---
title: ''
output: 
  bookdown::pdf_document2: default
---

# Limitations and future work 

      Despite the important strengths aforementioned, this study is not without limitations. One of the most significant obstacles to my work is the limited number of previous studies focusing on disinformation sharing behavior. Notably, it is particularly challenging to document and study social media users' information behaviors online. In fact, my research examines individual disinformation sharing intention rather than actual disinformation sharing behavior. The survey I designed simulated how individuals would react to and act on disinformation in a hypothetical social media environment. This is similar to polling research, where respondents may report for whom they may vote, but they may not cast the ballot eventually. In reality, individuals’ information sharing behaviors may be affected by many societal and psychological factors. For instance, people may be reluctant to share political messages in their social networks as it may have a negative impact on their relationships with family and friends. Fortunately, @mosleh2020self's study shows that self-reported sharing intentions demonstrated in online surveys, such as MTurk, are generally consistent with what actually would be shared on social media, which justifies the use of my research method. Since my research measures respondents’ sharing behavior rather than their attitudes, a possibility exists that respondents may have exhibited attitude-behavior incongruence with respect to disinformation acceptance in my study. In other words, individuals may believe the fabricated social media content presented to them but be reluctant to share it; or they may share it to others without actually believing it. This is another limitation that should be further investigated in future research. 

      In this project, disinformation is operationalized as a set of fabricated partisan claims in the format of Twitter posts. I used 16 question items to measure one's disinformation sharing behavior. To rule out the possible scenario in which respondents may have been exposed to existing disinformation or fake news stories before the survey, which may skew the survey results, I fabricated 16 social media posts and asked the respondents to report how they would react to such posts. For example, respondents who identified as Republicans may have received either the message "New study shows that Republicans are more likely than Democrats to be prone to science denial" or "New study shows that Democrats are more likely than Republicans to be prone to science denial." The wording of the first manipulation is obviously in favor of Democrats, and the second one is pro Republicans. In the results section, I congregated the responses by party alignment to perform statistical analysis. However, the experimental design in nature could potentially obscure a host of complex interactions as a Republican respondent's reaction to the first manipulation is highly likely to be different from his or her reaction to the second manipulation. 

In addition, one may argue that the research design does not precisely gauge disinformation behavior as the social media content is fabricated and fictional. Without an universal definition, disinformation is usually considered as any media content that is deceptive [@lazer2018science]. According to @allcott2017social, disinformation can be displayed as "news news articles that are intentionally and verifiably false, and could mislead readers." Therefore, in the case of my study, although the social media postings are fabricated by the researcher, they were designed in a fashion to intentionally and verifiably inaccurate, and therefore, should be safely classified as disinformation or fake news.   


      Another direction for future research is to understand the motives for individuals to share disinformation. My research investigates whether an individual is more or less likely to share fabricated messages, but does not examine the motivated reasoning for them to do so. Simply put, my research examines "how they share," but not "why they share." There could be many factors that drive an individual to share disinformation according to the political psychology literature: They may not believe the content of the information, but they share the information to inform others that it is suspicious information; Or they may believe the content of the information, and they share the information to promote it and let more people know about it. What prompts a social media user to share disinformation demands further research, but the literature on heuristics and elite persuasion may provide some theoretical insights into understanding (dis)information sharing behavior. Prior studies show that people usually use their endorsement in political figures as heuristics to guide political decision making and process unfamiliar information [@vis2019heuristics;@miler2009limitations;@steenbergen2018heuristics;@gilens2002elite], it is possible that they also use such heuristics to assess the veracity of disinformation. Furthermore, "disinformation" in this study is operationalized as fabricated social media messages that contain hyperpartisan information that likely represents a much larger proportion of Americans' social media diets. The actual on-platform exposure of social media users to real-world disinformation remains an open question.

     In my research, I experimentally manipulated research participants' party identification, levels of political knowledge, along with other factors, to investigate their disinformation sharing behaviors. The research results show that partisans are much more likely to dissimulate politically congenial disinformation. However, I did not manipulate prior factual beliefs and political motivations. Thus, observing a difference in information sharing across ideological lines may not be sufficient evidence to conclude that partisan identity causes the difference in (dis)information sharing. Additionally, sampling bias is another concern in the research design. Given that the panel of respondents were recruited using MTurk, the sample was not drawn from the entire American population. To address this issue, I used raking as the statistical method to adjust the segments of the target population in proportions that did not match the proportions of those segments in the target population. Even though it is crucial to adjust the sample to ensure representativeness of the U.S. population, survey weighting may also result in problems such as reduced accuracy, which could potentially skew the results and findings.