Blog Archive

Archives by Month:

October 2017

2020 Census Budget Challenges

2020 Census Budget Challenges

This is the third blog post in an in-depth series exploring the history and future of the U.S. Census. Explore our first post introducing readers to the history of the census and its importance or the second post introducing the census budget.

The Census Bureau has made great strides in ensuring that the 2020 Census is filled with cost-saving innovations and stays below its cost target of $12.5 billion.  Despite the progress, however, there are still many challenges ahead before implementing an efficient census. We’ll discuss the some of the Census 2020 solutions in the next blog, but first we detail some of the most pressing hurdles the bureau needs to overcome to stay on target.

Coordination and Alignment

In a recent report, the GAO commented that at that their core the Census Bureau and CEDCaP (a new data management system) were severely lacking alignment and coordination in scheduling, priorities, and risk assessment. The GAO also noted that this could lead to duplication of effort in some circumstances and put both organizations at risk of underperformance. Similarly Dave Powner from the GAO suspects the IT work could end up costing upwards of $1 billion extra on top of the $2.4 billion estimated currently due in part to these inefficiencies in coordination.  

Related to the topic of coordination, in May the Census Bureau’s former director John Thompson unexpectedly resigned after a string of questioning surrounding the difficult 2020 budgeting process. With only 3 years until 2020, not having an experienced director managing the operation could potentially lead to struggles in internal coordination and a new nominee will need to be chosen quickly if they are to be an effective replacement.

Additionally, in its analysis of the 2010 Census, one of the largest problems the GAO noted was lack of internal coordination and communication within the Census Bureau, leading to inefficient distribution of resources and less than ideal outcomes. The bureau is going to have to focus hard on coordinating its goals and processes across the organization in order achieve the intended cost savings of its innovations.

The U.S. Population is Becoming Harder to Count

Another upcoming and ongoing challenge that the bureau can do very little about is that the U.S. population is becoming harder to count in general. This is due to a variety of reasons, one being that the nation is becoming increasingly diverse, and diverse populations are harder to capture under a single strategy so more efforts need to be made to address the many diverse groups (a 3-part series on hard to count populations is forthcoming). Further complications might be found in the nation’s decreased trust in the government in general, possibly resulting in an increased reluctance to fill out the census questionnaire as there is doubt the government will use the information properly.

The population is also trending towards more complex and unconventional living arrangements that make targeting everyone in a household especially difficult. For example, houses with many adult roommates may not know if only one of them should fill out the form, or if they should each fill it out, perhaps leading to only half of the house being counted. This is further exasperated by the fact that the nation is increasingly mobile and the bureau cannot as easily keep track of where people move in and out of, potentially wasting significant resources on unoccupied addresses.

Rapidly Changing Technological Environments

A major challenge in budgeting for the 2020 Census is the rapid rate at which technological innovation occurs. This may initially seem like it would be a benefit to budgeting, but it can make planning quite difficult. The Census Bureau has an obligation to lay out its strategies and budgets for the coming Census as precisely as possible many years in advance. However, they also face a strong demand to make use of the latest/most efficient technology when conducting a census. This leads to challenges as the bureau cannot predict what technologies will be available come census time when they are releasing preliminary budgets and strategies many years in advance. This leads to a constant balance between using older technologies for consistency and using new technologies at costs that are potentially higher than estimated.

Constrained Fiscal Environments

In President Trump’s recently released 2018 budget, he highlighted the 2020 Census as a priority item, opting to provide an additional $130 million to the Bureau. This may appear like a commitment to ensure the census has the resources it takes for 2020, but the cyclical nature of the census budget means that the Bureau require more funding the closer they get to a decennial census year. In reality, the $130 million increase was a significant decrease from the bureau’s proposed 2018 budget.

Although the budget cuts to the Bureau are small relative to the total cost of the census, they have the potential to have significant long term costs. For example, in order to carry out an efficient and effective 2020 Census the bureau has to ensure that the system they have in place is going to function as intended. In order to do this they carry out field tests and studies in the years leading up to the Census to determine what the most efficient methods are in designing the Census plan. Recent concerns over the annual budget have caused the Bureau to cancel a variety of field tests in 2018, potentially leading to a less efficient operation in 2020. In general, increasingly constrained fiscal environments could make proper design and testing difficult.

