Is the Census underreported, and are there corrections?


I am aware that the census is not an exact population measurement as it is a survey.

However, I have been unable to find any information about the estimate of how under reported the census is believed to be, I say under reported because I have known a few people that have never filled out the form or returned it.

Now that the preface is out of the way the question is:

As the census is used to help determine district lines, public service funding, congressional seating, and various other things what is the estimated under reported rate and are there any corrective actions used when parsing the data, such as weighing a certain group or location more than another, in order to account for the difference and provide more accurate information or is it just used as is with direct number comparison?


Posted 2017-05-23T17:13:58.533

Reputation: 1 042

In certain metro areas, you will get the opposite effect with multiple census canvassers coming to your door. This is stated from personal experience. – K Dog – 2017-05-23T17:40:02.560


@KDog Generally one of the most difficult communities for a census to reach is the people without doors in the first place.

– origimbo – 2017-05-23T17:49:17.273

@origimbo that seems to be European and not totally relevant to the US. I don't have a source yet, but my understanding is that known camps are sent census takers who go tent to tent and count/ask just like with houses, and people sleeping in doorways are counted as living at that address. – None – 2017-05-23T19:00:30.810


@notstoreboughtdirt It seemed to give a nice overview of the problem, with some discussion of the major issues with street counts. If you want a purely US reference, there's always and

– origimbo – 2017-05-23T19:26:04.527



The Census Doesn't Work Like That

"The census" - as in the official decennial United States census is not based on a sample. The Census Bureau will attempt to collect information from every household in America.

You can read about this in their 2020 Operational Plan (pgs 8-9). They send forms to every household in the United States. People may complete their census form either online or on paper. Households which do not return a census form will be contacted in person by a field agent, who will attempt to survey them in person.

This is all just to say that although the decennial census is a survey, it is also a direct measurement of the population. It is not a sample-based design like many other surveys are. Therefore, there is no sampling error.


That being said, realistically not everyone will respond. In 2010 about 74% of households participated in the census (2010 Census - Take 10 Assessment Report, pg10). Although the Census Bureau has a Nonresponse Unit which attempts to drive up that number, they also use imputation deal with missing data.

The operational plan and CCM workshop documents both include imputation as a possible approach. For example (from the operational plan) says that the Response Processing unit (which deals with data that has been collected) may:

Create models based on established business rules to determine the appropriate course of enumeration action for cases (e.g., person visit, use of administrative records and third- party data, or imputation)

Imputation is a process for "filling in" missing data. Since imputation (and the rest of the work done by the Response Processing unit) is done prior to publication, all these corrections will have been made prior to the data being published.


Posted 2017-05-23T17:13:58.533

Reputation: 22 718

Probably belongs on different Q, but if they need Census to count population, how do they know who to send the Census forms to in the first place? (or, if they know who people are to send forms to, why bother with census as far as actual population counting exercise in constitutional context?) – user4012 – 2017-05-24T14:04:37.313

1@user4012 - That's outlined in the 2020 plan also. The first step of the census is to figure out where to count. They get lists of addresses to send the forms to. I kind of glossed over this section initially, but it's in there. – indigochild – 2017-05-24T14:05:59.603

Are the forms per-person or per-household? – user4012 – 2017-05-24T14:07:31.827

@user4012 - A single form covers everyone at the same address. – indigochild – 2017-05-24T14:11:54.673

The form that gets filled out is used more for demographic breakdowns than the raw headcount, I believe. – PoloHoleSet – 2017-05-24T14:22:25.320


This is a nice summary of how the Census works, however doesn't cover how the errors are covered. I have found , which takes into account their measurement errors and how they account for any bias. So far seems they account for race, age, gender, and housing miscounts. At the moment cannot tell if this post-enumeration is taken into the public results or just used for future formation of the census

– SCFi – 2017-05-24T15:08:43.617

@SCFi - Great find! I went ahead and edited based on your comment. I removed the stuff about the non-response unit because I think it's clear you aren't interested in that. The short answer is that they use imputation to fill-in the gaps in their data. – indigochild – 2017-05-24T16:04:52.183

1I know it's a fine point, but your source gives a participation rate of 74% of households, not 74% of people. – origimbo – 2017-05-24T16:47:00.150

1@user4012 they can get all of the mailing addresses from the postal registry, however, households can introduce new humans (births) and deaths occur in households, so knowing how many household mailing addresses there are does not tell them the exact population count. Knowing where to send them doesn't answer how many people there are. – Tommy – 2017-05-24T17:19:40.780

@origimbo - Agreed. Change made. – indigochild – 2017-05-24T18:10:47.613


As to why the Census doesn't work like that, you could mention the 2000 Census sampling controversy, and the Supreme Court's ruling in Commerce v. House that the law prohibits the use of sampling or certain statistical corrections in the count.

– Nate Eldredge – 2017-05-25T06:37:29.507