Support for BI in Finland

Public opinion is an elusive concept. It is very much affected by contextual factors. Therefore, it can change very rapidly. For example, while after the Chernobyl (1986) and Fukushima Daiichi (2011) nuclear disasters public support for nuclear power suddenly evaporated, the present-day climate discussion has re-established nuclear power as a serious policy option to achieve carbon neutrality. In a similar way, debates about and support for BI go in cycles. While the idea of BI fades away during periods of prosperity and full employment, it returns as an idea for providing a source of security during hard times, rising unemployment, poverty and insecurity of income and livelihood (Halmetoja et al., Reference Halmetoja, De Wispelaere and Perkiö2019).

Opinions on BI are also nested in national contexts. For example, Andersson and Kangas (Reference Andersson, Kangas, Kildal and Kuhnle2005) found that in Finland the most critical party (the Conservative Party) had more positive opinions on BI than the most positive party (the Greens) in Sweden. In fact, Finland is the only Nordic country where BI has been discussed as a policy option more seriously. BI has only vaguely been on the political agenda in Denmark, Norway and Sweden.

There are some comparative databases and analyses of cross-national differences in opinions on BI. In 2016, a question on BI was fielded in the European Social Survey. The results show large variations in Europe: the lowest support 34 per cent is found in Norway and the highest 81 per cent in Lithuania. Finland, with its 56 per cent%, is in the middle of these two extremes and close to the European mean of 54 per cent (ESS, 2018). The YouGov Survey (2019) produced rather similar results for Finland: 51 per cent of the Finns regarded BI as a very good or good idea.

Before the start of the Finnish BI experiment, two opinion surveys with different wordings and with different framings were conducted. In the first survey in 2015, the question on BI was posed at a general level: Is BI a good or a bad idea? The share of respondents who regarded BI as a very good or a good idea was as high as 69 per cent. The problem with such general questions is that they do not tell people about the possible costs. Therefore, a new survey which listed the necessary increases in the income tax rates was carried out. When the respondents were told what the level of flat-rate tax needed for financing different levels of BI would be, the support radically diminished. A lower BI of €500 with a flat rate tax of 40 per cent got 40 per cent support. A higher €800 BI with a 55 per cent tax rate was supported by only 29 per cent of the respondents (Airio et al., Reference Airio, Kangas, Koskenvuo and Laatu2016).

In 2017, yet another survey was carried out to estimate the support for BI. In the questionnaire, the level of BI and the most significant non-replaceable benefits were defined. According to the results, support for a BI model with €560 was at 51 per cent, while a full BI model guaranteeing €1,000 a month got only a very modest support of 20 per cent (Pulkka, Reference Pulkka2019; see also Crisp, Pulkka, & Rincon on this issue).

What explains the support for BI?

Previous studies have shown that there are several potential explanatory factors for the support for BI. These factors include macro-level variables like the extent of existing welfare policies and safety nets, and the level and coverage of existing social protection schemes (Roosma & Van Oorschot, Reference Roosma and Van Oorschot2020). Other potential macro-level factors include those that have successfully explained opinions on other social policy alternatives. These might include the industry structure, the developmental stage of the economy, labour market conditions, immigration, income inequality, political culture and ideology and political structures such as parties (cp. Lee, Reference Lee2018). Other factors to consider are the social policy preferences of the political elites and how these are mediated by the media and political publicity (eg. Campbell, Reference Campbell2002; Kangas et al., Reference Kangas, Niemelä and Varjonen2014). Political frames and the prevailing political discourses could also be possible explaining factors (cp. Chong & Druckman, Reference Chong and Druckman2007).

Within a country, socioeconomic and demographic micro-level variables, such as gender, income, socioeconomic status, ideology/party attachment and age, have been identified as important determinants. Low-income earners, unemployed people and those who view unemployment as a societal rather than an individual problem are more positive about BI (Andersson & Kangas, Reference Andersson and Kangas2002; Airio et al., Reference Airio, Kangas, Koskenvuo and Laatu2016; Pulkka, Reference Pulkka2019; Vlandas, Reference Vlandas2019). Other potential explaining factors include the economic insecurity of households and their problems in coping with their current income (see, eg., Lee, Reference Lee2018).

