This paper explores citizens' attitudes toward automatic decision-making in relation to Quality of Government (QoG) dimensions, such as impartiality and transparency. The research builds on material from an annual national survey in Sweden. The questions revolve around three domains of public decision-making (tax declaration, citizenship applications, and income support). Findings from the survey are presented via descriptive statistics. The results reveal that many citizens are unaware that government organizations are making automated decisions. There are clear differences in attitudes relating to the three domains. Citizens are generally more positive towards using automation in decisions regarding tax declaration compared to income support and citizenship applications. Through this analysis we contribute with new empirical findings regarding citizens' attitudes to automatic decision-making, its impact on quality of government and pave the way for additional explorations in this area. We identify a need to further analyze how these attitudes are shaped by socio-economic factors such as age, education, employment status, income, and digital competence on the one hand but also other values such as trust toward government, and political polarization.
Government organizations are increasingly relying on technologies such as robotic process automation (RPA) to automate decision-making in various areas [1, 2]. Public sector automation entails new challenges related to data quality, legal frameworks and human factors [3]. This calls for additional research on the topic, specifically in relation to different types of public services. At the same time, governments must provide high-quality services that citizens and others trust, both as users and to build a reliable and resilient society [4]. Quality of Government (QoG) is a theoretically based indicator of how a government ensures transparency, reduces corruption, fosters institutional stability, impacts economic development, social trust, and the effectiveness of public policies, shaping overall societal well-being [5]. In this paper, we specifically explore the impact of automation on QoG, by surveying attitudes to the automation of public decisions.
While extant research has investigated the impact of automated decision-making on organizations and public employees [6, 7, 8, 9], more research is needed on how citizens perceive how automated decision-making impacts quality of government. Automation is used both within the public administrative bodies and at the encounter where citizens are users of public services, which sometimes involves mandatory submissions of information and documents. Thus, it is important to study attitudes towards automatic decision-making not only in general, but also concerning different public service domains and contexts [10, 11].
Public encounters have traditionally been characterized by the contacts between street-level bureaucrats and their clients. However, automation of decision-making in particular and digitalization in general transforms these contacts to a more digital interface. Here, we argue that the political as well as administrative character of the public encounter are also brought into the digital encounter. Peters [12] argues that street-level bureaucrats are not only policy implementers but also political actors who shape policy through their decisions. They determine who receives government benefits and sanctions, exercising significant power over clients. Their discretion in interpreting and enforcing laws directly impacts equal treatment of citizens. When public services towards citizens are automated these public values and policy intentions must be translated through these services.
Quality of Government (QoG) is, as argued above, a key indicator of how well public agencies ensure fairness in governance in line with public values and policy. In this paper, we explore how automated decision-making affects QoG dimensions such as impartiality and transparency in different domains of public encounters.
The purpose of this paper is to investigate citizens’ attitudes toward automatic decision-making in three domains of public encounters (tax declaration, citizenship application, and income support) and their relations to QoG. We pose the following research question (RQ): What are citizens’ attitudes towards automatic decision-making in different public sector domains?
The research is conducted via a national survey administered to Swedish citizens - the SOM (society, opinion, media) dataset (https://www.gu.se/en/som-institute/the-som-surveys). The Swedish welfare state is extensive, providing social services, education and income redistribution. The redistributive logic of the Swedish welfare state builds on public values from a moral foundation and broad political support that sustain the system [13]. The public administration is also characterized by high trust in public services and institutions, reflecting a long-standing tradition of transparency, low corruption, and strong state-citizen relationships [14]. Thus, automation at the public encounter is a relevant setting for studying attitudes among citizens and their interpretations of the quality of services provided.
Additionally, Sweden has a highly digitally competent population, facilitated by extensive digital infrastructure and early adoption of digital services in general and e-government services in particular, as shown in the annual survey Internet and the Swedes (https://svenskarnaochinternet.se/rapporter/svenskarna-och-internet-2024/internetanvandning) The survey shows that almost 90% of the population are using Internet daily. Also, the use of public digital services is high, and we have in former studies [15] shown that attitudes towards automatization vary in relation to different QoG dimensions. In this paper we extend this study by studying attitudes towards automatic decision-making in three different domains.
