The Child Opportunity Index (COI) is a composite index of neighborhood opportunity published by diversitydatakids.org for all US census tracts. Information on individuals’ ZIP codes, however, is often easier to collect and analyze and more commonly available in many databases than information on individuals’ census tracts. Responding to numerous requests, we created ZIP code COI estimates to facilitate applications of the COI whenever census tract level analyses are not possible. While we generally recommend the use of census tract COI data if possible, our analyses also show that ZIP code COI estimates can be used instead of census tract data for certain purposes.
ZIP codes sometimes cover census tracts with different levels of opportunity
There are about 39,000 ZIP codes and 73,000 census tracts in the United States. ZIP codes are collections of mail delivery routes drawn by the U.S. Postal Service to facilitate efficient mail delivery. They do not have exactly defined geographic boundaries. Census tracts, on the other hand, have precise boundaries that are defined by the Census Bureau and divide the country into small areas with an average population size of 4,000 people. Addresses with the same ZIP code are often found in multiple census tracts. In these cases, ZIP code COI estimates are calculated as weighted averages across census tracts linked to a given ZIP code. Further details on ZIP codes and census tract COI data is available in the COI ZIP code data technical documentation.
When ZIP codes cover larger areas that do not differ in terms of the opportunities they provide for children, ZIP code level COI estimates will properly reflect the opportunity levels in that area’s census tracts. However, if ZIP codes cover larger areas that differ substantially in terms of their neighborhood opportunity, ZIP code level estimates will be a poor representation of opportunity levels in the neighborhoods they contain.
Results from empirical studies are consistent with both of these statements, so users may wonder: When is it okay to use ZIP code data?
ZIP code COI estimates are an appropriate substitute for census tract COI data for correlational analyses in large datasets
If using census tract COI data is not possible, it is appropriate to use ZIP code COI estimates for correlational analysis in large, nationally-representative datasets containing data from many ZIP codes—for example, to quantify the association between residential contexts and child outcomes. Two recently published studies have used ZIP code level COI estimates in this context: Bettenhausen et al. find pediatric hospital readmission to be elevated by a factor of 1.2 among patients residing in very low vs. very high opportunity neighborhoods; and Gastineau et al. find that hospital encounters due to gunshot wounds are 10 times higher among children residing in very low vs. very high opportunity neighborhoods.
In these cases, ZIP code COI estimates are appropriate substitutes for census tract COI data. Our analyses indicate small attenuation bias (bias towards zero) if ZIP code estimates are used instead of census tract estimates. Linear regression coefficients are about 10% smaller if residential contexts are measured at the ZIP code rather than census tract level. This suggests that by aggregating census tract COI data to ZIP codes, we lose some meaningful variation in neighborhood conditions. The loss, however, is small on average. We suspect that this holds true for analyses that are representative of large US states or metro areas, but it may not hold for datasets of smaller areas with relatively few ZIP codes.
Using ZIP code COI estimates for correlational analysis whenever census tract analyses are not feasible is also consistent with studies that have examined the association between area characteristics and health outcomes. These studies (Thomas et al., Lovasi et al., Fiscella and Franks) report small if any differences across estimates using census tract vs. ZIP code measures. Our recommendation is also consistent with estimates published in studies (on sexually transmitted infections, tuberculosis and violence; low birth weight and lead poisoning; and mortality and cancer incidence) that highlight limitations of ZIP code data. Data published in these studies shows that ZIP codes and census tract metrics are, on average, equally strongly associated with different health outcomes.
Using ZIP codes to define communities eligible for a program or intervention can result in high miss rates
Whenever ZIP codes cover areas that differ substantially in terms of their opportunity, it is problematic to use ZIP code data to target areas for investment or intervention or to determine program eligibility. Our analyses indicate that 66% of ZIP codes fall into two or more census tracts, 25% fall into six or more tracts, 10% fall into 11 or more tracts, and 1% fall into 20 or more tracts. We also found that ZIP codes tend to cover more census tracts in urban areas. This fact that ZIP codes cover many census tracts is not problematic on its own. However, it becomes a problem for spatial targeting—or defining eligibility at the community level—if the census tracts that a ZIP code covers differ substantially in terms of neighborhood opportunity.
Imagine, for example, an organization that considers using the COI to identify where to open a health clinic to serve children in a “high-need community,” using “very-low opportunity” as an eligibility criterion. If ZIP codes were used to identify high-need communities, some very low opportunity census tracts could be classified as ineligible because they are part of larger ZIP codes. A moderate opportunity ZIP code would be ineligible for the clinic even though it contains census tracts that are very low opportunity and that would be eligible if census tracts were used for targeting. So, even though a child’s home census tract is very low opportunity and therefore would meet the eligibility criterion, it would not be considered because it has been grouped into a ZIP code with other census tracts that have higher levels of opportunity. Using national data, we found a miss rate of 28% when ZIP codes rather than census tracts were used to determine “high need status”—in other words, 28% of children reside in very low opportunity census tracts that are not located in very low opportunity ZIP codes. If eligibility was defined at the ZIP code level, these children would be ineligible to receive services.
We do not think that ZIP codes are problematic for spatial targeting in all cases. There may be areas where ZIP codes are comprised of neighborhoods with homogenous opportunity levels. In these instances, ZIP codes are appropriate for targeting. However, it is not clear when and where this situation arises, so planners should draw on external information to ascertain whether or not ZIP codes are sufficiently homogenous in terms of opportunity. For example, they could compare ZIP code level and census tract level COI maps to assess to what extent ZIP code-level maps tell a different story about the location of very low opportunity populations.
ZIP code level data can make racial/ethnic inequities in access to opportunity appear smaller than they really are
Our analyses also show that ZIP code-based estimates tend to overstate access to neighborhood opportunity for Black and Hispanic children and understate neighborhood opportunity for White children. This occurs because within larger ZIP codes covering multiple census tracts, White children tend to live in areas with higher opportunity, while Black and Hispanic children tend to live in areas with lower opportunity. While the within-ZIP code racial/ethnic gaps in opportunity are small on average, they can be of considerable magnitude in specific ZIP codes. This will tend to shrink observed inequities in access to neighborhood opportunity.
To quantify this bias, we calculated Child Opportunity Scores for the neighborhoods where the typical Black and White child reside in for each of the 100 largest metro areas, first using census tract COI data and then using ZIP code COI estimates. We find that the Black-White opportunity gap for a given metro area tends to be smaller if calculated using ZIP code COI estimates, compared to the gap calculated using census tract COI data. This bias is small on average; in most large US metro areas, broad patterns of racial/ethnic inequity can be appropriately captured using ZIP code data. At the same time, the ZIP code COI estimates for individual ZIP codes can be a poor representation of the levels of opportunity experienced by children of different races/ethnicities residing in that ZIP code.