Administrative Complexity, Take-Up, and Targeting of Social Benefits

2026-06-01

2026-07-13

The primary outcome is an indicator for whether respondent i states that they would apply for the hypothetical social benefit shown in vignette v. The outcome equals one if the respondent answers “yes” to the application question and zero if the respondent answers “no.” Each respondent evaluates four hypothetical benefit programs, so the outcome is measured at the respondent-vignette level.

Stated reasons for non-application from structured question; grouped cost-channel indicators constructed from stated reasons for non-application; willingness-to-pay measures constructed from main-effect average marginal component effect estimates; interactions between administrative requirements and respondent characteristics.

Secondary outcomes are used to interpret application decisions and to study heterogeneity in the effects of administrative requirements.

For respondents who state that they would not apply for at least one hypothetical benefit, we elicit structured reasons for non-application. These reasons refer to the last vignette for which the respondent stated that they would not apply. We will group the stated reasons into broader cost-channel indicators. The main grouped cost channels are time and effort costs, psychological costs, cognitive costs, financial considerations or perceived ineligibility, documentation feasibility, and physical limitations.

We will also construct willingness-to-pay measures from the main-effect AMCE estimates. These measures express the estimated effect of an administrative requirement in monetary terms by relating the AMCE of the requirement to the estimated effect of the randomized benefit amount.

In addition, we will estimate interactions between administrative requirements and respondent characteristics to study whether different requirements deter different types of respondents. The respondent characteristics used for these interaction analyses are household labor income excluding transfers, wealth and liquidity, self-reported ability to get by in the hypothetical scenario, education, employment status, household composition, prior or current benefit receipt from administrative data, labor-market attachment from administrative data, poverty status defined as falling below 60 percent of median income, and the approximated benefit amount that the household would receive under the current SGB-II benchmark. This approximated SGB-II benchmark amount is the household-specific benchmark around which the vignette benefit level is randomized and will be used as a proxy for need under the current institutional rules.

We will use continuous measures as well as grouped versions, such as income or wealth bins, where useful. Because the hypothetical scenario differs by respondents’ equivalized household labor income, key heterogeneity analyses will be conducted both in the pooled sample and, where relevant, separately by hypothetical decision scenario. In the pooled analysis, respondents in the income-loss scenario are treated as having no labor income in the hypothetical situation when constructing income-based heterogeneity measures.

We implement a vignette-based factorial experiment in the IAB Online Panel for Labour Market Research (IAB-OPAL). Respondents evaluate hypothetical minimum-income benefit programs in an online survey. The hypothetical setting describes a reform in which existing social transfers are replaced by a single social benefit for living expenses and housing. Respondents are then shown four hypothetical benefit programs and state for each program whether they would apply.

The hypothetical scenario differs by respondents’ equivalized household labor income. Respondents with equivalized household labor income above a pre-specified threshold are asked to imagine that their household has no labor income and no state transfers. Respondents below the threshold are asked to imagine the same social-benefit reform while their household labor income remains in place. This assignment to scenario conditions is based on respondents’ income and is therefore not randomized. It is included to make the hypothetical benefit decision relevant also for higher-income respondents, who would otherwise have little reason to apply for the benefit irrespective of the administrative requirements.

Before the vignette decisions, respondents answer a question about how long their household could get by in the hypothetical scenario. This measure captures self-reported need in the hypothetical scenario. It also helps respondents apply the hypothetical scenario to their own household before evaluating the benefit profiles.

The vignette attributes are randomized at the respondent-vignette level. The randomized attributes are the monthly benefit amount, application obligations, form length, documentation type, documentation mode, language assistance, and application channel/travel time. The benefit amount is calculated around an SGB-II-style household-specific benchmark based on household composition, housing status, and income in the hypothetical scenario, with random variation added around this benchmark. Administrative requirements vary across vignettes to identify their causal effects on application intentions.

The main outcome is whether the respondent would apply for the hypothetical benefit. We estimate average marginal component effects of each administrative requirement relative to the least burdensome reference category of the same attribute. We also convert these effects into willingness-to-pay measures using the randomized benefit amount. Heterogeneity analyses examine whether administrative requirements deter respondents differently by measures of need and resource scarcity, including household labor income excluding transfers, wealth and liquidity, self-reported ability to get by in the hypothetical scenario, education, employment status, household composition, prior or current benefit receipt, labor-market attachment, poverty status, and the approximated benefit amount the household would receive under the current SGB-II benchmark.

For mechanism analysis, respondents who reject at least one vignette are asked about their reasons for non-application. We use these responses to construct cost-channel indicators, including perceived ineligibility or low financial relevance, documentation feasibility, cognitive costs, psychological costs, time and effort costs, and physical limitations.

Not available

Randomization of attribute values is implemented by computer within the online survey instrument administered by the Institute for Employment Research (IAB). The attributes of each hypothetical benefit program are randomly assigned at the respondent-vignette level. Each respondent evaluates four independently randomized vignette profiles. The randomized attributes include the benefit amount, application obligations, form length, documentation type, documentation mode, language assistance, and application channel/travel time.

The unit of randomization is the respondent-vignette. Each respondent evaluates four hypothetical benefit programs, and the vignette attributes are randomized separately for each vignette shown to a respondent. Thus, treatment variation occurs within respondents across vignettes. Standard errors will be clustered at the respondent level.

No

Approximately 7,000 respondents.

Because the treatment is not clustered, the number of clusters for treatment assignment is not applicable. For inference, observations will be clustered at the respondent level, yielding approximately 7,000 respondent-level clusters. Each respondent evaluates four vignette profiles, for an expected total of approximately 28,000 respondent-vignette observations.

Approximately 28,000 respondent-vignette observations.

The experiment uses a factorial vignette design rather than a small number of mutually exclusive treatment arms. The expected sample is approximately 7,000 respondents, each evaluating four vignette profiles, for approximately 28,000 respondent-vignette observations.

Randomization occurs at the respondent-vignette level across the following attributes: benefit amount, application obligations, form length, documentation type, documentation mode, language assistance, and application channel/travel time. Attribute levels are assigned by computer within the survey instrument. Because this is a factorial design, the expected number of observations per level depends on the number of levels of each attribute. For example, three-level attributes are expected to have approximately one third of respondent-vignette observations per level, while binary attributes are expected to have approximately one half of respondent-vignette observations per level.