Determinants of employment at older ages from the perspective of employers

Factorial survey experiment using hypothetical scenarios

2026-08-31

2026-09-30

The primary outcome is the age up to which the employer would retain the employee under the wage and non-wage job characteristics described in a vignette. The possible values range from 59 to 72 years. Each respondent evaluates four vignettes. We will examine whether and how the experimentally varied employee and job attributes affect the employer’s willingness to retain that employee. For each attribute, we will estimate the average marginal component effect (AMCE) on the retention age. To express the estimated effects of the non-wage attributes in monetary terms, we will additionally calculate wage-equivalent valuations by dividing each attribute coefficient by the coefficient on wages. These ratios indicate the wage change that would have an equivalent effect on the retention age. We will conduct several analyses of treatment-effect heterogeneity by testing interactions of the attributes with each other and interacting attributes with information on the firm, inquired after the vignette study.

We will construct two secondary outcomes. First, we will translate the reported retention age into full-time-equivalent years of continued employment after age 59. This outcome captures the total volume of labor that the employer would retain rather than only the duration of continued employment. Second, we will construct a binary indicator equal to one if the employer would retain the employee at least until the applicable statutory retirement age and zero otherwise.
Furthermore, after the fourth vignette, respondents will be asked to explain their retention decision of the fourth vignette in an open text question. We will code these responses using a hybrid procedure combining human coding and automated text classification. A response may be assigned to multiple themes. Our primary text outcomes will be binary indicators of whether a theme is mentioned. We will regress these indicators on the randomized attributes of the fourth vignette. These analyses will show whether particular vignette attributes affect the considerations that respondents report as salient.

We implement a factorial survey experiment with managers and HR decision-makers in German firms. We use this to study employers’ decisions regarding the continued employment of workers approaching retirement age. In the experiment, respondents evaluate four hypothetical scenarios describing a 59-year-old employee from the most common occupational group at their firm and the conditions under which this employee would prefer to continue working. For each scenario, employers indicate the age up to which they would continue employing the person under the stated conditions, assuming they could make their preferred choice. Each scenario is characterized by seven attributes. Two describe the employee; five attributes describe the conditions under which the employee would prefer to continue working:
• Replaceability of experience: Not easily replaceable; Easily replaceable
• Work performance: Average; Above average
• Gross monthly wage: Anchored to the firm's reported wage level ~U(0.8,1.3)
• Work-time flexibility: Yes; No
• Training opportunity: Yes; No
• Time and performance pressure: Low; High
• Working hours: 75% (with corresponding wage adjustment); 100%

Each respondent evaluates four scenarios, using a single profile design. One scenario is constructed using the respondent's previous answers about conditions in the firm's most common occupational group (wage, working-time flexibility, training opportunities, time
and performance pressure, working hours). If respondents do not answer the voluntary pre-vignette questions, we randomly select the corresponding attribute levels. The two employee attributes are set to “average” and “easily replaceable”. For the other three scenarios, all seven attributes are varied randomly. The gross monthly wage remains anchored to the wage level and is varied randomly around this value as in the scenario described above.
In each scenario, respondents use a slider from 59 to 72 years to indicate the age up to which they would continue employing the person under the described conditions. To study mechanisms, respondents revisit the fourth scenario after completing the main task. Respondents are asked to explain in an open-text response the reasons for their decisions, including any advantages and disadvantages of continuing to employ the person that they considered. We will use the open-ended responses to detect spillovers – that is whether an attribute evokes considerations not explicitly included, e.g., flexible working hours being interpreted as a signal for low intrinsic motivation.
The experiment is embedded in a questionnaire structured in several blocks, including preliminary questions that are feed into the factorial survey experiment as well as postexperimental questions that are used for heterogeneity analyses.

Not available

One of the four vignettes is constructed using respondents’ answers to preliminary questions about the work conditions in their firm's most common occupational group, including gross monthly wage, training opportunities, working-time flexibility, time/performance pressure and working hours. The employee characteristics are predefined. Because the preliminary questions are voluntary, missing attribute levels are also assigned randomly.
In the other three vignettes, all employee and job attributes are randomized using the survey software Tivian. The gross monthly wage is drawn randomly around the firmspecific wage level reported before the vignette experiment. If no wage is reported, the German average full-time wage is used as the reference value.
Employee characteristics are presented in a short text, whereas job characteristics are displayed in a table. The order of these two blocks is fixed, with employee characteristics presented first. Within each block, however, the order of the attributes varies across three randomly assigned respondent groups, allowing us to assess and mitigate potential attribute-order effects. In each scenario, respondents use a slider from 59 to 72 years to indicate the age up to which they would continue employing the person under the described conditions.

Randomization is at the individual choice scenario level (attribute values) and at the respondent level (order in which attributes are presented). The randomized attribute levels in three vignettes are assigned at the respondent-by vignette level. The employee characteristics in one vignette and any attribute levels replacing missing preliminary responses are assigned at the respondent level. The presentation sequence is also randomized at the respondent level.

No

None

Around 800 respondents

No treatment arms