1 of 32

Going deeper with Census microdata: IPUMS is your friend

March 6, 2025

Sandhya Kambhampati

Paul Overberg

David Van Riper

MaryJo Webster

2 of 32

ipums.org

3 of 32

Observation units of census & survey data

Geographic summary data: Areas

Microdata: Respondents

Households

Persons

+ Flexible classification, modeling & �cross-tabulation

- Restricted geographic detail

- Limited sample sizes

https://www.census.gov/geo/reference/hierarchy.html

4 of 32

Summary Data��

Age	Female	Male	Total
Under 5 years	9,598,316	10,051,876	19,650,192
5 to 9 years	9,787,932	10,191,107	19,979,039
10 to 14 years	10,310,162	10,797,748	21,107,910
15 to 17 years	6,143,574	6,416,023	12,559,597
18 and 19 years	4,214,584	4,400,774	8,615,358
20 years	2,177,131	2,310,954	4,488,085
21 years	2,155,180	2,270,025	4,425,205
22 to 24 years	6,307,618	6,599,470	12,907,088
25 to 29 years	11,411,800	11,850,355	23,262,155
30 to 34 years	10,998,357	11,224,653	22,223,010
35 to 39 years	10,687,259	10,658,796	21,346,055
40 to 44 years	10,034,534	9,966,088	20,000,622
45 to 49 years	10,395,144	10,174,825	20,569,969
50 to 54 years	10,650,734	10,320,033	20,970,767
55 to 59 years	11,202,165	10,583,556	21,785,721
60 and 61 years	4,427,289	4,115,170	8,542,459
62 to 64 years	6,150,445	5,622,814	11,773,259
65 and 66 years	3,850,294	3,467,860	7,318,154
67 to 69 years	5,303,633	4,679,985	9,983,618
70 to 74 years	7,130,592	6,115,586	13,246,178
75 to 79 years	5,030,479	4,083,235	9,113,714
80 to 84 years	3,520,631	2,558,706	6,079,337
85 years and over	4,262,925	2,358,891	6,621,816
Total	165,750,778	160,818,530	326,569,308

Age by Sex, 2016-2020 5-Year ACS Summary File (via IPUMS NHGIS)

5 of 32

https://www.sfchronicle.com/projects/2022/san-francisco-asian-population/

Geographic summary data: Areas

https://www.census.gov/geo/reference/hierarchy.html

6 of 32

Microdata: Respondents

Households

Persons

+ Flexible classification, modeling & �cross-tabulation

- Restricted geographic detail

- Limited sample sizes

https://www.washingtonpost.com/business/2025/02/21/are-federal-workers-lazy-lets-look-data/

7 of 32

Benefits of Using IPUMS

Harmonization
Customized data extracts
Integrated and supplemental documentation
Online data analysis tools
Additional data enhancements
User support

Benefits of using our data include harmonized data across time and space, which I will speak about in more detail on the next slide.

The ability to create your own customized data extract.

Integrated documentation that provides comprehensive detail about each variable and how they have changed over time as well as other helpful information for analyzing these data.

We provide online data analysis tools, so you can analyze our data without ever leaving our website.

We provide additional enhancements to the data that we offer, including consistent industry & occupation coding schemes, constructed family interrelationship variables, and consistent geographic footprint for constant spatial comparison (e.g., account for changes in boundaries)

And we provide a wide-range of user support services.

8 of 32

What is Harmonization?

Creating a single, consistent data series from datasets collected at different times
Applying a coding scheme to group broadly comparable categories but retain additional detail when available
Documenting these decisions and flagging potential comparability issues

9 of 32

Changes to detail in legal marital status in NHIS

B = Not in Universe

1 = Separated

2 = Divorced

3 = Married

4 = Single/Never Married

5 = Widowed

2004-forward Categories

B = Not in Universe

1 = Married – spouse present

2 = Married – spouse not in household

3 = Married – spouse in household unknown

4 = Widowed

5 = Divorced

6 = Separated

7 = Never Married

Pre-2004 Categories

10 of 32

Pre-2004 codes	2004-forward codes	Harmonized codes
B = Not in Universe	B = Not in Universe	00 = Not in Universe
	3 = Married	10 = Married
1 = Married – spouse present		11 = Married – spouse present
2 = Married – spouse not in household		12 = Married – spouse not in household
3 = Married – spouse in household unknown		13 = Married – spouse in household unknown
4 = Widowed	5 = Widowed	20 = Widowed
5 = Divorced	2 = Divorced	30 = Divorced
6 = Separated	1 = Separated	40 = Separated
7 = Never Married	4 = Never Married	50 = Never Married

The first digit of the code is consistent across all years�
The second digit of the code contains detail specific to pre-2004 data

11 of 32

Geography in PUMS

Regions, divisions, states & …
Public Use Microdata Areas (PUMAs):

At least 100,000 residents
“County groups” in 1970 & 1980
IPUMS has created 1960 PUMAs & “mini-PUMAs” �(> 50,000 residents)

12 of 32

IPUMS USA geographic resources

Supplementary variables, based on PUMAs

Counties, cities, metro areas, metro status, % metro

13 of 32

CITY “mismatch tolerance”

Minneapolis

1.3%�commission �error

Saint Paul

0% mismatch

14 of 32

CITY Comparability Documentation

https://usa.ipums.org/usa-action/variables/CITY#comparability_section

15 of 32

PUMA Match Summary by Large Place (75k+)ç

16 of 32

Create a customized dataset via website

17 of 32

Use the online tabulator

18 of 32

Use our API or R/Python packages

https://tech.popdata.org/ipumsr/

https://developer.ipums.org/

https://ipumspy.readthedocs.io/en/latest/

19 of 32

Read the Documentation

Harmonizing data…

makes it easier to look at broadly comparable concepts/variables across time/space
makes it easier to make a mistake

IPUMS documents changes to…

response categories, universe definitions, question phrasing

20 of 32

Weights in Microdata

Not everyone included in the data file has the same probability of being selected into the sample
Weights account for how many people in the total population each observation in the sample population represents
https://blog.popdata.org/ipums-faqs-sample-weights/

21 of 32

Small Sample Sizes

Microdata allow for identification of (highly) specific subgroups; these groups can be “too small”
There is no singular threshold for “too small”
Sampling error around estimated statistics may be too large to allow for meaningful interpretation
Options: pooling multiple samples, aggregating categories

22 of 32

23 of 32

24 of 32

25 of 32

26 of 32

27 of 32

28 of 32

29 of 32

“Between 2018 and 2022, the share of households with annual incomes of more than $750,000 that rented rose to 10.5%, according to census data from IPUMS at the University of Minnesota analyzed by The Wall Street Journal, the highest level since the survey began in the mid-2000s. It was 8.4% in the previous five-year period.”

30 of 32

31 of 32

32 of 32

Questions?

Please go up to the microphone to ask a question. Thank you.

Sandhya Kambhampati, sandhya@latimes.com

Paul Overberg, poverberg2@gmail.com

David Van Riper, vanriper@umn.edu

MaryJo Webster, maryjo.webster@startribune.com