A former FiveThirtyEight data journalist, Walt Hickey, found that Fandango tended to round up movie ratings in their system in a more generous way than what is typical. For instance, a 4.1 rating score would get bumped up to 4.5. In other words, Fandango rounded up to the next highest 0.5 increment rather than round to the nearest one.
This project sets out to explore Fandango ratings in more detail, specifically if the ratings have changed since Hickey’s article.
The samples used for the Fandango and other movie ratings datasets were not random. There were concrete parameters that spelled out which movies would be included (e.g., movies with at least 30 reviews).
Hickey’s article came out in 2015, so we want to compare Fandango’s ratings compared to those in 2016. This can be viewed as a pre/post-test.
Null: In a comparison of Fandango movie ratings, there will be no significant difference between those submitted in 2015 and 2016. Alternative: In a comparison of Fandango movie ratings, the 2016 ratings will be, on average, lower than the 2015 ratings.
The kernel density plot above shows us a few key things. First is that in 2015, movies were rated at 4.5 or 5.0 at a higher rate than what it was in 2016. Both years have a negative skew, though 2016 appears to be more normally distributed than 2015.
The research question at hand is whether Hickey’s article caused a change in the Fandango ratings between 2015 and 2016. This kernel density plot provides an initial suggestion that the ratings are slightly different and in the negative direction as hypothesized.
## # A tibble: 5 x 4
## Fandango_Stars Freq Proportion Percentage
## <dbl> <dbl> <dbl> <dbl>
## 1 3 11 0.085 8.5
## 2 3.5 23 0.178 17.8
## 3 4 37 0.287 28.7
## 4 4.5 49 0.38 38
## 5 5 9 0.07 7
## # A tibble: 6 x 4
## fandango Freq Proportion Percentage
## <dbl> <dbl> <dbl> <dbl>
## 1 2.5 6 0.031 3.1
## 2 3 14 0.073 7.3
## 3 3.5 46 0.241 24.1
## 4 4 77 0.403 40.3
## 5 4.5 47 0.246 24.6
## 6 5 1 0.005 0.5
Figure 2 further illustrates that the differences between 2015 and 2016 are marginal. That said, the mean in 2015 is slightly higher than 2016, showing potential alignment with the alternative hypothesis.
The ratings of 2015 and 2016 movies are two different samples, so We run a two sample t-test.
It does appear that the mean rating scores from 2015 (4.085) and 2016 (3.887) are statistically significant (p < 0.01), suggesting that this difference is unlikely due to random chance.
While we can’t be sure that Hickey’s article was the catalyst for the change, it does appear that Fandango did adjust their rating rounding methodology.