It’s Positively Biased

February 8, 2009inf2241_classfi22411

Pandora’s box

Rounds two and three in the lab, and this time we’ve got a new plan. Our team came ready with some design ideas (including sketches!), we settled on an idea early, and we subdivided our tasks so we could accomplish as much as possible during the lab time. We have three sessions to get everything done; what follows is a summary of our work.

early sketch of the biased voting system by Antonio

early sketch of the biased voting system by Antonio

early prototype from week 4

early prototype from week 4

We were asked to create a biased voting system. There are a lot of ways to do this, particularly considering the loaded meaning of ‘biased’ (thinking back to ‘moral’ from earlier in the term). After some discussion, we decided to create a light-activated voting machine. The voter would be asked to select the amount which they support a candidate/issue. The voter would turn a knob (potentiometer) which would send a varying amount of current to activate an LED. The amount of light coming from the LED would be recorded by a light sensor. That number would then be converted into a degree of acceptance for the candidate/issue. Our code is as follows.

int lightPin = 2;    // select the input pin for the light sensor
int potPin = 4;    // select the input pin for the potentiometer
int ledPin = 13;   // select the pin for the LED
int sensor = 0;       // variable to store the value coming from the sensor
int potent = 0;       // variable to store the value coming from the potentiometer

void setup() {
pinMode(ledPin, OUTPUT);  // declare the ledPin as an OUTPUT
Serial.begin(9600); // initialize serial port

void loop() {
potent = analogRead(potPin);    //read the value from the potentiometer
sensor = analogRead(lightPin);    // read the value from the sensor
digitalWrite(ledPin, HIGH);    // turn the ledPin on
delay(potent);                  // stop the program for some time
digitalWrite(ledPin, LOW);   // turn the ledPin off
Serial.println(sensor);  // writing the val

By the end of our second week in the lab, we finally had the system working, more or less. We housed the contents in a box to limit external light sources. However, we quickly noticed a problem – the little lights on the Arduino board were messing with our numbers! This may just be a calibration issue… we could manipulate the Var “potent” in the code somehow (we’re currently thinking of multiplying it to excacerbate the small difference to the point where we can usefully employ them).

Breadboard/Arduino sandwich... yummy!

Breadboard/Arduino sandwich… yummy!

Another issue that we may tackle next week (time permitting) is what we do with the numbers generated by each voter. We would need an output in order to allow the voter to interact more with the system – displaying a score or percentage would allow them to continue to attempt to change their vote by small degrees. We’re currently using the Serial.println function, but may attempt to use the Processor language for more complexity and to allow us to store the numbers (kinda important in a real voting system).

Update (fourth week)

This time, we came prepared. As a solution to the Arduino light messing up our results, we brought a box that had a built-in separator, allowing us to isolate the light sensor and LED from the Arduino board. Moreover, by the end of the third week, we noticed that something was causing a short circuit and, as a result, the board would overheat to the point of discouraging any manipulation. This is what we were hoping to fix on our last week as soon as we walked in. And so we did. We had to move the light sensor resistor upstream, before the sensor itself and add a second resistor before the potentiometer. These sub-circuits, in turn, were not electrically fed by the first and last row of the breadboard. We moved them to the center of the breadboard and linked them with a jumper.

Comparison between the 3rd and 4th week

Comparison between the 3rd and 4th week (click image to see larger)

Inside the box

Inside the box

Final result

Final result

A Discussion on Bias

One of the most interesting aspects of our system is how it would introduce a gray area into an otherwise binary decision.  A voter could give (what they believe) is a more accurate representation of what they feel about a candidate or issue. Thus, they might support candidateA 83%. They would (attempt to) use the potentiometer to give candidateA “83% light”. These percentages would then be tallied… however the outcome would be, once again, binary.

Does this introduce a bias? We discussed this, and are, as yet, unsure. If a system allows more accuracy than the traditional system, is it biased? In some positive way? Maybe it’s biased towards accuracy? Heh. We’ll definitely need to continue looking at this question in the coming weeks.

Of course, many people would just decide to cast their vote at one end of the spectrum (meaning that they choose to support a particular choice 100%). This aspect of the system introduces another type of bias; different types of people will make stronger choices than others. For example, we would guess that a more educated person is less likely to see things as black and white, which would be reflected in their voting patterns (with more choices in the middle of the spectrum). We had a discussion about if/how to compensate for that, possibly by introducing a counterbalancing technical bias. We plan on looking into that more next week.

There might also be a potential for bias as a result of how the questions are asked. Phrasing, order, and types of language would all need to be considered. This would be particularly relevant if the system were used to ask about policy as opposed to people. The need for careful phrasing is more important with our proposed system, as it does not record binary votes… shades of meaning may ultimately make a large difference in the outcome of a poll. Finally, questions we’ll need to continue to address: Is increased accuracy a positive bias? Does adding more randomness into the system make the system more biased? Relatedly, do we need to bias in a direction? If so, what kinds of issues do we want to address? Alternatively, do we bias against a certain type of person?

A Discussion on Bias… updated

Our original idea with this device was to create a non-binary voting system where people had to chose from a gradient of options and not necessarily only between two set choices. As we observed our final result, we decided that there were actually different options for voting with our system. In a binary format, individuals might be asked to chose between two options by carefully selecting the correct corresponding value using the potentiometer within 10 seconds. For example, to vote for John Kerry, a voter could turn the potentiometer to reach the number 80 (as displayed on the serial window of the Arduino software). This point was interesting to us, as it is difficult to manipulate the potentiometer to obtain (and sustain) a very specific number. Moreover, the time limit (10 seconds) results in additional pressure for the voter who must get acquainted with the system, figure out how it works and finally cast a vote within the time limit. The biases involved in this option pertain to the physicality of the device itself (the potentiometer is hard to manipulate for larger hands) and to the inexperience of some voters who are automatically disadvantaged in this process. Older voters that are used to a more traditional voting system might be very confused with this new configuration.

