Archive for November, 2013

#GISDay was a great success this year, at least for me. A chance to renew contacts I made last year, meet new ones and hang out with friends all day. Four of us piled into the truck and headed to Bloomington this morning from IUPUI Campus and descended upon the Wells Library at IU. Compared to last year, there was more energy, more talk, and a lot more going on. The big news is the LiDAR data that is in the final stages of post-processing and will be completed this year. This will be LiDAR coverage of the entire state of Indiana, giving everyone in the geospatial community a never-before-seen resolution and accuracy data set. Projects such as subsidence tracking from old mining sites, public health issues, and forestry applications, both urban and rural, are just a few of the projects overheard or talked about today.

Thanks to my traveling companions for making the trip to and from a lot more interesting, and to all those who took their time to talk with us and share the GIS experience.

Now if I can just scrape together the cash for a trip to the ESRI UC this summer I will be all set . . .

Advertisements

Autism Spectrum Disorder (ASD) in Indiana schoolchildren appears to follow the same trend as that reported by the CDC with roughly two percent of the children enrolled in Indiana Public Schools reported as having been diagnosed with an ASD.

For my research methods and stats modeling class this semester, we are required to write a research proposal. For this project I chose to look at autism rates in Indiana due to having the data readily at hand from an earlier request to the Indiana Department of Education (IDOE), and also because I am taking an Intro to Epidemiology class. The idea is to take a geospatial look at the distribution of children enrolled in the Indiana Public School system that have been reported as having been diagnosed with an ASD. I was able to get a polygon shapefile of the school districts in Indiana and joined that with the table sent to me by IDOE. The data from IDOE only shows school district and a count of children reported to have an ASD. In the reporting, if a school district had less than 5 children, no data was reported. For these districts, the null value was replaced with 2.5, which is half the minimum reported value. This is the same technique used by other projects I have worked on. This allowed me to create a quick map of the “observed” ASD rates reported for each school district as shown in Figure 1.

Figure 1 - Observed ASD by School District

Figure 1 – Observed ASD by School District

This map shows the reported values, without any adjustment, just the raw numbers. It shows clusters of higher rates around Marion, Allen, St Joseph, and Vigo counties. But are there really clusters in these areas? We need to look deeper into the data and adjust these raw numbers for enrollment, as schools with significantly higher numbers of students, should naturally reflect a higher number of children with an ASD. So in Figure 2 we have the same data mapped out using the crude rate of number of children diagnosed per 1000 enrolled children per school district.

Figure 2 - Crude ASD rate per 1000 Enrolled Students

Figure 2 – Crude ASD rate per 1000 Enrolled Students

Now we see much less variation across the state, but there is still what appears to be a clustering in several areas. The very dark blue up near Ft Wayne and the other nearly as dark near Evansville are outliers and very well could be an incorrectly entered counts. I would need to recheck these with the IDOE to verify. So, is this the whole story? Well, it could be, but there is another technique we can apply to look even deeper at the data. We can use an empirical Bayesian estimation method. This method takes into account the ASD rate and variance of the surrounding school districts to adjust the value for each district. We do this because we really want to see what the dispersion is across the state without the manmade district lines. In Figure 3 we see the result of the Bayesian analysis.

Figure 3 - Bayes Adjusted ASD rate

Figure 3 – Bayes Adjusted ASD rate

Now we see a relatively even and fairly random pattern, meaning that ASD is fairly evenly dispersed across the state. The effect of the two previously reported outliers is clearly evident and needs to be addressed as to whether the data was reported accurately for those two locations. If I were doing this as a thesis/dissertation, I would contact the IDOE and we could compared the reported values with the previous year’s values to see if it is a significant change.

The idea for this post is to show how important it is to look critically at charts, graphs and maps that are presented in reports, and especially in the media, as to how they present the data. Are they showing us raw numbers, crude rates, or something else? The crude rate is very often used as it is quick, easy, and does a good job of reflecting the data, whereas the Bayesian method is much more time consuming and therefore costly, even though it does the best job of reporting data like this as it looks at the data more spatially. It all depends on the needs and questions the research is trying to answer.

Cartography and Ringmaps

Posted: November 9, 2013 in GIS
Tags: , , , , ,

The latest issue of ArcUser magazine (Fall 2013) has a great article and step-by-step instructions on how to create ringmaps (see “Looking at Temporal Changes”) along with a toolbox that can be added in to ArcMap. The article is well written and the toolbox that goes along with it is very clean, simple, and quick. The thing to watch for, especially with these kinds of maps, is the cartography. Making the map is great, but getting the right symbology, legends, etc. on the map is also critical, and ringmaps have some unique features.

The first and most obvious are the rings themselves. Identifying what each ring stands for is very helpful. In the example I created based off of the article, the inner ring stands for 2006 data and the outer ring for 2010, but how best to show it? I chose to insert the years into the rightmost rings themselves as text objects. This denotes them without detracting from the map or taking up space with another legend item. Another issue that has to be watched is the polylines that connect the rings to the map. I converted these to graphics and then adjusted them. I also annotated the labels for the counties and adjusted both these and the polylines to be as near the center of the county as practicle, but also watch and make sure that no two polylines crossed near where on ended, making sure it was clear where each line ended. I added the standard north arrow (I went fancy here just for show, the simpler the better with north arrows), the small text and information on where the data came from, and a scale. The scale in this case is not really necessary, but fills in some extra whitespace and helps balance the area. The last item is the legend. Ringmaps are mostly used to show trends, as this one does, so having the actual values in the legend is not as important as identifying the trend. In that case, I went with a low to high method which simplified the legend, helps the reader understand the trend, and makes for a cleaner design. If this was being used in a research paper or report, then I would assume the actual numbers would be found in a table or discussed in the body of the text.

Ringmaps are great and I have plans to use them in a upcoming paper I am working on and so this article and toolbox was very helpful and timely. The things I have shown here are not the only or best way, but work well for me. Try it yourself and let me know what results you come up with and where you see I could improve on this one.

test

What do you see in the image below? A series of black blobs, or the words MAIL BOX?

mailbox

 

 

I recently started reading the book “Gödel, Escher, Back: an Eternal Golden Braid” by Douglas Hofstadter, and Chapter III really got me thinking. The chapter is on Figure and Ground, which mainly relates to art where the figure is the subject of the artwork and the ground is the background. Hofstadter’s book, though, is not so much about art as it is about logic and the way we think and perceive the world.

I started thinking about figure and ground from a geospatial point of view. As an example, let’s say we are analyzing an area of interest trying to determine a specific type of habitat or feature. Depending on the type of analysis, the features we are looking for may be small, complex, or not neatly defined. But what if we looked at it from the point of view of what is not our target feature? Then, like in the image above, the ground becomes the figure and the figure becomes the ground. Similar to the way astronomers used to look at the glass plate exposures from their telescopes. They would look at the negative image where the stars were black on a clear or white background, thereby making them easier to see because of the contrast. Using this same idea applied to geospatial analysis, in some cases, might useful in highlighting the areas we are interested in by looking at what it is not.

References

Hofstadter, D. R. (2000). Gödel, Escher, Bach: An eternal golden braid : 20th-aniversary edition with a new preface by the author. N.Y: Penguin Books.