Go to any discussion group or conference workshop entitled “Women in (Blank)” and you’ll hear high-achieving women discuss how they are experiencing imposter syndrome. Imposter syndrome is feeling like a fraud, that your successes are the result of luck as opposed to talent and that someone is going to find out and expose you.
Perfectionism, fear of failure, and undermining one’s own achievements already run rampant in very technical and objective sciences. However, in my experience, this is especially true for women in STEM and male-dominated fields. Being in a field that is heavily gendered makes you feel “other”-ed and that you might not belong. Coupled together, it is not surprising that many women in data science experience imposter syndrome.
I myself have experienced imposter syndrome along with an almost crippling fear of failure. As my career has progressed, the lessons in failure have helped me overcome my imposter syndrome. Since then I have realized that my voice is not only valid, but worthy of being listened to.
When I started my very first data analyst job, the analysis I was tasked with was determining the results of an AB test. I thought they were interested in the magnitude of change rather than the direction and the magnitude of change.When I found out that my take was not in line with what business objectives were,this mistake is an embarassing one. The only excuse I had is that I was young and inexperienced.
My manager caught my mistake and brought me in to discuss. He explained to me what I did wrong and why it was wrong. I had lost the company money and time and I felt terrible. To me, it felt like I just proved that I was not enough and now everyone knew. I thought I would be fired. But I wasn’t.
It wasn’t the end of the world. Certainly I had more oversight on the next couple of analyses that I did. But this failure taught me a valuable lesson:
To understand the why of a problem rather than the what.
This mistake taught me that an analysis only makes sense based on the decisions that one would like to take from it. That means that the interpretation of the analysis is vitally important. Your findings should be clear and answer a real business need.
This post is part of DSSe, an initiative to elevate the voices of women in data science.
At the same company, the business intelligence team that I was a part of was tasked to determine if a new ad feature would increase our revenue. We had predictive models that worked well and our AB testing process was well-defined. We started our tests with a small but statistically significant population in one country. The results were very promising. We did another AB test on a bigger population in multiple countries. Again, the feature change yielded better monetization metrics in the test group compared to the control group. It had better click-through-rates and consequently better revenue. We looked at retention as well where no significant difference could be found. We wanted to thread the fine-line between showing enough ads to increase monetization while keeping users happy and coming back.We kept the AB test going for a month and the results still showed positive. The conclusion was to release the feature.
However, six months later, global revenue was down by 20%. By that time, we had released many other features. It was a monumental effort to figure out the source of the problem, but we eventually narrowed it down to a reasonable guess.While the feature did increase monetization at first, we ended up losing in retention. People were leaving and never coming back to the application. The ads were too intrusive and the benefits of the product were overshadowed but strangely, users were leaving two months later and not immediately.
Even though our AB tests were correct, we could’ve recommended to wait longer. But how long? We had already looked at a month. All the metrics that we tested were showing the right signs. It’s true that we could’ve kept a very small control group without the feature but again for how long? To catch this, we would’ve needed to keep the tests not only running longer but also we could not implement other features, which delayed us in finding the source since those features increased noise. When the retention pattern is unnoticeable until day 45, you’re asking for no feature implementation for 45 days at the very least for something that, at the time, you had no evidence that would happen. Again, for this feature it was day 45 where we started seeing problems but for other products or other features, there is no way to know how much time is necessary. There are business tradeoffs for certainty that many companies and startups, in particular, can’t afford to make. This is where I learned that being sure that you are right may not be the smart choice.
To this day, I don’t think we made the wrong decision. We made the best decision with the information we had at the time. It ended up being wrong, but it’s being able to identify what is going wrong, how to fix it, and to fix it fast that ends up mattering the most.
During my time, I was a good analyst. My work was of high quality and I was a valuable member of the team. Despite the two failures I outlined above, I had many more accomplishments and successes. So when evaluations were up, I believed I should get a raise.
When I did get a raise, it was less than what I had envisioned.In that moment, between hearing the number and signing the letter, I forgot what I brought to the table. When you internalize the feeling of being an imposter, it is hard to stand up for yourself. I didn’t fight for that raise and my biggest regret is not the money.. My biggest regret was the fact that I didn’t defend myself by celebrating my accomplishments and making by bosses take notice. My inability to acknowledge my successes for myself made it so that I could not justify my worth to someone else. I have never made that mistake again. Now, I make note of every win that I have and if I am not being compensated, I fight for it.
Experience in failure will help you put your feelings of inadequacies in perspective. Once you view failing as a way in which you can learn more about yourself or gain more experience, it becomes less scary to take risks. Being afraid to fail will do you no good in your career and in your life. You’ll miss opportunities and you’ll let yourself down. I encourage every budding data scientist to fail, in big ways and in small. You’ll learn far more about yourself. And it is this confidence in who you are that stops the nagging voice in your head that tells you that you are unworthy.
This post is part of DSSe, an initiative to elevate the voices of women in data science.