In state we trust: Evidence from Poland on why we undersave for retirement

ABSTRACT

With a possible decline in public pension benefits, private savings will become critical for maintaining desired living standards, yet most people undersave. We focus on one of the possible explanations for this: trust in the state. Using data from a dedicated survey and applying structural equation modeling, this study demonstrates that Polish citizens cherish the unconditional, albeit subconscious, belief in the state as a potential rescuer of last resort in life-threatening situations. This trust plays an essential role in people’s propensity to save for the future. We discuss potential explanations of this phenomenon and its consequences for social policymaking.

KEYWORDS:

• Trust
• public pensions
• private pensions
• welfare state
• structural equation modeling (SEM)
• survey study

Introduction

Standard economic theories model people as far-sighted economic agents, aware of their future needs and available resources and optimizing their lifelong consumption. One prominent example of this is the life cycle theory (Modigliani and Brumberg 1954), which predicts that people will aim to smoothen their consumption and save money during their working career to use it for consumption when they are too old to work.

However, there is a plethora of empirical evidence (Crawford and O’Dea 2020; Suh 2021) demonstrating that actual private savings are far too low to allow future pensioners to maintain their desired living standard. This undersaving is usually explained using behavioral arguments: people do not save enough because they are myopic, poorly organized, or procrastinatory (Benartzi and Thaler 2007). This study focuses on a different explanation of this observed phenomenon. We argue that one of the reasons why people undersave is their deep trust in the state. Contrary to most literature devoted to this topic, we demonstrate that this trust is not a simple function of people’s satisfaction with the state’s institutions nor is it a product of rational expectations. Indeed, it is quite the opposite. Trust in the state, which we model here, is subconscious and deep as faith. It is defined as the people’s belief that the state can be the rescuer of the last resort and will always help a person in a life-threatening situation. We show that trust defined in the following way is widespread and almost independent of personal characteristics.

To what is owed this belief? We speculate that it might be a product of people’s experience with the welfare state. All contemporary poles and almost all contemporary Europeans have lived their entire lives under some forms of state welfare. Thus, they might look at the state’s omnipotence as a form of the law of nature. Judging by their experience, the state has never let anyone die of hunger or lack of basic medicine; therefore, it might look as if the state is, and always will be, able to help. In our analysis, we put aside questions on the roots of this deep trust. Instead, we acknowledge its existence and demonstrate that it plays a significant role in shaping people’s saving behavior.

Interestingly, it turns out that trusting the state on pension savings depends on the people’s education level. Even though less-educated as well as highly educated people trust the state, it is only in the first group that we observe a significant (negative) impact on the propensity to save. There are a few potential explanations for this finding. First, we hypothesize that people who are less educated are characterized by a relatively lower potential to correctly assess the consequences of their decisions. They are subconsciously guided by the intuition that the state will not leave them behind. Therefore, there is no need to save additional money privately. Second, even though highly educated people also trust the state, they have higher expectations concerning living standards under retirement. Aware that the state cannot guarantee these standards, they must put some money aside; therefore, in their case, the link between trust and savings is weaker than that of others.

The paper’s main contribution is the proposal of a new concept, Trust in State, which describes the deep, subconscious belief that the state should intervene in extreme situations. Besides defining the concept, we attempt to analyze the impact of this trust on old-age saving behavior and discover characteristics that condition trust–saving relationships.

The remainder of this paper is organized as follows: Section 1 reviews the literature, Section 2 presents the local pension context, Section 3 provides information on the data used and the methods applied, Section 4 provides an analysis of the results, and Section 5 draws the conclusions and discusses the implications of the results for state policy.

Literature Review

The concept of trust has a long tradition in social sciences, so unsurprisingly the academic work is highly diversified in this field. To understand this heterogeneity, we begin with a discussion of the trust meaning. Following Bauer and Freitag (2017), trust attitude means that a truster A trusts a trustee B with regard to some behavior X in context Y at time t. Depending on the parameters (A, B, X, Y, and t), trust can take various forms. Referring to potential trustees (B), literature usually distinguishes two general types of trusts: social and political. The first trust type recognizes a group of people as trustees (e.g., family and society members), while the latter refers to some political actors like government. At this stage, our Trust in State concept is definitely a political trust type.

