How does runtime influence IMDb ratings, and how is this relationship moderated by genre (Adventure, Action and Comedy) when comparing films released before 2015 to those released after 2015?

Motivation

In today’s world, movie ratings play an integral part when viewers decide which movie to watch. Among the many factors that influence the rating of a movie, runtime stands out as a relevant one when it comes to audience evaluation. Previously, a study by Choudhary et al. (2024) found runtime to be a statistically significant variable, influencing the rating of a movie across different genres, though the magnitude of this effect varied between genres. This raises the question about exploring not only the magnitude, but also the direction of the runtime effect. On the one hand, longer movies allow for more complex storytelling and character development, but on the other hand, decreasing attention spans (Hayes, 2024) among people may actually drive longer movie ratings down.

Previous research suggests a significant relationship between movie genres and their ratings, however often note that isolation of the genre variable is not clearly attainable (Choudhary et al., 2024). While there is no clear consensus in the literature on which genres consistently receive the highest ratings, Matthews (2021) argues that audience expectations are more structured for mainstream genres such as drama, action, and comedy than for niche categories. In these popular genres, ratings often reflect not only film quality but also how well a movie aligns with established genre conventions.

Building on this, the present research focuses specifically on adventure, action, and comedy, as these genres have consistently generated the highest box office revenues in North America over the past decades (Statista, 2025). Adventure and action dominate due to their large-scale productions, international appeal, and substantial commercial success, while comedy plays a complementary role by reflecting cultural preferences and offering insights into audience diversity. Together, these genres allow for a balanced analysis that captures both the financial strength of blockbuster categories and the cultural relevance of humor-driven films.

In addition to examining the relationship between runtime, genre, and IMDb ratings, it is also important to define a clear time interval for the research. The global Netflix subscriber data provides a useful benchmark: in early 2013, Netflix had around 30 million paid subscribers, but by 2015 that number had more than doubled to over 70 million (Netflix, 2025). This rapid acceleration reflects the point at which streaming shifted from an emerging model to a mainstream mode of media consumption, making 2015 a suitable cutoff year. Streaming services mark a natural cutoff point because they changed how runtimes are perceived: before streaming, films were optimized for theatrical showings and ticket sales, while streaming enabled more flexibility in length. As audiences gained on-demand access, tolerance for both shorter and longer runtimes shifted, making streaming adoption a key turning point for analyzing runtime effects on IMDb ratings.

To ensure balanced group sizes around our cutoff, we divided the data into two equal-length periods: 2011–2015 and 2016–2020. This allows us to compare runtime–rating relationships across time while keeping the number of films per period reasonably similar. Accordingly, the research will compare films released between 2011 and 2015 with those released between 2016 and 2020.

Thus, the research question for this project is defined as “How does runtime influence IMDb ratings, and how is this relationship moderated by genre (Adventure, Action and Comedy) when comparing films released between 2011 and 2015 to those released between 2016 and 2020?”.

This research addresses a gap in the currently existing literature of factors that influence audience reception of movies such as IMDb ratings by investigating how the release year (between 2011 and 2015 versus between 2016 and 2020) and genre (Adventure, Action and Comedy) have a moderating effect on the relationship between runtime and IMDb ratings of movies.

Previous research looked at the individual effects of runtime, release year and genre (Horror, Comedy and Action) on IMDb ratings while this research also looks at the relative effects of these variables and how they interact with each other. Further, this research is relevant to different marketing stakeholders in the movie industry such as marketing managers of movie studios, streaming platforms and cinemas. By providing valuable insights on the runtime preferences of audiences these can be used for example by movie studios to create movies across different genres with an optimal runtime and by streaming platforms and cinemas to find the optimal marketing strategy to movies of different runtimes and genres.

Data

We are using two files from the IMDB database:

1. title.basics.tsv.gz (contains unique identifier of the title, the type/format of the title, original title, the release year of a title, TV Series end year, primary runtime of the title, genres)
2. title.ratings.tsv.gz (contains unique identifier of the title, averageRating and number of votes the title has received)

• The final dataset includes 20144 observations. A variable description / operationalisation table is below.

Variable Description and Operationalisation

Variable	Type	Source	Operationalisation (How it is measured/defined)
Rating	Dependent	title.ratings	averageRating (on a scale of 1-10, continuous)
Runtime10	Independent	title.basics	(runtimeMinutes - mean(runtimeMinutes))/10
Genre	Moderator	title.basics	Comedy (reference), Adventure (dummy), Action (dummy)
YearGroup	Moderator	title.basics	2011-2015 (reference), 2016-2020 (dummy)
numVotes	Control	title.rating	Total IMDb votes
logVotes	Control	Derived	log10(numVotes). Interpreted as +1 = 10x more votes
Intercept	Derived	Model	Comedy movie, 2011-2015, mean runtimeMinutes, mean logVotes

To ensure data quality and meaningful analysis, we applied the following filters to the dataset:

1. Runtime filter: We excluded movies with a runtime below 30 minutes.

Rationale: Very short films (e.g., shorts, experimental pieces) are structurally different from feature-length movies, and including them would bias our analysis of how runtime affects ratings.