This paper proceeds as follows: Section 2 presents theoretical background, followed by materials and methods in Section 3. Findings are presented in Section 4. After a discussion in Section 5, conclusions are drawn in Section 6.
To build a theoretical frame for analyzing citizens' attitudes towards automated decision-making, we position the analysis in relation to key concepts within the theory quality of government (QoG). Furthermore, we discuss citizens’ access to public digital services in Sweden and the broader implications for QoG and the legitimacy of automated decision making [16, 17]. Automatization of public services is embedded in a governmental institutional framing; thus we have operationalized the analysis and the survey questions around five QoG key concepts presented in Section 2.2 (reliability, impartiality, individual considerations, transparency, and administrative burden for citizens). It should be noted that while corruption is commonly also seen as a key component decreasing the quality of a government [18], we do not focus on this dimension explicitly in this paper. However, as measured and visible corruption indicates that governmental processes and services are not transparent or impartial, our chosen dimensions may serve as proxies in this regard.
To explore attitudes among citizens in relation to automatization of different types of services we constructed statements relating to automatization in three domains of public encounters (tax declaration, citizenship applications, and income support) further outlined in Section 2.3. 2.1.
Quality of Government (QoG) refers to the degree to which public institutions function in an impartial, reliable, and corruption-free manner, ensuring effective governance and legitimacy. Based on research on state capacity and institutional quality, QoG is generally linked to governance effectiveness, trust in government, and democratic stability [5]. Public services provided to the citizens play important roles in enhancing the quality of government. Standards and governance practices can improve quality and build trust [19]. Agnafors [20] argues that there is a need for a richer definition that accounts for moral content and also encompasses a plurality of public values and interpretations in line with a public ethos. He argues that this public ethos is found among citizens in their experiences of and relations to government agencies through the rule of law, efficiency, stability, and a principle of beneficence. Meanwhile, the attitudes to public services are formed in relation to governance effectiveness, trust in government, and democratic stability [21].
QoG is a key for public governance in all public organizations, and it has been shown to generate legitimacy to a higher extent than the electoral input side of government [22]. High-quality government institutions operate based on universalistic principles rather than favoritism, ensuring that public administration is both fair and predictable. Previous research presents mixed findings related to using AI technologies for automatic decision-making. Although a general trend is that automated decisions are perceived as fairer than decisions made by humans, these results vary between different social groups, and there are also concerns about the effects of extensive use of these technologies [23, 24].
The universalistic principles of QoG [5, 25, 26] are in line with the general ideas of how digitalization can transform bureaucratic processes and citizen-state relations. These relations can be defined as the “public encounter” [27]. Here, Lindgren et al. [27] request research on how digital public services relate to citizens' quality of life. As we focus on how citizens interpret quality of the government they interact with, and QoG relates to trust in society and the public ethos, it can be considered an indicator of citizens' quality of life.
Automated decision-making can be seen as a common expression of digital tools in public encounters. Lindgren [3] discusses how automation of public service, entails new challenges and requires careful consideration when implemented to support the public encounter. She highlights risks when automated systems replace traditional processes as new and unforeseen errors emerge; that it forms new types of interfaces between humans and the automated system; and “creates new tasks, roles, and responsibilities on multiple levels of the organization related to development, implementation, monitoring, use, maintenance, and management of the automated system” (p. 9). Thus, public organizations must both create well-designed human-system interfaces and ensure transparency and prevent critical decision-making from becoming black boxed [3]. While automation holds significant potential for improving public services within the public organization, we turn to its impact on QoG by addressing the citizens’ interpretations of their (non)-experience of automated public decision-making. The attitudes regarding a public service form trust not just in the service as such but also contribute to general trust in government. 2.2.
