Case selection and data collection

We investigate how voters respond to electoral promises that benefit specific social groups compared to those that benefit the population at large. Germany provides a relevant and broadly representative case for such an inquiry. As a parliamentary democracy with a proportional electoral system, multi-party competition, and coalition governments, it shares key features with many other advanced democracies. German parties regularly issue targeted appeals to specific constituencies alongside broad-based promises. Moreover, moderate levels of political polarization and the continued relevance of programmatic competition make Germany a suitable setting for studying the effects of targeted pledges in a context that is neither highly polarized nor entirely candidate-centered. These characteristics allow insights from Germany to inform our understanding of targeted campaign appeals in other multi-party systems.

Our analysis draws on a large-scale experiment embedded in an original online survey conducted between 25 January and 8 February 2024. The sample of 3,500 respondents was drawn by the survey company Bilendi and is quota-representative of the German electorate in terms of age, gender, region, and education.Footnote ³ Respondents were recruited from Bilendi’s panel and compensated based on the length of the questionnaire. All responses were collected from eligible German citizens aged 18 or older.

Experimental design

We implemented a randomized vignette experiment to examine how voters prospectively evaluate group-targeted versus broad-based campaign pledges. Respondents were randomly assigned to one of four conditions that differ in the group addressed by a fictional electoral promise.

The experiment builds on existing research on campaign targeting in the US context (Hersh and Schaffner Reference Hersh and Schaffner2013), but uses policy pledges – rather than rhetorical appeals – presented as stylized campaign posters mirroring real electoral materials, thereby enhancing external validity.

All respondents first read a short introductory text setting the scene for a hypothetical federal election. They were told that several parties (A, B, C, and D) were competitive, and that the next screen would show a campaign poster from Party A. Respondents were then randomly shown one of four posters presenting Party A’s core campaign promise. The posters varied only in the beneficiary of the pledge: either a broad-based pledge to improve healthcare (control condition) or a group-targeted pledge benefiting (1) rural residents, (2) parents, or (3) pensioners (treatment conditions Ta–Tc).Footnote ⁴ The pledges were made as similar as possible by adhering to a common format focused on increasing public spending, in order to avoid biases linked to reactions to specific policy instruments. Since different social groups are affected by different types of policies, the pledges necessarily differ in their policy domain. It would make little sense to promise pensions to families or child benefits to rural residents. Our aim was therefore not to construct artificial, interchangeable pledges, but to approximate how realistic policy appeals operate in practice while keeping their format, framing, and fiscal logic consistent. All four pledges involve higher public expenditure rather than tax cuts, regulatory changes, or symbolic recognitions, ensuring a comparable distributive structure across treatments. To address concerns that respondents might react primarily to policy domains rather than to group targeting, we conduct robustness analyses showing that our results hold when accounting for differences in issue importance across policy areas.

We selected the three target groups based on two main criteria. First, their size ensures sufficient statistical power to detect group membership and group identity effects. Second, these groups tend to elicit varied voter responses without triggering strong partisan associations. In the German context, they are perceived as differently deserving, but not in ways that are sharply polarized along ideological lines. Deiss-Helbig, Guinaudeau, and Matthieß (Reference Deiss-Helbig, Guinaudeau and Matthieß2026) show that the standard deviation of perceived deservingness for our groups is comparable to that of other frequently studied groups, such as the poor, unemployed, employers, and women. However, targeting groups like women or the poor can more readily activate partisan cues, as such pledges often align with specific parties (left-leaning parties in these cases). By contrast, our selected groups – parents, rural residents, and pensioners – are routinely addressed by parties across the ideological spectrum and not strongly associated with any of them (Deiss-Helbig, Guinaudeau and Matthieß et al. Reference Deiss-Helbig, Guinaudeau, Matthieß and Rentrop2024). This makes them comparatively uncontroversial cases where parties could, in principle, hope to mobilize support without incurring strong negative reactions.

For the broad-based pledge, we selected a promise to allocate more resources to improve the healthcare system. In Germany, this issue is not unusually salient – healthcare has not ranked among the most prominent political concerns in recent decades.Footnote ⁵ However, it has universal relevance, as healthcare affects all individuals when they fall ill. Our manipulation checks (see below) confirm that this pledge is perceived as universalistic, with no single group seen as its exclusive beneficiary.

