A study about season variability in beer popularity
Imagine being on vacation near the beach with a group of friends, savoring the moment as the sun appears to dive into the sea during the beautiful sunset.
In your hand, you hold a beer to accompany this serene scene. What kind of beer is it? There is a high chance
that your answer is: ‘a light blonde refreshing beer obviously!’. Summer makes us crave a lighter blond beer,
while winter is a season for brown beers, high on alcohol to warm us up.
But … is everyone thinking like you do ? What do people all over the world think ? Is it a stereotype?
Is this idea verifiable via statistics? That is what we want to explore in our analysis.
Many websites exist where people can give their opinion on the beer they tasted :
RateBeer and BeerAdvocate are two of the main popular ones.
Analyzing the evolution of parameters such as the given grade or the number of ratings could offer
valuable insights into the seasonal trends of various beer styles.
Which website should be considered? Let’s examine the country distribution of the two websites:
As displayed by the map, BeerAdvocate seems to be very popular in the US, while the distribution of the number of ratings in the countries in RateBeer is wider. Therefore, RateBeer data are more representative of the worldwide beer tastes and will be the only data used to try to identify seasonal trends.
RateBeer dataset offers insights on the beer ratings of the website between 2000 and 2016. Data from 01-01-2003 to 01-01-2017 will be considered, due to a lack of ratings before. Among the most rated beer styles on RateBeer, we retrieve 3 beer styles that catches our attention and might be interesting to explore:
These are the 3 main characters: 'The Famous, the Cool, and the Strong'! Considering them, one would particularly
anticipate observing summer trends for Pilsener and winter trends for Belgian Strong Ale.
What could show that these beer styles are seasonal or not?
The RateBeer dataset notably provides the ratings given by the users. The user provides an overall grade
for the tasted beer, but also a grade for the aroma, palate, taste, appearance of the beer, and the mean of
all these grades gives the rating. These 6 characteristics can be grouped together and considered as ‘user specific’.
It can be supposed that a beer receives different ratings depending on when it is consumed in winter or summer:
‘User specific characteristics':
Trends from the evolution of criteria over time are not evident to see for any of the three beer styles. A good idea would be to focus on one characteristic of one beer style: overall score of IPAs, and to use STL (Seasonal-Trend decomposition using LOESS) statistical model. The latter is useful to decompose the plot in 3 subplots: the general trend, the seasonal and the residuals, therefore, potentially revealing seasonal patterns.
Unfortunately, the residuals show the same amplitude as the ‘seasonal’ part of the plots.
For each of the characteristics and each of the beer styles, the same idea was applied: no patterns were discernable.
Well, the users do not feel a
beer differently depending on the season. Should we give up? Not yet.
Normalised number of ratings
Indeed, it could be supposed that Pilsener gets more ratings during the summer and Belgian Strong Ale during the winter. Assuming that people rate during the same period when they drink the beer, variations in the number of ratings could be representative of variations in beer consumption.
In order not to be influenced by the global evolution of ratings on the website, let’s consider the normalized number of ratings for these 3 beer styles (Note: for the following plots, only the evolution from 01-01-2010 to 01-01-2017 was displayed, as it was easier to visualize.):
Remarkably clear trends emerge! Based on the inference that number of ratings mirror consumption, Pilseners appear to be more consumed during the summer, while Belgian Strong Ales are favored during the winter! Interestingly, a recurring trend with IPAs is observed, although it's not as pronounced as with the other two styles. From February to August (roughly spring and summer), there's a rise in consumption, followed by a dip between September and January (autumn and winter). This pattern suggests that even if a beer style isn't expected to be seasonal, its popularity might still be influenced by the time of year. But why? What makes Pilseners good beers for the summer? Why do people gradually neglect IPAs from autumn onwards, until spring returns? One might consider exploring the reasons behind these specific seasonal patterns.
Could these variations be influenced by differences in alcohol content? Let's find out together !
Pilsener, with clear trends in summer are the lighter beers, while Belgian Strong Ale, showing peaks in winter, are stronger. IPA’s alcohol degrees are in the middle, consistent with the more spread and less brutal trend. Are lighter beers more commonly consumed in the summer, while stronger beers are preferred in winter? Let’s split the whole dataset into two categories: light beers and strong beers, and see if the number of ratings is periodic and thus depends on the season. But wait. What is considered as a light or strong beer ? Focusing on the distribution of alcohol degree for all beers:
With this information, the definition of 2 new characters becomes possible:
As before, the objective is to observe the evolution of the normalized number of ratings to identify potential seasonal trends. Additionally, it is essential to evaluate whether there exists a statistical difference between the normalized number of ratings in the summer and winter for both beer categories.
As expected, there is a clear tendency for light beers during the summer and strong beers during the winter. Moreover, the difference of normalized number of ratings is statistically significant !
However, is alcohol content the only determinant of beer choice during a season? Picture this: it's December, and cold outside. You step into a bar to order a beer. Tired from the semester, seeking the perfect brew for comfort.
The alcohol degree is probably not the only thing that will influence the choice of beer. For example, a preference might lean towards a brown, dark beer rather than a pale one! Alternatively, a preference might be for a bitter beer over a sweet, mild one!
