Film Success Analysis

Summary

• The project brings in data from The Movie Database API, cleans the data, and analyzes some of the factors that make films successful
• The following conclusions can be drawn from the analysis:
  • On average, profitable films have higher ratings than unprofitable films
  • On average, profitable films have higher budgets than unprofitable films
  • Film budget is positively correlated with level of critical acclaim

How to Reproduce the Analysis

1. Enter The Movie Database API key into the api_keys.py file
2. Run the data_cleanup.ipynb file to pull data from the API, clean the data, and export the cleaned data into a new CSV file
3. Run the analysis.ipynb file to reproduce the analysis

Data Extraction and Cleaning

• A list of film IDs found on Kaggle is used to pull fresh data from The Movie Database API
• Data is cleaned in the following ways:
  • Date columns are converted to datetime format
  • The dataset contains a few rows with N/A'd metrics. These rows only make up a small percentage of the data so all rows with N/As are removed
  • Movies shorter than one hour and longer than four hours generally represent films that are too different than those in the rest of the dataset. Therefore, films shorter than one hour and longer than four hours are removed
  • Films with budgets less than $100,000 are either very old or are not films that many people commonly see and are thus too different from films in the rest of the dataset. Therefore, films with budgets less than $100,000 are removed
  • All films that have 0 reviews are removed from the analysis
  • Films released before the internet age generally have much fewer reviews than films released during the internet age. To prevent there from being inherent bias against older films, only movies released after 2005 are included in this analysis

Calculated Metrics

• A film's "profit" is calculated as: "revenue" - "budget"
  • "Profitable" films are films with "profit" greater than $0 and "not profitable" films are films with "profit" less than or equal to $0
• A film's "critical acclaim" is calculated as: ("average film rating score" * "quantity of reviews") / ("days since film's release")

Analysis

Question 1: Do profitable movies tend to have higher or lower average review scores (on a scale of 1 to 10) than unprofitable movies?

• A two-tailed two-sample t-test yields statistically significant results at a 95% level of significance (p-value < 0.025). On average, profitable films have higher ratings than unprofitable films
• A Welch's test confirms the results found in the two-tailed two-sample t-test

Question 2: Do profitable movies tend to have higher or lower budgets than unprofitable movies?

• A two-tailed two-sample t-test yields statistically significant results at a 95% level of significance (p-value < 0.025). On average, profitable films have higher budgets than unprofitable films
• A Welch's test confirms the results found in the two-tailed two-sample t-test

Question 3: Does film budget impact level of critical acclaim?

• A single variable regression shows that film budget is positively correlated with level of critical acclaim. This means that films with larger budgets tend to have more critical acclaim

Analysis Weaknesses and Opportunities for Further Research

• The initial dataset is created from The Movie Database API using a list of film IDs from a Kaggle dataset. It is unclear how this list of IDs in the Kaggle dataset was initially gathered so it may not be a random sample
• The analysis studies the difference between films with positive profit versus films with zero or negative profit. This could be misleading because films with close to zero profit are included in both the "profitable" and "not profitable" groups
• The analysis covers films with a broad range of budgets ($100K and up). It could be interesting to explore differences in budget ranges because a film with a $100K budget is likely very different than a film with a $300M budget
• The single variable regression only explains about 25% of the variation in the dependent variable (log critical acclaim). Adding more variables to the model could improve its explanatory power