Movies Box-office analysis between series and non-series
Project GitHub | LinkedIn: Niki Moritz Hao-Wei Matthew Oren
The skills we demoed here can be learned through taking Data Science with Machine Learning bootcamp with NYC Data Science Academy.
Do you feel the number of movie series increase the last ten years? Nine out of the top ten domestic gross movies in 2019 (until July) are sequels, live-action remakes or shared universe movies. If we look at the number one worldwide gross movies since 1989, there are only three movies that are non-series movies.
While franchise movies tend to bring in million, their production budgets are also incredibly high. That raises the following questions: Do those movies make more profit than other non-series movies? If yes, how much profit does an original film make so that the production companies decide to make a sequel? Furthermore, what makes a movie successes so that audiences are still willing to watch its sequel?
This project is going to answer the questions above by scraping the boxofficemojo.com, the-numbers.com, and rottentomatoes.com, and conducting data visualization, numerical analysis, natural language processing (NLP), and sentiment analysis using Pandas. All Python script, data, and more figures can be found in my Github repository.
I used Scrapy, which is based on Python, to conduct web scraping. Most of the data for this project came from boxofficemojo.com. Fig. 1(a) shows the page of the yearly box office. Because there are too many missing data before 1981, I decide to scrape the top 100 domestic gross movies from 1982 to 2019. Clicking the year tag reveals a table of the top one hundred gross films in that year. Fig. 1(b) is a sample movie webpage. The boxes in Fig. 1(b) are the data collected. The advantage of this website is that it provides a table to indicate if a movie is a movie series.
However, the production budgets are missing for some movies. It is necessary to obtain more data on the budget to compare the profit between series and non-series movies. Therefore, I scraped the top 4000 films ordered by the production budget from the-numbers.com to fill the missing data.
Fig. 2 shows a screenshot of movie reviews from rottentomatoes.com. I collected the reviews of the first film of the movie series where their rotten tomatoes critics score is higher than 60, which is the baseline of a "fresh" movie. The reviews are collected to shed light on the reason why a movie is a success. The "fresh" and "rotten" signs are also gathered to conduct the sentiment analysis.
Figure 3 illustrates the numbers of series movies released each year from 1982 to 2018. This plot clearly shows that the numbers of series movie increase from 10 – 15 in the 80s and 90s to 30 – 45 after 2010. Almost half of the top 100 domestic gross movies are series movies. The audience still goes to movie theaters to watch the film with familiar characters they have come to love.
Figure 4 presents the comparison of the worldwide gross between the series and non-series movies using a combination of histogram and kernel density estimate (KDE) plot. The x-axis is the global box office in log scale ranges from $1 million to $10 billion. The mean for the series and non-series movies are $180 million and $54 million, respectively. The medians are $214 million and $55 million, respectively. The p-value of the two-sample t-test is 3.2e-140. This p-value indicates that the difference in series and non-series movies worldwide gross are statistically different.
However, it might be unfair by comparing the total box office, since the series movies usually have a higher budget than those are not. Figure 5 compares the profit of the series and non-series movies. Because the data of budget on commercial and the profit shared with theater are challenging to acquire, I calculate profit as (gross - budget)/budget. As can be seen in Fig. 5, the franchise movies do make more profit than the non-series movies. The mean profit of the franchise movies is 4.29, whereas it is 2.45 for the non-series movies. The p-value of the two-sample t-test is 1.25e-28, which is much less than 0.05. This result shows that the profit between these two groups has a significant difference.
Figure 6. demonstrates the profit distribution of the first movie of each movie series. The average profit of these movies is 5.33, which is higher than the overall franchise movies. This result is consistent with the impression that the first movie of the series is usually the best. The reason for this could be either that the audience is getting tired with that series or the production budget is higher than that of the first movie.
I apply natural language processing to generate a word cloud, as shown in Fig. 7. This word cloud is made from more than 10000 reviews from the first of the movies series that received positive reviews. This word cloud indicates that the critics are looking for a movie with a great story and characters. They want to have fun while watching the movie. These parameters might be the keys for a film to receive positive reviews so that the film might have a chance to make a considerable profit.
Each critic review at the Rottentomatoes is assigned with a "Fresh" or "Rotten" sign with it (Fig. 2). I conduct sentiment analysis to evaluate if the emotion of the "Fresh" reviews is positive, and the feeling of the "Rotten" reviews is negative. Figure 8(a) shows the polarity of the reviews that have the "Fresh" sign for the different critic score. The data points are more condensed in the region with positive polarity. These results might indicate that the critics might mention some drawback of the movie, but they still express the positive emotions overall.
Figure 8(b) shows the emotion of the reviews that have the "Rotten" sign with the different critics' score. Unlike the "Fresh" reviews, most of the data points of polarity distribute between 0.5 and -0.5. These results show that the critics give relative natural reviews even though they feel the movie is not good enough.
Based on the data acquired from the boxofficemojo.com, there are more and more top 100 domestic gross movies that are either sequels, live-action remake, or shared universe movies.
The data show that if a film makes a 500% profit, it might have a sequel or be remade in the future. That might be the reason why series movies are making more profit than non-series movies. However, we can see that the sequels usually earn less gain than the first movie in the series.
Nevertheless, not every franchise movie is a financial success. The results from the natural language processing show that a fun movie with good storytelling or characters that the audience connects to tends to have positive reviews. Therefore, those movies would have the potential to make a good profit.
Will the numbers of the series movie become even higher in the future? I want to keep tracking the trend of series movies. When does the audience get tired of them, or they never get tired? It will be interesting to analyze the reason why some sequels are still very successful while other sequels are terminated due to the bad review or financial failure.
In this project, there are still some missing production budget data. More comprehensive budget data is needed to make the analysis more accurate.
I only used NLP to analyze the reason why some movies are successful. Therefore, NLP could be used to find the reason why the movies received negative reviews. Finding other vital features that make a movie outstanding would also be worthwhile to investigate.