Hi! I’m analyzing the characteristics of long-term homeowners in Baltimore in IPUMS, and I’m confused by why I’m seeing a big difference between IPUMS and the official Census stats in some cases.
For example, I’m interested in trying to analyze when homeowners moved into their units. According to the SDA for ACS 2024 5Y ( IPUMS Login ) there were ~9000 homeowners who moved in the “12 months or less” period; according to the Census 5Y, we have ~2700 homeowners who moved in in 2023 or later ( Explore Census Data ). The total is very similar in both cases (~121,000) but this category is wildly different; wouldn’t the “in the 12 months or less” period number have to smaller than the “2023 or later” number. Am I doing something wrong?
A 5-year sample includes responses from 5 years of surveys. The 2024 5-year sample includes responses from 2020 through 2024. Many respondents from 2020 through ~2023 who reported moving “in the last 12 months or less” would have moved before 2023.
I.e., the “last 12 months” is benchmarked to the date of the survey response, not to the last year of the 5-year period.
Thank you for the reply. That’s helpful! There’s no way using IPUMS data to create categories by year the way the Census releases these stats, is that right? I could use a single year’s worth of ACS data — the 1 year, for example — and that would give me a fixed time period, right? ( IPUMS Login ) I was hoping to include a chart on the tenure of homeowners in the city.
My analysis is largely focused on long-term homeowners, who we’ve defined as having resided in their homes for 20+ years (so I would collapse MOVEDIN 6 and 7 into one category for 20+ years in analyzing the ACS in IPUMS). That yields 45,699 homeowners in the 5Y ACS, or about 37% of homeowners.
(That’s also in line with what I see in the 5Y ACS Census table I linked earlier, where we had 33% of homeowners moved in 1999 or earlier — 39360. I would expect that number to be a little lower, given that this time period is 25+ years before 2024 — not 20 exactly). I assume I’m fine to say that a third of homeowners were long-term homeowners, per our definition?
I will then be analyzing the 45,699 homeowners by race, gender, etc. Do you see any issues with this? I figured I should use 5Y data for a smaller geography like a city?
It’s not possible to exactly recreate the table using the publicly available ACS data. The Census Bureau typically uses their internal versions of the data to produce official estimates like the table you referenced. They may have used information about the exact survey month to create the table, and they may have used more precise information about the date of the move. However, you can certainly benchmark the date ranges to actual months and years based on the survey year. There are some considerations for doing so.
The ACS is fielded on a rolling basis throughout the year. The MOVEDIN codes in the publicly available data do not reference absolute years, but rather reference specific numbers of years or months in the past. Since it’s not possible to determine the month the ACS was conducted for any respondent, the move-in time ranges you can estimate will be greater than what’s reflected in the codes.
For example, a respondent in the 2005 1-year ACS with MOVEDIN=2 moved into their home between 13 and 23 months ago. They were surveyed at some point between January 2005 and December 2005. This means they moved in sometime between February 2003 (January 2005 minus 23 months) and November 2004 (December 2005 minus 13 months). Lacking information on the survey month means that the range of possible dates on which they moved in is much longer than if you had access to the survey month.
Pooling multiple 1-year ACS samples together or using a multiyear file can be beneficial when working with a smaller population, like a particular geographic area. However, analyzing multiple years together would prevent you from having as much precision in the absolute year and month in MOVEDIN.
You could use either the 5-year or 1-year ACS samples for this analysis. If you prefer to work with the 5-year samples but want to look at single survey years, see the MULTYEAR variable, which reports the actual year each data point is from.
Thank you. I want to better understand how I would analyze “longtime homeowners”. I filter for OWNERSHIP=1 (residence is owned), RELATE=1 (householder) and use PERWT (because I want to discuss people, not households); is this fair? As I said, I’m interested in the MOVEDIN variable, which I believe is only recorded for head of the household. How would I characterize my results? I feel confident that this is a universe of homeowners, but possibly a very conservative one that assumes no joint ownership? Or does using PERWT account for this somehow?
In cases of joint ownership, is there a gender bias in whether men or women would be recorded at the head of the household?
This is the table I’m hoping to describe: IPUMS Login ; at the moment I say “A third of Baltimore homeowners have lived in their homes for more than 20 years**”** but please let me know if that’s incorrect.
Additionally I wanted to say how many homeowners there were in the city; I was going to say “nearly 130,000”, would that work given the table I linked?
Your request for assistance characterizing the results of your analysis of IPUMS data falls outside the scope of IPUMS User Support. Below I am providing some information below that may be helpful as you work with our data.
You seem to have produced a table estimating the number of householders by how many years ago they moved into their home. The weight PERWT should be used with most person-level analyses of IPUMS USA data. In a person-level analysis, the individual is the unit of analysis. When you restrict your analysis to only householders (i.e., RELATE == 1), keep in mind that the unit of analysis then becomes householders, rather than all individuals. You are correct that the concept of householder is not identical to the concept of homeowner, due to joint ownership. Joint ownership is not identified in the ACS, as only one person is designated as the householder, and the household as a whole is described as living in an owned versus rented dwelling. It is worth thinking through the implications of restricting to householders for your specific analysis and the information you are trying to convey.
Generally, the householder is the person in whose name the dwelling is rented or owned, and you are correct that this could become biased in the case of joint ownership. As a starting point, you could determine what share of householders are male versus female. This article from the Urban Institute digs into different ways to think about gender bias in householder identification and understanding home ownership.