the county. 

Does the mapping platform display how opportunity varies by race/ethnicity within the county? 

Yes. In the Location Details panel, click on the carrot next to the overall index or domains. Each section includes a pop-out chart that shows the distribution of children across different Opportunity Levels within the county, both for the overall child population and by race/ethnicity. The racial/ethnic inequities in access to opportunity can be substantial. For example, in Cook County, IL, 51% of Black children reside in very low-opportunity neighborhoods, compared to only 2% of White or Asian children.

State by state, how well do county-level estimates reflect the distribution of opportunity across neighborhoods within those counties?

Measuring opportunity at the county level risks obscuring inequities in children’s access to neighborhood opportunity whenever their local neighborhood opportunity meaningfully differs from the county average. We analyze the extent of this potential misclassification in two ways. Underlying our analyses is a (hypothetical) policy application in which state policymakers use county-level—instead of neighborhood-level—estimates of opportunity to target resources to children residing in very low-opportunity environments.

One way to assess the extent to which county-level estimates accurately represent children living in very low-opportunity neighborhoods is to examine the “true positive rate,” defined as the percentage of children residing in very low-opportunity neighborhoods who are located within very low-opportunity counties. Specifically, the true positive rate is calculated as: 100 x (the number of children in very low-opportunity tracts who are also in very low-opportunity counties / the number of children in very low-opportunity tracts). If a state policy intended to reach children in very low-opportunity neighborhoods, but it measured neighborhood opportunity at the county level, the true positive rate captures the percentage of children in very low-opportunity neighborhoods who would actually be reached. If all children living in very low-opportunity neighborhoods also lived in very low-opportunity counties, the true positive rate would be 100%.

A second way to think about the mismatch between local neighborhood opportunity and county opportunity is the “false positive rate,” or the percentage of children who do not reside in very low-opportunity neighborhoods but do reside in very low-opportunity counties. The false positive rate captures the percentage of children who become eligible for resources even though they would be ineligible had opportunity been measured at the neighborhood level. Specifically, the true false positive rate is calculated as: 100 x [(the number of children who do not reside in very low opportunity census tracts but do reside in counties classified as very low opportunity) / (the number of children not residing in very low opportunity census tracts)]. If all children living in very low-opportunity neighborhoods also lived in very low-opportunity counties, the false positive rate would be 0%. 

While the true positive rate captures the percentage of the target population that is actually reached, the false positive rate captures the percentage of the population that is not actually eligible but ends up receiving resources. 

Figure 1 shows the true positive and false positive rates of county-level estimates of neighborhood opportunity nationally and for each state. For 22 states, the true positive rate is less than 50%, meaning that fewer than half of children in very low-opportunity census tracts reside in very low-opportunity counties. If state governments in those states used county Opportunity Levels to target resources to children living in very low-opportunity neighborhoods, they would miss more than half of their target population. 

The national true positive rate is 47%. The highest true positive rate for any state is only 74% (South Dakota and Maryland). The lowest true positive rates are observed in Illinois (22%), Florida (29%) and Oklahoma (31%). In these states, targeting resources to very low-opportunity counties would reach less than one-third of children residing in very low-opportunity neighborhoods. 

The national false positive rate is 14%. False positive rates range from 9% (Virginia) to 30% (Arizona) So, between 9% and 30% of children not residing in very low-opportunity neighborhoods would become eligible for resources if states used county-level opportunity to target resources. Figure 2 (below) displays true positive and false negative rates nationally and for all states, ranked from smallest to largest true positive rates.

In sum, relying on county-wide measures instead of localized neighborhood data risks both under-serving children who need help and over-extending resources where they may not be needed, both nationally and at the state-level. 

Are all versions of COI 3.0 available at the county level?

County-level COI metrics are only available as state-normed (for comparison of counties within states) and nationally-normed (for comparisons of counties nationwide) versions. There are no metro-normed county estimates. Users can map the overall COI and its three domains (education, health and environment, social and economic) at the county level. Both nationally-normed and state-normed county Opportunity Levels are available for domains and subdomains. Nationally-normed county Opportunity Scores are available for domains and subdomains. State-normed county Opportunity Scores are not available. 

The county-level estimates shown on your mapping platform don’t match the ones provided in the data files. Why is that?

All county-level estimates in our mapping platform are computed from 2010 census tract data. If you request county-level COI 3.0 data files from us, however, data from 2020 onwards is computed using 2020 census tract data.