The aforementioned dichotomy is also relevant for the foundations of trust and its consequences. There is some disagreement about whether social trust depends more on repeated interactions with other people (Paxton and Ressler 2017) or is rather shaped by good democratic institutions that reduce the risk of dishonest behavior between society members (Warren 2017). Nevertheless, it evolves slowly. Contrastingly, political trust is believed to be driven mainly by the current state of the economy and the output of the welfare state (Kumlin, Stadelmann-Steffen, and Haugsgjerd 2017). That is why social, unlike political trust, is generally believed to be fairly stable over time. However, Bouckaert and Walle (2003) present an exceptionally different view. The authors argue that political trust should be separated from current policy effects because trust is entirely unrelated to what the government is or does. People form expectations of government actions with respect to very unlikely (wars, natural disasters) or very distant (old-age pension) events, in case they do not have any experience of “satisfaction.” However, the authors did not introduce any measure of their deep trust. We claim in our study that this type of trust exists, and we propose a specific measurement scale that would, hopefully, fill in the gap.

There are also diverse consequences with respect to both trust types. Life satisfaction, mortality, and suicide rates (Uslaner et al. 2017) or even vaccination coverage (Kawachi 2017) have been associated with social trust. On the other hand, political trust determines participation and support for incumbent political leaders (Hooghe 2017) and tax compliance behavior (Anderson 2017).

Research, whether people’s willingness to save for retirement is related to some form of political trust, is scarce and frequently indirect. Ljunge (2012) delivered consistent cross-country evidence that younger generations had a higher acceptance of claiming public benefits they were not entitled to, according to the World Value Survey. Ljunge (2012) proposes the adaptation mechanism hypothesis – the behavior of people born after the World War II has been shaped by their exposure to welfare state institutions throughout their lives. This finding is especially relevant for our study as the vast majority of current working generations may think about the state’s support and, in particular, public pensions as a law of nature, impacting trust and reducing individuals’ self-precautionary behavior.

The consequences of trust were also analyzed by Edlund (2006). He is one of the few that directly measured trust and related it to people’s willingness to save. To answer the research question, “Is distrust of institutional capability an important prerequisite for the general welfare state support withdrawal?” Edlund (2006) used two scales. The first scale diagnosed respondents with the perceived risk that the welfare state would fail to deliver particular services, and the second, in contrast, asked about the perceived need to complement the public net of social protection with private insurance. In a preliminary analysis, Edlund (2006) explored the relationship between the scales employed and noted that it was mixed for particular social risks. However, important for our study is that the link was rather clear in the case of distrust of old age welfare (Edlund 2006, 401) – people who doubted governmental support in this field demanded private pension insurance more than others. However, this was not the main research objective of the study, and the findings should be treated with caution. In particular, we are uncertain if the declared preference of the “perceived need to complement the public welfare system with private insurance” translates into any real-life actions (i.e., decisions) (Edlund 2006). This relationship, potentially, could have been better diagnosed when matched with other scales, and respondents’ characteristics, but this was outside the scope of the paper.

The link between trust and saving for old age has also been investigated by Perek-Białas (2017) and Buchholtz, Gąska, and Marek (2021). In both studies, trust was approximated using a single item. Perek-Białas (2017) focused on institutional trust (“Do you trust the social security office?”). Buchholtz, Gąska, and Marek (2021) analyzed whether government should subsidize minimum pensions. Similar to Edlund (2006), both questions represent some form of trust but not the deep trust that we explore, which is unrelated to any particular area of public policy. In the first case, it is also largely unclear whether the respondent shares an opinion on service quality or presents a perspective on the financial capacity of the public insurer. In the latter, the answer may present an opinion rather than a belief that the state will support older people. Nevertheless, a negative correlation between these trust-related items and declared retirement savings was noted in both cases, which grounds our preliminary intuition.

