Mean SAT Scores & Participation for 2013 by State

This project involved displaying two datasets on one map, the mean SAT scores by state for 2013 as well as the rates of participation. The scores are displayed as a choropleth map with the participation rate overlaid as bar graphs that show the ratio of test-takers to non test-takers for each state. Care was taken that the bar graphs for each state were clearly assigned and readable.

There seems to be a clear inverse correlation between rates of participation and mean scores for each state. So much so that for the highest two classes for SAT scores (i.e. 1601 and above), the highest participation rate reaches only 13% (Ohio). Meanwhile, the states that have 100% participation or close to it (Idaho, Maine, Delaware) have mean scores among the lowest in the country. All three are in the bottom five along with another high participation region - Washington DC at 81% - and a middle of the road state when it comes to participation, Hawaii.

The low participation states seem to have a population of motivated test takers who have a vested interest in the outcome of the test, while the high participation states have made taking the SAT mandatory for all graduating high school students, thus diluting their results with a large volume of students with no stake in their test results.

The primary takeaway from this map, then, is that mean SAT scores by state is not a useful metric for comparing educational outcomes across states.

Mean SAT score totals by state for 2013

Population Density Dot Map

This mapping project involved displaying population density by dot density. Provided with a base layer map of the counties of South Florida and a .xls file with population information, I joined the excel table to the map layer and then had ArcGIS display the population of the area by overlaying the basemap with dots representing a quantity of population.

In order to better represent the actual lived population density of the area, the size of the dots and the quantum of population they represented was adjusted manually until an appropriate relation was found (where in the densest portions of the map, the dots would begin to overlap each other). Also, a mask layer was employed that represented the urban areas of South Florida and all dots were constrained such that they had to fall within the limits of that mask, providing the result below.

Further work was done adding further geographical context, with large cities and some roads being identified on the map. However, ArcGIS' dot density feature proved to be problematic once masking was employed. Not only did redrawing the dot layer consume a lot of time every time a change was made to the base layer (scale, location, etc.), but when there were other layers in the .mxd file - cities and roads - turning on the masking feature caused ArcGIS to crash. Several workarounds were attempted, but there was nothing for it, so I simply redid the map base layer with the dot density information and produced a much sparer map than anticipated.

Population Density in South Florida

Proportional Flowline Mapping

This week's lab assignment turned out better than I had feared. My task was to create a proportional flowline map showing the relative proportion of immigration to the U.S. from various parts of the world, but I had to do all my work entirely within Adobe Illustrator. Fortunately, we were provided with a basemap in .ai format, which had layers prepared in good order, so I did not have to worry about interpenetrating features like I've encountered in the past when I've exported to .ai format from ArcGIS.

The flow of immigration is depicted in the map below by arrows that are sized proportionally to the relative amount of immigrants that entered the U.S. in 2007 from each continental region (excluding Antarctica). The proportional width of the arrows was based on the square roots of the immigrant population (thus tempering the proportional differences, else the difference between Asia's immigrant population and Oceania's would have been two orders of magnitude rather than one).

2007 Immigration to the United States

Isarithmic Mapping

This week's lab assignment involved two varieties of isarithmic maps, which are used to depict continuous phenomena that can be fairly said to have a value at every point on a surface. Examples of continuous data that can be shown isarithmically include precipitation levels, temperature and elevation levels.

The two maps created this week show the same data in slightly different ways. The continuous tone map takes the precipitation data provided in the raster layer and simply assigns it a color value along the range of the color map in direct relation to the precipitation value on the number line. The data is unclassified. The raster layer is also set up to provide elevation levels, which are depicted via using hillshading.

The hypsometric tints map depicts the data in discrete classes, which were determined manually. To allow this to work, the raster data on precipitation levels were converted from floating point data to integers. The different classes were then further emphasized by creating contour lines using the same class intervals as were used to create the tint classes.

Average Annual Precipitation (Continuous Tone)

Average Annual Precipitation (Hypsometric Tint)

Proportional Symbol Mapping

This week involved creating maps that displayed data via proportional symbols, i.e. for each enumeration area that data was presented for on a map, a symbol (a circle, square or possibly an icon of some sort, such as a wine bottle) is displayed with its size proportional to the value of the data point relative to the rest of the data.

Two maps were made, one in ArcMap and one in Illustrator. Illustrator allows a great deal of design control and can take .ai formatted files that were created in ArcMap, thus allowing ArcMap's strengths to be leveraged before going to Illustrator to gain finer design controls. Unfortunately, ArcMap seems to have problems organizing layers in a logical fashion when exporting to .ai files, often mixing up objects in a single layer that have nothing to do with each other.

