1 of 37

Understanding the �2020 U.S. Census�

International Association for Social Science Information and Technology

William P. O’Hare

billohare1@gmail.com

September 29, 2021

2 of 37

Outline of Presentation

How the 2020 Census was conducted

Key results of the 2020 Census

Measuring quality of the U.S. Census

(Some points oversimplified in presentation)

4 of 37

Importance of the Census

The Census is mentioned in the 5^th sentence of the Constitution.

Political power - reapportionment and redistricting

Funding - $1.5+ trillion per year to states and localities

Decisions by government and private sector

Research

5 of 37

Conducting the 2020 U.S. Census

6 of 37

2020 Census - Large and Complex

Costs $15.8 billion -- $3 billion less than repeating 2010 method

150 million housing units

Hired more than 500,000 temporary workers

Largest U.S. mobilization outside of wartime

15+ trillion statistics produced

7 of 37

Most important data from the Decennial Census are data for small areas�

3,200 Counties
19,000 Villages and Towns
10,000 School districts
175,000 Voting Precincts
84,000 Neighborhoods (census tracts)
8 million census blocks

8 of 37

Topics covered in the 2020 Census

Age
Sex
Race
Ethnicity (Hispanic Origin)
Relationship to the householder (What is a householder??)
Tenure (rent or own)

10 of 37

Implications of Race/Hispanic Categories

People can mark more than one race category:

Race Alone or
Race Alone or in Combination
Specific Combinations (White and Black and Asian)

Hispanic Origin and Race are different concepts;

Hispanic Origin by Race

12 of 37

2020 Census Operations

April 1, 2020, was Census day --- usual place of residence concept

In mid-March 2020 the Census Bureau sent out letters to most households inviting them to participate in the 2020 Census

80% were “Internet First”-- Letter with unique code for internet response but no paper questionnaire
20% were “Internet Choice” -- code plus paper questionnaire

Mid-March to end of April five more letters

Enumerators sent to get data from households that did not self-respond

13 of 37

Data collected differently for population in housing units and group quarters

Population living in Housing Units (about 150 million housing units)

Houses,
apartments,
condos,
mobile homes,
RVs,
boats

Population living in Group Quarters (8.3 million people in 2020)

College Dormitories
Prisons
Military barracks
Nursing homes

Pandemic impacted Both

14 of 37

Impact of Pandemic�

Postponed sending census enumerators into the field

Many stakeholder groups (trusted voices) were not able to hold events to promote Census

People did not want to open door to unknown people (enumerators)

Colleges and nursing homes closed down

15 of 37

How people in households were Included in the 2020 Census

Self-Response Rate - what percent of households returned a completed census questionnaire (65% of households) – Mostly internet
Non-Response Followup – census enumerators sent out to visit households that did not self-respond (35% of households)

Non-Response Follow Up Interview with household (11% of households)

Proxy Response – data from proxy (neighbor, landlord, etc.) (18% of households)

Administrative records – use data from other federal sources like tax records to count people in a household (5 % of households)

Statistical Imputation – statistical estimates of people and characteristics when none were reported (1% of households)

16 of 37

Group Quarters Data Collection

Records summited by the Group Quarters facility

Census Bureau enumerators are allowed into the facility to collect data

Facility workers distributed and collected census questionnaires

17 of 37

Problems with Group Quarters in 2020 Census

College towns (on campus and off campus)

Students left town before Census enumerators could get there
Colleges did not have much information on students living off campus

Nursing homes

Census Bureau could not get access because of Covid
Nursing homes shut down before Census enumerators could get there

18 of 37

Key Findings from 2020 Census �

Slow pace of total population growth from 2010 to 2020

Population increased 7% from 2010… slowest rate since 1930s

Decline in white population (2.6%) + big increases in

Hispanics (23%)
Asians (36%)
Some Other Race (180%)
Two or more races (127%)

More diverse population

�

Number of children (ages 0 to 17) was 1.5 million lower than 2010 Census

20 of 37

State Figures – Expectations and Census Counts on Reapportionment

Texas expected 3 new seats but got 2
Florida expected 2 new seats but got 1
Arizona expected 1 new seat but got 0

New York expected to lose 2 seats but only lost 1
Rhode Island expected to lose a seat but did not
Minnesota expected to lose a seat but did not

21 of 37

Measuring Census Accuracy

22 of 37

Census Quality Measurement

Operational metrics (How the Census was conducted)

Accuracy of the Census

Census Coverage (Net Undercounts and Overcounts)
Omissions

23 of 37

Thoughts on Quality Measures

The importance of differential quality – highly aggregated data mask important differences among subgroups

Small area data is the most important data from the Decennial Census…but we have NO measures of accuracy for small areas.

24 of 37

Differential Privacy�

New in 2020 Census

What is the problem?

What was done?

What impact does it have?

25 of 37

How do we know about Census accuracy?

Demographic Analysis (DA) compares census counts to expected numbers based on births, deaths and net immigration

Post-Enumeration Survey (PES) compares census counts to results of second survey conducted in a sample of places

Net coverage
Components of coverage – omissions

26 of 37

Net undercount is NOT the same as people missed!

The net undercount is the balance between population missed and population counted twice

People missed in the Census are called omissions

27 of 37

Omissions (population missed)

“People who should have been included in the Census, but were not”
Conceptually, this is the focus of most people
Only source of data is from the Post-Enumeration Survey
Omissions quite different than net coverage

28 of 37

Demographic Analysis

Compares census counts to expected population based on births, deaths and net immigration by

age, sex, race, and Hispanic Origin

Net Coverage Only (no data on omissions)

National only

Limited data on race and Hispanic Origin

Black and Non-Black only
Only Hispanic under age 30

29 of 37

Post-Enumeration Survey (PES)

A survey conducted in sample of places right after the census is completed

Records from the second sample are compared to the census records to see who was missed and who was counted twice

Some data for states and large counties

Only source of data on omissions

36 of 37

Summary

Decennial Census is a large and complex undertaking
Most important Census data are small area data
The 2020 Census was complicated by Covid Pandemic
Initial data suggest 2020 Census accuracy is somewhat worse the 2010 Census
Most important accuracy measures are differential accuracy

37 of 37

Thank You

Bill O’Hare

Billohare1@gmail.com