ACS Data Users Group

I have two poverty dashboards based on ACS5 data, one has a state filter, one just a full listing of counties. No prior year because of COVID etc, so I'll wait till 2022 comes out and put 2022 vs 2021 data in. I might put in a little more effort, but I also want a housing dashboard, gov programs (SSI, UI,SNAP etc.), Race, Grandparents and maybe a few more. Not looking for perfect, but just some good recaps.What I really hoped to find when I first look for Census data. They both look about the same so I'm only posting one.

https://public.tableau.com/views/Poverty_by_County2/Poverty_by_County?:language=en-US&publish=yes&:display_count=n&:origin=viz_share_link

Thanks!

Looks all right, nice to see all the counties in one place. I wonder if you've considered using SAIPE data, for at least the overall rates and the under-18 group. It's annual data rather than 5-year. SAIPE does have under-5 annual poverty data, but only for states and school districts. But it also has ages 5-17 in families for all levels, along with MHI.

https://www.census.gov/data-tools/demo/saipe/#/

Thanks for the feedback. Several people mentioned SAIPE, which has some clear advantages. So, it's a possibility down the road, but I really want some solid ACS dashboards first. I also want to cover cdc wonder mortality bfrss and a few others. I mostly use weekends for all this and I need a few for the other stuff (friends, family, camping etc), so maybe another 6 months to get it all covered.

It might be helpful to know who your prospective audience and use case for this is. It's helpful that you provided baseline US numbers for context (state might be useful as well). I'd consider including counts in addition to percents; more people (or is it households?) in poverty live in Contra Costa County, CA (at 8.2%) than Trinity County, CA (at 22.5%). And it would be helpful to have definitions of your columns (what does "H" mean? Are these households or individuals?) I don't think the "Key Diff" columns are necessary, since they essentially repeat information already provided. Also, from a policy perspective, I've never found the Gini index to be useful at a local level (as opposed to national). Higher wealth disparities (seems bad) is another way of saying less wealth segregation (seems good).

Also as a reminder, you shouldn't compare 5-year estimates with overlapping years (ie. you can compare 2018-2022 with 2013-2017, but not with 2017-2022). You're essentially comparing the sample of the last year of one estimate with that of the first year of the other, and it's too small a sample to be significant.

MOEs should be reported, too.

Thanks for the feedback. Right now I'm sharing it with country health departments in CA, but am hoping it gets Broad use. I need to be more clear on pop vs households and H means really high vs US and also needs clarification or something. Column width sucks in tableau. Some people really like the repeated US numbers and diff Columns, but I'm flexible to all feedback. Gini is just put in there to make poverty more complete, I may drop it. When 2022 comes out I will compare 2022 acs5 vs 2021 acs5, not a single year, and SAIPE might come later down the line.

[quote userid="6235" url="~/discussion-forum/f/forum/1268/poverty-dashboard---any-feedback-is-appreciated/3410#3410"]When 2022 comes out I will compare 2022 acs5 vs 2021 acs5[/quote]

You really should not do that.

People who know what they're talking about are taking the time to give you good advice, at your request.

If you're going to do whatever what you want regardless, why ask for the feedback?

OK I misunderstood/ misread your comment, I thought you were saying don't use 5 year vs 1 year. I am completely flexible to using the last 5 year range, and value all feedback especially in this forum. I do appreciate it. Responding to these on my phone during my dinner break (I work 1-9:30. Not ideal, but generates progress.

The issue might be clearer if you refer to each of them with their full range, 2018-2022 and 2017-2021. If you compare the two, you're not comparing 2022 with 2021. Sometimes we'll use the final year as a shorthand, but not on anything public. It's a five-year estimate and each year carries equal weight. The reason these are five-year estimates isn't because people just like vaguer time frames, it's because that's the number of years of survey responses required to achieve a sample size big enough to make meaningful data. And if you compare overlapping five-year time frames, the only differences are in the non-overlapping years. And so the difference between the overlapping time frames doesn't have the (already meager) five-year sample size.
tl;dr- Refer to the data by their full five-year range.

Yes, thanks and 2021 acs5 is not clear for many individuals, and I think for the label 2017-2021 acs5. Even great data charted is worthless without clear labels. As for the complexity of using prior?? I'm thinking at this point, maybe for the upcoming 2022, I'll just skip comparison to prior because no matter how I do it and label it, it'll probably confuse some people and I want first impressions to be very clear.

A few comments:

1. If possible have the All button checked when the dashboard is opened. Without that being the case if I want to look at one state I first need to check the box the uncheck the box to the deselect all states then pick the state I want. Alternatively, open the dashboard as a single value list.

2. Its unclear to me what the H flag denotes

3. If possible find a way to remove the US value columns. Would it be possible to have the US always appear as a geography no matter what the filter? If so then you could dispense with several columns and make your font size a bit larger.

I have another dashboard with 1 state selected, sorting by state and a few other elements work better. H was a last minute decision and means really big difference vs US, I need to fix it somehow or drop it. I can easily drop the repeated US columns, and get more space, but some people really like them (???). I will have a menu or method of going between the dashboards and have a US country map that covers this poverty data, but I haven't figured out which fields to use. I really like heat maps that highlight ranges.

The issue of "county based" poverty numbers or even any numbers from all of the counties is that there are very many small counties in the United States. In fact, only about 1,300 are large than 20,000. So when you look at the Poverty population, which is say 20% in some counties, you are looking a universe size of 260, and a sample size of possibly as small as 20 to 40, SAIPE since it is model based is more likely to be more accurate for this estimate, and Other estimates may be better for other purposes rather than the ACs Sample. Which is around 10% at best for the five year.

Yes, you are correct, and I might add a population filter to help. SAIPE is better in a number of ways and might be something I do after bfrss and cdc wonder mortality. There is just something my mind likes about data for every single country from ACS. Everything is an estimate and I want to hit a home run in ACS data (maybe 10 dashboards) before I fight the next windmill.

I'm working on a R package that will give the Supplemental Poverty Measure (SPM) at the tract level. Eventually I hope to post the package on GItHub. The method uses PUMS data including the SPM PUMS data https://www.census.gov/data/datasets/time-series/demo/supplemental-poverty-measure/acs-research-files.html. The SPM is a research measure that, unlike the Census poverty threshold measure, includes income/support from government programs such as SNAP as well as expenses based on variables that vary across geographies (read "local cost of living.") https://crsreports.congress.gov/product/pdf/R/R45031 The statistical method is small area estimation using IPF to adjust a PUMA level cross tabulation (13 variables) to ACS 5-year marginal tract level tables. I've done all the census tracts in Virginia (about 2000 tracts), which takes several days on my PC (20gb memory). You can then aggregate the tract level synthetic data to higher level geographies. All 13 covariates (including SPM) are carried along. It is possible to add replicate weights to the synthetic (output) data file. Any comments welcome.

Dave