Group Project by Kathryn G. Roland, Tommaso Lombardi, Sheena Dilixiati
Project Overview
We designed a voter recommender system for ballot question one: nurse-patient ratio. In this system, a voter is provided with voting “recommendations” based on her or his preferences. We incorporated automation, information theory and other concepts covered in class, In our first meeting, we decided what we want to focus.
We came up with:
- Background Research: What these preferences are and how to support the preference.
- Define Problem: How do we find out the decision making process.
- Design: How to inform them of the recommendations
Correspondingly, we had several requirements and guidance for our final prototype:
- Unbiased, non-leading question setting
- A reasonable weighting system of result score.
- User-friendly interface design for asking the question
Limit the Number of Patients Assigned to Each Nurse ?
Should there be a legal cap on the number of patients assigned to one nurse? If voted into law, this measure would limit how many patients one nurse could care for in hospitals across Massachusetts. Question 1 was designed to establish patient assignment limits for registered nurses working in hospitals.
Read WGBH News' one-page voter guide on Ballot Question 1 to get a quick overview.
Initial design
Additionally to basic background research, to gather the initial information about the reasons why supporters are planning to vote ‘Yes’ and opposers are planning to vote ‘No’ , we conducted in-person interviews with college students and graduates.
For these interviews, we stated, “If Question 1 on the Massachusetts Ballot passes, there would be legal limits on the number of patients that could be assigned to each nurse in Massachusetts hospitals. This number would vary based on the hospital unit and patients' conditions. Do you plan to support or oppose the initiative stated in Question 1, and Why?”
Based on the interview result and our background research, we designed an initial question set for voting preference system.
We choose to used Qualtrics Online Survey Software to distribute a digital questionnaire to users and record their preferences to a series of questions based off of those initial interviews. Using scoring function and survey flow, we can design the system which would automatically show the final recommendation to users. The questions are all single-choice and closed-ended, in an effort to maximize the efficiency of analysis.
For detailed question design and weighting system, we used the card sorting technique to process and group information, make it easier to analyze. We limited the number of questions not to overwhelm the user. To do that, we created five categories that we deemed were most important to draw questions from.
Prototype Design
Final Response-Recommendation Logic
To get our final questionnaire, we gave the questions to the original voters and seeing if the questionnaire recommends the vote that they previously stated they intended to support. We also asked those original users for feedback about the survey. Modified the survey to clear question ambiguity, include demographic questions, and present the users with background information about the question initiative.
Disclaimer: the software we used, unfortunately, cannot learn from the users. Therefore we had to give the system a predetermined final result based on scoring the questions. The final score the user gets is then what determines whether or not they are recommended a yes or no.
Modification
After distributing the survey, we had some feedback regarding the survey design. We made some changes to the survey design:
- Added more about Introduction section
- Changed font
For the final prototype, we had introduction part for survey overview, icons and explanation for the final result. We included demographic questions, in the beginning, to have a more holistic view of the user - this aspect could be beneficial for an eventual machine learning algorithm to help with the results of the survey.
Final Survey Link:
https://tufts.qualtrics.com/jfe/form/SV_bKIUcsEqFaAJw1v
Discussion
If we could implement machine learning, deep learning, pattern recognition and other forms of artificial intelligence to the data analysis, there will be a considerable difference.
Predict Result
Using some weird algorithms and techniques, we could change a massive tedious amount of data into some practical recommendation and solutions. According to the research(Want to predict how someone will vote? Identity may matter more than policy). It is possible that we could predict the result before the voting. As it mentioned in the research (Parsimonious data: How a single Facebook like predicts voting behavior in multiparty systems), we could even predict from how people were liking politicians’ public Facebook posts.
Considering how our life is presenting by different kinds of data. Prediction of the voting result or even what might be the next vote will be an easy piece of work.
Multidimensional Data Analysis
Since our identity may matter more than policy, we could extend our survey data to other types of survey or prediction. It is possible that we could predict what people going to buy on the black Friday using the same data.
Public Service
Because we have unlimited and diversity data set. It will be reasonable that we know why people are choosing or voting using big data analysis.
As a result, we could use those analyses for public service. from the government perspective, since we know the reason why people make decisions, it's easier to control problems and improve the current situation. For example, if most of the people live in an area abstain from voting or complaining about one specific problem, the government should raise attention to it as soon as possible.
Comments