People’s receptivity towards BI is likely also determined by how the BI is designed in relation to the current welfare system (cp. Lee, Reference Lee2018). For Finland, this factor became visible in the studies by Airio et al. (Reference Airio, Kangas, Koskenvuo and Laatu2016) and Pulkka (Reference Pulkka2019). In these studies, different definitions and model specifications seemed to result in different levels of popular support. We find this a fruitful line of research when trying to analyse the actual support for and political feasibility of BI. However, it is perhaps not too surprising that policies tend to gather broad support when presented at an abstract level, but when the actual policy design is given and, eg., the costs of introducing the policy take shape, the attitudes towards it become more polarised (cp. Hiilamo & Kangas, Reference Hiilamo and Kangas2009). Providing the respondents with more information on the hypothetical policy proposal is one way to approach this problem – exposing them to a real-life experience is another.

Opinions on BI

In this paper, our purpose is to explore and estimate the relationship between people’s participation in a real-life experiment and (O1) their support for BI. We will also investigate (O2) their opinions on BI’s ability to make it easier to find employment. People’s participation in an experiment should provide them with an opportunity to learn about the policy through their own experience and to develop their opinions on it further.

The relationship between this participation and popular support is, however, ambiguous. If the participants experience the policy positively, we can expect that they are more likely to support the policy than their peers who have not experienced it, and vice versa. Based on the design of the Finnish experiment, we know that for many participants the experiment should have provided them with a regular and foreseeable minimum income, reduced the necessity to transact with the employment and benefit officials, increased their work incentives and, if they found employment, even increased their income. Our hypothesis is that because of these design-based features (H1) participation in a BI experiment should increase the level of support within the target population. Similarly, (H2) participation should make participants more convinced that BI makes it easier to find employment.

There are several reasons for why the relationship between participation and popular support might be different in different subgroups. The main motivation for studying the heterogeneity of this relationship is that previous studies show that a number of different background factors are linked to political attitudes and opinions on policy options (eg. Ervasti et al., Reference Ervasti, Goul-Andersen, Fridberg and Ringdal2012; Larsen, Reference Larsen2006; Taylor-Gooby & Leruth, Reference Taylor-Gooby and Leruth2018; Vlandas, Reference Vlandas2019; Van Oorschot et al., Reference Van Oorschot, Roosma, Meuleman and Reeskens2017). We will explore to what extent – if any – these factors explain the tendency to switch opinions based on daily-life experiences. This will give us a more detailed picture of how different population groups might react to personal experience with a BI.

Ability to express opinion

It is frequently difficult for respondents to express their opinions on abstract policy choices. Therefore, the share of people who say that they do not know is high in many survey studies. In the ESS, the share of “I do not know” answers was 4 per cent for Finland and 9 per cent for all countries in the sample (ESS, 2018). The YouGov Survey (2019) displayed surprisingly high levels of ignorance in Finland and elsewhere: 18 per cent in Finland, 23 per cent in the United Kingdom, 22 per cent in Sweden and 14 per cent in Germany.

Moreover, the quantity and level of background information and the framing of the question matter (Kangas et al., Reference Kangas, Niemelä and Varjonen2014). Whereas in some cases adding information will diminish the level of ignorance, in other cases more information increases insecurity and leads to cognitive dissonance, a psychological stress that is triggered by new and contradictory information (Festinger, Reference Festinger1957). For example, in a previous survey where the respondents were asked whether BI is a good or a bad idea, 4 per cent of the respondents could not express their opinion. When additional information about taxes was given, this share increased to 6 per cent (Airio et al., Reference Airio, Kangas, Koskenvuo and Laatu2016).

In this paper, we are interested in the (O3) ability of individuals to express their opinion. Participation in the experiment provides much more personal and concrete background information for the respondents to evaluate the BI than any information given in an opinion survey, regardless of how accurate it is. Therefore, our hypothesis is that (H3) those who have received a BI are better able to express their opinion on BI than those in the control group.

Design of the survey study: sampling and data collection methodology

The sample of the study consisted of all 2,000 participants of the experiment and 5,000 randomly chosen people from the original control group. The sampling for the survey study was conducted in September 2018. In October 2018, an additional 600 people were randomly sampled from the control group to compensate for the attrition caused by incomplete register data in the sampling framework. Those who had died during the experiment, those who had a special security status and those who had no telephone number in the customer registry of the Social Insurance Institution were dropped from the sample (551 people). The rest of the chosen people (7,049) were first informed about the study via a paper letter. The research data were collected via computer-assisted telephone interviews. The interviews were conducted in October to December 2018, in four calling rounds.

Attrition

Of the people who were contacted with a letter (7,049 persons), around 23.2 per cent (1,633) were successfully interviewed (see Table 1). The attrition in the study was a result of three different sources: dropping out of the sample (see the above section), refusal to respond or being unreachable after sending the information letter (66) and refusal to respond (2,337) or being unreachable (3,013) when calling by telephone.

