Guiding Managers to Increase Wage Equity: A Field Experiment in a Company

We implement a decision guidance into the software tool used by managers when they determine the annual salary increases (the “merit increase”) for their direct subordinates. The guidance tool provides a proposed merit increase value for each employee. The guidance value is calculated based on this employee's position in wage range, which refers to a pre-specified salary band for each combination of career level and job function in the company. The higher the employee's position in wage range, the lower is the guidance value. If the managers follow the guidance this should lead to a more compressed wage distribution within the pre-specified salary bands. However, managers still have their full discretion and can deviate from the guidance as they wish and are made aware of this. To allow managers to follow the guidance, we also adjust the budgeting approach as compared to previous salary increase cycles such that the sum of guidance values equals the available team budget.

2020-02-05

2020-02-14

Managers' merit increase decisions

Salary increase (in percent). We are interested in studying whether and how the treatments affect the extent to which salary increases depend on the employees' position in the salary range for the respective job.

(a) survey measures;
(b) utilization rate;

(a) Survey data can help us to understand potential side effects of the intervention. In particular, this might include the feeling of accountability of managers during their reward decisions, potential justification behavior of managers as well as the fairness perception of employees.
(b) For a subset of the employees who work in a consulting unit within the company, we receive performance data (utilization rates) which may allow to analyze to what extend the intervention changed the relationship between merit increases and performance.

The intervention affects managers from different countries (US, India, and several European countries) in several units of the firm. We randomly assign managers into one of four treatment groups.

(1) STATUS QUO: No guidance value is provided. The merit process follows the same budgeting approach as in the previous year. No information on the average team budget is provided.

(2) CONTROL: No guidance value is provided. The merit process follows the new budgeting approach that is based on the guidance values (sum of guidance values equals team budget). Managers receive information on the average budget per team member.

(3) GUIDANCE VALUE: A guidance value is provided. The merit process follows the new budgeting approach that is based on the guidance values (sum of guidance values equals team budget). Managers receive information on the average budget per team member.

(4) GUIDANCE RANGE: A guidance range is provided ([-10% of guidance value, +10% of guidance value]). The merit process follows the new budgeting approach that is based on the guidance values (sum of guidance values equals team budget). Managers receive information on the average budget per team member.

Comparing (2) and (3) shows the causal effects of the guidance value.
Comparing (1) and (2) shows the causal effects of the new budgeting approach.
Comparing (3) and (4) shows the causal effects of using the guidance value versus using the guidance range allowing to study differences in manager’s accountability for the merit increase.

Stratified randomization (stratification by country and board area) done in office by a computer using stata.

Randomization takes place on the manager level. Each manager makes several salary increase decisions such that we cluster on the manager level.

Yes

624 managers

We observe merit increase decisions for 9,073 employees.

STATUS QUO: 72 managers
CONTROL: 186 managers
GUIDANCE VALUE: 185 managers
GUIDANCE RANGE: 181 managers

For our power calculation we assume that merit increases in CONTROL will be a weighted avarge of merit increases in the previous year (70%) and the team average value presented to managers (30%). Under this assumption and given our sample size, we can for instance detect effect sizes above 25% compliance with the GUIDANCE VALUE treatment with a power of 80% in a specification where we regress ln(merit_increase) on the position in range interacted with the GUIDANCE VALUE treatment controlling for team size and country.