Why people cancel Netflix
This was a team survey project on subscriber churn. Our team designed the questionnaire and collected responses through Qualtrics, and I conducted the statistical analysis in R and presented the results. The findings show that the paid-sharing policy has a stronger effect on cancellation than the recommendation algorithm.
Research question
Netflix faces three issues at the same time: complaints about its recommendation feed, new fees for password sharing, and increasing competition. The goal of the study was to find out which of these has the greatest effect on a subscriber's decision to cancel. Each question used a 0 to 10 scale, and the outcome variable was the likelihood of cancelling or downgrading within the next six months.
Drivers of cancellation
I tested each item against cancellation intent using a Pearson correlation. Five of the ten items were significant at the 5% level. Feeling overwhelmed by choice or seeing repetitive titles had no significant effect on its own, while watching less because of that fatigue did.
Subscriber attitudes
I used one-sample t-tests against the scale midpoint of 5. Respondents rated the fairness of the paid-sharing policy at 3.0, well below neutral (p < 0.0001), and reported that it reduced their trust in Netflix. Willingness to switch to another service averaged 7.4. Live sports and events, an area Netflix has been expanding, scored below the midpoint.
Preferred recommendation layout
Respondents ranked five recommendation layouts from best to worst. Model 2 was the clear preference, with a mean rank of 2.0, and 52% of respondents ranked it first. A Friedman test confirmed that the rankings were not random (p < 0.0001), which suggests that the layout is worth testing as a way to reduce recommendation fatigue.
Full model
A multiple regression using all items to predict cancellation intent explains 66% of the variance (adjusted R² of 0.56, n = 44). Five predictors are significant: the influence of paid sharing on the subscriber's plan, perceived fairness of the policy, willingness to switch, perceived quality compared with competitors, and reduced viewing due to fatigue. Paid sharing has the largest coefficient.
Limitations: the study used a convenience sample of 44 people, so the results indicate what should be tested further rather than the views of all subscribers. Because the regression includes ten predictors on 44 observations, I relied on the adjusted R² and the individual correlations rather than on the headline R² alone.
Tools
I used R to clean the Qualtrics export and to run the correlations, t-tests, Friedman test and regression, and Python (matplotlib) for the presentation charts. The charts on this page were redrawn from the same survey data.