Our next blog post will discuss some of the innovations and efficiencies that the Census Bureau has created for Census 2020.

Census 2020 Blog Posts

  1. What is the Census?
  2. U.S. Census Budget Introduction
  3. 2020 Census Budget Challenges
  4. 2020 Cost-Saving Innovations
  5. How Does the Census Use Local Administrative Data?
  6. Data Security and the Census
  7. Redistricting and the Census


Constructing an Index of Food Insecurity

The Issue

As part of a project for Forgotten Harvest, a Detroit area food bank, Data Driven Detroit was asked to construct a single measure that would quantify food insecurity levels by census tract in the tri-county area of southeast Michigan (Wayne, Oakland, and Macomb Counties).  The measure, dubbed the Food Insecurity Index, had to satisfy four criteria.  It had to

  • be based on publicly available data at the census tract level;
  • incorporate multiple factors affecting food insecurity;
  • be independent of direct measures of food insecurity such as interview data with residents; and
  • be methodologically sound but readily understood.


Defining the indicators in the index

Household finances are recognized as central to a household’s ability to provide food for its members 1.  Accordingly, household finances became the measure around which we selected other variables to broaden the index.  As a first step, then, we constructed four indicators at the census tract level of household finances.  These four were:

  • median household income;
  • count and percentage of households receiving Supplemental Nutrition SNAP benefits;
  • count and percentage of households below the poverty level; and
  • count and percentage of households with housing costs greater than 30% of household income, which we labeled “housing burden.”  


In order to select one of the four measures of household finances we examined each measure’s relationships with other potential measures of food insecurity such as lack of a personal vehicle.  Our goal was to select the household finances indicator that was related to the other potential indicators but not so strongly related that the index would essentially consist of multiple measures of the same concept.  The housing burden variable best fit these criteria. Ultimately, the Food Insecurity Index was composed of the housing burden measure and the following five measures, four of which were drawn from the 2010-2014 Five-Year American Community Survey:   

  • households with no vehicle – the count or percentage of households in a tract with no vehicle available
  • individuals with less than a  high school diploma – the count or percentage of individuals age 25 years or older in a tract with less than a  high school diploma
  • low-income movers — the percentage or count of individuals in a tract with incomes less than $25,000 who moved in the past year
  • beyond 1/4 mile of transit– the percentage of land area in a tract more than  one-quarter mile from a bus stop
  • single-parent households – the percentage of families in a tract with children younger than 18 years of age that are headed by a single parent


Method of constructing the index

To construct the index we utilized a slightly modified version of the methodology used by the Census Bureau in the mid-2000s to identify census tracts predicted to have low response rates in the 2010 census 2. For the Food Insecurity Index, each one of the six indicators was sorted from low to high across the 1100+census tracts.  For each indicator, each of the tracts was assigned a number from 1 to 11, with a “1” indicating a low likelihood of low food insecurity and “11” a high likelihood of food insecurity.  Finally, for each census tract, the scores were summed across the six variables to get the Index value, with values ranging from 6 to 66.

The final step in constructing the Index was deciding whether to represent all six indicators as percentages or to rely on numbers for four of the indicators and percentages for the ¼ mile from transit and single-parent household indictors.  We constructed the Index both ways and compared the results.  While the two indexes painted similar pictures of the geography of food insecurity in the tri-county area, we decided that the index that included numbers for the first four variables rather than percentages gave a clearer picture of the magnitude of the problem.


Scores on the Index

Figure 1.

Figure 1 illustrates the distribution of scores of the Index for the three counties.  The lowest score for any census tract was 8 (indicative of low food insecurity), while the highest score was 65.  The median score was 38.  The distribution looks roughly bell shaped.

An advantage of the methodology used to construct the Index is that it recognizes and incorporates variation within census tracts.  For example, there are large portions of Wayne, Oakland, and Macomb Counties that are not served by buses.   Census tracts in those areas are therefore likely to have a relatively high score on the “beyond quarter mile of transit” indicator even though they may have low scores on the other measures that would otherwise indicate food insecurity.  Due to this their overall Food Insecurity Index score would be in the lower range of the Index despite their lack of proximity to a bus stop.

Figures 2 through 4 show the distribution of Food Insecurity Index scores for each of the three counties.  In comparing the three county graphs, the salient feature is the shape of the distributions, not the total number of tracts per Index value, since the three counties have different numbers of tracts.