Figure 1 displays the campaign materials shown to respondents. The posters were shown in German; the English translations below reflect the original wording.

Figure 1.

Posters with election pledges.

Note: Posters are shown to respondents in German. English translations are provided in the main text.

The experiment varies the target group addressed by otherwise comparable spending pledges, allowing us to assess how voters respond to realistic forms of group-targeting while keeping the general format and distributive logic as consistent as possible.

To assess whether respondents perceived the treatment as intended, we included two manipulation checks, presented in the Appendix. First, immediately after viewing the campaign poster, respondents answered an open-ended question: ‘Who do you think primarily benefits from Party A’s promise?’ Figure A2 shows that most respondents correctly identified the target group, particularly for the child care and pension pledges. Misidentifications were more frequent in the rural and control conditions but remained limited overall.

Second, respondents indicated how personally affected they felt by the pledge, using a 7-point scale from 1 (‘very negatively affected’) through 4 (‘not affected’) to 7 (‘very positively affected’). Table A4 summarizes the distribution of responses by condition. Consistent with our expectations, the control pledge on healthcare is perceived as broadly beneficial: three-quarters of respondents report feeling positively affected. As Figures A4–A6 in the Appendix show, this pattern holds across subgroups defined by health status and age as well as the groups considered in our study (i.e., parents, rural residents, and pensioners). Associations between perceived affectedness and respondents’ self-reported health condition or age (measured based on Spearman’s rank correlation coefficients) are statistically significant but weak, suggesting that these factors account for little variation in reactions to the pledge. Similarly, non-parametric comparisons between parents and non-parents, pensioners and non-pensioners, and rural and non-rural residents reveal no significant differences in perceived affectedness. This reinforces that the healthcare pledge is broadly perceived as beneficial, and does not subjectively overaffect certain segments of people. In contrast, and as expected, the targeted pledges elicit more polarized reactions. Large proportions of respondents, especially in the parents’ and pensioners’ conditions, report feeling unaffected or even negatively affected. Figure A3 further shows how perceived affectedness varied by whether respondents were part of the intended target group. As expected, those in the target group were more likely to report stronger personal benefit. However, the distributions also indicate that a non-negligible share of untargeted respondents perceived the pledge as relevant to them, underscoring the blurred boundaries of self-identification.

To operationalize our outcome variable, we use respondents’ propensity to vote (PTV) for the pledging party, measured on a 7-point scale ranging from 1 (‘very unlikely’) to 7 (‘very likely’), based on the question: ‘On a scale from 1 to 7, how likely are you to vote for this party?’

We test the impact of the hypothesized moderating variables – group membership, group identification, and perceived deservingness of the targeted group – using several items administered at the very beginning of the questionnaire, well before the experiment, to minimize potential priming effects (Heath, Smith, Gilby et al. Reference Heath, Smith, Gilby and Hoolahan2015). The item batteries for group identification and perceived deservingness included eleven additional groups, and item order was randomized. Group membership (H1a, H1b) is measured as a binary variable based on respondents’ socio-demographic characteristics, identifying membership in the group of rural residents, parents, and pensioners. The exact wording of the questions used to identify these groups is provided in Table A2 in the Appendix.

Group identification (H2a, H2b) is measured by asking respondents the extent to which they feel part of various social groups – among others pensioners, rural residents, and parents. Specifically, we ask: ‘We all have different ideas about which social groups make up our society, and we can each belong to several such groups. To what extent do you feel you belong—or do not belong—to the following groups?’ (response options: not at all belonging; rather not belonging; rather belonging; very much belonging; don’t know). For the analysis, we recode responses into a binary indicator. This item captures the self-categorization dimension of social identity – the minimal condition for perceiving a group appeal as personally relevant – rather than affective attachment or pride.Footnote ⁶

