The unexpected fact that I found out by interviewing sex by voice changer

ByBill Strain

A service that connects people who aim for technology jobs such as programmers and developers and companies that want to employ those people and provides interview practice "Interviewing.ioHowever, when examining data from interview exercises received by many users, it was found that there was a difference in interview result depending on gender, and in order to investigate the cause, conceal the sex of the applicant of the interview with the voice changer As a result of experiments, surprising facts have come to light.

We built a voice modulation to mask gender in technical interviews. Here's what happened. - interviewing.io blog
http://blog.interviewing.io/we-built-voice-modulation-to-mask-gender-in-technical-interviews-heres-what-happened/

Interviewing.io is simply a practice platform for computer technical interviews. It is a system that allows users to interact with users who want to practice interviews and interviews, and users can receive feedback from companies, sometimes they may be adopted as they are.

Companies listed in interviewing.io vary from large IT companies giants such as Google, Facebook, Twitch to small startups. When interviewing.io meets the conditions of users and companies, interviews are conducted using calls and chat, but Alain Learner, who is in charge of human resources at interviewing.io, is investigating user interview data It turned out that the ratio of proceeding to the next interview is different for men and women users.

To put it more precisely, the number of men who could proceed to the next interview was 1.4 times that of the female, and the average score of the feedback after interview by the company was 3 points on average with 4 points for men, and 2.5 points on average for females It turned out that men learned good results by interview. Regarding this result, Mr. Learner says "I think maybe a woman may be biased by weakness in IT", and he tried experimenting as to what would happen when he interviewed by hiding sex with a voice changer That's it.

For this experiment I created a voice changer application. You can check "Mr. Learner's voice not using voice changer" and "Mr. Learner's voice using voice changer" from the following two movies.

Interview Without Voice Modulation - YouTube


Interview With Voice Modulation - YouTube


When you see the second movie, you can see that Mr. Lerner's voice sounds like a man. Mr. Learner announced the use of the voice changer only to the user who receives the interview, and after paying attention that "Do not say gender during the interview", we conducted the experiment after 19:00 every Tuesday. In the experiment a total of 243 users were interviewed, two thirds were men and one third were women. These 243 people are divided into three groups "No voice changer", "No voice change with voice changer", "Voice changer with pitch change", and so on.

When analyzing the data collected from 243 users, the result opposite to Mr. Lerner 's expectation that "the interview is filtered by sex" came out. Specifically, a man who was interviewed as a female in a voice changer recorded a higher average score than a man who did not use a voice changer, or a woman who was interviewed as a male by a voice changer did not use a voice changer He said that he recorded a lower average score than women.

Mr. Lerner, who had predicted that "the interviewer is affecting some kind of bias, such as women are weak to IT," the interview results between men and women are expected to make a difference, somehow, trying to find the cause Analyze all the interview data obtained in the experiment. Then, you will find one incompatible data. That means that a female user who failed in the interview has withdrawn interviewing.io without having to challenge another interview afterwards. Moreover, the number was about seven times more than male what.

In the following bar graph, the vertical axis is "Percentage of persons who left unsubscribed from interviewing.io after interview", and the horizontal axis shows "Number of failed interviews". About 33.3% of females and 4.7% of men who withdrew from interviewing.io after failing one interview only. The proportion of people who left unsuccessful after failing the interview twice is 14.8% for women and 4.2% for men, and the difference is smaller than for the first interview.


It may be possible that some people have left unsuccessful because they have decided to work for a different company after the first interview and some people have used interviewing.io as a one-time trial, but there is still a difference between men and women It is an interesting result. Perhaps the woman is not good at recovering from failure ... ... Mr. Lerner thought, but the specific cause is unknown. However, it seems that the result that the bias due to the difference in sex of men and women was not in the interview is very positive.

Mr. Learner will continue to conduct the same experiments this time to collect more data and want to explore "the reason why the difference will be born due to the difference in sex by the difference in sex" which was not found in this experiment I will.