Another bias could be found in the binary format by introducing yet another way to vote for the 2 candidates. For example, a vote for John Kerry could be only registered if you reach the value “80”, whereas a vote for George Bush can be registered if the values fall anywhere between 240 and 280. This clearly introduces a skewed favoritism for George Bush. Considering the physicality of the machine and the difficulty to operate it, one might settle for the easier choice to achieve, or simply void their vote.

Conversely, the voting system could be used to vote in a non-binary manner on qualitative questions. For example, a question like “Do you agree that abortion is murder” could be answered in a variety of ways from strongly disagree to strongly agree, passing by all shades of opinions and neutrality, each using the potentiometer.  In this sense, no vote can be voided. An obvious bias stemming from the question itself is the phrasing and the possible order of the questions which might influence voters. The abortion questions above is in no way neutral and is effectively leading the voter to agree to the statement. In this scenario, the physical and experience biases that applied to the binary voting are still valid here. However, we have introduced perhaps a more positive bias, one of accuracy. The voters, instead of having 3 stark choices (two options and void), are faced with a continuum of options to represent the intensity of their emotions/feelings about a specific subject. While this is more accurate for the voter, the voting system itself hides a lack of precision of the choices and selection process. Inside the ominous green box lives an overly complicated system (for what it actually does). Manipulating the potentiometer is not the only action that happens here; turning the knob does not directly manipulate the numbers output on the Arduino serial screen. Rather, the varied input is part of a chain of translations within the box. The potentiometer actually controls a LED in one section of the box. The varying degree of light is then captured by a light sensor which feeds the readings to the Arduino. The precision of the final number can be seriously doubted as some of the voltage is lost in translation (for example, the LED to light sensor translation is very finicky and the distance between the two can considerably effect the final result).

We did some further research on the subject of biased voting systems and found several articles helpful, listed below:

Tomz, M., & Van Houweling, R. (2003). “How Does Voting Equipment Affect the Racial Gap in Voided Ballots?” American Journal of Political Science 47 (1): 46–60.

Calvo, E., Escolar, M. & Pomares, J. (in press). Ballot design and split ticket voting in multiparty systems: Experimental evidence on information effects and vote choice. Electoral Studies.

Herron, M. C., Wand, J. (2007). Assessing partisan bias in voting technology: The case of the 2004 New Hampshire recount. Electoral Studies, 26, 247-261.

The Tomz & Van Houweling was particularly useful, as it was well written, current, and dealt explicitly with bias. As a result of the electoral issues in the US during the 2000 elections, voting equipment receives a great deal of attention in media and academic circles. Tomz and Van Howeling are particularly interested in how different types of voting systems produce different likelihoods of invalid votes, when examining in regards to race. They note that studies have shown that African Americans cast invalid presidential votes at a higher rate than whites, and then examine reasons why that might be the case. They use data from the 2000 presidential elections and account for present vs. absentee voting. They also attempt to validate their findings with opinion surveys that asked specifically about race and exit polls to estimate levels of intentional undervoting, i.e. Finally, they note that the Black/White gap may exist due to factors outside of their scope. They specifically mention three interesting factors; socioeconomic discrepancies, as black voters may be less educated and thus more prone to error; familiarity with the voting equipment, as election drives had produced a higher than normal number of first time voters; and finally, previous issues of racism (which might discourage African Americans from asking for help).

Tomz and Van Houweling found that the black-white discrepancy for invalid votes is widest where punch cards and optical scanners are used. They suggest that these forms of voting equipment are highly vulnerable to human error (hanging chads, confusing layout, no method to verify vote, etc.). They suggest that lever machines and Direct Recording Electronic Machines reduce this gap to the point where most of it can be explained by intentional voiding of ballots. They conclude that DRE and lever machines should be employed in future elections, unless cost or speed are too much of an issue.

Two other articles were assessed and sparked interesting discussions among our team. In their article, Calvo, Escolar and Pomares (in press) looked at how information displays might effect split-ticket voting. They clearly showed that voters are very receptive to information cues on the ballots and might alter their voting in response to this. Their major finding is that e-voting technologies that only displayed the candidates by name (candidate-centric) had a higher risk of split-ticket voting (3). Conversely, systems that emphasized party-centric cues such as party colors and logos reduced the amount of split ticket voting (3). An interesting point is their emphasis on the cognitive demand of voters where candidate-centric ballots demanded more cognitive capacities for voters in regards to recalling the corollary information associated with a name (3). In this sense, this issue became central to our discussion on the bias of experience where older voters accustomed to paper ballots will necessarily need to demonstrate more cognitive abilities in order to figure out how to work with the potentiometer.

Herron and Wand (2007) conversely observed how technology was wrongly blamed in New Hampshire. Following the 2004 elections, some New Hampshire precincts using the Accuvote machines reported unusually low numbers for John Kerry. Immediately, some began to blame the technology used and Ralph Nader himself paid for a manual recount of those precincts (248). However, the manual recounts yielded the same numbers as the Accuvote results. The article was not directly linked to our biases and our voting system, yet it was interesting to observe how the public now expects the technology to fail or to be biased. In this regard, even the most professional looking and seemingly accurate machine will attract the same criticism as the voting system we built which was obviously biased in a very specific manner.

You can follow any responses to this entry through the RSS feed. You can leave a response, or trackback from your own site.

One Comment

  1. » Blog Archive » RGB (Red, Green, Blue) Voting SystemFebruary 21, 2009 at 8:32 pm

    […] It’s Positively Biased […]

Leave a reply

Follow Us