Tracing Disinformation Trajectories from the 2010 Deepwater Horizon Oil Spill

In 2010, as part of a team at the University of Colorado, I organized a crowdsourcing effort to map tweets from the Deepwater Horizon Oil Spill. For more than three months that summer, I collected #oilspill tweets. I also interacted via Twitter with digital volunteers and local citizens (along the Gulf Coast), as we worked together to create a crisis map of the event. At the time, I noted the emotional impact that those involved with the effort were experiencing, especially those who lived within or near the affected areas. I recall exchanging messages one evening with a woman as she walked along her favorite beach and tweeted reports of oil impacts to the place she loved so much. And I will never forget the direct message (DM) I received late one night from a woman who lived in southern Louisiana. She told me she had read an article that said there was an imminent threat to people along the coast, that the ocean floor would collapse, sending a tsunami of water and oil onto shore. She sent me a link to the article. She asked me if I thought she should leave her home. The article was frightening, but there was something not quite right about it. It cited Russian scientists. The website had a strange structure. And the information was all very confusing. And very, very scary. I told her (via DM) that I couldn’t say what to do, but that maybe she should trust her intuition (and go visit family elsewhere for awhile), at the very least to reduce her own anxiety.

Two years later (2012), as part of a team of researchers based at the University of Washington, I received a small grant from the Coastal Resource and Resiliency Center (CRRC)to study how people in the online crowd understood the risks and benefits of the use of chemical dispersants to clean up the oil spill. My graduate student (Dharma Dailey) and I returned to the #oilspill data that I had collected (>600,000 tweets) and conducted mixed-methods research to understand the online conversation around that event. This included identifying accounts of people who lived near the affected area and doing extensive qualitative analysis of their tweets. We approached the activity we were examining (information sharing by the online crowd) as one of collective sensemaking (Shibutani, 1966) — people working together to make sense of available information, which included considerable scientific uncertainty and conflicting views. You can find our final paper here.

As part of that research, we also conducted a “link analysis” — qualitatively coding external online content that was linked-to in the tweets. We sampled 500 tweets and read all of the articles referenced in these tweets (more than 100 unique articles). To do this, we relied heavily on the Internet Archive’s Wayback Machine to revisit web content that had disappeared or changed in the years since the spill. We tagged each article according to who was cited, including scientists and political figures. It was painstaking work, but incredibly enlightening.

We noted plenty of strange things, but I’ll just share the ones that are relevant today. First, we saw a considerable amount of politicized content, mostly highly critical of the Obama administration and its response to the spill. Here’s a few semi-random tweets (of thousands) from that subset:

@anonymized1: I don’t know if the #oilspill is Obama’s Katrina, but Obama is certainly America’s Katrina! #tcot #p2

@anonymized2: The “great exaggerator” is the one that owns this; its Obama’s lack of response to the #oilspill that makes it HIS Katrina.

@Sami_Shamieh: Report: Obama killed nearly as many pelicans as he did jobs in June http://bit.ly/c8wzvA #tcot #p2 #oilspill

The @Sami_Shamieh account was particularly prolific in terms of tweet content, sending 1230 #oilspill tweets during the three month period, and that content was particularly dense in terms of political themes. Many (perhaps most) of the tweets from this account were highly critical of Obama, including some accusations that echoed widespread sentiment of disappointment as well as other more slanted claims, some insults, and a few references to unrelated political criticisms/attacks on the President. Here’s a selection:

@Sami_Shamieh: Obama is lying about the amount of oil pouring into the Gulf to save his popularity #oilspill #tcot

@Sami_Shamieh: Obama is so stupid, he thinks “recovery” means kicking his cocaine habit #obamafail #tcot #oilspill

@Sami_Shamieh: Finding liberal voters is getting harder than finding Obamas birth certificate #oilspill #p2 #tcot

That #tcot tag, which appears often in tweets from this account, is a hashtag used to designate a certain type of political content: “Top Conservatives on Twitter”. The #p2 tag is ostensibly a similar tag for liberals, but in my crisis-related data, it is more likely to appear with #tcot on right-slanted content than to appear alone on left-slanted content.

The @Sami_Shamieh account was the 8th-most retweeted in the entire #oilspill collection (retweeted 3259 times). We ran additional analyses to assess the extent to which this political content was circulating within the mainstream Oil Spill conversation — i.e. among the Twitter users who were “local” to the event as well as other influencers in the broader conversation. As part of this, we constructed a network graph (below) to map information trajectories across accounts. We used the Gephi application to create the graph.

Information Flow Network Generated by Retweets with #oilspill Tweets

In this graph, nodes are Twitter accounts and edges (connections between the nodes) are created when one account retweets another. Edges become stronger when there are more retweets. And accounts are sized by the overall number of retweets. Actually, we use the log(number of retweets). Accounts are colored using a method (Louvain) for detecting communities — accounts with similar connections have the same color. From this graph, you can get a sense of how information flowed during the Oil Spill. For example, the bright green group includes local Twitter users, NGOs, and media who were active in the affected area.