Well, let’s see if beers with different colors or bitterness are drunk at different seasons:
- The most common value used (in the US at least) to measure beer’s color is the Standard Reference Method,
or SRM scale for short. Globally, a low SRM value corresponds to a pale beer, while a high SRM value corresponds
to dark beers.
- To measure bitterness, the International Bitterness Unit (IBU) is widely accepted as the standard. Mild beers typically
have a low IBU, while, conversely, bitter beers tend to have a higher IBU.
Unfortunately, it was too difficult to categorize these characteristics due to the different names used for each site.
Therefore, an alternative method was chosen: The beer styles were provided to chatGPT, and it was asked to provide a range for the SRM and IBU. A table with ranges for both values and each beer style was given and the mean of the range was computed to obtain a single value. In the end, tables with estimated values for SRM and IBU were obtained.
As before, for each characteristics we can split our data into two parts:
To determine what qualifies as a pale, dark, mild, or bitter beer, it's imperative to examine the distribution of the data and make decisions on how to split the data accordingly :
Considering these distributions, the following choices are made:
Examining if seasonal trends can be identified:
Very interesting! Combining all the informations, it can be concluded that during the summer, the most rated (and possibly consumed) beers tend to be light, pale, and mild, whereas during the winter, the preference shifts towards strong, dark, and bitter beers. However, reviewing the beer menu in the bar, one might notice that maybe pale beers are often lighter than dark ones. Similarly, strong beers tend to be more bitter compared to lighter ones. Is there any correlation between these factors that could explain the similarity in their tendencies?
Statistically, the 3 characteristics ABV, SRM, and IBU could simply be correlated. Computing the Pearson's correlation coefficient and displaying the scatter plots between these characteristics could be relevant:
Indeed, there is a moderate correlation between these 3 factors. Light beers are more likely to be pale and mild,
while dark beers have more chance to be dark and bitter. We see that bitterness and alcohol degree are the most
correlated characteristics.
In this analysis, the exploration delved into the seasonality of various beer styles and the reasons behind these periodic trends. We're thirsty now!
Now, the goal is to determine which style is the most seasonal, meaning which beer style is truly specific to a season.
To achieve this, a seasonality degree was computed to quantify the seasonality of each beer style for winter and summer.
This degree is first calculated for each chosen year, and the mean for every year provides the overall seasonality degree.
The more likely a beer style is consumed during summer, the higher its seasonality degree will be along the positive y-axis.
The more likely a beer style is consumed during winter, the higher its seasonality degree will be along the negative y-axis.
The seasonality metrics is defined by:
$$100\cdot\frac{std(\text{seasonality}) - std(\text{residuals})}{\text{normalized number of ratings}}\cdot \text{peak month}\{^{-1 \text{ if winter, } +1 \text{ if summer}}_{\text{otherwise 0}}$$
The resulting ranking is as follows:
The analysis yields a curated list of beer styles that shows the seasonal preferences, distinguishing between winter and summer favorites. This insightful data aligns well with previous findings. Notably, the English Strong Ale emerges as THE summer seasonal beer, while the Old Ale stands out as THE premier winter selection.
When looking at the box plot above, a clear distinction can be seen in the standardized ABV of the summer oriented beers
compared with the winter oriented ones.
A winter beer style will tend to a higher ABV than a summer beer style.
The end?
Mmh...
Wait a minute… Since the beginning, the analysis was only conducted considering the time factor. What about geography ?
Indeed, in the analysis, the entire world was considered, even though raters do not come from the same country. However, seasons are not
experienced uniformly in every country, especially considering the reversal of seasons in the southern hemisphere compared to the northern
hemisphere. Nevertheless, nice trends were obtained for certain beer styles. How is this possible? As shown in the RateBeer rating map at the beginning, about 40% of the ratings are
from the US,
and above all, northern hemisphere countries represent more than 95% of the data. This explains the observed trends corresponding to a June
to August summer and a December to February winter.
It would still be interesting to see if the seasonal trends are indeed reversed for the southern hemisphere.
Thus, 3 regions are selected: North America (USA, Canada), Europe (33 countries) and Oceania (Australia, New Zealand).
The choice of Oceania is because they are the only southern hemisphere countries with a relevant number of ratings.
Let's explore this question for a beer style that was identified as being suited for summer
(according to the degree of seasonality degree metric): Pale Lager.
The plot above shows that the Europe and North America number of ratings tends to reach their peaks in the summer, as expected. However, when looking at the Oceania plot, the number of ratings tends to reach their peaks shifted by 6 months later than in the northern hemisphere.
This gives us more confidence about the hypothesis that the Pale Lager is more often consummed during summer, independently of the location.
However, this result also shows that taking into account the location of the rating when analyzing could lead to less biased conclusions.
On the other hand, comparing the normalized number of ratings from the southern hemisphere to the northern hemisphere shows that the effect is really negligeable.
Moreover, in the case of the Pale Lager, there are 25 times less ratings in Oceania compared to Europe and North America. It explains, why the Oceania's seasonal patterns are noisier than the other continents.