On Message Sources, Part 1

Recently I’ve been looking into the problem of pro-health messaging on twitter regarding the opioid epidemic. This presents a natural analogy to contemporary pedagogy as we are looking for optimal strategies to educate a population, albeit distant and often unwilling in this case. A rather important point of consideration in dong so is how effective such campaigns really are. More so, are individuals more likely to share pro-health messages if they are seen to be coming from the right source, and do certain sources carry certain messages with more gravitas?

In order to explore these dynamics, I’ve conducted a pilot study analyzing 5,403 original, non-retweet tweets collected with keywords relevant to opioid abuse. These tweets were collected from Virginia and Georgia over an 8 month span. Each tweet was manually labelled for having one or more messages regarding opioid abuse or marked irrelevant if no such messages were present. Message types included generic avoidance of drugs, discussion of health consequences, discussion of legal consequences, public interventions, reports of contamination and/ or misrepresentation of drug contents, and use witnessed/ experienced. Users profiles for tweets not marked as irrelevant were flagged as coming from individuals, social pages, organizational pages, law enforcement, public agencies, and media pages.

For each tweet, a sum potential exposure was calculated as the sum of the original tweeter’s friends and followers counts, plus the sum of the friends and followers counts for each retweet of the original. Using these sum exposures, message virality was defined as the sum of all exposures for a given message divided by the sum of the friends and followers counts of the original posters. Plotting these out, several trends become apparent:

Firstly, looking at total messages expressed, we see that individual posters, social pages, and deleted users make up a vast proportion of the bulk of all shared messages. We also see that they really like to talk about their personal drug use or drug use they’ve witnessed. Their relative proportions of messages are also quite similar. One may infer that the deleted pages would mostly represent individuals, and that social pages generally followed the structures and interests of the individuals captured within this data set, rather than adhering to organizational content guidelines.

This is evermore apparent when you compare the total messages expressed per profile category. However, an interesting thing happens when you start weighting figures by the total exposures rather than just sheer message output:

Suddenly those news pages become extremely prominent. We also note that more health-positive/ abuse-negative messages rise in weight of exposure across the board. News agencies do very well with general fear stoking, as discussion of health and legal consequences of opioid abuse flourished there. Individuals also significantly boosted the exposures for messages related to contaminated drugs.  Law enforcement pages face mixed results when talking about consequences of abuse, but they do very well when conveying messages about avoidance and public interventions.

Looking back to that pi chart, once weighted by exposure we see the News category suddenly gains a large upper-hand over individuals, and Law Enforcement gains significantly as well.

Finally, we come to message virality, or the odds of a message transmitting well beyond its initial audience. Within virility, we find a nice surprise in that law enforcement pages did the best of all profile types in spreading messages with regards to avoidance, public interventions, and health consequences. Secondly as expected, news pages do well when discussing the consequences of opioid abuse. Individuals discussing what or how things should be fixed got a fair bit of representation as well in spite of their relatively low follower counts. However, while individual messages of contamination saw fairly solid exposure rates, they actually were not retweeted within this data set. This would suggest that many of those reports of contaminated drugs were made by individuals whose follower bases were great in number yet reluctant to share messages. Finally, a strong disappointment here is present within the organizational category. While there were some companies and advocacy groups within this category, most of these tweets were from K-12 public schools hosting anti-bullying and anti-drugs days to positively influence their students.  As these messages showed poor virality, it seems that people really aren’t spreading these kinds of messages from these kinds of sources.

There’s a lot of take home messages from this with regards to pedagogy, and more posts will follow on the matter. I think it’s important to note that who your audience perceives you to be will have drastic effects upon how well your messages are conveyed. No one wants Law Enforcement entities to wag the finger of enforcement at them, yet when they’re perceived as kind and helpful people are extremely receptive. News agencies meanwhile do best for doom and gloom to the detriment of most other subjects. Finally, people like to feel that the experts are in control and that solutions are within reach. Posts regarding public interventions were quite successful coming from individuals, law enforcement, and to a lesser degree, news sources.