Interestingly, the @Sami_Shamieh node is outside of the main #oilspill conversation. That account is highly retweeted, but often by people who only engaged with the Oil Spill conversation once or twice — and only by retweeting that account. However, that’s not to say that information from @Sami_Shamieh (and other politicized posts like it) never made it into the mainstream. On occasion, it did. Four locals in our qualitative sample retweeted @Sami_Shamieh a total of 11 times. And other members of the broader conversation spread info from that account as well. In the graph, you can see a few tentacles (I mean edges!) extend from that red cluster on the right into the larger cluster. We can think of them, potentially, as “weak ties” (Granovetter, 1973), which are powerful in that they can connect more tightly clustered groups with information outside their borders. We can also perhaps imagine that the spreading mechanism in this case came not through following relationships but through Twitter’s search functionality: people following the #oilspill hashtag may have seen these tweets and retweeted them without ever following @Sami_Shamieh (or even one of that account’s followers).

Note: By 2013 (when we completed our analysis), the @Sami_Shamieh account had disappeared. It is unclear if the current account by that name is at all related to the one that was tweeting in 2010.

With hindsight, we can interpret some of the actions of this account as mirroring what we now see within alt-right communities online — spreading highly politicized news and commentary. However, it is important to note that there were also left-leaning accounts among those critical of Obama and the response.

A second interesting feature of this event that we noted during analysis (but didn’t dwell on at the time) were a few strange stories that were circulating about potentially (and in some cases imminent) catastrophic consequences of the spill. These stories were often posted from accounts focused on the political content, but they were also shared by some of the locals who were trying hard to understand the event (its causes and impacts), and they reminded me of the one that Twitter friend of mine had shared with me, via DM, while the spill was still happening. Here are a few examples:

@Anonymized4: TOXIC RAIN/COREXIT COULD DESTROY AMERICA http://www.eutimes.net/2010/05/toxic-oil-spill-rains-warned-could-destroy-north-america/ #oilspill

@Anonymized5: GULF EVACUATION ALERT: #Russia recommends to #USA they nuke the #BP #Oilspill? http://bit.ly/aTpS9k oil + toxic chemicals + nuke? Dirty!

@blackgoldnews: #oilspill #bp Warning* Russian Scientists :Sea Floor Fractured Beyond Repair http://url4.eu/5Hh4v

One common element of these most-frightening stories were links to external websites that contained articles referencing Russian scientists or Russian Ministry reports supporting the claims. For example, the first tweet above links to an article in the EU Times that includes the following claim:

“A dire report prepared for President Medvedev by Russia’s Ministry of Natural Resources is warning today that the British Petroleum (BP) oil and gas leak in the Gulf of Mexico is about to become the worst environmental catastrophe in all of human history threatening the entire eastern half of the North American continent with “total destruction.”

The EU Times is an online “news” site that often features neo-Nazi themes. In 2009, the Southern Poverty Law center attributed the website to a woman living in Florida who self-identifies as a “white racialist”.

This website was one of several that were spreading similar stories about the Oil Spill. Many of the articles on these sites linked back to other similar sites. For example, an aptly named “disinfo.com” site shared this same story, excerpting from the EU Times in their version.

Perhaps not surprisingly, the political cluster of accounts (the red group on the right in the Figure) played a role in propagating some of these stories. @Sami_Shamieh spread at least two of them.

@Sami_Shamieh: Could Toxic Rain from Obamas #OilSpill Destroy North America? http://www.silverbearcafe.com/private/06.10/toxicrain.html #p2 #tcot

@Sami_Shamieh: Reports Say Huge Explosions Cause Seafloor Collapse Beneath Gulf #OilSpill http://bit.ly/cj1PFB #tcot #p2

And sadly, some of these stories eventually reached the accounts of locals, who were emotionally impacted by the information within them. The tweets that we studied demonstrated considerable emotional distress, and though causality is impossible to assign, it is possible that the information spreading online about potential environmental and public health impacts (much of which was accurate, but some of which was not) contributed to this distress.

So what does this all mean? I think we have evidence (from 2010) of disinformation trajectories that have origins in Russian propaganda and that move through alternative “news” sites and social media accounts into mainstream conversations. I think we have evidence that some of these flows did affect how people experienced that event — they confused people, they scared people. I do not think that we have evidence here that these flows are coordinated or orchestrated by an organization or foreign state. (We also have plenty of disinformation trajectories that do not have origins in Russian propaganda.) It is just as likely that these flows reflect emergent effects of multiple actors performing information-sharing functions with very different motivations. However, it is also possible that if someone understands the dynamics, they can utilize these flows to shape what people see online — which is increasingly tied to how people experience the world. I’d make a pretty high wager that someone (or some many) does understand these dynamics and is utilizing these flows in this way — if not back then, then certainly now.

References:

Shibutani, T. (1966). Improvised news: A sociological study of rumor. Ardent Media.

Granovetter, Mark S. “The strength of weak ties.” American journal of sociology (1973): 1360–1380.