I am calculating income ratios for California using POVERTY (0-99, 100-199, etc.). The poverty rate for all prior years from 2015 and back essentially are the same as the published AFF rates, but substantially different from the 2016 data–17.4% vs. around 14.4%. I noticed this was an issue with some previous ACS updates. Help?
The first thing to note is that statistics reported in American Fact Finder use full-count census data, whereas IPUMS USA provides a sample of the data (unless you are using full-count samples from 1850-1940). Therefore, you can expect your values to vary slightly, but usually within the margin of error.
The larger difference that you are observing is likely a product of differences in the way IPUMS and the Census Bureau calculate the POVERTY variable. The following is provided in the description of POVERTY on IPUMS USA:
The original PUMS samples for years prior to 1990 did not include a poverty variable. Original PUMS samples from 1990 onward included poverty values, but IPUMS poverty values differ from the original PUMS values in a key way. The original PUMS samples treated all households members unrelated to the head as one-person families when assigning poverty values, even if such persons were part of a secondary family (i.e., persons living with their own relatives but not related to the household head). Thus, the original PUMS poverty measures do not account for the presence of children (or any other aspect of family size and composition) in secondary families. For example, in the original 1990 PUMS sample, a woman unrelated to the householder who has a child would receive a poverty value appropriate for a single person with a given income, rather than for a two-person family with a child. Consequently, the original PUMS samples from 1990 onwards tend to underestimate poverty. In the IPUMS, by contrast, the POVERTY value would be based on the threshold fitting the secondary family consisting of both the mother and the child. The IPUMS samples also round to the nearest poverty value, while the original census PUMS samples always round up.
As a result, calculations made using POVERTY from IPUMS USA data will inevitably vary from statistics reported by American Fact Finder.
I hope this helps.
Understand those differences. But previous year calculations only show a minor variation. The 2016 data is substantially higher:
| | AFF | IPUMS 1-100% |
| 2005: 2005 | 13.3 | 13.8 |
| 2006: 2006 | 13.1 | 13.3 |
| 2007: 2007 | 12.4 | 12.6 |
| 2008: 2008 | 13.3 | 13.5 |
| 2009: 2009 | 14.2 | 14.4 |
| 2010: 2010 | 15.8 | 16.0 |
| 2011: 2011 | 16.6 | 16.8 |
| 2012: 2012 | 17.0 | 17.1 |
| 2013: 2013 | 16.8 | 17.0 |
| 2014: 2014 | 16.4 | 16.6 |
| 2015: 2015 | 15.3 | 15.5 |
| 2016: 2016 | 14.3 | 17.6 |
Sorry, should have checked it before sending. Here’s the table:
Thank you for providing additional data. Are you working with an IPUMS USA extract or SDA? Additionally, would you mind sharing the American Fact Finder link where you have generated the values for California? I’m seeing slighltly different values than what you’ve listed.
If it’s easier, you can also send an email message to firstname.lastname@example.org and we can work with you to figure out the difference.
Working with SDA. The AFF data is from Table S1701 for California. I also ran the numbers on the DataFerrett app–prior year numbers come out essentially the same as from SDA, but the 2016 percentage comes out closer to the AFF–14.5% vs. the 17.4% from the SDA.
Below are the rates for 2014-2016 I have calculated using SDA with a re-coded variable:
2016 ACS: 16.3
2015 ACS: 14.8
2014 ACS: 15.6
I created a re-coded variable (povcat) where poverty (1-99) represents those below the poverty level. I then used this re-coded variable to run frequencies with a person weight (perwt).
Perhaps you can share the process you are using in SDA? It appears that there are difference across the years, but as mentioned earlier, it is not unexpected.
I’ve been using poverty(r: 1-99;100-199;200-299;300-399;400-499;500-501 with perwt and frequency. As indicated above, this gets different (per your explanation above) but considerably closer estimates to the AFF numbers for every year but 2016.
Thanks for your perseverance.
While there does appear to be a larger difference in the most recent data, at this time we do not suspect there to be an error with the data. If variables are edited, a note will be posted here.
For ease of analysis, if you have access to a statistical tool such as Stata, SAS, SPSS, etc., you could also run your analysis by creating a data extract. Information about getting start with the extract system is available here.
After you pointed out this oddity, we looked into the data more and discovered a bug in the poverty construct. The issue has been fixed and the corrected 2016 ACS file should be available very soon.
We like to reward users who inform us about errors with a free IPUMS coffee mug. To claim yours please email email@example.com with your mailing address and we will send one to you as soon as possible.
The fixed data is now live on IPUMS USA. Data available through SDA will be updated soon.