Do Drivers Really Hate UBER?
UBER, from the very start, has always been in the spotlight of controversies. One of the biggest contributors to this has been their rapid and aggressive expansion plans. With the IPO about to come on May 9th, 2019 the company yet faces another challenge of drivers going on a strike. With it’s biggest competitor Lyft always holding a more driver-friendly image, do drivers really hate UBER?
From a literature perspective, you will find an abundance of literature talking about the customer side of these gig economies with changing dynamics of the system. While it does affect the demand side of this economy, these policies significantly impact the supply side of the network (drivers) . To dig deeper into this domain, we studied the effects of the change in dynamics of the gig economies on the sentiments of the supply side of the economy.
Where did we get the data?
There is no readily available data from UBER which would be useful to quantify or formulate a substantial hypothesis. However, there are extensive amounts of unstructured data sources available out there in terms of driver reviews and comments about the company. We collected the data by scraping the Uber People website and Indeed reviews from the drivers. By performing Natural Language Processing (NLP) and building a sentiment classifier using Machine Learning techniques we generated sentiment labels for the data gathered using Uber People website. The processes and models build for this purpose are beyond the scope of this blog, but I would soon be covering the same in a follow-up blog.
We further use advanced analytics techniques to study the sentiments and correlate them with the changes in the dynamics of UBER policies and mobile application features. One of the biggest concerns or a pain-point for the drivers was the lack of In-App tipping services. We particularly study the impact of the introduction of In-App tipping service on the sentiments of drivers towards the company.
The basic assumptions we make about the data are as follows:
- The users participating each have just one profile and are genuine drivers
- The sample of data taken is representative of the all UBER driver data
- Even though we observe more negative sentiments than positive sentiments the relative sentiment is representative of the trend
Does the Introduction of In-App tipping services change the game for UBER?
Using the scraped data and predicted sentiments we plot a 5 monthly moving average of the percentage of positive reviews for reviews related to tips, we come up with an interesting trend.
The first evident feature that we notice is a cyclic trend, where the sentiments around tips increase in fall but start to decrease during spring. While the reason for this is still being analyzed we sure do have some other interesting results to be shared.
Build better voice apps. Get more articles & interviews from voice technology experts at voicetechpodcast.com
Initial fall in sentiment score!
We perform regression discontinuity to study the average treatment effect of the introduction of In-App tipping on driver sentiments. The equations of the two lines are as follows:
- Before In-App tipping: Percentage positive reviews = 0.3*Time + 26.99
- After In-App tipping: Percentage positive reviews = 0.364*Time + 23.10
Thus, we see a 21.3% increase in the slope after the introduction of In-App tipping.
Interestingly we see an initial dip in the sentiment score before it rises again. Further analysis of this phenomenon revealed an interesting story behind the picture. By performing text mining we found that the users were happy with the introduction on In-App tipping but were unhappy about the fact that the percentage of tips were low.
Looking at just these findings does not give us a complete picture of what is going on. In the period that the In-App tipping was introduced, UBER was undergoing a major makeover in terms of firing their CEO and their 180 days of change initiative. To get a complete picture of the results we are trying to interpret, we compared the sentiments related to Tips topic with the general sentiment of the UBER drivers.
We see that the general sentiment trend is on a constant decrease while the trend of positive reviews is on a constant increase. To further back our claim, we also study the system by introducing dummy variables in the system and performing regression on the system again.
We chose the following events as significant events:
- CEO Stepping Down
- In-App Tipping
- Introduction of UBER Eats
- Class Action lawsuit in which UBER had to pay their drivers a significant amount of money for not having reimbursed them enough for their services
We chose these events because in the same time frame as the introduction of In-App tipping which may have a significant impact on the increase in the sentiment observed in the In-App tipping data. The observation points towards In-App tipping and the CEO stepping down as being the most significant predictors for the increase in the driver sentiments. To further investigate this we plot the frequency of the words to ensure what people are talking about.
The above word cloud gives us a clear picture that the reviews we consider for the analysis are exclusively related to Tips and not anything else.
What is so special about this analysis?
This analysis lies at the intersection of Machine Learning and Econometric models to describe the underlying dynamics of the gig economy systems. It gives a quick way to analyze the impact of policy changes on user sentiments. This analysis can easily be translated to any kind of industry and application.
We will also look at more advanced techniques and in-depth analysis for this data.
P.S: The code for this analysis is available at my GitHub repository.