By framing the analysis and designing the survey based on theories of Quality of Government, we focus on five central aspects: 1. The reliability of decision-making, which refers to the extent to which automated decisions are consistent and predictable. 2. Impartiality, which has been a fundamental principle in previous QoG studies and ensures that decision-making processes do not favor particular individuals or groups. 3. Impartiality can also be translated into the less individual considerations in decision making, a formulation that made up our third item. 4. Transparency, i.e., that both the rationale behind decision making and the decision-making process itself should be clear. 5. The hypothesis that automatization would increase the administrative burden for the user, the citizen, and that more administrative burdens would decrease the potential gains in respect to quality of government through automatization.
More reliable decisions can significantly enhance quality of government (QoG) by improving the effectiveness, efficiency, and accountability of public administration. A main argument is that decisions based on evidence and standards are more reliable [28]. Sanderson [29] argued that the attempt to ground policymaking in more reliable knowledge of what works retains its relevance and importance. Citizens are served both by the single more reliable decision they receive when using a service, and by the benefits they gain collectively from a more effective, efficient, and trustworthy government.
Impartiality is a general key component of QoG, building on fair and unbiased implementation of public policies and services, which in turn is crucial for maintaining political support and democratic legitimacy [5, 22]. More impartial decisions are in general seen as an indicator of QoG and it has also shown to increase support for higher taxes and social spending [30]. Thus, when it comes to public decision-making in citizen related services the potential to improve impartiality can be an argument for more automatization.
A core aspect of impartiality is equal treatment; however, decision makers also must take individual considerations into account. This is a core competence of the professional street-level bureaucrat that involves translating and adopting policies into local practices by adjusting to individual circumstances [31] where professional discretion plays a key role [32]. This is particularly relevant when it comes to means tested public services such as child poverty support systems [33], illustrated in our study by income support services.
Democratic public administration relies on high transparency in both motives for decisions and regarding the process of decision-making. Transparency thus becomes a key component in QoG [5]. Digital technologies are generally seen as tools to increase transparency and in also in particular for more transparent digital government that can further be enhanced by high quality design [34].
Previous research has highlighted “hidden” inequalities [35] and costs [36] in the wake of digital government development. For example, Hoglund-Ryden and De Andrade [37] note that the creation of digital self-service transfers administrative burden to citizens, which affects vulnerable citizens to a large extent. Administrative burden examines and measures the burden, costs, and benefits of digital public services [38] rather than examining the literacy (ability and knowledge) needed to use and understand public digital services. Thus, examining the effect of automation on administrative burden for the individual is an important part of our study. 2.3.
To explore citizens’ attitudes to automated decision-making we choose three different domains of public service: tax declaration, citizenship applications, and income support. These domains represent various levels of government (income support is managed by local authorities and the other two domains are the responsibility of two dedicated national agencies – the Tax Agency and the Migration Agency). The domains also differ in type where tax declaration is something all adult citizens do on a yearly basis, while income support is tied to specific life situations and means tested, citizenship is a unique and critical issue based on fundamental democratic values but made only by a few people and once a life.
Tax declaration is compulsory in Sweden and submitted digitally or on paper by all citizens and most residents. There are different levels of this service. The simplest level is a confirmation of the information already entered by other actors (e.g., employers), and from data known by the government. This confirmation can be made by signing the distributed form, or a digital signature with e-identification or a single-entry pin-code. Based on these data the Swedish Tax Agency (Skatteverket) makes decisions about the taxes for the individual. They have gradually gained more and more discretion for automatization of the control of data and the final decision on taxes, through development of the regulation in the amendment to the Tax Procedure Ordinance. Through recent regulatory changes, the Tax Agency is given the conditions to increase the degree of automation in certain decision-making processes by removing the requirement to specify a decision-maker in cases where decisions are made automatically. This regulatory change is intended to streamline administrative processes and enhance efficiency in tax administration.
Citizenship application is open for those fulfilling eligibility criteria on a minimum period of legal residence (typically five years, or three years for stateless persons and refugees), a permanent residence permit, and good conduct. The process of applying for citizenship involves submitting an application to the Swedish Migration Agency (Migrationsverket), either through an online system or on a paper form. If approved, citizenship is granted, allowing full political and social rights and access to a Swedish passport.
