BARD BLOG

Contents

• Subject application
• Comparison to alternatives
• Obstacles and glitches
• Chronological journal of sessions
• Links

Subject

Better Automated ReDistricting (BARD) is software developed by Michael P. McDonald and Micah Altman.

Chronological journal of sessions

• 30 December 2010
• 3 January 2011
• 7 January 2011
• 11 January 2011
• 13 January 2011
• 18 January 2011
• 27 January 2011
• 31 January 2011
• 7 February 2011
• 9 February 2011

Obstacles, glitches, and inefficiencies

• There was a security obstacle concerning registration.
  
  
• An error message appeared on an MIE browser.
  
  
• A report did not appear after a wait of 40 minutes.
  
  
• The website can forget client plans in between sessions.
  
  
• Street names did not appear.
  
  
• It was difficult to find the borders of the civil divisions (townships, cities outside townships, and villages outside townships) legally necessary to build state legislative districts in split counties. It was necessary to consult the American Fact Finder provided by the Bureau of Census.
  
  
• The emphasis on tracts obscures local civil divisions.
  
  
• It seemed impossible to keep stable the territory displayed on the map while switching legislative chambers.
  
  
  • There was no way to overlay various sets of districts on one map, to prevent splintering into tiny voting precincts.
    
    
  • Although State Senate districts must include whole State House districts, there was no easy way to find discrepancies between the two chambers' districts.
  
  
• It was unclear whether the shading for ethnic groups represented absolute or relative population.
  
  
• It was difficult to read light print on a dark background. This was true of some menu commands and district numbers on large counties on chloropleth maps.
  
  
• We could not find leftover land or its population. The server seems able to find it. How do we?
  
  
• We could only see the population of one geographic unit (county, tract, or block) at a time.
  
  
• Rectangles appeared in only one alignment, along compass coordinates.
  
  
• An attempt to pick up leftover geographic units (counties, tracts, or blocks) can accidentally grab units that the client intends to leave in other districts.
  
  
• We found no way to draw a circle or ellipse around land to be moved
  
  
• It was unclear how to use the "anchor" function to move a number of geographic units at once.
  
  
• Cohesive communities of interest could not easily be assembled for later combination into districts.
  
  
• Several attempts to move a block from one district to another failed.
  
  
• The BARD demo can only create districts within the client's imagination. It cannot suggest novel solutions.

30 December 2010

We tried the demo on Thursday evening.

The first browser we tried, Microsoft Internet Explorer, soon produced a pop-up message: "Stop running this script? A script on this page is causing Internet Explorer to run slowly. If it continues to run, your computer might become unresponsive."

Switching to Mozilla Firefox eliminated the interruptions, allowing the browser to run slowly without affecting the time-sharing capability of Windows XP.

A chloropleth map appeared that we recognized as the state of Ohio, on the south shore of Lake Erie. Counties were outlined and shaded according to total absolute population (as distinguished from density).

Ohio is a reasonable choice, Ohio being a rather complex state to cleave, although not as complex as Florida or Texas. One point of confusion is that cities and villages may be located within townships or outside any township (as a county subdivision). The two types of cities and villages are sometimes found in the same county.

We chose the option of seeing existing congressional districts. The outlines became visible, with district numbers. It was impossible to read two of the district numbers on the dark background in populous Cuyahoga County.

After giving up on existing districts, we started again with a blank map.

The menu offered to "snap" the map to show counties, tracts, or blocks. There were no options for civil divisions (such as cities, townships, and villages), planning areas, or neighborhoods. Intending to see if the city of Cleveland had an appropriate population and characteristics to serve as the basis of the first district, we looked in vain for the city line. We saw no indication that cities like Cleveland, Lakewood, Euclid, Parma, and our hometown of Shaker Heights even exist, much less the means of combining a number of cities to form a district. Nor could we find the Cuyahoga River, which we employed as the only natural boundary between the eastern and western Cleveland districts in 1985, when that city's population still exceeded a single district.

The menu offered three tools for adding territory to a district:

• one-by-one selection of geographic units (counties, tracts, and blocks),
• drawing a rectangle, or
• drawing a polygon.

This is how those three tools could be employed:

• Any geographic unit (county, tract, or block) intersecting a rectangle or polygon would be included.
• The rectangles were compass directional, with sides apparently falling along parallels of latitude and meridians of longitude. (A rectangle's sides ran parallel to the edges of the screen. Because of the scale, we could not tell whether the map projection was Mercator or not.) Compass directions would suffice for much of the Midwest, where land was mapped and sold in squares before it was surveyed. We do not see how compass-true rectangles could be applied as efficiently in New Jersey or South Carolina, where boundaries often run parallel or perpendicular to the shore. (We have found that a simple formula entered into a column of a spreadsheet can sort centroids on a diagonal.)
• The select function required pulling down the menu after choosing each unit to add; we saw no option to quickly click the mouse on each of a group of units and assign them as a cluster, which would have speeded the procedure.