Wine Consumption in Europe [done in ArcMap]

Wine Consumption in Western Europe [done in Adobe Illustrator

Choropleth Mapping

This lab involved creating two choropleth maps drawing from the same dataset. A choropleth map is a type that should be familiar to most people these days. It's a thematic map that uses color, shaded or pattern fills of bounded areas (often the borders of counties, states or nations) to show the variation in the data being considered. The dataset that I worked with in the lab was the change in population by state from the year 1990 to the year 2000, expressed by the percentage change over that interval.

The critical portions of this assignment involved determining how to divide the data points up into a classification scheme and then creating a color scheme to express that classification. How should the data be chunked into groups (if it should be, at all)? How many groups (or classes)? How should the fills for each class be determined? None of these questions have straightforward answers, and details like these, if chosen poorly, can lead to maps that mislead rather than inform.

For a map showing growth rates by state, to be displayed in color, I chose a classification scheme based on an algorithm that attempts to create classes that match natural groupings in the data, in theory matching like with like. I only made my choice after trying several methods to ascertain how well they expressed the data. I also viewed the data as a histogram in order to see how the data was spread out. The data displayed a significant skew to the left (the lower range of growth rates), with a few strong outliers to the right. I wanted to make sure that those outliers were visible on the maps and not obscured by being classed with regions with significantly lower rates of change.

I also made the choice to remove the District of Columbia's data from the classification calculations, although I do show it and its negative growth rate on the map. As it's not only an outlier, but also the only negative growth area on the map, I picked a color well outside the color scheme I chose for the classification scheme used with the rest of the data.

The other map I produced calculated growth rates over the same time period by arbitrary regions. For this map, I was instructed to build my classification scheme outside of ArcGIS, and thus had to calculate it by hand. This was not as difficult as it might have been as the data was limited to nine regions. I ended up with a basic scheme that placed two regions in all but the highest growth region, which got a class all its own.

--

The base map for this assignment uses a conical equal area projection. As a result, the direction of north on the map varies greatly by longitude. Because of this, I chose to indicate the orientation of the map without using a north arrow. Instead, I drew a graticule (grid lines showing a range of latitude and longitude) on the main map and also in the inset maps for Alaska and Hawaii. The area covered for the Washington D.C. inset was of a large enough scale that a small north arrow within the inset was appropriate.

Color Map of Population Growth in the United States from 1990 to 2000 by State

Displayed as Percentage Change from 1990 Levels

Greyscale Map of Population Growth in the United States from 1990 to 2000 by Region

Displayed as Percentage Change from 1990 Levels

Data Classification

This week, our lab assignment centered on implementing and assessing a variety of data classification methods. This involves taking a data set - in this particular case the percentage of population in each US census tract in Escambia County, Florida that was reported as black in the 2000 census - and determining the best way to divide the range of data values into categories that can be symbolized on a map. Four different classifications were used.

• Equal Interval classification simply divides the entire range of values from minimum to maximum into identically sized sub-ranges without regard to how the data is clustered along the number line.
• Standard Deviation classification bases its classes on the standard deviation from the mean value of the data.
• Natural Breaks classification uses an algorithm to find an 'ideal' set of ranges that respects how the data is aggregated along the number line.
• Quantile classification creates sub-ranges that each contain the same number of values in each, or at least as close as possible based on how well the number of sub-ranges factors into the number of values. In this case, number of classifications was five, so it is referred to on the map as Quintile.

Two maps were created. The first displays all four of these data classification methods in separate maps for comparison purposes. The second map only features one map that uses the best option for classifying the particular data set being inspected. In this case, the Natural Breaks method was chosen. The data values supplied from the census data showed a heavy positive skew - a majority of values were clustered in the lower quarter of the possible range of values, with a long tail extending to the left. Such a heavy skew indicates that this data does not conform to a normal (or gaussian) distribution. That makes the Standard Deviation classification a poor fit for the data. The skewing also makes the Equal Interval classification a poor choice. Note in the map below that most of the census tracts have the same classification on the equal interval map. The Natural Breaks and Quantile methods are both suitable for this data, but the latter results in the lower ranges being extremely narrow. That is potentially misleading to a map user who doesn't look closely at the legend. The Natural Breaks classification does have narrow ranges on the left side of the number line as well, but not to the same extremes.

Four Classification Methods Compared

Population Percentage using Natural Breaks classification.

Getting Started with Spatial Statistics

This week featured an introduction to best practices for preparing a map and dataset for a spatial statistical analysis. This included evaluating the spatial distribution of data collected (as per the map linked below), including calculating the mean center and median center of the data points. If they are in close proximity, it is an indicator that the data may be normally distributed, whereas if they are not the data likely has some sort of spatial skewing going on. The oval representing the directional distribution of the data points below is calculated based on 1 standard deviation of the data point's location calculated along the x and the y axes. In this case, the data tends to be spread out further along a more or less east to west axis, although the axis have a slight tilt towards the southwest from the northeast. Normally distributed data would also tend to have about 68% of its data points lying within such an ellipse.

