Gender Discrimination in (Online) Negotiations

I study gender discrimination in online negotiations on a classified ads platform. I post classified advertisements on the largest platform for such classifieds in Germany and collect offers. I then ask for a discount and randomly sign the replies with a male or a female
name, respectively. In the negotiations, I try to isolate statistical discrimination by using two treatment arms. The first treatment consists of the name only, whereas the second one signals knowledgability with the topic and negotiation experience. I study how often potential partners send a lower second offer as a reaction to the initiation of a negotiation and, if they send one, how large the discount is. Since all communication takes place via chat, I can apply Natural Language Processing (NLP) methods, text analysis methods, and potentially third-party ratings to study potentially different communication with women and men.

2023-12-03

2024-08-31

My main outcome variable is a binary variable equaling one if, as a reaction to the initiated negotiation, the counterpart makes a second offer which is strictly lower than the initial one. Some people do not formulate an exact second offer, but signal that they are willing to discuss prices and engage in a negotiation. I will construct a secondary outcome on willingness to negotiate from this. Another secondary outcome is the height of the discount offered by those platform users who send a second offer. Further secondary outcomes will be generated from the chat protocols.

I post generic job advertisements on the largest online platform for classified advertisements in Germany. These ads usually describe small tasks to be done in someone's home. I will post a painting job for two rooms in an apartment. The ad does not contain information on the gender of the "employer'' posting it and asks potential contractual partners to send an offer. I will apply my treatment to all platform users who send an offer. If users send a first message that does not contain an offer, I ask them to send one (the message asking for an offer does not reveal the "employer's'' gender).

Respondents receive one out of two possible replies: 1) an inquiry for a discount or 2) an inquiry for a discount including a justification for the inquiry. I randomly vary the gender of the name with which I sign the inquiries (i.e., gender is the treatment). Given that potential respondents do not know the gender of the person who posted the ad until they communicate (i.e., until after they made an initial offer), I do not expect variation in initial offers conditional on treatment, as this only occurs at a later stage. Different reactions to negotiation attempts by gender might have two reasons: Discrimination could be taste-based, i.e., negotiation partners have a distaste against negotiations with a particular gender and thus refuse to accept a lower price suggested by, e.g., a woman. The other case in which the observed discrimination is statistical is if potential contractors believe women and men to have different degrees of knowledgeability about the posted task or different negotiation abilities. I tackle this question by sending out reply 2) which contains information about how well the person who posted the ad is informed about, e.g., the complexity of the task, how much it should cost to do the job, and the way negotiations take place on the platform.

Not available

Treatment is assigned via two coin flips. The coin flips determine whether the first message received will be answered with the "plain" or the "justification" treatment and whether the name signing it will be male or female. All other treatments are then automatically set because the second reply will be signed by a name with the other gender. The following two replies after that will be signed with the same gender sequence and use the second treatment. For example: if the coin toss determines "female" and "plain" to be the first reply to be sent, the second, third, and fourth messages will be replied with "male"/"plain", "female"/"justification", and "male"/"justification", respectively.

The list of the cities to post the ad in was also drawn randomly. I took a list of all German cities and selected a random sample of 200 cities (stratified by city size in four categories and region, i.e., West and East Germany). The algorithm selecting the cities ensured that there is always a minimum distance of 125 km between the cities within a week and at least 35 km between cities of the current week and those of the previous week.

individual. All platform users responding to the ad are treated, it is randomized which treatment will be applied to the first message, all following treatments are determined by that first one, see above.

No

200 cities or 10,000 individuals, whichever is reached first.

200 cities or 10,000 individuals, whichever is reached first.

2500 individuals in each treatment.

Main Outcome: Obtaining a counteroffer (binary) I am planning to gather data in 200 cities or for 10,000 observations, whichever is reached first. For the case that I gather fewer than 50 observations per city on average, I additionally report minimum detectable effect sizes for smaller sample sizes. By gender. The pilot yielded a mean likelihood of receiving a counteroffer for female “employers” of 0.45, with a standard deviation of 0.50. With a sample of 10,000 (8000, 6000) observations, the minimum detectable effect size for the gender difference in obtaining a counteroffer is 0.028 (0.031, 0.036). By negotiation treatment. The pilot yielded a mean likelihood of receiving a counteroffer for “employers” using the plain negotiation of 0.41, with a standard deviation of 0.49. With a sample of 10,000 (8000, 6000) observations, the minimum detectable effect size for the gender difference in obtaining a counteroffer is 0.028 (0.031, 0.036). By gender—no negotiation treatment. The pilot yielded a mean likelihood of receiving a counteroffer for female “employers” using the plain negotiation of 0.37, with a standard deviation of 0.49. With a sample of 10,000 (8000, 6000) observations, the minimum detectable effect size for the gender difference in obtaining a counteroffer is 0.028 (0.031, 0.036). By gender—negotiation treatment. The pilot yielded a mean likelihood of receiving a counteroffer for female “employers” using the negotiation justification of 0.52, with a standard deviation of 0.50. With a sample of 10,000 (8000, 6000) observations, the minimum detectable effect size for the gender difference in obtaining a counteroffer is 0.028 (0.031, 0.036).