Table 1. Attrition and unit non-response in the treatment and control groups.

The reasons for refusals at the phase of telephone interviews were the following: not able to find a suitable time for the interview (268), not interested (58), not able to participate or too long an interview (19), not participating because of a general principle (18), problem with the language (187), interrupted (23) and no registered reason or other reason (1,764).

Weighing the data to correct for bias caused by attrition

The attrition, and especially the unit non-response, in the experiment might lead to an unrepresentative sample and more importantly to an imbalance between the treatment and the control groups. To analyse this possible bias, we use register data from the sampling framework (benefits register from the year before the experiment) combined with unit-level information on responding to the telephone survey.

Compared to the target population, the respondents differ statistically significantly in the amount of their earnings and unemployment benefits, their age, their mother tongue and in their number of dependent children. In Table 2, we see that, on average, the respondents earned more and received less unemployment benefits than the target population. They are also older and more commonly speak Finnish or Swedish as their mother tongue (which are the official languages in Finland). Those who have only one dependent child are under-represented in the respondents.

Table 2. Background factors of the respondents in the treatment and control groups and in the target population before the experiment (register data).

Note: Standard deviations shown in parentheses.

The imbalance between the treatment and the control groups reduces the credibility of drawing causal conclusions about the relationship between the tested policy and the outcomes. Between the groups, only the mother tongue of the respondents differs statistically (see p values in Table 2). The proportion of those who speak languages other than Finnish or Swedish is greater in the treatment group than in the control group. However, there might also be other unobserved characteristics that are not balanced between the groups.

We try to correct the bias caused by attrition in the sample by weighing the data. We calculate the weights by fitting a model that predicts the probability of responding to the survey to the sampling framework. We model the weights separately for the treatment group and for the control group. The covariates in the model are gender, age, spouse, foreign language, type of unemployment benefits and region. The variables represent the situation before the experiment started, and they come from administrative registers. Table 3 shows that weighing the data corrects the distribution of the selected background factors in the right direction.

Table 3. Background factors of the target population and respondents before and after weighing (register data), %.

Methods for analysing the impact of the experiment

In our analysis of the potential impact of the experiment, we utilise weighted distributions of the outcome variables for both the treatment and the control groups, and logistic regressions of the outcome variables. The main outcome of the Finnish experiment was set to be employment. Attitudes towards BI are considered a secondary outcome, and thus the analysis here is more exploratory in its nature.

The logistic regression model that we use in the statistical analysis is

$$ \mathit{\ln}\frac{\pi }{\left(1-\pi \right)}=\upalpha +\updelta t+{\beta}_1{x}_1+\dots +{\beta}_p{x}_p, $$

where $ t $ is the treatment indicator, $ {x}_1\dots {x}_p $ are the controlling (dummy) variables and $ \pi $ is the probability of the event that we want to investigate (the outcome variable having the value 1).

The controlling factors have been derived from the previous literature. They include gender, age, education, socioeconomic status before entering unemployment, family type and type of municipality. These factors have been successfully used in explaining people’s attitudes towards BI in previous studies (see the “Previous studies on the popularity of BI” section). All the covariates are coded as dummy variables in the model.

The covariates are based on self-reported data from the survey. We assume that they are not dependent on the treatment. The regression estimates for covariates are shown in the appendix (Table A1).

Operationalisation of the studied outcomes

The support for BI and the opinions on its ability to make it easier to find employment are measured with the following operationalisation:

The available alternatives for answering each statement are strongly agree, strongly disagree, neither agree nor disagree, somewhat disagree, strongly disagree and cannot say.

In the regressions, strong and modest agreement are coded as 1, and indifference, strong and modest disagreement are coded as 0. “Cannot say” answers are not included. However, we regress the “Cannot say” answers separately in an index that counts the occurrence of “Cannot say” answers for all four statements. In this regression, the value 1 represents one or more “Cannot say” answers, and 0 represents no “Cannot say” answers.

While the meaning of the questions Q1, Q3 and Q4 seem to be pretty clear and straightforward, Q2 may contain some ambiguity. We think that it is not entirely evident which bureaucracy is being referred to here. However, in the Finnish public discussions, the system of benefits is often described as too complex, intransparent, unforeseeable and discouraging (bureaucratic) for those beneficiaries who would have the option to accept gig-like, part-time or temporary job offers. It is assumed that the level of bureaucracy discourages beneficiaries from accepting those jobs that are available. Thus, we believe that although the meaning of the statement is not entirely clear, we are still able to represent the respondents’ opinions on this potential deficit of the system of benefits.