Meanwhile, there was no simple way to draw a circle or ellipse, not by inscription, not by superscription, not by drawing a radius, not by marking three points of arc.

A geographic unit (or those within a rectangle or other polygon) could be added to Districts 1-19 or placed in the "unassigned" list. Population was reported for the numbered districts. The "unassigned" population included only units the client subtracted from districts, not the entire population of the state or county left outside numbered districts. There was no option for compiling portions of districts and keeping them for subsequent combination into districts. (For example, one might wish to group a few closely-related suburbs together to see whether their inclusion might make or break a city district.)

We chose to see tracts.

We switched the theme from total to black population. Most of the tracts lightened, but a forked cluster near the center of the screen appeared dark. It was unclear whether that indicated absolute or relative black population. In other words, we had no immediate indication of whether a tract with a black majority of 2,000 and a white minority of 1,000 would be displayed lighter or darker than a tract with a white majority of 4,000 and a black minority of 3,000.

We then started District #2. Although we tried to collect only land outside District #1, we accidentally grabbed part of District #1.

3 January 2011

On Monday, we took a stab at populous Franklin County, near the center of the state. In 1985, we drew one district entirely east of the Scioto River, leaving the rest of the county grouped with some neighbors. The gaping holes in the Swiss-cheese-like city of Columbus posed an unusual challenge that tends to be associated with haphazard annexation where it is permitted by state law. We vaguely remember townships bisected by strips of the city and several census tracts that straddled the city line. We kept a map from the American Fact Finder open for consultation, with Columbus city selected for shading. Otherwise, we would have had no idea how to find the city limits. Even with this reference, it was difficult to find the same location on both maps at the same time.

We drew District 1 east of the Scioto River. It appeared to be mostly in Columbus.

Among the statistics displayed on the screen is the population of unassigned land, but it includes only those geographic units (counties, tracts, and blocks) selected to be unassigned. If a portion of the state is left over at the end, how will we know, and how will we find it?

We requested a report with all the statistics. A pop-up notice appeared: "Working/Please wait while your report is created. This will take a few minutes." This notice remained on the screen for 40 minutes (about 7:50-8:30 p.m.), until we walked away from the computer.

We moved a few tracts into District 2. District 1's population hit 629,382, just shy of the target of 630,730. Is there a way to get closer? As most tracts are larger, we would probably have to move blocks. We can use the mouse to find the population for one block at a time, but how can we see the populations on a number of blocks at once? We are used to seeing each block's populations printed on the block on printed maps produced by government agencies.

We added several counties to raise District 2 to 628,149. We later found out they were Delaware, Madison, and Union counties. Why did we add those counties? They were close, and they fit. Could we have added other counties instead? Probably, but we don't really know.

The three counties we added had an aggregate population of 191,111 in 2000. Franklin County's population was 1,068,978. Together, all four counties had 1,260,089. Yet we only had 1,257,531 in the two districts. Where were the other 2,558? Probably in Franklin County, although we did not see where.

The BARD demo seems unlikely to produce a district that the client does not think of. Dozens of earnest people might innocently overlook the possibility of combining communities one way rather than another. That blind spot might become an obstacle when making a legal argument that one plan is among the best that can be drawn, as well as depriving the state of cohesiveness in its electorates.

7 January 2011

We saw a tool that would drag geographic units (counties, tracts, or blocks) into another district. Is there a tool that will drag a district line? We'll have to check.

11 January 2011

We ran into a security obstacle having to do with registration. That may need looking into. We chose the option of making an exception for this URL.

We decided to try state legislative districts. Although allowed a fairly wide population range, they are required to include whole municipalities or city wards.

The map only offered to show boundaries by two options: None and County. When we got to Franklin County, we found census tract displayed, some of them divided by the lines of existing districts. When we tried to the portion of a tract in one district, the portion in the other district came along for the ride.

A map of the Columbus area published by the Bureau of census about 1981 shows Census Tract 63.91 consisting of portions of Perry township, Sharon township, and Columbus city. To create districts that comply with the Ohio constitution, it may be necessary to divide the tract between the two townships. Except for a few blocks of Linworth Road in the vicinity of the Outer Belt highway, the Perry-Sharon township line does not seem to be marked by visible features. We have not seen the names of townships or streets on the BARD demo. How are we supposed to find the right locations?

The utility of census tracts varies. In Chicago, Detroit, Jacksonville, and Denver, they are the building blocks of larger communities, such as planning districts. In New York City, running willy-nilly like chaotic brickwork across boulevards into disparate neighborhoods, they mean jack squat; there are even blocks (within tracts) that straddle railroads that form the boundaries of community districts. In Los Angeles, their practicality along the ridge of the Santa Monica mountains diminishes as they descend into the basin.

