What do Tweets Reveal about Sentiment toward Obamacare vs Affordable Care Act?
Gary Carini
Vice Provost for Artificial Intelligence & Institutional Research at Baylor University
In their book entitled: The Second Machine Age, Brynjolfsson and McAfee present a compelling case for the surge of technological influence in everything we do. At the heart computing is the algorithm – code that captures sophisticated inputs in order to produce complex output. When we think of a computer going to medical school, we wonder how. Algorithms enable computers to store millions of volumes of medical information such that, when we type in symptoms, the computer returns a diagnosis. We still need medical personnel to treat us, though.
Many have said the computers can recombine and process information, but cannot (yet) be the source of ideas. So, humans have the “secret sauce,” as we are needed to interpret the output provided by the algorithm. In addition, Levy & Murnane (2004) suggest that we have a great capacity for interpreting complexity in a way that computers do not. Sure, they can present relationships between two data points, such as quality of care and morbidity, but cannot present us with the true meaning of the correlation.
Dr. Patricia Norman (coauthor here and with this research) and I have begun to explore patterns of moods and tones within the healthcare industry. We sought to analyze tweets associated with the Affordable Care Act and Obamacare, pre and post inauguration. We gathered tweets from January 18th and 25th. This web crawling yielded over 28,000 tweets over these two days.
By looking at positive and negative sentiment, we can detect overall tone in tweets or other documents. We observed the pre and post inauguration sentiments below. The Affordable Care Act tweets show an upward trend for both positive and negative sentiment. However, sentiment is more negative before and more positive after the inauguration. For Obamacare tweets, the negative sentiment is greater than the positive sentiment pre and post inauguration. Moreover, the difference becomes larger after the inauguration.
The pattern diagrams immediately below are based on 21,593 tweets containing #Obamacare extracted through web crawling. Focusing on Obamacare (contrasted with the ACA later), we observe more emotion that focuses on the individual level impact of repeal or replace. The analysis allows us to see the keywords in the context of the tweets.
Specific references to the impact on the patient were more closely associated with the emotion words such as scare, opposition, death, loss, and breakdown. Repeal, interestingly enough, appeared on January 25th but not the 18th.
#Obamacare Tweets from January 18, 2017
#Obamacare Tweets from January 25, 2017
Consider the diagram below, based on tweets with #Affordablecareact, obtained through web crawling for January 18th & 25th, 2017. With a focus on fear, again, the patterns we observed gave more attention to various aspects of the potential impact to policy. This reflects the patterns observed from 6,580 tweets across these two days.
#Affordablecareact from January 18, 2017
#Affordablecareact from Janaury 25, 2017
In the future, we will investigate other themes within healthcare based on real time data, including other tweets and RSS feeds from a variety of sources. Thematic changes within Obamacare, the Affordable Care Act, and the American Health Care Act tweets over time is of interest, since it has implications for shaping healthcare systems level changes. To date, we have collected 365,000 tweets and can extract a variety of societal, policy, and organization level information. We are also extracting patterns from documents for publicly traded healthcare systems and can contrast that with non-profit systems to discern areas of emphasis and resources allocation.
Product leader. Currently focusing on my own product ideas and maximizing ikigai.
8yVery interesting. Agree with the comments above that it would be even more interesting to look into demographics and political leanings after removing bot generated tweets. Maybe even look into how the sentiment about ACA and Obamacare has changed over the years since 2012 (hopefully the Twitter API would allow for looking back into 5-year old data). Algorithms are a reflection of the data humans create and humans, along with the information sources they rely on to form an opinion or sentiment, can be biased.
Helping Business Owners Achieve Intentional Success® | The #1 Executive Coach and Advisor in the AV Production Industry
8yThis was interesting. I'm curious whether your analysis considered (or if it was even a factor) if the tweet was itself computer generated (ostensibly for propaganda purposes) or came directly from an individual? Given the propagation of algorithm-generated propaganda in the past eighteen months, are tweets untainted enough to be considered usable data anymore?
Retired
8yWhat does this prove?
Principal at North Aspect
8yNice work! This could be a new path toward a Healthcare Consumer Confidence Index, or possibly a Government Services Consumer Confidence Index that could serve as a leading indicator of taxpayer sentiment toward proposed legislation. It’s a great way to passively gather random data from a large number of participants, and monitor how effective communication efforts have been. It would be interesting to see what, if any, demographic differences exist between users that use the term “Obamacare” vs. those that use “Affordable Care Act.” Does one group identify more as liberal vs. conservative? Is one term used more often in a specific region? Are there education, income, race, gender, or age differences that show up as outliers in the data? Interesting that “fear” is at the heart of both. I’m sure you can knock out that extra analysis in your abundant free time...I expect it by the end of the week!