Research Question: How Accurately Does Murder, She Wrote Depict Murder in America?
Throughout the pandemic, I’ve found myself wanting to rewatch TV I’ve seen many times before—a trend I think many people are experiencing given the popularity (or at least the amount of news coverage) of such shows as Friends and Seinfeld coming to streaming services. My show of choice is Murder, She Wrote (MSW) starring Angela Lansbury as teacher, mystery author, and super sleuth Jessica B. Fletcher. I’ve been watching this show to “escape reality,” but it’s got me wondering how much reality I am actually escaping. How accurately does MSW depict murder in America?
To explore this question, I have created a dataset about the murders committed in MSW and compared it with the data gathered by the FBI’s Uniform Crime Reporting Data project, which documents all known murders committed in America since in 1939.
Audience
Exploring how well media represents real life is an important area of inquiry for media studies scholars. And media concerning murder seems especially trendy given the continued popularity of case-of-the-week police procedurals and murder and true crime podcasts. My project audience also includes all my fellow fans of the Fletcher-verse.
Description of the Visualizations
My murder board opens with a word cloud of episode titles. Unsurprisingly, Murder is the most used word in the episode titles. Dead words (Death/Dead/Deadly/Die) all have a strong presence as well.
The next two pages examine the genders of the victims and offenders in each of the datasets through waffle charts. The genders of the victims are very similar between the two datasets; much less so for the genders of the offenders. MSW writers had more many more female murders than real life.
Next, I compared the relationship of the victim to the offender between the datasets via bar graphs. The FBI dataset includes many more relationship types than what I deduced from MSW episodes; as not all murders are solved, it also has a high number of unspecified relationships, which was not something I encountered in MSW. J.B. Fletcher has a perfect solve rate. The largest category in real life was that of acquaintance, followed by stranger. In the MSW episodes, other known is by far the largest category, followed by acquaintance. Overall, in real life and the episodes, people are much more likely to be killed by someone they know.
Lastly, I compared the weapons used through stacked bar graphs. I knew the MSW data would likely lack any pattern between the years, but this is made all the more striking given how consistent the weapon breakdown is in the FBI data. Guns are by far the most common weapon, followed by knives/cutting instruments. MSW has a fair share of both, but much more blunt force trauma (blunt object) than real life.
Data and Design Decisions
I have two datasets. The first dataset is the FBI’s Uniform Crime Reporting Program Data, which details all murders in the United States from 1939 through 2019. I am very luck to have a friend who worked with their dataset for her PhD dissertation, and she provided me with her cleaned version of the dataset (she worked with the data from 1980 through 2015). She also advised me that it was easier to work with murders that had one victim and one offender, so I filtered out all other murders with multiple victims and/or offenders. From her dataset, I also filtered the years to match the years that MSW episodes aired: 1984 through 1996.
The second dataset is one I’ve created. It includes the season, episode number, title, episode description, and air date for all 12 seasons of MSW (264 episodes). I did not include data for the four MSW movies. In my dataset, I’ve noted the genders of the victims and offenders (assuming a gender binary), the relationship of the victim to the offender, the weapon used, the circumstance of the murder (felony vs. non-felony categories), and notes about the murder (further clarifying circumstance and motive). For the relationship, weapon, and circumstance categories, I have mirrored the language used by the FBI’s codebook for their dataset. I excluded all episodes that take place abroad, outside of the “present day” represented in the show (e.g., episodes or murders within episodes re-examining cold cases), and outside of the “reality” of the show (e.g., episodes that represent the plot of a book or a dream), as well as murders that are not single-victim/single-offender. Murders that turned out to be suicides or were deemed justifiable (e.g., self-defense) have also been excluded (as they have been from the FBI dataset). For any storyline that took place over more than one episode, and only one single-victim/single-offender murder occurred place, the first episode has been removed. In the rare instances where an episode showed two unrelated single-victim/single-offender murders, this episode was listed twice, once for each murder. So every line of data in both the MSW and FBI datasets represents one single-victim/single-offender murder. I compiled the MSW dataset from my own watching as well as the following sites: IMDB, Wikipedia, Murder, She Blogged, and the Murder, She Wrote Fandom Wiki.
