In their analysis of newspaper coverage of opioids in the United States, McGinty et al divided the United States into four geographical regions (Northeast, South, Midwest, and West) and chose one newspaper from each region. The present study follows the same format, choosing the New York Times (Northeast), Tampa Bay Times (South), St. Louis Post-Dispatch (Midwest), and San Diego Union-Tribune (West). Examining all of these newspapers in total could potentially mitigate regional differences in sentiment toward psilocybin. Some of these newspapers have absorbed other newspapers over the time period being examined (January 1, 1989 to December 31, 2019). Following the example of McGinty et al, the archives of those absorbed newspapers have been included under the titles of their current publishers.

The four newspapers above were also chosen because the online media database Nexis Uni (the academic research portal of the LexisNexis database) had archives of their articles over the time period being examined. Sets of articles were created by searching Nexis Uni for all articles that mentioned psilocybin. McGinty et al and Zhang et al also used aggregated newspaper databases to compile their articles, as opposed to using the databases of the newspapers themselves (LexisNexis in the case of McGinty et al and newspaperarchive.com in the case of Zhang et al). In addition, my study used Nexis Uni in order to avoid any variations in the quality of databases and search engines administered by the newspapers themselves.

When creating sets of articles from Nexis Uni, it was necessary to search for a variety of terms in addition to “psilocybin.” Although synthetic psilocybin is available, the compound is often ingested via the mushrooms which naturally contain it. Unlike many other psychoactive substances, psilocybin is therefore frequently referred to by its delivery container (mushrooms), even in journalistic contexts. After extensive reading of psilocybin journalism, the present study chose four search terms for psilocybin: “psilocybin,” “magic mushrooms,” “hallucinogenic mushrooms,” and “psychedelic mushrooms.” Searches were made for both plural and singular forms. If other common journalistic terms for psilocybin are discovered which are not represented in the present study, this could be a topic for future research. There are also many slang terms for psilocybin, such as “shrooms” or “boomers.” However, most of these slang terms also have other meanings and seemed rarely used by the newspapers in this study unless other terms for psilocybin were also used. For example, a search for “shrooms” in the New York Times over the study’s time period only found 5 articles using this term for psilocybin, but 25 non-psilocybin uses. Therefore, I did not include these slang terms in my Nexis Uni searches.

There are also a small number of non-psilocybin mushroom species that have hallucinogenic compounds (such as muscimol, which is found in some Amanita species). There would therefore seem to be at least some risk that terms such as “hallucinogenic mushrooms” or “psychedelic mushrooms” might refer to these species rather than psilocybin species. For example, 4 out of 308 New York Times articles in the Nexis Uni search results referred to these mushrooms and had no references to psilocybin mushrooms. To avoid this problem, I removed articles from the Nexis Uni sets if they only referred to non-psilocybin mushrooms.

Duplicate articles were removed from the sets, even if they had different date lines. Nexis Uni has a functionality to filter out duplicate articles, but it was only sporadically successful, so manual removal of duplicates was also necessary. However, multiple articles about the same news event were included as long as the articles seemed independently written (for example, following new developments in the event or approaching it from a different perspective). The number of articles in which a term occurred was counted, not each individual occurrence of a term. Some articles used more than one term, which is why the overall number of articles in each newspaper’s set is lower than the number resulting from adding together the articles in each term category. It is also important to note that many of these articles do not have psilocybin as their central topic. Many journalistic mentions of psilocybin are made only as parts of articles about different topics.

After the searches and data cleaning described above, the final article set consisted of 534 total articles for all four newspapers. Of these, 304 articles were from the New York Times, 89 articles were from the San Diego Union-Tribune, 67 articles were from the St. Louis Post-Dispatch, and 74 articles were from the Tampa Bay Times.

Determining the sentiment of articles presented challenges. Nexis Uni, for example, has a feature called “Negative News” that uses algorithms to create a set of only negative articles for a particular search term. However, the Negative News results for psilocybin were often inaccurate. As just one example, the Negative News algorithms chose the New York Times article “Cancer Study of Hallucinogen Hints at New Role for Illegal Drug,” (Hoffman) which described studies that showed psilocybin could help treat depression and anxiety. Yet an article describing patients using psilocybin to reduce mental illness would seem to be a positive article. I suspect that Negative News chose this article due to its frequent use of words that often have negative connotations such as “depression” and “anxiety.” Many more examples of incorrect algorithmic choices could be listed here.