Figure 2.

The median Index score for Macomb County census tracts was 40, with scores skewed to the higher values.  

Figure 3.

Oakland County, in contrast, had a median score of 33.  Its distribution is skewed toward lower scores, indicating low food insecurity.  Nonetheless, there were numerous tracts with high scores.

Figure 4.

Wayne County’s median score of 39 masks the wide variation in scores across the county, denoting quite a few tracts that are less likely to face food insecurity but far more where it was a problem.


Map of Food Index Insecurity Index Values in Wayne, Oakland, and Macomb Counties

While Figures 1 through 4 illustrate the variety of levels of food insecurity in the three counties of southeast Michigan, they do not locate the differences geographically.  That is displayed in Figure 5.

Figure 5.

Overall, these populations are heavily concentrated around Detroit and Pontiac, often in areas that are fairly well served by existing and potential Forgotten Harvest (FH) and Gleaners Community Food Bank (GCFB) agencies. However, the Food Insecurity Index also identified several other areas of potentially high need, including several that are not currently well-covered by the existing FH/GCFB network. These areas include:

  • Madison Heights, between I-75 and the Oakland County Line
  • Wixom, between Pontiac Trail and I-96
  • Romulus, east and south of Detroit Metropolitan Airport
  • Roseville and Clinton Township, along Gratiot Avenue
  • Southfield, near the I-696/US-24 interchange
  • Van Buren and Huron Townships in southwestern Wayne County
  • Portions of Chesterfield Township and New Baltimore in Macomb County


Of the Top 20% of tracts that were most likely to be food-insecure (having Index scores of 49-65), 63 of the 211 were located in Detroit.  This represented over 20% of the city’s census tracts.  However, 148 of the 852 census tracts in outer suburbs also placed into this range, demonstrating that food insecurity is a regional problem.  

For example, Novi and Wixom in southwestern Oakland County are often considered affluent areas with few residents in need of assistance, but the Index data showed that there is actually quite a bit of need that may not be being met.  Those two tracts had index scores of 60 and 54, respectively, out of a potential 66.



The Food Insecurity Index was constructed for two main purposes:  First, Forgotten Harvest wanted to identify food insecure areas in Wayne, Oakland, and Macomb Counties that are not served by either Forgotten Harvest or Gleaners.  Second, Forgotten Harvest wanted to know whether there were areas currently served by FH or GCFB that are food secure enough that FH resources could more profitably be deployed elsewhere.  

The Food Insecurity Index addresses these purposes in multiple ways.  By analyzing the data by census tract, the Index can isolate levels of food insecurity at a lower level of geography than has been done before and adjust the service areas to better meet the need.

Moreover, the Index, composed of publicly available data measures, recognizes that food insecurity is multidimensional.  It incorporates census tract level measures of households’ financial ability to buy food and the amount of financial risk they have, indicated by the housing burden measure; their access to means of traveling to places where they can obtain food; their residential stability (for low-income households); their level of education; and an aspect of their household composition (percentage of single parent families).

In addition, by its multidimensional nature, the Index recognizes that a census tract is not necessarily a homogeneous entity.  For example, while some tracts may be uniformly well off or poorly off on all the measures included in the Index, many are mixtures.  Examining the scores on each of the six variables for high risk food insecurity census tracts can help identify the factors associated with the insecurity.  

The Food Insecurity Index numbers are just the starting points for understanding food insecurity in an area.  Beyond examining the scores on the individual components of the Index, the analyst can bring in other information such as the unemployment rate; use of SNAP benefits; average family size; ethnic composition; and availability of full service grocery stores.  And it would benefit understanding to look for common factors in contiguous tracts with similar scores.

Finally, Index scores can be used as conversation starters with people in food insecure areas.  Such conversations frequently bring to light factors that researchers and organizations have not known about or considered but that may be important contributors to food insecurity.

See for example. “Household Food Security in the United States in 2015, ERR-215. Economic Research Service/USDA.

2 Antonio Bruce and J. Gregory Robinson, “Tract Level Planning Database with Census 2000 Data.”  U.S. Census Bureau March 2007.

Introduction to the US Census Budget

Introduction to the US Census Budget

This is the second blog post in an in-depth series exploring the history and future of the U.S. Census. Explore our first post introducing readers to the history of the census and its importance.

