American Community Survey Data Users Group
ACS Data Products & Resources
ACS Public Use Microdata Samples (PUMS)
ACS Summary Files
Application Programming Interface (API)
ACS Data Issues
ACS Data for Small Geographic Areas & Population Subgroups
Mapping ACS Data/GIS Applications
Measuring Trends Over Time with ACS Data
Mapping ACS Data…
Mapping ACS Data / GIS Applications…
Visualizing ACS multi-year…
This group requires membership for participation - click to join
Aug 20, 2014 5:11 PM
Visualizing ACS multi-year estimates
I saw a post earlier in this group that suggested this group discuss Mapping as well as general ACS data visualization. Hopefully that's ok, because I've got a question about general data visualization...
I was wondering if anyone has any thoughts or advice to share about how to appropriately, but also comprehensibly, visualize ACS multi-year estimates for lay public audiences in graphs, in a way that:
- Allows users to see trends over time,
- Gives them access to the most recent data,
- That facilitates wise use of data, and
- Strikes a good balance between full data disclosure and ease of use
I know the Bureau says that comparing non-overlapping estimate periods is the easiest, but as you all are well-aware rural areas only get the 5-year estimates and well, that means we'll have to wait till 2015 to get the first unique 5-year period (2010-14) since the first released 05-09 period. Also, it's certainly not impossible to compare overlapping estimate periods -- it just requires a bit more mental gymnastics and training of users (which I have been providing for the last few years, and am prepared to continue providing). So I've been struggling with how to visually display overlapping estimate periods for the rural leaders and practitioners with whom I work and provide services, and if I should at all.
Here are four visual display options that I've identified (attached are the graphs). Perhaps you know of different options? Is there some standard option that someone or some group has identified?
1. In Display Option 1, I report all the spanned period estimates and show the estimates that correspond to each year of the period
2. In Display Option 2, I only report the most recent 5-year period estimate (for all the years to which they correspond)
3. In Display Option 3, I only report the non-overlapping multiyear estimates, like the Census Bureau recommends doing for multiyear comparisons. The problem is that I won't be able to do this until 2015, when we get the 2010-2014 estimates.
4. In Display Option 4, I report all the available multiyear estimates, but do not show the estimates that correspond to each year of the period
Of course, I haven't even touched the issue of displaying MOE error bars in these graphs, which would add more complexity to the display options and the decision. This is certainly also in the back of my head for this consideration...
Thanks so much,
Julie Black AICP
Aug 21, 2014 10:47 AM
I don't have an answer for you, Lena, but THANK YOU for asking the question! Our agency has also been struggling with this! I will note that your charts don't display the margin of error, which is perhaps one complication too many.
Aug 22, 2014 8:18 AM
Hi Lena -
I also have no firm answer. As a data practitioner, I'm inclined to suggest Option 1. (When in doubt, show all the data!) But... thinking about translating the data for lay audiences, I think there is too much overlap, which makes chart 1 hard to read, and the public expects an answer more like option 4.
Could you add margin of error bars to option 4, so people can see that any "trend" incorporates both actual change and sampling error? (Not to mention the overlapping year issue...)
Also - another option would be to show something akin to Option 1, but showing only every other year, or every third year?
Aug 22, 2014 9:55 AM
Good question. I suspect there are many acceptable answers. Obviously Option 4 is the easiest to read. Perhaps Option 4 style with only the non-overlapping data points from Option 3 tells the story the best. That also leave room for margin of error bars.
Aug 22, 2014 10:09 AM
I am currently working with census data for geographical and temporal analytics. My suggestion is to provide a time series (toggle to display or not) so that the user can "slide" through time. In the time series, you can aggregate information per region using a zooming user interface in your geographical map. For example, you can tally female population in southern states per year. We have developed applications (w/ D3) and users enjoy the interaction.
Hope this helps,
Dr. K. Simpson
Aug 22, 2014 11:19 AM
Each one of your graphs is a "hypothetical" representation, but 4 is the easiest to follow. A brief footnote can explain the source of your estimates. I would certainly make reference to the margin of error, either graphically or in the footnote/source. Dr. Simpson's suggestion above, regarding the use of time series is certainly worth pursuing. Thank you for asking the question.
Aug 22, 2014 1:32 PM
Thanks everyone, so far, for your thoughtful responses. Yes, the plan is to integrate MOE information in the chart in some way for users (perhaps error bars or values that display when a user mouses over the point in the chart -- these charts will exist as a web application). I just wanted to get some ideas about the basic display of the multi-year estimates at this point.
It sounds like you all are gravitating toward Option 4, with some variations. This is what I've done in the past, and people seem to understand it. I'll keep playing with the options that display the data for all years in the multi-year estimate period, maybe even put it in front of some lay data users to see how they handle it. If I do put it in front of people, I'll report my findings back to this group!
If anyone else has any other suggestions for this visualization, please contribute. I don't think consensus has been reached on this yet!