The Vote-o-graph

Experiments with Touch Screen Voting

Part of http://homepage.cs.uiowa.edu/~dwjones/voting/
by Douglas W. Jones
THE UNIVERSITY OF IOWA Department of Computer Science

This material is based, in part, upon work supported by the National Science Foundation under Grant No. CNS-052431 (ACCURATE) from October 2005 to Sept 2011

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation or the University of Iowa or any other agency.

Contents

  1. Summary
  2. The Vote-o-graph
  3. The Experimental Data
  4. Publications
  5. The Name

Summary

The Vote-o-graph experiment was an investigation of touch-screen voting machines conceived in the fall of 2007 to study how people interact with misbehaving touch-screen voting machines. By the middle of 2010, exactly 100 subjects had been exposed to the experimental voting machine, dubbed the Vote-o-graph to explore how they responded to miscalibrated touch screens, insensitive touch screens, poor ballot layout and dishonesty in the machine.

Among our important results, we note the following: If the machine "flips votes", that is, shows a summary screen that does not accurately reflect the selections made by voters in the top race on the ballot, one third of the voters do not appear to notice, but rates of navigation skyrocket. This is because voters who do notice the flip go back to change it and in addition many go back to check other races on the ballot.

When voters touch a "button" on the screen, the vertical coordinates of the touch relative to the button are sharply focused with a normal distribution centered below the midline of the button with most touches falling within a centimeter-wide band. This is a sharp enough focus that measurements of touch position relative to the buttons on the screen can be used to measure the extent of touch screen miscalibration.

When the touch screen response is delayed, simulating insensitivity, voters keep their fingers on the "button" much longer than they do when the machine responds promptly to a light touch. A delay of 0.1 seconds causes a measurable effect, and a delay of 0.25 seconds causes a big effect. Furthermore, subjects who experienced the 0.25 second delay engaged in significantly less ballot proofreading, abstained from voting in more races, and complained about the insensitivity of the screen.

The following have been involved in some way in the Voteograph experiments:

The Vote-o-graph

The Vote-o-graph was an off-the-shelf HP Pavillion tx200 series Entertainment Notebook PC. This has a touch-screen, and the screen can be closed over the keyboard in a tablet configuration, with a form factor comparable to many existing touch-screen voting machines. The Java software and shell scripts used for the Vote-o-graph is included here in both Zipped and Unix tar formats, along with documentation and the ballot definition files used in our experiments.

The Experimental Data

The data gathered by the Vote-o-graph system is in the form of event logs. The logs are in XML format, and they record the time (relative to the session) of each touch and release and the type of touch (candidate selection, navigation and background). Touch coordinates are stored relative to the "button" touched on the screen, so that, if such a log was used in a real voting machine, release of the log would not violate voter privacy. During early runs of the experiment, we encountered problems that led us to change the log format, including increasing the precision of time measurements. The data exported here has all been converted to the final format. As with the Vote-o-graph software, the data is available in either Zipped or Unix tar format, along with the XML schema and additional documentation.

Annotations in the Data

Unfortunately, the Vote-o-graph software contained several errors. As noted above, most of these were corrected on-the-fly, but one had to be dealt with after the fact. Our logs did not report the difference between write-in voting buttons and normal candidate selection buttons on the screen. We went through the logs after the fact to infer which events exposed and hid the write-in keyboard. All touches made between write-in keyboard exposure and hiding were keypresses on the virtual keyboard. The final touch that caused the keyboard to be hidden was always an accept or cancel keypress. An annotation has been added to the event notices associate with write-in keyboard exposure and hiding, noting "writeIn show" and "writeIn hide" as shown below:

95.468: Screen touched at 4759458984233 
95.588: Button Feedback. writeIn show
95.588: Hit Button 
95.737: Screen released after 269413315 

115.659: Screen touched at 4779649716531 
115.777: Selected a candidate. Feedback Given. writeIn hide
115.778: Missed Button
116.004: Screen released after 345313392 

Each of the above examples shows a sequence of four events, starting with the event logging the start of a touch and ending with the event logging the end of the touch. The added annotations are shown in bold face on the second event in each case. Note that the erroneous report of a missed button after each write-in hide event. In fact, all touches made while the write-in screen is displayed were incorrectly reported as misses in the event logs of the first two experimental runs, covering a total of 20 subjects are incorrectly reported as missed buttons.

As a result, in our analysis, we ignored all missed button events when the write-in screen screen was displayed. Any attempt to analyze the behavior of subjects entering data on the write-in screen in relation to missed buttons should ignore our first 2 runs (20 subjects). The event log file names are the dates of the experimental run.

Publications

Finally, all of the publications resulting from the Vote-o-graph experiment are indexed here.

The Name

The name Vote-o-graph was inspired by the name of a long forgotten 19th century machine. To quote from the article on voting machines The Universal Cyclopaedia, Vol. XII, A. J. Johnson, 1900; page 248, column 2:

The votograph, or American ballot protector and recorder, formerly known as the Rhines machine, was made legally usable in Michigan in 1893. It is a box with a horizontal face on which the names of the candidates are arrenged, as with the Myers machine, by parties and offices. Slips bearing the names are inserted in the push-buttons themselves. Below are separate tally-rolls for each candidate, with serial numbers printed upon them in a vertical row. Pushing the button places a punch in position for each name desired, so that when all candidates have been selected the closing of the machine lid puts a hole through each roll. ...

J. W. Rhines received U.S. Patent 422,891 on Mar. 4, 1890 for his vote recording machine. There are significant differences between the machine described in the patent and the Votograph described in the encyclopedia article, evidence that Rhines continued development between the time his patent was granted and the time the encyclopedia article was written.

In adopting the name under which Rhines' machine was marketed, we've changed the spelling and hyphenation. Nonetheless, Rhines is the one who should get credit for the name.