Few people doubt the importance of having an accurate census count every 10 years, and many would not hesitate in advocating for procedures to be put in place to ensure each census performs with a greater deal of accuracy. However, despite popular consensus valuing accuracy, people often forget to realize the tradeoffs that need to be considered when budgeting for the census. In the next three blog posts about the census budget, we will first explore introductory census budget information, discuss the challenges facing the 2020 Census, and examine some of the innovative ideas the Census Bureau is implementing this cycle.

The Census Budget

The Census Bureau’s budget operates significantly different from most government agencies. This is largely because of its cyclical nature, whereas most agencies operate on a mostly consistent annual budget. The Census Bureau does the majority of its spending every 10 years during the actual implementation of the Decennial Census, with the annual budget dropping sharply afterwards and then slowly creeping back up.  The Decennial Census is the largest peacetime mobilization and federal employment spike that the nation sees. 


Complete national censuses are extremely expensive to administer, and the primary function of the Census Bureau is to carry out the official census every 10 years. As will be discussed in future posts, the census counts affect representation in the House of Representatives and is used as a guide to distribute funding and other resources. Although the Census Bureau conducts a host of other surveys and population counts, none are as extensive as the Decennial Census which cannot be watered down because it is constitutionally required.

Because of the constitutional requirement for the census, Congress has less influence than usual in determining the budget for it. This is not to say that Congress is without influence though, as they certainly set the course, especially in regards to the upcoming 2020 Census.

How Expensive Have Past Censuses Been?

The census has historically always been a costly endeavor for the United States; however the cost of executing it has been increasing at a significant rate over the last 50 years. For example, adjusting census costs to 2020 dollars, the 1970 Census cost a total of $1.1 billion to carry out, the following census in 1980 cost a total of $3 billion. In 1990 and 2000 the cost went up to $4.7 billion and $9.4 billion respectively. The dramatic increase in cost is not just attributed to growing population, as the cost per household has been increasing at a similar rate.

Source: The Census Bureau

The 2010 Census

The 2010 Census’ budget increased even more, costing a total of $13 billion, which was the most expensive census ever in real dollars, both in terms of total cost and the cost per household, which was nearly $100. The GAO High Risk report suggests that the cost per household increases has been caused by a decline in census participation by mail, requiring more door to door efforts to obtain accurate counts. The preparation and execution of the 2010 Census required an unrivaled 3.8 million hires which significantly contributed to the cost.

The 2010 Census ended up costing so much that the oversight committee suggested that the fundamental design is no longer capable of delivering a cost-effective headcount given the nation’s diverse trends. This leaves future censuses in a tough situation as they will be forced to adapt the design in order to provide a financially sustainable model that can still produce accurate results.

What are the Cost Estimates for the 2020 Census?

In 2011 the National Research Council examined the tremendous cost of the 2010 Census and gave a recommendation that the 2020 Census should be carried out at a lower cost per household than in 2010. The resulting book, Envisioning the 2020 Census, is a wealth of information for further study of history, budgeting, and statistical developments. They claimed this was a reasonable request due to improving methods of data collection including increased reliance on the internet. A few years later Congress noted that the 2010 Census was so expensive that they decided to set a mandate for the bureau stating that the 2020 Census is not to exceed the cost of 2010. This was the first time a mandate of this kind was set, and created a target budget for 2020 of under $12.5 billion in spending.

Despite the Census Bureau stating they will be able to achieve cost goals, staying below the cost of the 2010 Census both in total cost and cost per household, some people are skeptical. Recently, former Census Bureau director John Thompson raised uncertainty on meeting the estimates, noting that the new Census Enterprise Data Collection and Processing program (CEDCaP) has overrun its cost estimates by over $400 million so far.  Additionally, the GAO’s placement of the 2020 Census on its high risk report indicates a serious concern that the project runs a significant risk of overspending.  The 2020 Census faces many challenges in sticking to the appropriations it received from Congress.  Our next post discusses those in more detail before exploring some potential solutions.

Census 2020 Blog Posts

  1. What is the Census?
  2. U.S. Census Budget Introduction
  3. 2020 Census Budget Challenges
  4. 2020 Cost-Saving Innovations
  5. How Does the Census Use Local Administrative Data?
  6. Data Security and the Census
  7. Redistricting and the Census