BITlab: Behavior Information Technology

BITlab
404 Wilson Rd. Room 251
Communication Arts & Sciences
Michigan State University
East Lansing, MI 48824

BITLab REU Interns have posters at Mid-SURE 2015

By: Emilee Rader

The annual summer Mid-SURE (Mid-Michigan Symposium for Undergraduate Research Experiences) event will be held on July 22, 1:00-4:00 PM at the Breslin Center.

The BITLab will be represented by 6 student posters, featuring projects these students have worked on during their 10-week NSF REU (Research Experiences for Undergraduates) Internships with us.

Title: Modeling the Diffusion of Shared Votes on Facebook
Authors: Paul Rosemurgy (Michigan State University), Chankyung Pak, Dr. Emilee Rader
Project: Algorithmic Curation
Time: 3pm

There has been much discussion concerning the potential for Facebook to influence the outcome of an election through social signals about voting behaviors, made possible by an “I Voted” button that users can click to let their friends know that they voted. Bond et al. (2012) found that receiving a social message about others’ voting behavior resulted in greater voter turnout. This has raised a question about whether Facebook has the power to differentially mobilize voters and influence which political party receives more votes. We created an agent-based model to simulate how the diffusion of social signals of voting behavior results in network-level voting patterns. We simulated different scenarios by varying initial conditions such as the size of the network, average number of friends, and percent of participants who will initially share their vote to match data taken from Facebook’s network. Our simulation measures network level outcomes such as the total number of votes induced by the interaction of individual agents within the network. When individuals in a social network recursively influence their neighbors by announcing that they voted, network-level voting patterns appear. An increase in the sharing of voting behavior is directly related to an increase in the number of votes. The model can also be utilized to suggest that Facebook not only has the power to influence the number of votes, but also which political party will vote more when the option to share votes is given to exclusively one political party.

Title: Flocking Together: Social Influence and Homophily Interact to Create Grouping Behavior
Authors: James Finch (Michigan State University), Dr. Emilee Rader
Project: Algorithmic Curation
Time: 1pm

A typical property of social networks is autocorrelation, the tendency for one person’s characteristics to predict the characteristics of those they have social relations with. Determining to what extent social autocorrelation is caused by similar people forming relationships (homophily) or by related people becoming similar (social influence) is an ongoing area of study. While previous research has studied the effects of homophily and social influence with real-world data, the lack of detail regarding subjects’ social decision-making processes in these data prevents these studies from explaining social influence and homophily’s impact on autocorrelation in terms of individual decisions and motives. The present study sought a more detailed, mechanistic understanding of social influence and homophily’s effects on social networks through the construction of an agent-based model. By testing different social decision-making paradigms in a model social network and measuring the corresponding homophily and social influence behaviors that emerge, individual-level operationalizations of these two constructs were formed (e.g., “people avoid relationships with dissimilar others” is a decision paradigm that operationalizes homophily). These operationalizations then became conditions in experiments designed to assess their effects on autocorrelation, the overall structure of the network, and how information travels across it. Results indicate that decision-making paradigms analogous to homophily cause isolation between dissimilar people in social networks, while paradigms analogous to social influence cause clustering among related people. Broadly, these results suggest that social influence and homophily are fundamental and interacting social forces that uniquely structure our social world.

Title: Understanding Online Security Behavior
Authors: Ruth Berman (Macalester College), Dr. Richard Wash, Dr. Emilee Rader
Project: Security
Time: 1pm

From email to online banking to Facebook accounts, passwords are an integral part of protecting online information. Yet, many people engage in insecure password practices, such as using the same password for multiple accounts or choosing commonly used passwords like ‘123456’ and ‘password’. These practices leave people vulnerable to online attacks that compromise personal information. Using a dataset containing the online activity and security practices of over 120 users of Microsoft Windows 7 and 8 collected over the course of six weeks, we analyze in which circumstances users adopt better password security practices. We use a measure of password security called ‘entropy’, which is a score assigned to a password based on its length and uniqueness of characters. Along with entropy, we look at whether users choose different passwords for different accounts. We then compare this information to users’ understanding of online security and their descriptions of their own security practices, compiled from a survey completed by all participants both before and after the six week data collection period. We hope to better understand which systems of thought lead to safer password practices and more secure online behavior. Insight into these factors can help us to encourage better online practices and password behavior in the future.