Income support is offered, as a final complement to extensive social insurances and welfare benefits, to individuals who are unable to financially support themselves through work or other means. It is managed on a local government level by Swedish municipalities. Income support is applied for on a monthly basis and covers basic needs such as accommodation costs and food. This particular area has been popular to automate through RPA techniques capable of looking up information about applicants’ income from other government registries [39]. Here, Germundsson et al. [40] highlight tensions between using automatic decision-making in domains that, by law, should be subject to individual judgments by human caseworkers, which makes this an interesting domain to include in our survey.
In the following section, we outline how we study citizens’ attitudes towards automation in relation to these three domains.
This paper presents analyses using material from an annual national survey in Sweden administrated by the SOM-institute (Society, Opinion and Media) at Gothenburg University. The survey includes items in this survey to measure attitudes toward automated decision-making. We have developed the survey items that are included in this part of the survey. This work is an extension of research conducted on previous versions of this survey [15]. A major addition from previous years is that we now include measures attitudes towards three areas of automated decision-making (tax declaration, citizenship application, and income support) in addition to general attitudes in this area. 3.1.
The survey sample included 3,750 persons who were randomly selected from residents of Sweden between 16 and 90 years of age. The sample included Swedish and foreign citizens who, according to the Swedish Tax Agency, have a primary address in Sweden. Postal questionnaires were sent to the selected individuals. The natural dropout (e.g., deceased individuals, or people who have moved abroad) was 145 individuals, which gives a net sample of 3,605 respondents. The collection of data started in September 2024 and ended in December 2024. The SOM-institute sent five reminders to the respondents during the period. The respondent could send the answers through post or digital platforms. In total, 1,801 respondents answered the survey, giving a relatively high response rate of 50 percent. There are differences in response rate related to gender and age. Women (53 percent) has higher response rate than men (47 percent) and younger have lower rate than older. The lowest response rate has persons between 20-24 years (30 percent) and the highest found among persons between 70-79 years (67 percent). Although these differences, an analysis of the response rates has concluded that the survey gives a good demographic representative of the Swedish residents [41].
The first question assessed whether citizens were aware that government entities use automatic decision-making (yes / no). The following questions ask the respondents if they think automated decision-making regarding tax declaration, citizenship, and income support is appropriate. They could answer that they agree totally, agree to some extent, hardly agree, or do not agree at all to the statements that automated decision-making is appropriate in these areas.
To operationalize the theoretical background on automated decision-making, we formulated questions in the survey concerning the beliefs about automated decision-making in general and in the three areas of tax declaration, income support, and citizenship. The answers reflect how the respondents expect decision-making in the public sector to change if computers make decisions instead of public servants regarding five aspects: more legally secure decisions, more impartial decisions, less consideration of people’s situation, less transparent decisions, and more work for the individual. The questions were accompanied by a 4-point Likert scale ranging between “completely agree”, “partially agree”, “hardly agree”, and “Do not agree at all”. 3.2.
Our analytical approach rests mainly on presenting descriptive statistics and frequencies in tables with balance metrics, which is the share with positive answers (agree total + agree partial) minus the negative answers (agree not at all + agree hardly). It provides an indication of overall support for the statement and ranges between +100 (total agreement) and -100 (disagreements).
In this section, we present frequency tables with balance metrics to illustrate the findings from the survey. As seen in Table 1, a majority of the respondents (76.9%) are unaware of the fact that government organizations use automatic decision-making.
