By Ed Kissam and J. Gregory Robinson
Standard Deviations blog posts represent the views of the author/organization, but not necessarily those of The Census Project.
Although policy discussion of census equity has mostly focused on national level of differential census undercount in relation to OMB race/ethnic categories, our recent analysis of patterns of 2020 Census self-response in California focusing on Fresno County indicates there are still deeper pockets of undercount not evident in standard analyses based on the PES (post-enumeration survey) or DA (demographic analysis). We show how this negatively impacts neighborhoods and communities with concentrations of low-income Hispanic households and immigrants.
Relying on Census Bureau historical data we show that the level of census response in each racial/ethnic group has always been associated with differential undercount as indicated by the PES. We go onward to highlight the Census Bureau finding that, in 2020, household omission was three times higher (10.7%) in the decile of lowest-responding tracts nationally than in the highest-responding decile (3.3%).
This indicates that non-response follow-up (NRFU) was simply unable to fully overcome the problem of low response. Lower levels of tract-level response probably are ameliorated by NRFU operations but, inevitably, transformed into differential undercount. Recent Census Bureau research (Rothbaum et. al. 2021 and Brown et. al. 2023) strongly suggests that the same structural community characteristics that result in lower census and ACS response make both NRFU procedures and ACS weighting very problematic in many of the same areas where response is low.
Looking at patterns of response and presumed undercount in Fresno County, a Hispanic-majority county with a population of about 1 million, we have shown that the differential in tract-to-tract self-response within the county (and resulting undercount in some small communities) are comparable to the state-level differences in self-response when we compare the state with highest level of self-response (Minnesota) and the lowest (Alaska).�
In a state such as California, where the largest single ethnic population is Hispanic, equitable allocation of resources based on census-derived data needs to look deeper than race/ethnic profile of tracts to effectively direct efforts toward equity. Consequently, California mounted a vigorous Get-Out-The-Count campaign, targeted toward the tracts with high HTC (hard-to-count) index scores based on an algorithm developed by the state that incorporated 13 ACS variables and state data on broadband connectivity to previous ones developed by the Census Bureau.
Happily, the overall Fresno County Census 2020 self-response rate ended up being slightly higher than the national average. But there was still a 19.6% gap in self-response between the easiest-to-count quintile of tracts and the hardest-to-count tracts in the county. Despite the census promotion efforts, many of the high HTC areas continued to respond at lower rates. California efforts appear to have significantly improved overall census response but failed to assure equity in many local neighborhoods and communities.
Our streamlined regression analysis highlights the relationship between the �structural� characteristics of census tracts and census self-response and shows that, as has always been the case, socioeconomic factors such as higher proportions of renters, or higher prevalence of poverty, were associated with lower response.
But other factors played a role too. Most disturbingly Census Bureau introduction of the online response mode of data collection, while working well overall, broadened the gap between the digital �haves� and �have nots.� Our analysis shows that the level of broadband access already known to be lower in low-income than in affluent tech-savvy neighborhoods, was a prominent factor in uneven 2020 census response throughout California and especially in rural areas of Fresno County.
Concentration of non-citizens was also a significant additional factor in local level of 2020 self-response and ultimately, is likely to have contributed to undercount in areas with a higher proportion of recent immigrants, in part because NRFU is particularly challenging in such areas due to limitations of administrative records and reluctance to participate in proxy interviews.
It was not surprising to see that level of American Community Survey (ACS) response rate in a tract was correlated with Census 2020 self-response rate. Consequently, a sub-county region of Fresno County referred to as the �Westside,� with a much lower census response rate than the rest of the county (54%) also had an extraordinarily low ACS response rate (~26%)�only slightly more than half of the Fresno county average.
Key factors associated with low ACS response were % of adults with less than a high school education, prevalence of poverty, and concentration of limited-English households�key features of socioeconomic disadvantage.
The bottom line is that the response patterns we analyzed in Fresno County and in California show that census omission is �multi-dimensional,� that tracts, neighborhoods, and communities left out of the 2020 Census and ACS, are those that most urgently need the social program funding allocated on the basis of census-derived data.
We considered how the combination of low census response and low ACS response in the Westside and other sub-county areas of Fresno County might affect distribution of ESEA Title IA funding and educational equity within Fresno County where the program provided about $124 million in FY21-22 funding since Title IA funding is based on # of children 5-18 in poverty in a jurisdiction. We are not yet prepared to definitively quantify the projected impact for our case study area but the stakes are in the millions of dollars per year for the marginalized communities of the Westside (which consist of 9 low-response census tracts) where education is the key to future community well-being and in the billions nationally.
Our Fresno County case study looks at only a small area in a tremendously diverse nation, but it highlights the fact that national-level and state-level estimates of differential undercount do not adequately ameliorate the funding inequity stemming from differential undercount.
We believe the local dynamics of census and ACS response affect many disadvantaged neighborhoods across the United States in similar ways we have observed in Fresno County and other areas of California (Los Angeles County) and that, practically speaking, equitable allocation of federal (and other) social program funding should take into account the fine-grained variations in local census self-response as an indicator of underrepresentation in census and ACS data. Further research on distinct local factors affecting response level in other hard-to-count areas would help in better understanding the magnitude and extent of the problem.
Our referenced paper provides a detailed discussion of the analysis and offers suggestions for Census Bureau research in preparation for 2030.
