Ranking companies in Bangalore using Glassdoor ratings and Confidence Intervals
Why job-search is horrific
Job-search involves searching through Linkedin and other such websites to figure out where to apply. This is often frustrating, because with new companies cropping up left, right, and center, it is often difficult to judge how “good” a company is.
Enter Glassdoor, the de facto standard for checking out a company’s reputation. A company that treats its employees right may trump other factors like the type of work and pay. As such, checking out the Glassdoor rating of a company before applying for them is of paramount importance.
However, as mentioned earlier, with new companies coming up all the time, sometimes you don’t have enough information to judge that firm. What do you do when a startup has a single but high rating? How do you compare that to another company with a not-so-high average rating but a large number of ratings?
We can use a component of statistics called confidence intervals, which help us figure out, with a 95% confidence, in what interval the “actual” rating of a company would lie, given its average rating
r on a sample of
The range is given by the following formula:
t denotes the t-distribution value at 95% confidence for
n-1 degrees of freedom and σ is the standard deviation of ratings in the sample.
Now we have a better way to judge a company, even if it has fewer ratings. We have a tool to compare startups to big companies.
So let's get on with it. Some web scraping using
BeautifulSoup will help us get a list of companies in Bangalore (from this Wikipedia article) and the ratings of each company (from the Glassdoor review page). Confidence intervals will help us get the lower bound for the “actual” rating of a company.
The Wikipedia article has 133 companies, out of which 104 were considered.
The final ranks of the companies with their lower bound rating value:
- Pratilipi (4.8)
- Glance Networks (4.6)
- Zapr Media Labs (4.5)
- Zerodha (4.2)
- Hasura (4.2)
- Flipkart (4.1)
- Infosys (4.1)
- Wipro (4.0)
- MyGate (4.0)
- Intuit India (3.9)
Some well-known companies missing in the top 10 are Swiggy (3.9), Myntra (3.7), PhonePe (3.6), Commonfloor (3.5), and InMobi (3.5).
Note regarding scraping
I’d like to thank Glassdoor for providing the data. In order to ensure that I don’t overload their servers with my requests, I scraped a maximum of 1000 ratings for each company (only the rating value, not the review text). This may have biased the results. In addition, only English reviews are considered.
Note regarding companies considered
The companies considered are those mentioned in the Wikipedia article. Since it is not an exhaustive list, several firms may be missing.
Note regarding sampling and Confidence Intervals
There could be further bias in the results because the sample of ratings taken for a company is not a Simple Random Sample (SRS), but rather a convenience sample since every employee of a company doesn’t have the same chance of writing a Glassdoor review. The results could be improved by grouping the reviews by employee type and getting a Random Multi-Stage Cluster sample (by considering a random subset of employee types) or a Stratified Random Sample (by taking a random subset of employees from each employee type).