On an overall level, as illustrated by the balance values in Table 2, the respondents agree that automated decision-making leads to more impartial decisions and less work for the individual. However, the balance values also reflect concerns about legality and transparency, and that automatic decision-making will lead to less consideration of people’s situation. …more legally secure decisions …more impartial decisions …less consideration of people’s situation …less transparent decisions …more work for the individual
4.1% 17.2% 41.1% 32% 14.9%
some extent 34%
Hardly agree Do not agree Balance
at all 40.4% 21.5% -23.7 47.9% 35.7% 39.8% 31.4% 21.1% 13.3% 17.6% 34.2% 13.8% 9.9% 10.6% 19.5% 30.2 53.6 43.6 -7.4 Regarding the three domains, Table 3 shows a sharp contrast between respondents’ attitudes towards automatic decision-making for tax declaration compared to citizenship and income support respectively. A majority of the respondents (76.6%) agree that it is appropriate for automated decision-making about tax declaration, while a majority disagree with automated decision-making concerning citizenship applications (75.3%) and income support (71.5%). The largest difference as shown in Table 4-6 regards “legally secure decisions”, which shows a positive balance for tax declaration, but negative for citizenship and income support. …more legally secure decisions …more impartial decisions …less consideration of people’s situation …less transparent decisions …more work for the individual Table 6 When decisions about income support are made by computers instead of civil servants, it leads to...
Agree total Agree to Hardly agree Do not agree Balance
some extent at all …more legally secure 10.4% 35.5% 31.4% 22.7% -8.2 decisions …more impartial decisions 23.6% 43.2% 18% 15.2% 33.6 …less consideration of 42.7% 31.9% 14.6% 10.7% 49.3 people’s situation …less transparent decisions 28.4% 40.6% 19.7% 11.4% 37.9
30.1% 34.7% Our study provides insights into citizens’ attitudes towards automated decision-making in two main ways: first by showing how a relatively large number of citizens are unaware that government entities are conducting automated decisions. Second, by unveiling how perceptions of quality of government dimensions show a large variance dependent on the domain where automation is applied.
Given the overall result that many citizens are unaware of the fact that government organizations are automating decision-making, it can be questioned whether such initiatives are implemented based on citizen needs, or on orientations towards internal efficiency. It should be noted here that we used the term “computer” in the questions here, as the survey builds on a common set of questions that are similar each year. Lately, the societal discourse surrounding automation has changed though, and if we used terms such as “AI” or “robot assistants”, the results might have been slightly different [23, 24].
A general observation is that citizens generally have a more positive attitude towards tax declaration than income support and citizenship applications respectively. This difference in attitudes requires further research, but it is possible to outline some potential underlying causes. For example, the Swedish Tax Agency enjoys a relatively high degree of trust, and tax declaration is also a service that most citizens have experience of. The standardization of tax declaration form may also be seen as a service that can be automated without increasing the burden for the client.
Meanwhile, tax declaration is an activity all citizens conduct on a yearly basis, while the other two domains are more situational, bound to contexts where public servants use their professional discretion to make judgements based on several factors that vary much between individuals, thus the demands for professional desecration in decision-making is higher here [31, 32].
A key challenge of this analysis is that while we have examined attitudes through survey responses, these attitudes are inherently shaped by respondents' interpretations and reasoning processes. To account for this complexity, we have formulated and structured the survey questions around different domains, allowing for a more contextual interpretation of the responses. Here we have, as outlined in Section 2, formed the questions in the survey around three domains to capture variations in public service decision-making.
However, an important limitation remains since it is still unclear whether respondents' interpretations are grounded in personal experience or reflect a more generalized perception of how society functions. This distinction is particularly relevant given the differences between the domains. For example, most citizens have experienced submitting a tax declaration, making it a more universally understood process, whereas applying for citizenship is a rare and often one-time event. Such disparities in direct experience can influence how respondents conceptualize and evaluate public service interactions. These challenges are inherent in survey-based research, and even further confounded by the evolving role of AI-technologies in public administration and generally in society. As automation continues to transform government services, it is important to examine how citizens interpret these changes and whether they perceive them as enhancing the quality of government as the increasing use of AI, and not at least how automatization commonly in public discourses is referred to as AI. Future studies should therefore explore in greater depth how individuals construct their understanding of digitalization and automation in governance, shedding light on the mechanisms that shape public trust and engagement with automated decision-making processes.