13 January 2011

This time we tried a Gateway computer with a Microsoft Internet Explorer browser. (We will have to check whether the usual brand is Hewlett-Packard or something else.)

We could not find any of the congressional plans drawn last week. Perhaps a software upgrade eradicated them.

We switched to the State Senate.

Zeroing in on Franklin County, we moved three tracts from District 3 to 15. The screen urged, "Please wait while new districts are computed." Then we moved three tracts from District 15 to 3. Although slow, BARD produced the desired result. Perhaps the operation seemed slow because the computation was performed on all 33 districts from the start (the existing 33 districts of the State Senate).

Block 1005 of Tract 27.10 remained part of District 3, although surrounded by District 15. Somehow, it had been missed. We tried to move it from District 3 to 15 at 7:23 p.m. (EST). Several times, the display correctly confirmed inclusion of the block, but failed to move it. Eventually, the command worked. District 3 was contiguous. It was not yet the right size.

18 January 2011

We encountered a security problem. A message appeared on the screen informing us that the site's identity could not be verified because of an invalid security certificate. We were offered the opportunity to make an exception, which we did.

We chose to see the State House of Representatives, which has 99 districts. Each State Senate district is supposed to include three entire State House districts. We could see no easy way to display both the House districts and the Senate districts at the same time, to make sure that any change in one would be reflected in the other. Nor, when we looked at Senate districts, did we see options for grouping House districts.

27 January 2011

This time, we tried to find any connection between the two state legislative chambers by switching between them during the work session. We chose to see a map of existing districts of the State House of Representatives. It started as a small-scale statewide map that we clicked to home in on Franklin County. Then we switched to the State Senate. Again, we had to narrow from the statwide map to Franklin County. When we switched back to the State House of Representatives, we again found a statewide map. The demo does not seem to be set up to allow switching between different sets of districts in the same geography. To compare how the different districts affect an area, it would be necessary to copy the screen when each map is displayed and compare those images. It would still be a chore to overlay the sets of districts.

Why is it necessary to compare the different sets of districts?

• The state constitution requires each State Senate district is supposed to include three entire State House districts.
  
  
• To contain costs and protect the secret ballot, it makes sense to avoid splintering territory into undersize election precincts of one or two blocks. That is done by making sure the congressional and state legislative district lines either coincide or fall far enough apart to place entire precincts between them. How can that precaution be taken without seeing the lines on the same map?

31 January 2011

• There is an "anchor" function that is supposed to allow multiple geographic units to be moved at once. We could not figure out how to use it. The description was in small, light print on a dark background, making it impossible to read.

7 February 2011

• It seems that Dublin city, a county subdivision in Franklin County, not only straddles the Scioto River, but it isolates pieces of townships on both sides. It lassos chunks of Perry township on the east bank and Washington township on the west. As these three governmental units are too small to split, Ohio law makes it necessary to group them together in the same state legislative districts. Let's see if BARD can help with that.

9 February 2011

• Today we wrote to the support staff. After mentioning this page, we wrote,
  • "At this writing, information is needed that would clarify how the BARD demo can be used to create state legislative districts that meet all legal requirements. Further instruction would be appreciated."
  
  
• We hope to get an answer soon.

Here are some comparisons between Better Automated ReDistricting (BARD) and Ward Cleaver's methods:

Feature	BARD demo	Ward Cleaver's proposed website	Ward Cleaver's proven traditional methods
form	fully automated software to assist client in drawing legislative districts	semi-automated website to draw legislative districts to order	various paper and virtual maps, spreadsheets, and tables
system requirements	21st century browser	any browser that can handle input forms	pencil, paper, calculator, census documents, microfiche reader
geographic range	Ohio, apparently to be extended to USA	intially USA, eventually expanding to other countries and continents	mostly USA
levels of government available for partition into districts	federal and state, as far as we can tell	all conceivable, including, hamlet village borough town city township county state federal continental planetary	town city county state federal
geographic units (defined pieces of land) available for building districts	block census tract county	where available, block neighborhood (from Neighborhood Statistics Program of US census) neighborhood (as defined locally) hamlet village borough town community planning district city township magisterial district county province state region country	block block group neighborhood (from Neighborhood Statistics Program of US census) neighborhood (as defined locally) hamlet village borough town community planning district city township magisterial district county state
population standards	rigidly applied but not explicit	chosen by client from list	±2% usual; ±3% in Maryland for 2000
input	client's choice of display criteria geographic units (defined pieces of land) to comprise each district report features	client's choice of standards for finished plan from input menu	population statistics
metes and bounds as output	not mentioned	optional	standard
orientation of maps	north on top	chosen by client	usually north on top, because of limited availability of base maps
algorithm for building districts	land added as chosen by client	partitioning, limited exhaustive search, staff operator-chosen, etc.	partitioning, trial and error, occasional exhaustive search