As it can be seen, the aforementioned arguments have been based on quantitative empirical studies so the measurement scales of trust become a clue point in this story. There are two general approaches in this field: survey questions and experiments. In case of survey items, people are usually asked whether they trust other people (social trust) or governmental institutions (political trust). However, survey questions are potentially prone to a fundamental methodological problem of incentive incompatibility as respondents have no motivation to uncover their true views. Once financially incentivized experiments are performed, the problem of incompatibility becomes limited. However, it is not easy to ascertain which approach is definitely better in predicting trust-related behaviors (Glaeser et al. 2000; Fehr et al. 2005). It is clear that incentive compatibility is a great advantage of experimental approaches, but it is rather difficult to simulate experimentally the attitude toward governmental institutions, so external validity may be questionable in this particular area. On the other hand, there is empirical evidence that some survey-based self-reported measures of trust predict real-life behavior (Monica, Lanier, and Meer 2008). For this reason, Bauer and Freitag (2017, 24) advise to formulate survey items in a more concrete way with respect to trustees (B), expected behavior (X) and context (Y) by providing, for example, a description of the B’s situation. Some other works also suggest to use multi-item latent constructs to measure trust (Freitag and Bauer 2013; Poznyak et al. 2014), as they minimize a measurement error relative to single-item scales. In our work, we follow these recommendations to support the robustness of our trust estimates.

To sum up, the knowledge of what a deep political trust is, how it should be measured, and its role in shaping people’s willingness to save for retirement is fragmented and indirect. We believe that our study will fill in an important research gap, expanding the literature in these three areas.

First, we characterize a new variable, Trust in State, defined as people’s belief in the state’s capability to serve as a rescuer of last resort. We demonstrate that this belief is general and cannot be limited to any particular area of public policy and/or perceived satisfaction from it. Therefore, it is distinct from similar political trust concepts analyzed in the literature (Kumlin, Stadelmann-Steffen, and Haugsgjerd 2017).

Second, using the data from the dedicated survey, we will demonstrate that trust in the state is an important factor affecting individuals’ precautionary behavior and, specifically, plays a significant role in shaping people’s willingness to save for old age. It is not the only driver of people’s willingness to save, but we will show that it is important.

Finally, we discover an important heterogeneity in the trust–savings relationship. Even if trust in the state is a society-wide phenomenon, it shapes the savings behavior of less-educated citizens. This finding has important implications for policymaking and questions the efficiency of programs aimed at increasing voluntary participation in pension-savings funds.

Local Pension Context

A wave of parallel pension system reforms has been widespread across Central and Eastern Europe since the late 1990s (e.g., Latvia in 1995, Hungary in 1998, and Romania in 2000). In 1999, the old Polish public pension system of non-financial-defined benefit (NDB) type was replaced by two compulsory pension pillars: non-financial- and financial-defined contribution (NDC and FDC) pillars. Since then, old age benefits have become tightly linked with the individual pension contributions accumulated over working life.

The contribution rate in the public pension system has settled at 19.52% of gross salary. Worsening demographic projections and low statutory retirement age (60 for women, 65 for men) mean that Polish pension benefits are set to be progressively inadequate. Based on the data from the European Commission (2021), the replacement rate¹ for Poland would decrease from 54% in 2019 to 25% in 2050. According to the International Labor Organization Convention on Social Security, this ratio should be above 40% to ensure pensioners’ well-being. Alarming forecasts are frequently discussed in the media. The Social Security Office informs participants once a year about their individual pension projections, so they are likely aware of their old-age financial prospects.

It is also necessary to highlight the existence of minimum pension benefit guarantee. In case a participant does not accumulate enough capital but has been contributing to the pension system long enough (20 years for women, 25 years for men), the state will subsidize the gap between the amount guaranteeing the minimum consumption floor and the fair value of the pension benefit in actuarial terms. In 2019, the value of this benefit was 1100 PLN (approximately 255 EUR) per month for a single-pensioner household, which is even less than the social minimum threshold² (1191.40 PLN in 2019). In the same year, 5% of the retirees under the public pension system were receiving benefits below this statutory minimum as their contribution period was below the established threshold.

In this backdrop, future retirees may justifiably feel the need to take care of their old-age requirements on their own. On the other hand, several studies (Jedynak 2016; Szczepański 2021) confirm that the pension savings in Poland are inadequate. Therefore, the limited self-precautionary saving behavior amid warning projections deserve research attention to help calibrate an effective policy response.