Title: Cleaning Computer Security Data
Authors: Robert Plant Pinto Santos (Senior Federal University of Ceará, Brazil), Dr. Richard Wash, Dr. Emilee Rader
Project: Security
Time: 3pm

Does how people think about computer security affects how they actually use their computers? Answering this question will help to create more efficient approaches to improving computer security and make people less vulnerable. To answer this question each person completed a survey about how they view their behavior and then they installed software on their personal computer that recorded every process that was executed by the computer over the course of 6 weeks. A computer can execute hundreds of processes in a minute and every computer runs in a different way, not only according to its hardware and software but also by the way the person uses it. This means that there is a lot of data to be analyzed and that data is messy. The data needs to be cleaned and reorganized to extract the information that we need to match computer and survey data. It is necessary to clean and reorganize the data to be able to analyze it and to avoid misleading results. Data cleaning includes examining the data and finding patterns in it, as well as search for cases that do not agree with those patterns. Then we must determine if these weird cases should be cleaned. This is unclear because a data point data does not follow a pattern is not necessarily inaccurate. We need to decide when the data is ready to be analyzed because it is possible to get caught in a loop of analysis and cleaning.

Title: How Does the Number of Members Impact the Perception of Online Communities?
Authors: Sean McNeil (Cornell University), Chankyung Pak, Dr. Richard Wash
Project: Crowdfunding
Time: 3pm

Donorschoose.com is a non-profit crowdfunding platform where teachers can request donations to facilitate classroom projects. This study seeks to identify which words, phrases, and topics in a project description are more likely to lead to a project getting funded. Beyond that, we hope to discover whether teachers learn these words, phrases, and topics over time and become more successful at funding their projects by incorporating these features. Previous studies on text analysis not only lend insight on how to approach this kind of work, but also reveal that the different situational context impacts how the text performs. The same text has varying results in different fields and situations. This suggests we need to study the results that projects descriptions on donorschoose.com produce specifically. Using data from donorchoose.com between 2007 and 2012, we were able to analyze hundreds of thousands of text descriptions. We investigated several text analysis strategies including Linguistic Inquiry and Word Count, topic models, and n-gram frequency to select the best plan for studying these kinds of descriptions. We also sought to study whether teachers learned (i.e. whether they incorporated more qualities of successful projects and less qualities of unsuccessful projects over time) by comparing the project description of a teacher’s first project to that of their latter projects. If we can better understand the reasons donors decide to contribute to certain projects and how project creators learn over time, we can better educate project creators on how to construct successful projects.

Title: The Secret to Success in Crowdfunding: Word Choice
Authors: Ellen Light (University of Wisconsin-Madison), Jacob Solomon, Dr. Richard Wash
Project: Online Communities
Time: 2pm

Online communities like Wikipedia, WebMD, Twitter, and GoFundMe generate, collect, and display knowledge. However, sustainable and productive online communities are difficult to create and maintain. Understanding how people think about online communities will help us understand how to create successful ones. One factor in a community’s success may be its size, measured by the number of members. Previous studies suggest that large communities experience more movement in and out of the community than small communities: more people join, but more people also leave in the same amount of time. However, little research has been done on the thinking process behind user participation. This study attempts to characterize the differences, if they exist, between how people think about large communities and small communities. To do this, we constructed a survey that questions people about a fake online community where members discuss health issues. Participants will either observe this community with many members or with few members, then be asked about how they see the site, including the purpose and characteristics of the community, whether they would consider joining, and whether they noticed the number of members. We expect that large communities will seem more sustainable and less demanding, making people more likely to join. Small communities may seem to require a larger sense of personal responsibility and longer-term commitment, making it harder to attract new members. We hope that the results of this survey will give us insight into the success of online communities.