The FBI dataset was overwhelming in detail. It included much more information than I’ve ultimately decided to use, namely age, race, and ethnicity. These are details that are not always easily expressed in dialogue, so I knew it would be too hard to include in my dataset. I collected information about the circumstances of the murders in each episode, but I am no legal expert, and I really struggled to know whether these murders were meeting the legal definitions of felony murder versus non-felony murder, and the differences between such felonies as robbery, larceny, and burglary. I have such little confidence in this column in my dataset that I’ve decided it is best not to try and create any comparisons with the FBI data for this category.
Furthermore, the FBI dataset offers 29 different types of relationships of the victim to the offender; yet I still found these categories to be lacking. For instance, it notes employer and employee, but not co-worker or business partner, which seemed much more common a relationship in MSW episodes. Also, lover was a common relationship, but that seems different from boyfriend/girlfriend, friend, and acquaintance. Ultimately I classified all of these as “other known.” I’m unsure if this matches with the FBI’s approach to their “other known” category. The FBI relationships also categorize “homosexual relationship” as different from other familial relationships; indeed it is grouped under “Outside family but known to victim”, which makes their dataset feel very dated. Their gender categories are actually classified as “sex,” and they only include male, female, and unknown. To match, I used the same binary to classify the genders from the episodes–assuming from names, gender presentations, and pronouns.
Perhaps my favorite design decision is that I found a free font that matches that used in the MSW opening credits. It is admittedly more hard to read than many other fonts, so I relegated it only to the titles and made sure to keep it very large.
As I was creating my MSW dataset, I was very drawn to the titles, and really wanted to make a word cloud. However, when I used the titles from my main MSW dataset and turned them into single words and pivoted, it created so many duplicate lines of data, which would throw off my other visualizations. So I created a separate dataset of the titles in a separate Tableau workbook, a screenshot of which is included in this storyboard. I tried just the titles I included in the rest of my data at first, but I decided including all episodes ultimately created a richer word cloud. I excluded commas, colons, hyphens, and ellipses in my dataset. I filtered out words used only once, as well as the words “Part,” “1,” and “2” as these were only used to indicate multipart episodes. I only listed multipart episodes once in my dataset. Lastly, I excluded all articles (a/an/the).
Another workaround I used was in the color of the Weapons board. The FBI has five different categories for gun; but it was not always easy to distinguish the type of gun in the MSW episodes, so I chose to group them all in one category. For the bar graphs, I grouped the FBI gun types together by simply making them all the same color. Similarly, for the relationship graphs, I wanted to highlight that most relationships were known, so I offset stranger and unspecified in different colors.
The FBI dataset is much larger, with many more cases than all of MSW. To normalize the data, I therefore had all of the bar graphs show the data as a percentage of total number of cases (case number in the FBI data; number in the MSW, a field I created after it turned out I had two episodes with multiple single-victim/single-offender murders and so episode number stopped being a unique identifier).
Next Steps
I struggled to link my word cloud dataset with the other datasets. In my final iteration, I hope to fix this so that the word cloud and its tooltips can live all within on workbook and storyboard.
I’d also like to address the multiple gun categories in the FBI dataset and condense it all into one to avoid my current workaround in the weapons bar chart. I’d also consider creating calculated fields for the relationships, perhaps condensing all of them into four categories instead of the many I have now: family, known (outside family), stranger, and unspecified.
With more time, I’d love to talk more with my friend about what would be needed to analyze cases with multiple victims and offenders. That would pull a lot of episodes back into my dataset and I think would ultimately better answer my original question of how well the show represents murder. I would also look at the movies. If any were aired during the regular run of the series, I would also include them. I excluded them assuming that they aired in years different from the regular season.