Data and Methods

Data Source

The main objective of this study was to assess the role of trust in the welfare state in shaping individuals’ propensity to save for retirement. To determine this, we ran a dedicated questionnaire survey. The study was conducted in November 2019 using the Computer Assisted Personal Interviewing (CAPI) method on a sample of 1069 adult Poles by a reputable research agency.³ The response rate for the study was 37.3%, which is the expected level for this type of study. Therefore, we applied weights to the individual observations to reflect the population structure correctly. These weights, provided by the research agency, were used to calculate all the estimates presented in the paper.

The study was carried out on a representative, nationwide address sample, taking into account the voivodeship, spatial dispersion, and the category of a place of living (village, small towns, and so on). Within the household, the respondent was randomly drawn using the Kisch grid, and 105 interviews were verified ex-post (9.8% of the interviews). During the inspection, no basic irregularities to reject any interview were found. Respondents were 18 and over. No variable other than age was controlled, therefore, some deviations in the sample structure are inevitable. Women are overrepresented in the sample, giving 57.3% of the answers, compared to 42.7% for men. According to the Eurostat data (Table EDAT_LFS_9903), in Poland, the percentage of people (aged 18–74) with an education level lower than secondary and secondary (we use such division in part of the further analysis) was 72.8% in 2019, while those with higher education comprised 27.2%. In our sample, these percentages were 67.7% and 32.3%, respectively, showing a small overrepresentation of better-educated respondents. However, considering the scale of the discrepancies, we expect they would not adversely affect the results quality.

Methods

Having reviewed the literature, we created a conceptual model describing the relationships between latent constructs and defined their measurement models. When shaping the questionnaire, we decided to extend the list of questions that define the constructs, allowing for potential redundancy. Therefore, we performed confirmatory factor analysis (CFA) as the first step. We used the Kaiser–Meyer–Olkin (KMO) test of sampling adequacy⁴ and Bartlett’s test of sphericity⁵ to assess factorability and extracted latent constructs using principal component analysis (PCA). In PCA, we applied Promax rotation⁶ because the considered phenomena were potentially correlated. We used CFA to obtain the final list of latent variables and extract the items that best characterize them, showing the need to split some preplanned constructs, combine others (e.g., Trust in State and Trust in NGOs; see next section), and eliminate some items. For the constructs extracted in this way, Cronbach’s alpha values were calculated as well as Jöreskog’s rho,⁷ which is adequate for structural equation modeling because it is based on the loading rather than the correlations observed between the observed (Demo et al. 2012). Composite reliability (CR) was used to measure internal consistency reliability. These measures allowed us to verify the reliability of each factor. The cutoff value for Cronbach’s alpha was 0.7⁸ for CR and 0.7–0.95.⁹

Next, we assessed the convergent validity of the considered constructs using the average variance extracted (AVE). The desired AVE values are 0.50 or more, which indicates that the construct explains at least 50% of the variance of its items.

After creating and initially verifying the quality of the proposed constructs, covariance-based structural equation modeling (SEM) was used to assess the direction and strength of the relationship between latent variables. The model employed only reflective indicators, which were expected to be highly correlated. In the modeling, we allowed for correlation of error terms due to the similar design and common subject area of the questions in the questionnaire.¹⁰

The model was estimated in two stages. The assumption of normality of the multivariate distribution was not met; hence, a simplified rule was adopted stating that if asymmetry and kurtosis are not excessive,¹¹ deviations from the normal distribution should not distort the results (McDonald and Bollen 1990). This implies that the maximum likelihood method is justified as it guarantees consistent and asymptotically effective estimators when applicability conditions are met. However, as shown in Table 1, simplifying conditions were not fully met for all the variables.¹² Therefore, the obtained estimation results were verified by applying Bayesian estimation with non-informative prior distributions. The obtained results did not differ significantly between the two methods, proving their robustness against normality violations.

Table 1. Latent concepts and descriptive statistics of the corresponding items.

