A Field Experiment on Salary Negotiations

We recruit individuals from the tech sector who are on the job market. They either are looking for a job, or they already have a job but are looking for offers from other companies. Our survey consists of two waves: a baseline survey and a follow-up survey. These surveys collect rich information on the job market history of these employees: whether they have a job already, whether they have interviews with other companies, whether they receive offers, whether they attempted to negotiate those offers, whether the terms of the offers changed, and so on and so forth. The goal of the study is twofold. First, we want to provide descriptive analysis on how negotiations unravel and whether there are differences by gender. The second part of the study is experimental, by cross-randomizing two treatments in the baseline survey.

In the baseline survey, we cross-randomize two treatments:

(1) A nudge to negotiate job offers, including information about the share of people who negotiate and their success rate in increasing salaries conditional on negotiating.

(2) A special deal for Levels.fyi's negotiation coaching services. The service typically costs $650 upfront, and the client can request a refund if the total compensation does not increase by at least $1,200. In our special offer, the subject does not have to pay anything up front (and does not even need to enter their credit card information). The subject pays $550 ex-post, if and only if they negotiated an increase in their offer of at least $1,200.

2023-05-24

2023-08-25

Some of the outcomes of interest are measured in the baseline survey:
1) The intention to negotiate future offers and expected success.
And for the negotiation nudge:
2) The willingness to pay for the coaching services.
3) The take-up of the coaching offer (among those who are given that option).
The rest of the outcomes of interest are measured in the follow-up survey:
4) Whether the subject had interviews and/or received offers.
5) Conditional on receiving an offer, the probability of attempting to negotiate the offer.
6) Conditional on receiving an offer, the final compensation terms.
7) Conditional on receiving an offer, the probability of accepting the offer and the quality of the employment match.

We measure the outcomes in the baseline and follow-up surveys. Additionally, if possible, we will complement the survey data with other datasets: e.g., data from the individual's profiles in LinkedIn, data provided by the coaches at levels.fyi.

We cross randomize our two interventions (information and coaching services).

We included a number of survey questions to be used for heterogeneity analysis, to disentangle underlying causal mechanisms, and also to characterize the marginal individual who chooses to negotiate.

1) Heterogeneity by prior beliefs: The information treatment teaches individuals about the average returns to negotiating an offer. Before the information-provision stage, we elicit the respondent's prior beliefs about the returns to negotiating offers. We expect this prior belief to mediate the effects of the intervention. For example, some subjects will find out that the returns to negotiating an offer are higher than what they thought initially, thus making them more likely to negotiate an offer. On the contrary, if some subjects were over-estimating the returns to negotiations, they may become less likely to negotiate a future offer. The information treatment also teaches individuals about the share of job candidates who negotiate their offers, so we also elicited the respondent's prior belief about that share. Third, the information treatment also teaches individuals that they should not feel awkward about negotiating an offer. Thus, we elicited a prior belief about whether the individual feels uncomfortable negotiating offers, and another question about whether the individual is a social extrovert or introvert.

2) Heterogeneity by gender: we seek to measure whether there is a significant gender difference in the propensity to negotiate an offer and, if so, to document its causes and consequences.

3) Other forms of heterogeneity: we want to identify the characteristics of the marginal individual that is affected by the treatments. With that goal in mind, we included a series of questions in the baseline survey. For example, it is possible that individuals with no prior negotiation experience are the most responsive to the negotiation nudge. To test that hypothesis, the baseline survey includes a battery of questions designed to measure the individual's prior negotiation experience (e.g., whether the respondent negotiated an offer in the past or took a negotiation course).

Randomization is done by a computer.

Randomization occurs at the level of the individual participant (i.e., job candidate).

No

We plan to recruit at least 500 participants.

It is difficult to predict the final sample size, as it depends on a lot of factors that are largely out of our control (e.g., the response rates to the baseline and follow up surveys). We plan to recruit at least 500 participants, but will attempt to recruit more if possible.

Within each treatment arm, each participant has a 50% chance of being assigned to the treatment. So there is a 25% change that the subject is assigned to the information treatment, a 25% chance that the subject is assigned to the coaching service treatment, a 25% chance that the subject is not assigned to any treatments, and a 25% chance that the subject is assigned to both treatments.

The minimum detectable effects are different for different outcomes of interest, and they depend on a number of assumptions that are difficult to predict (e.g., the share of subjects who take-up the coaching services when provided with the special offer). The goal of recruiting at least 500 subjects is informed by the power calculations. For instance, under conservative assumptions, we expect to be able to detect effects of around 5 percentage points in the probability of negotiating.