2. Vote count filter: We excluded movies with fewer than 50 votes.

Rationale: IMDb ratings for movies with very few votes are often unstable and unreliable. Setting a threshold of 50 votes ensures that our dataset contains movies with sufficient audience engagement to provide a more representative measure of audience opinion.

Handling Missing Values (Runtime)

During the data preparation phase, we observed that some movies in the IMDb dataset had missing values for runtimeMinutes. Further exploration revealed that these missing values were not randomly distributed: they tended to occur more frequently in certain genres and time periods. Because runtime is our main independent variable, we could not simply drop these movies, as that might bias our analysis.

To address this, we applied a median imputation strategy grouped by genre and release period. Specifically: Movies were grouped by genre (Comedy, Action, Adventure) and release year group (2011–2015 vs. 2016–2020). Within each group, missing runtimes were replaced with the median runtime of that group. (Median was chosen instead of the mean because it is more robust to outliers.) This approach ensures that our dataset remains complete without artificially inflating runtimes or discarding a large number of observations.

Method

• For this research we will perform a multiple linear regression with interaction terms to find ou t whether the runtime of a movie (continuous) influences its IMDb rating (continuous) and whether a film’s genre (Comedy, Adventure or Action) (categorical) and the release period (2011-2015 vs. 2016-2020) (categorical) influence this relationship. The runtime is the independent variable, the IMDb rating is the dependent variable and the two moderators are Genre (Comedy vs. Adventure vs. Action) and Release Period (2011-2015 vs. 2016-2020). Further, we will include the number of IMDb votes (expressed as a log-scaled variable) as a control variable since ratings based on more votes are usually more stable and reliable (Xie & Lui, 2013). We have chosen for a multiple linear regression with interaction terms as this is the most suitable way to combine these variable types, a continuous independent and dependent variable and two categorical moderators, into one model. This will lead to the following model: Rating = X₀ + β₁·Runtime10 + β₂·Adventure + β₃·Action + β₄·Yeargroup_2016–2020 + β₅·(Runtime10 × Adventure) + β₆·(Runtime10 × Action) + β₇·(Runtime10 × Yeargroup_2016–2020) + β₈·log₁₀(Votes) + ϵ

where:

• β₁ = How the effect of +10 minutes runtime on IMDb ratings changes for Comedy movies released in 2011–2015
• β₂, β₃ = How the effect of +10 minutes runtime on IMDb ratings changes for Adventure and Action movies compared to Comedy movies
• β₄ = How the IMDb ratings change between movies released in _2016–2020 compared to movies released in 2011–2015
• β₅, β₆ = How the effect of +10 minutes runtime on IMDb ratings changes for Adventure and Action movies compared to Comedy movies
• β₇ = How the effect of +10 minutes runtime on IMDb ratings changes for movies released in 2016–2020 compared to 2011–2015
• β₈ = How the IMDb ratings change between movies when the number of votes increases by a factor of 10

Preview of Findings

Runtime influences IMDb ratings only modestly overall, but the effect varies by genre and release period. Adventure films tolerate longer runtimes best, Action films tend to be penalized for them, and movies released in 2016–2020 received slightly lower ratings overall. Audience size (number of votes) is a strong positive predictor of ratings.

The results from this project can provide data-driven insights for the film industry. Studios can optimize runtimes by genre to improve audience satisfaction. Streaming platforms can adjust recommendation algorithms. Cinemas and marketers can use runtime as a strategic element in positioning films to target audiences.

Repository Overview

.
├── .gitignore
├── README.md
├── makefile
└── src/
    ├── analysis/
    │   ├── Makefile
    │   ├── desc_analysis.R
    │   ├── install_packages_for_analysis.R
    │   ├── make_reg_outputs.R
    │   └── regression.R
    ├── data-preparation/
    │   ├── DataExploration.Rmd
    │   ├── Makefile
    │   ├── clean_and_merge.R
    │   ├── download.R
    │   └── install_packages_for_data_preparation.R
    └── reporting/
        ├── Makefile
        └── report.Rmd

Dependencies

This project runs in R (≥ 4.0) and requires Make (included with macOS/Linux; available via Git Bash or Rtools on Windows) and Pandoc Installation guide. All necessary R packages are listed and installed through the setup scripts (install_packages_for_data_preparation.R and install_packages_for_analysis.R) included in the data preparation and analysis folders. All required R packages are installed automatically when you run make.

Running Instructions

1. Clone the repo and open a terminal in the project root (where makefile is).

2. Run the full pipeline by typing "make" in the terminal.

3. Results appear in gen/output/ (intermediates in gen/temp/).

4. Clean generated files by typing "make clean" in the terminal

About

This project is set up as part of the Master's course Data Preparation & Workflow Management at the Department of Marketing, Tilburg University, the Netherlands.

The project is implemented by team 7 members:

• Berkay Ustundag - 2044582
• Eva Kortezova - 2163260
• Berk Yilmaz - 2164297
• Yaz Ucar - 2153214
• Victor van Rossum - 2042008