Even if attempts were made to adjust the negative weight of words in order to fit the paradigm of psilocybin, algorithms might still have difficulty with nuances. For example, in a New York Times review (Kakutani) of the book Prozac Nation, the word “depression” was used with the same definition as the positive article mentioned above, i.e. as a clinical psychological issue. However, in this context, “depression” was a negative word because the review implied that the author used psilocybin as a way to ineffectively escape her depression, not to treat it.

To avoid problems with using algorithms to determine the sentiment of articles, for the purposes of this study, I decided that human judgement was a better option. Yet there were still significant risks with this option. For example, the present study is being conducted by only one researcher, which could result in significant bias about sentiment. Future studies in this area might be improved by having more than one researcher determine the sentiment of articles. In addition, the same researcher could potentially give different sentiment ratings to an article depending on the condition of their mind at a given moment. A full discussion of the debate between machine and human judgment is beyond the scope of the present study, touching on the fields of philosophy, computer science, and logic.

To help mitigate these risks, I decided to read and rate all 534 articles twice. The two periods of reading and rating were separated by roughly two months, to allow for variation in mood or focus possibly caused by external events. The sentiment scores for each period were then added together and averaged. For those interested, a data table with my sentiment scores for each period is available in the Results section of this website (through the link “Sentiment-(Newspaper)”). This is still far from a perfect methodology, but I believe that the results provide at least some insights.

After settling on human judgment, other challenges presented themselves. Take, for example, the New York Times article “Ideas Unlimited, Built to Order” (Schiesel). This article praised the pleasant effects that psilocybin could have for its users, which would seem to be positive, but the article also called psilocybin “toxic,” which would seem to be negative. In addition, even articles that were clearly determined to be positive or negative might have had different degrees of positivity or negativity. For example, a brief article describing an arrest for psilocybin possession might be negative, but it could be argued that there is more negative impact from a longer article about a psilocybin arrest that also details the damaging impact of that arrest on the life of the accused. It would therefore seem negligent to treat these two articles identically.

To address these problems, I found inspiration in previous studies about media portrayals of psychoactive substances. McGinty et al and Zhang et al created different thematic categories in which each article could be assessed. Since their thematic categories were not exactly applicable to psilocybin, I identified five thematic categories to rate sentiment about psilocybin. For each category, each article was given a score of -1 (negative), 0 (neutral), or +1 (positive). The total scores in each thematic category were then calculated for each year for each newspaper. The scores in each category were also then added together to provide a single sentiment score for each year.

The five thematic categories in the present study were:
1) Legality: Is using psilocybin seen merely as a criminal act (negative) or only unfortunately illegal (positive)?
2) Mode of Action: Is psilocybin seen as physically damaging (negative) or physically beneficial (positive)?
3) Personal Impact: Is psilocybin seen as psychologically detrimental (negative) or psychologically beneficial (positive) to the user?
4) Scientific Integrity: Is psilocybin research seen as pseudoscience (negative) or rigorous and serious (positive)?
5) Social Impact: Is psilocybin seen as detrimental (negative) or beneficial (positive) to society beyond the user?

To further clarify the difference between mode of action and personal impact, mode of action concerns physical effects of psilocybin while personal impact concerns psychological effects. Returning to the example mentioned above (“Ideas Unlimited, Built to Order” [Schiesel]) the praise for psilocybin’s pleasant psychological effects received a +1 for personal benefit, but the reference to psilocybin as “toxic” received a -1 for mode of action.

My study is presented in a website built with the content management system WordPress and a heavily modified version of the WordPress theme ResponsiveBlogily. ResponsiveBlogily was not exactly suited for presenting a master’s capstone, so I had to hand-code substantial changes to the HTML and CSS. I hand-coded most of the website’s data visualizations with a JSON grammar called Vega-Lite. I used GitHub to store the data tables that were accessed by the Vega-Lite data visualizations. Code other than HTML and CSS cannot be inserted directly into WordPress, so I needed to insert the Vega-Lite into the ResponsiveBlogily JavaScript files. The few data visualizations not coded in Vega-Lite were built with Tableau software, and the Tableau images were inserted directly into the WordPress text content.

<< Previous Section   Next Section >>