Finally, perceived deservingness (H3a, H3b) is assessed by asking whether respondents believe the government should take the needs of various social groups (inter alia pensioners, rural residents, and parents) more, less, or equally into account.Footnote ⁷ Responses were given on an eleven-point scale ranging from −5 = ‘much less’ to +5 = ‘much more’. For the analyses, we recode them into a binary variable distinguishing whether respondents deemed each group ‘deserving’ (scores $\! \gt 0$ ) or ‘not deserving’ (scores $\! \le 0$ ) on the original −5 to +5 scale. While deservingness is often measured through items capturing concern for the living conditions of specific groups (e.g., van Oorschot Reference van Oorschot2000), our operationalization focuses more closely on the priority or entitlement dimension – namely, whether respondents believe a group should receive greater or lesser governmental consideration than it currently does. This formulation acknowledges that people may feel concerned about certain groups without necessarily thinking that the government ought to do more for them. In this respect, our measure differs from indicators of group affinity (Robison and Moskowitz Reference Robison and Moskowitz2019; Bornschier, Häusermann, Zollinger et al. Reference Bornschier, Häusermann, Zollinger and Colombo2021) or concrete policy support (Kootstra Reference Kootstra2016; Bell Reference Bell2021), which capture related but distinct dimensions. It thus provides a direct and interpretable gauge of whether a group is viewed as meriting additional policy attention, aligning with the ‘should receive more support’ notion in the deservingness literature (Meuleman, Roosma, and van Oorschot Reference Meuleman, Roosma, van Oorschot, van Oorschot, Roosma, Meuleman and Reeskens2017).

Our experimental design deliberately balances realism and control to examine the effects of group-targeted versus universalistic campaign promises. By using mock campaign posters, we simulate realistic campaign exposure – enhancing external validity – while maintaining substantial control over pledge format – crucial for internal validity. An alternative design using real party pledges from a recent election might have increased external validity further, but at the cost of internal validity. Different groups are offered policies that vary in nature, design, and ambition: using these kinds of pledges would not have allowed us to vary the target group while holding other aspects of the pledges as constant as possible. Actual campaign pledges differ not only in their target audience but also in perceived substantive importance, public salience, budgetary implications, policy design, and media visibility. This makes it difficult to isolate the effects of group targeting from confounding reactions to these other attributes and, additionally, introduces the risk of ‘pretreatment’ (Druckman and Leeper Reference Druckman and Leeper2012), particularly for well-known pledges that respondents may have encountered prior to the study. Using pledges from real parties would also introduce potential confounds related to party identification or voters’ associations of specific parties with certain social groups, which are mitigated in our design by the use of a fictional ‘Party A’. Our experimental setting provides a controlled context for detecting positive reactions to group-targeted pledges: by removing partisan cues and external noise, it reduces potential confounds while enhancing the clarity of the stimuli by presenting the appeals in a clear, tangible form. We return to the implications of this methodological choice in the discussion section.

Main analyses

Figure 2 presents the average effects of group-targeted pledges on respondents’ vote intention, compared to the broad-based control condition. The upper panel shows the mean PTV for Party A in each treatment group, on a 7-point scale. Respondents exposed to the broad-based healthcare pledge report the highest level of vote intention (mean = 4.65), followed by those in the pensioners (4.28) and rural (4.11) treatment conditions. Support is notably lower in the parent condition (3.80). The lower panel displays average treatment effects (ATEs) relative to the control group, based on a linear model with the broad-based condition as reference. All three targeted conditions yield negative coefficients, indicating lower vote intention compared to the universalistic pledge. The negative effect is largest for the parent condition (b = −0.85), followed by rural (b = −0.55) and pensioners (b = −0.37). The confidence intervals suggest that these differences are statistically significant at conventional levels. These findings provide no evidence that group-targeted promises yield greater overall support for the pledging party than a broad-based appeal. On the contrary, targeted pledges appear to reduce average vote intention, even for target groups widely viewed as deserving, such as parents or pensioners. This motivates closer examination of whether these aggregate effects conceal heterogeneity across voters with different relationships to the targeted group.

Figure 2.

Treatment effects of targeting.

Note: The upper panel shows raw sample means (dots) for each group. Colored bars represent 90% (thick line) and 95% (thin line) confidence intervals. The lower panel displays average treatment effects from a linear model with the control group as reference. Coefficients indicate the average difference in propensity to vote compared to the control condition (broad-based pledge).

