Introduction
When first exploring different datasets on various platforms, our group began having trouble selecting legitimate data and trends that made logical sense. We started with the “Top Streamed Songs on Spotify” dataset, and instantly, by the names of songs and artists listed and by confirming our suspicions through research, we determined that the dataset was inaccurate. Next, our group navigated to the “Global Income Inequality” dataset. Similarly, by viewing the first few columns, separating high and low income groups and their average income during given years, the average salaries did not align after searching the web for this basic information. Finally, our group came across the “Bike Sales Data of 100k” dataset. At first glance, we came to the understanding that this dataset could be valid, however, with 100,000 rows of data, it was impossible to determine if this dataset was authentic. When thinking about analyzing data, humans are generally better visual learners and can interpret more complex information with visualized graphs as compared to tabular data.
Authentic Data V.S. Synthetic Data
After experiencing difficulty with our dataset selection, we decided to trust our instincts and begin sketching graph ideas for the bike sales dataset. Sketching is good for creating individual interfaces, however, it does not assist in confirming whether a dataset is authentic or synthetic. We chose to use the five design-sheet method as detailed in Roberts, Headland, and Ritsos, 2015 (Sketching designs using the Five Design-Sheet methodology). In short, the five design-sheet method uses five pages of five sections each. You initially explore your topic with rough sketches on the first page, then hone in ideas on pages 2–4 with more realistic drawings and ultimately finalize your ideas with the fifth sheet.
Through a variety of different variables to consider in the dataset, our group decided to unpack trends related to bikes sold by city, most popular bike models, total sales by month, the distribution of bike prices bought, as well as incorporating bike buyers by gender. Stagnant trends from 2020–2024 in all these categories revealed the flawed nature of our dataset. Our new task was to disprove all trends displayed in our graphs. Consequently, we decided to research various events that may have influenced sales to increase or decrease in addition to deriving analytical results about bike consumerism in cities, patterns in bike purchases related to gender, and consumerism based on age groups.
Analytical Sketching
We began our analytical sketching using our five design-sheet method by exploring as many possible visualizations as we could. We were looking to create visualizations that were simple and readable but also allowed users to understand complex trends in the dataset. We were particularly interested in the changes over time of metrics such as average price, locational metrics, gender differences, and even expanding these across the different bike types. We were able to then narrow down these ideas into a few bar charts and line graphs that we felt showed the most promise in our data set. Namely, the quantity of different types of bikes sold, the distribution of bikes across different price points, and thirdly the difference between the number of types of bikes sold to men and women.
Storytelling Sketching
Identical to the analytical sketching, we used our five design-sheet methods to explore possible stories we could tell with our data. We were particularly interested in telling the story of changes in bike sales due to the COVID-19 epidemic. After finding out bike sales increased 64% during the pandemic, according to AP News, we hoped to uncover how this change affected different parts of the biking market. With this in mind, we explored correlation matrix, types of bikes sold over time, number of sales over time, and age distribution based on types of bikes sold. For our second through fourth sheets, we narrowed down our data to be changes in average price over time, difference in gender purchasing bikes, and number of bikes sold over time.
Bike Sales Data
According to our synthetic data displayed in the “Bikes Sold by City” bar graph, New York sold the most bikes compared to all the other major cities in the US. Through our analytical research about bike consumerism in major cities, this was the most accurate interpretation of the dataset as a whole. According to the article, “The Most and Least Bike-Friendly Cities in America”, New York, New York, ranked in the top 12 bike-friendly cities, with none of the opposing cities in the bar graph being listed in the relative domain. However, visually analyzing this graph still raises suspicion due to the uniform level of bikes sold in all the other major cities mentioned.
Similarly, in our “Total Sales by Month” and “Total Quantity of Bikes Sold by San Antonio” line graphs, the trend is relatively constant from 2020–2024, which does not make sense in the context of events like COVID-19 and winter storms. This aspect of our data does reveal some truth, however, displaying a drop in sales during the months of January, indicating low demand for bikes at the start of winter. In the “Total Sales by Month” graph, the line suddenly decreases in September of 2024, which is due to the data being cut off in the beginning of the month. Similarly in the “Total Quantity of Bikes Sold by San Antonio Store (Monthly)” graph, the data is cut short in July of 2024.
With the implications of COVID-19 in mind, we figured that there had to be a sudden increase or decrease in bike sales due to the pandemic. After finding an AP News Article, we confirmed our suspicions, revealing a 64% increase in bike sales in March of 2020. Obviously, our graphs do not even remotely reflect that increase and remain at normal levels.
After our research on the winter storm, Uri hitting Texas and causing power outages, we concluded that our data might have correctly mirrored this event in the line trend, however, there is no definite way for us to determine if this graph corresponds to the actual bike purchasing levels during this time.
Since our other graphs already display constant trends, viewing the additional “Most Popular Bike Models” bar graph was easy to distinguish as synthetic due to the unusual consistency shown across all different bike models. Additionally, the bar graph demonstrates the relationship between male and female purchases with respect to each bike model and it is odd that they all have the same correlation. Our synthetic data shows that females are actually buying more bikes than males and this seemed rather unexpected. To double check the authenticity of this claim, we looked into the Velotric article stating that men are three times more likely to purchase a bike compared to women, which contrasts our dataset.
Our final graph on our “Bike Sales Dashboard” is the “Distribution of Bike Prices” bar graph indicating the amount of bikes bought at each price. Obviously, this data cannot be authentic because people are not going to be buying as many expensive bikes in comparison to more affordable options. This data was especially hard to research, and we could not get a rough estimate of which options are being bought at the highest quantity. However, we can most likely assume that bikes from the lower to mid price range are probably being bought at the highest rate.
At this point, this data is clearly impossible. Using a sample of 100,000 bike sales, there is no way that people ages 18–24 spend the exact same amount of money on bikes as people 35–44 or 65+. According to this study by the U.S. Bureau of Labor Statistics, persons 35–44 years old spend the most on recreation, but the rate decreases before and after this group. This is not reflected at all in our data, since each group spends the exact same amount on bikes. With all of this in mind, it is clear that the data that we previously thought was real is in fact synthetic and computer-generated!
Conclusion
Our exploration of the various datasets ultimately highlighted the critical importance of verifying data authenticity before drawing conclusions. Despite initial optimism with the “Bike Sales Data of 100k” dataset, a deeper investigation revealed inconsistencies and unrealistic trends that pointed to synthetic origins. Our findings demonstrated that the unusual stability in bike sales and pricing, particularly during significant events like the COVID-19 pandemic and winter storm Uri, did not reflect real-world dynamics.
By employing the five design-sheet method, we were able to visualize potential insights, yet our analyses consistently uncovered discrepancies between our visualizations and established knowledge regarding bike consumer behavior. The stark contrasts we observed, such as the improbable equality in spending across age groups and the unexpected gender purchasing patterns, further solidified our conclusion that the dataset was not a reliable source of information.
Leave a Reply