The story begins the 15th of December 1791. On this famous day, the United States ratified one of the Bill of rights: the famous Second Amendment:

‘A well regulated Militia, being necessary to the security of a free State, the right of the people to keep and bear Arms, shall not be infringed.’

To make it simple, theses 27 words assure the right of an american citizen to own and bear arms in the country. It consequently gives them the ability to rebel against a tyrannical government and to protect themselves without restrictions from the government. Nowadays the gun is really part of the American culture with 32% of the american owning one, which makes the USA the country with the most guns, but also with one of the highest rate of gun homicide in the world. The numerous and deadly mass shootings have risen a debate in the american society: does the Second Amendment protect us or does it threaten us? A true division among the population occurs in the United States. This is where we act! We will investigate correlations between social characteristics and the opinion of an american citizen about gun control.

The dataset used in this project is extracted from Quotebank, a corpus of quotations found in English news articles, published from January 2015 to April 2020. As the study takes interest into gun owning, a lexical field is used to select only relevant quotations, in other words, quotations containing one of the following words : ‘mass shooting’, ‘gun control’, ‘gun violence’, ‘2nd amendment’, ‘second amendment’, ‘gun owning’, ‘firearm’.

The first step of our investigation consists in analyzing the quotations content. To see which lexical field is used when talking about the Second Amendment, the occurrences of each word present in the 69 963 quotations is estimated. The 40 most frequent words (without the stop words) are visible in the word cloud, where the size of the word in the cloud is proportional to its occurrence in the list of quotations. As expected, the most frequent words are the one of the initial lexical field, however other words such as: law, rights, people, state, police and protect are recurrent and correspond well to the debate of the gun permit.

First, a timeline of the quotations is visualized to get an idea of the quotations repartition in time. The timeline takes into account repetition of quotations in several articles and are classified according to the date of publication of the corresponding articles. The frequency-timeline of ‘gunquotes’ is completed with a timeline of mass shootings, that were selected according to their number of deaths. Our guess is that the high peaks in frequency are correlated with shocking events, like mass-shooting which imply more than 10 deaths. As you can see, the number of mass-shooting events in the United States is quite huge and the subject of gun owning in the English newspapers is quite predominant.
The study will focus more precisely on some events, see the graph for the months of October 2017 and February 2018, corresponding to two of the three highest peaks of our timeline. These peaks also correspond to two important mass shootings that occurred in Las Vegas and in a high school in Florida, shocking events that relaunched the debate on the Second Amendment.

Is there a way to predict if an American citizen will support or not the Second Amendment? Is there a direct correlation between characteristic of someone (gender, age, political party) and their opinions? Let’s find out! Our hypothesis here is that there exists a linear correlation between the result of the sentiment analysis and the fact that people are Republican or not, male or not and aged of more than 60 or not. The values of the model parameters associated to each feature are displayed in the graph below:

Positive parameters are associated to the features ‘party_Republican’ and ‘gender_male’. Thus, a male is expected to be more pro-gun than a female, as well as someone belonging to the Republican party compared to someone belonging to the Democratic party.
The parameters associated to the feature ‘age_more60’ is negative, which means that an American citizen, whose age is higher than 60, will tend to be opposed to the gun permit. However, the p value of this coefficient is high, which means that it is not statistically significant, and we can get rid of the age parameter in our model.
Moreover, one can notice that the coefficient associated to the political party is higher than the one of the gender. So, the fact that someone is Republican will have more impact on its opinion than the fact that she may be a female aged of more than 60.
The obtained value of R2 score is 0.098, so the quality of our linear regression is low. It is certainly due to a lack of information. Indeed, we didn’t have access to features such as the place of living, skin color which could have improved the model. Moreover we didn’t have enough quotations of people of less than 40 years old.

Thank you for reading our Datastory. We hope you enjoyed reading it and we wish you all a
Merry Christmas.