Figure 3 presents the test of our preregistered hypotheses (H1a–H3b), which concern average reactions to group-targeted versus broad-based pledges depending on respondents’ relationship to the targeted group. We estimate separate linear models to test whether support for the pledge-making party differs between targeted and broad-based conditions on average (in gray), and depending on objective group membership (purple), subjective group identification (blue), and perceived deservingness of the targeted group (pink). None of the three moderators significantly reverses the negative ATE. While the negative effects are attenuated among respondents aligned with the group, the coefficients for group members (b = −0.18), identifiers (b = −0.32), and those with positive deservingness assessments (b = −0.34) remain negative compared to the broad-based pledge. In other words, contrary to hypotheses H1a-H3a, targeted pledges do not lead to increased support even among those whom parties may hope to mobilize. This finding should be interpreted with some caution. The broad-based healthcare pledge may serve as an especially appealing baseline, which could limit the room for group-specific promises to generate additional electoral gains – even among likely beneficiaries. We explore this potential interpretation in robustness checks accounting for issue importance (see below).

Figure 3.

Treatment effect of targeting, by subgroup.

Note: The figure displays average treatment effects from linear models comparing each subgroup exposed to a targeted pledge (rural, parent, or pensioner) to the control group (broad-based healthcare pledge). Subgroups are defined by group membership (purple), group identity (blue), or perceived deservingness (pink). For each concept, estimates are shown for respondents with the attribute (1, represented by a square) and without it (0, represented by a diamond). ‘Full sample’ (represented by a circle) represents the average treatment effect measured for all respondents exposed to a targeted pledge, regardless of subgroup status. Points indicate estimated effect sizes; colored bars represent 90% (thick) and 95% (thin) confidence intervals.

By contrast, hypotheses H1b–H3b find empirical support: targeted pledges significantly reduce the party’s appeal among voters who are not part of the targeted group, who do not identify with it, or who consider it undeserving. These voters report substantially lower vote intention for the pledge-making party. The estimated effects are substantial and statistically significant: b = −0.81 for nonmembers, b = −0.86 for non-identifiers, and b = −0.91 for respondents with low deservingness ratings. These results reveal an asymmetry: while the potential electoral benefits of targeting are limited compared to the broad-based condition – even among intended beneficiaries – the political costs among outgroup voters are more pronounced. These findings underscore the potential risks of targeting. When voters feel excluded or question the targeted group’s deservingness, targeted appeals may backfire.

Discussion: potential limitations, robustness, and scope conditions

A full understanding of our results requires reflection on the methodological choices that underlie them. We first examine a key concern – whether our treatments differ in substance independently of the group-targeting dimension – before turning to broader methodological considerations.

A potential limitation of our design derives from the fact that different social groups tend to be addressed through distinct policy schemes reflecting their specific circumstances and needs. Even policies targeting the same group may differ in ways that shape voters’ reactions. Different instruments – such as additional spending, tax cuts, regulatory measures, the extension of rights, or concrete outcomes – may trigger distinct responses, for instance, because of an aversion to spending or regulations as such. Likewise, policies may differ in perceived importance or normative appeal, irrespective of the group affected. To minimize such confounds while keeping the treatments realistic, all pledges take the form of additional public spending, avoiding differences in the instrument used to deliver benefits.

Still, our interpretation assumes that the observed gaps in vote intentions are driven primarily by the targeting focus, and not by reactions to the policy domains themselves. To examine this possibility, we assess whether the main results hold when accounting for differences in perceived issue importance. Respondents were asked to rate the importance of several policy areas (including those featured in our treatments) on a five-point scale ranging from ‘not important at all’ to ‘very important’. As shown in Figure A7, the healthcare system ranks as the most important issue overall, followed by old-age provision, family support, and rural infrastructure. This raises a potential concern that the higher support for the healthcare pledge observed in our main analyses partly reflects the general importance of that domain, rather than the absence of group targeting per se.