From there, we looked at ways to determine how the values of the data are distributed. Both histograms and qq plots were looked at as ways to evaluate whether the data values are normally distributed and also whether there are any outliers in the data. From there, we looked at ways to explore the variation in our data, with an eye toward making sure that the data has a locational element to it. This involved voronoi maps and semivariogram clouds. These also act, again, as methods that can help point out outliers in the data that may need to be removed from the dataset.

Finally, we were exposed to methods that can be used to do a trend analysis to look for any spatially oriented trends inherent in the data.

An Analysis of the Spatial Distribution of Temperature Data in Western Europe

Mapping Typography

This week's cartography exercise focused on the complexities involved in using text labels in mapping. I was given an Adobe Illustrator file that contained only an unlabeled mapped area and a separate list of diverse features to locate and label on said map. These features included islands, bodies of water, cities, an airport and some recreational features. It was my task to label these features according to established typography guidelines for maps and to add the other necessary map elements that would make the map legible to the map user. I placed text labels according to the type of feature labeled, e.g. placing labels for hydrographic features within spaces representing those features, in all caps, spanning the feature when appropriate and colored them carefully so that they stood out legibly from the blue background of the map area, but did not overwhelm the rest of the map or the labels for the land based features. Care was taken with label placement, type size, type positioning and choice of type family. The resulting map, below, provides a general overview of the layout of one chain of the lower Florida Keys, Marathon.

The Islands of Marathon, FL

Cartographic Skills

This week, we worked on making a visually coherent map and also sharpened our Adobe Illustrator skills. We were given a incoherently colored map of South Florida depicting the percentage of Hispanic population by county along with a pair of smaller scale maps to provide geological context for those who don't know where Florida is. I fixed the colors by generating a monochromatic range of colors via Color Brewer and manually editing the colors for each percentage range. A blue background helps the orange tones stand out, while the two inset maps have been left with color schemes that tend to blend in to the background.

Introduction to Adobe Illustrator...

This map is output from a simple introduction to using Adobe Illustrator with output exported from ArcGIS in *.ai format. On the surface, one can't really distinguish it from an export to .jpg from ArcGIS as most of the work was done by organizing the map output in the layers panel, primarily in such as way as to ensure that the scale bar and the geographic output always stayed the same relative to each other. I've done enough Illustrator map work on my own to appreciate this viscerally.

Two Map Critiques

Example of Good Design:

This map succinctly depicts the status of the opposing alliances of the Cold War - NATO and the Warsaw Pact - as they were comprised in 1970. The design minimizes clutter, placing the rest of the world in the background (or simply not depicting it due to the use of an orthographic projection). The map layout, with the North Pole tilted toward the viewer, helps emphasize the proximity of the rival powers.

Map source: From Nuclear Weapons and NATO: Analytical Survey of Literature by United States Department of the Army, 1970. 

From the Perry-Castañeda Library Map Collection, University of Texas

Example of Poor Design:

At first glance, this map is rather striking. It purports to show growth trends among the world’s “Mega-Cities”, which it defines as those with a population of 10 million or greater. However, on closer inspection, the data displayed on the map suffers from being incomplete and incoherent. The megacities displayed on the map have a graphic representation of their populations from 1950, 2000 (an estimate) and 2015 (a projection). One can infer that Mega-City status is conferred based on city populations in 2000, as many of the cities listed had populations much lower than 10 million in 1950. However, this is never defined. However, if that inference is intentional, then this map is woefully incomplete. Several cities that qualified for “Mega-City” status in 2000 are not included on this map. These include Seoul, Delhi, Baghdad, Teheran, Istanbul, Manila and Moscow. Meanwhile, the time increments selected for the population growth graphics for the cities included are inconsistent, spanning fifty years and then just fifteen. The fifty-year span from 1950 to 2000 is also used to depict national population growth trends with a single metric depicted chloroplethically. The metric used is (presumably) on the population difference for each nation from 2000 to 1950. Since each nation had a different starting population, population density and land area, this metric is meaningless. The fifty year timespan is simply too great to show any meaningful trends, anyway, even if the mapmaker had used a per capita growth rate or something else more appropriate.

Map source: From Global Trends 2015: A Dialogue About the Future With Nongovernment Experts [page 29], National Intelligence Council, 2000 

From the Perry-Castañeda Library Map Collection, University of Texas 

Introduction

I'm here by way of Providence, RI, where I recently got back into IT with a unix sysadmin gig that I'm doing to get out of the debt I've accumulated since I got my masters in landscape archaeology back in 2008.  Since then, I've had the privilege of digging in London and Iceland for a bit before I had to move back to the States.  I'm hoping to get into a fully funded PhD program in a couple of years, although I wouldn't be all that broken up if I just ended up working for a CRM firm instead.

I've been interested in GIS since I started the process of switching careers from IT to archaeology, but didn't get the kind of in depth GIS training I'd expected from my Masters program, so I thought the archaeology track here looked like a good option (especially since I couldn't find any appealing GIS programs near where I am).