This study has highlighted the importance of examining the formation of attitudes of automation of services at the public encounter towards citizens. We have focused on the overall RQ: ‘What are citizens’ attitudes towards automatic decision-making in different public sector domains?’
This study has revealed differences in attitudes depending on the domain where automated decision-making is applied. These findings require further investigations of underlying causes, but also how they relate to other variables. The SOM dataset, where our study is conducted, is rich and allows for several comparisons between groups of respondents, including demographics. Exploring differences between these groups would be a natural extension of this research, including how the different variables interplay with dimensions such as political leaning alignment and interpersonal trust. Aspects of digital literacy, competences, inclusion, socio-economic background, and political attitudes are therefore essential for developing inclusive governance strategies that maintain high levels of institutional trust and contribute to quality of government.
Thus, there is a need to further relate these responses to factors such as age, education, employment status, income, and digital competence, since they may play a significant role in shaping these attitudes. Other dimensions we want to elaborate on further based on the data set is to include broader societal values, political ideologies [43], and trust in institutions, which also contribute to the evolving perceptions of government. A particularly pressing area for further research is the impact of increasing societal polarization on these attitudes. In many democracies, including Sweden, trust in government institutions has traditionally been high, yet recent trends indicate growing divisions along ideological and socio-economic lines. This polarization may influence how different groups perceive public services, policy interventions, and digital governance initiatives, as indicated by a recent literature review [42]. Another potential avenue for further research would be cross-national comparisons as previous research has highlighted national differences in attitudes toward automated decision-making in government [10].
This research was supported by the Swedish Research Council, grant number 2023-00793. [6] G. Juell-Skielse, E.O. Güner, S. Han, S, Adoption of Robotic Process Automation in the Public Sector: A Survey Study in Sweden. In International Conference on Electronic Government (2022:1) (pp. 336-352). Cham: Springer International Publishing. [7] G. Juell-Skielse, I. Lindgren, M. Åkesson, U. Melin, A. Ranerup, B. Johansson, ... & M. Jabbari,
Service Automation in the Public Sector. Springer International Publishing. (2022:2)
[8] I. Lindgren, M. Åkesson, M. Thomsen, D. Toll, Organizing for robotic process automation in local
government: observations from two case studies of robotic process automation implementation
in Swedish municipalities. In Service Automation in the Public Sector: Concepts, Empirical
Examples and Challenges (2022) (pp. 189-203). Cham: Springer International Publishing.
[9] D. Toll, I. Lindgren, U. Melin, Process automation as enabler of prioritized values in local
government–A stakeholder analysis. In International Conference on Electronic Government
(2021) (pp. 288-300). Cham: Springer International Publishing.
[10] A. Kaun, A.O. Larsson, A. Masso, Automating public administration: citizens’ attitudes towards
automated decision-making across Estonia, Sweden, and Germany. Information,
Communication & Society, 27(
Information Polity, 27(
Science Review, 107(
Edward Elgar Publishing (2012).
[23] N. Helberger, T. Araujo, C. de Vreese, Who is the fairest of them all? Public attitudes and
expectations regarding automated decision-making. Computer Law & Security Review, 39
(2020).
[24] T. Araujo, N. Helberger, S. Kruikemeier, C.H. De Vreese, In AI we trust? Perceptions about
automated decision-making by artificial intelligence. AI & society, 35(
European Union. International Journal of Managing Projects in Business, 8(
European comparison. European Political Science Review, 5(
Innovative Social Sciences Journal, 8(
Öhberg & A. Bergström (eds.), I rörelse. Gothenburg University: The SOM institute. (2025). [42] S. Koniordos, A. Yfanti, Literature Review on the Impact of Digitalization of Work on Political
Attitudes, Participation and Trust. (2024) [43] L. Sundberg, E. Wihlborg, T. Denk, Attityder till automatiserat beslutsfattande i offentlig sektor: I gränslandet mellan teknik och ideologi. In B. Rönnerstrand, A. Carlander, P. Öhberg & A. Bergström (eds.), I rörelse. Gothenburg University: The SOM institute. (2025).