We therefore re-estimate treatment effects separately for two strata: respondents who rate the targeted domain as at least as important as healthcare, and respondents who rate it lower (Figures A8 and A9). These exploratory analyses do not suggest that relative issue importance alone accounts for the main findings. If domain preferences were decisive, targeted pledges should outperform the universalistic healthcare pledge among respondents who prioritize the corresponding issue (and the effect of targeting should become robustly positive). This pattern does not emerge. Across groups, treatment effects remain mostly negative even when the targeted issue is rated as at least as important as healthcare. The results show that higher importance of the targeted domain can attenuate the disadvantage of targeted pledges, but do not provide consistent evidence that this disadvantage is reversed. Positive estimates appear in a small number of subgroup specifications, especially among respondents who belong to or identify with the targeted group, but these findings are not consistent across treatments and are mostly not statistically significant. Overall, these robustness checks do not suggest that the main findings are driven primarily by differences in issue importance.

Three further aspects of the experimental design are particularly relevant for assessing the range of the observed effects: its prioritization of internal validity through comparable pledges that differ mainly by target group and are presented without partisan or contextual cues; the use of policy-based rather than symbolic appeals; and the focus on comparatively less contentious target groups.

First, our design seeks to enhance internal validity by using realistic pledges that differ mainly in their target group, within a preregistered setting providing as much control as possible. This control necessarily comes to some extent at the expense of external realism: real campaigns expose voters to multiple, often cross-cutting messages, partisan reputations, and repeated cues that can shape interpretation. By removing party labels and contextual noise, the experiment seeks to isolate as far as possible the causal effect of group targeting and to elicit distinct voter reactions linked to group membership, identification, and deservingness. This means, however, that we cannot capture how party images or media frames interact with such appeals. The hypothetical setting stripped of partisan and contextual signals could be seen as particularly conducive to responses to targeting. If so, our results would suggest that mobilizing voters through targeted promises is even less likely in real campaigns, given the absence of added value compared to universalistic pledges in our controlled environment. Yet partisan labels may also moderate the asymmetry we observe – dampening backlash among outsiders who trust the messenger, amplifying attraction among targeted loyalists, or generating effects of ‘surprise’. Indeed, Hersh and Schaffner (Reference Hersh and Schaffner2013) find that reactions among ingroup members can be especially strong when an appeal comes from a source not traditionally associated with their group. Analyzing such interactions between attitudes towards parties and group targeting thus constitutes a relevant avenue for future research.

Second, our experiment focuses on substantive, policy-based appeals rather than symbolic references to groups. Previous research has largely examined symbolic cues with mixed findings, leading scholars to question whether adding a material dimension would increase electoral returns to group targeting (Hersh and Schaffner Reference Hersh and Schaffner2013). Our design directly tests this proposition by presenting concrete policy pledges that deliver visible benefits to defined groups and reveal that even when targeted promises are made tangible, they do not outperform a universalistic alternative and can provoke significant backlash among outsiders. In this sense, our findings suggest that the limited payoff of group appeals in prior symbolic experiments does not simply stem from their vagueness or lack of policy substance. Rather, they point to a more general constraint on the electoral rewards of group targeting.

Third, our study focuses on broadly accepted, less politicized groups: parents, pensioners, and rural residents. This choice follows the broader design logic of neutrality: as explained, we selected groups that are frequently addressed in electoral programs (Thau Reference Thau2019; Deiss-Helbig, Guinaudeau, and Matthieß et al. Reference Deiss-Helbig, Guinaudeau, Matthieß and Rentrop2024), but carry limited partisan connotations, ensuring that the fictional party would not be automatically linked to real parties representing those constituencies. This decision could entail a conservative test of potential effects: because these groups are widely viewed as legitimate recipients of policy support, negative reactions among outsiders are likely attenuated compared to the sharper resentment that appeals to more polarized constituencies might provoke. This conservative design likely yields a lower bound of the polarization that group-targeted pledges can generate. At the same time, prior studies suggest that members of more politicized constituencies – such as those defined by class, religion, or ethnicity – may react more strongly to group cues, both positively and negatively (Robison and Moskowitz Reference Robison and Moskowitz2019; Dassonneville, Stubager, and Thau, Reference Dassonneville, Stubager and Thau2025). Appeals to such groups could therefore produce clearer ingroup mobilization, but also more intense outgroup backlash, especially when moral deservingness or partisan counter-identities are salient. Extending the analysis to more contested social categories would thus help identify the conditions under which group targeting mobilizes supporters, alienates outsiders, or deepens social divisions.