Is lack of usability an inherent problem in residual vote rates?

After reviewing a big pile of recent research about voting and usability (what I’ve been doing rather than posting to this blog the last couple of weeks), I was left with the question: Are residual votes inherently usability problems?

I realize that counting is a technical problem, and could result in residual votes – undervoting or overvoting.

But more questions come to mind:

Why do people overvote? (That is, make more selections for an office than are allowed.)

Why do people undervote? (That is, making fewer choices for an office than are allowed.)

Why do voters undervote?

We assume in both over- and undervoting that people are doing this unintentionally for the most part. Although I have read a report of a voter saying that he voted for two candidates for state representative because he had promised both he would, obviously, there are times when voters would want to undervote. I don’t want to worry about intentional undervoting – I wonder about unintentional undervoting.

There are many motivations for intentional undervoting: I don’t vote for judges because I know nothing about them. I only vote for the two candidates when I could have voted for more candidates because I want my vote to be weighted toward those two candidates. I’m just not interested in voting for anything except President. I don’t think my vote counts for anything in the presidential race.

Well, okay. But what about accidental undervoting? This must come down to problems with design of the polling place, voting systems, the ballot, and how pollworkers are trained and function. Bear with me while I explore in a sort of thought experiment the type of problems there might be with the information and visual design of ballots. Yes, this is conjecture. We need more research.

What are some usability problems in using ballots?

As the post below asserts, a number of researchers have identified a variety of usability problems. Most of the research I have reviewed about usability and voting is conducted from a social sciences or political sciences point of view. Based on research in areas such as cognitive psychology, human-computer interaction, information design, linguistics, reading, and technical writing Ginny Redish identified likely problems with the design of ballots, focusing on instructions on ballots. Her review of ballots from all 50 states and D.C. shows that instructions

• Are inconsistent
  
• Don’t anticipate likely mistakes
  
• Don’t cover important situations, such as straight-party voting and writing in candidates
  
• Use computer jargon
  
• Use voting jargon
  
• Use other words that may confuse voters
  
• Name buttons in ways that may not be explicit enough
  
• Signal voters to vote before they’ve completed the ballot
  
• Put the action before the context
  
• Are all in one place, not where they’re needed
  
• Are not in logical order
  
• Come too early or late to be useful
  
• Are in paragraphs, not separate lines
  
• Are statements, not directions
  
• Use gender-based pronouns
  
• Threaten rather than help voters
  
• Use double negatives rather than the positive

See http://vote.nist.gov/instructiongap.pdf to get a copy of Ginny’s report, entitled “Review of the gap between instruction for voting and best practice in providing instructions.”

Looking at design heuristics for forms and user interfaces, possibly these problems cause undervoting, too:

• It may not be obvious what his touchable or markable on a ballot.
  
• Colors, contrast, or other visual cues may be missing or too subtle for voters to know that their choice has been completed.
  
• The touchable area may not be large enough to hit easily.
  
• It may not be easy to get back to a screen on the Direct Recording Electronic (DRE) machine that the voter has been to already; it may not be easy to distinguish which cards or pages or sides of cards are for which types of offices or decisions.
  
• Messages may not be worded plainly or assume that voters knows more about the system than they do.
  
• The divisions among offices may not be obvious enough, or they may be so close to the bottom edges of the screen that voters with vision problems may miss them.
  
• The design of the page or the wording of the instructions may demand that voters make inferences about something, which they may not be able to do.

While researchers have collected data on the numbers of over- and undervotes, it is difficult to say even from observational studies what the specific problems are that cause residual votes.

Looking at these likely problems, what else might be causing over and undervotes, and are they different on optical scan ballots versus online on a DRE? Here are my preliminary thoughts about causes:

Optical

Over

• Voter makes a mistake marking intended candidate
• Doesn’t understand limitations on multi-candidate races

Under

• Voter is unaware of multiple candidate option
• Doesn’t understand multiple candidate option
• Doesn’t find all candidates she wants to vote for, either because of rotation (not alphabetical) or some other reason
• Doesn’t see the race at all (perhaps because of the position on the card)

DRE

Over

Probably not possible on a DRE – implications are that voter doesn’t vote for intended candidate though, because she can’t figure out how to change her vote

Under

• Voter misunderstands instructions
• Doesn’t see that it is a multi-candidate race
• Can’t find the name she wants to vote for, either because of rotation or some other reason
• Cannot physically select candidate (because of bugs or disability)
• Doesn’t see the name because of glare
• Missed a race divider so didn’t see that there was another race to vote in

Are voters’ tasks supported by information design and language in ballots and voting systems?

Since much of the research about usability and voting seems to say that error rates are similar across voting systems when the voting population is fairly homogeneous (minority, low income, and low education populations have higher residual vote rates than white, affluent, highly educated voters), let’s think again about the types of problems that all those voting systems might have. See the posting for Friday, August 25.

Many of the known and suspected problems I think come from information design. Why would these things happen? Redish’s review suggests that some problems are probably due to placement and order of information along with language. I suggest, based on other research that I’ve done, that other problems come from visual design interacting with information design, such as typography.

A voter may not be able to find what they are looking for because of

• Rotation order of candidates.
  
• Visually, two (or more) offices run together, or one office runs over two pages.
  
• There are no obvious instructions (the instructions are in a separate place or not visible within the voter’s field of vision).
  
• Items are too close to the bottom or top of the screen.

A voter may not be able to see things because

• The type is too small.
  
• There isn’t enough contrast between the type and the background.
  
• The line height is too tight, so things are too close vertically.
  
• Information is in all caps, which turns the words into a solid block visually that is ignored.
  
• Shaded areas to divide offices on an optical scan ballot may turn them into blank spots that are ignored.

Then there are ergonomic issues that may commingle with the information design problems. The setup of the physical system might mean that

• Lighting at the polling place creates glare on the DRE screen.
  
• The angle of a screen leaves parts of the ballot not visible.
  
• The height of the booth makes it difficult to reach the screen or uncomfortable to mark an optical scan ballot.

Then there is language. Language-action theory, says that “action is initiated and shaped by conversations.”^[1] The language of ballots must make up at least part of the “conversation” that the voter is having while voting. Bill Killam says he’s seen interaction “between the ballot instructions and the voter’s perception of both computerized systems and elections as official activities. The effect was observed in how they interpreted the instructions (therefore how they acted) as well as how important it was to them (think performance anxiety).” This cognitive load suggests that Redish is right about emphasizing that instructions should be made up of “short, succinct, clear sentences” written in “gender-neutral, simple, and consistent language.” As I said, more research is needed.

---

[1] Denning & Dunham, Innovation as language action, Communications of the ACM, Volume 49, Number 5 (2006), pages 47-52