ID	Questions	Mean	Standard deviation	Skewness	Kurtosis
Latent concept: Trust in State
Imagine you found yourself in a long-term, difficult life situation (e.g., health problems or job loss). Whom would you expect to get material support from, knowing that you will be unable to pay it back? * And where, in your opinion, support like that should generally come from?
Q1	State institutions	5.38	1.50	−1.02	0.75
Q2	Charity organizations	4.67	1.59	−0.58	−0.15
Think about the following scenario. Your friend, who used to work illegally, got in an accident and will be unable to work for several years. Your friend has no savings. Whom should they ask for support?
Q3	Non-government organizations (e.g. funds)	4.86	1.68	−0.71	−0.13
Q4	State institutions for sickness pension or social assistance benefits	5.23	1.59	−0.79	0.07
Latent concept: Satisfaction with Welfare State
Are you satisfied with the way the Polish state handles
Q5	Health care	3.05	1.85	0.55	−0.88
Q6	Support for elderly, including pensions	3.19	1.74	0.30	−1.04
Q7	Support for children (nurseries, kindergartens, paid parental leaves)	3.77	1.84	−0.10	−1.07
Q8	Investment in residential real estate	3.40	1.64	0.18	−0.75
Latent concept: Distrust in Market
Do you agree with the following statements
Q9	Government should set the prices of basic food products.	4.35	1.78	−0.30	−0.92
Q10	When it comes to the development of new technologies state companies are more efficient than the private ones.	4.05	1.64	−0.11	−0.69
Q11	Work conditions are better in state companies than in the private ones.	4.30	1.69	−0.29	−0.64
Q12	Government should decide about strategic directions of the economic development.	4.79	1.47	−0.50	−0.02
Latent concept: Trust in Public Institutions
When considering public institutions, I trust
Q13	Government	3.19	1.77	0.28	−1.09
Q14	National social insurance fund	3.20	1.72	0.22	−1.03
Q15	National social insurance fund for farmers	3.36	1.67	0.10	−0.87
Latent concept: Pension Saving Propensity
Q16	I have a plan how to save for retirement	3.01	1.44	0.41	−0.68
Q17	I believe that I will be able to save the amount of money high enough to increase my pension in a significant way.	3.25	1.39	0.07	−0.81
Latent concept: Long-term Saving
What are the saving goals of your household?
Q18	To secure children’s future	0.25	0.43	1.15	−0.69
Q19	To secure the future (in old age)	0.31	0.46	0.84	−1.30
Q20	To be able to leave a bequest	0.15	0.36	1.98	1.92

* The italicized fragment was part of the preceding question. However, presenting this phrase is necessary to form a context for subsequent questions.

The fit of the model was verified in two stages. The values of item loadings for all latent constructs were verified in the first stage. A value of 0.4 was adopted as the minimum threshold. Then, the values of the standard measures of the model fit were verified in the second stage. The Tucker-Lewis index (TLI) and the comparative fit index (CFI) that allow for a comparison with an independent model were used for this purpose. For TLI and CFI, we assumed a threshold value of 0.95. We also verified the root mean square error of approximation (RMSEA) with a cutoff level of 0.08 and Chi²/df with a maximum acceptable value of 3.¹³

All calculations were performed using weights (weighted correlation matrices), which allow, in conjunction with the appropriate sampling frame, the generalization of the analyzed results to the population. The calculations were performed using IBM SPSS Statistics 26.0 and IBM SPSS AMOS 26.0.

Definition of Latent Variables

As already mentioned, the main area of analysis is a description of trust and assessing its impact on the propensity to save for retirement. The estimated model will provide a precise definition of our crucial variable, Trust in State, and will allow us to demonstrate the relationship of this latent variable with other concepts of trust and the propensity to save for retirement. Due to the nature of the dependencies assessed, particularly regarding retirement savings, the model was estimated for a subsample that comprised people who do not yet receive pensions. The final sample size was 827, sufficient to estimate the model (Kline 2011).

The key concept analyzed in this article is Trust in State. We understand it as a person’s conviction that, if needed, the state will serve as a “rescuer of the last resort.” Such conviction might be deep and subconscious and, therefore, cannot be measured by a single, direct question: do you trust the state? Building solely on this question raises many potential problems with the interpretation of trust.¹⁴ Respondents could declare distrust of the state while simultaneously being convinced that state intervention will occur in extreme life situations. The latter could be a consequence of the welfare state’s experience, the lack of which is unimaginable for most citizens in developed countries. A crucial feature of such trust is expectations around actions taken in crises in which external assistance is required: who should act as the ultimate guarantor of security in extremis?

Therefore, the items considered as candidates to form the Trust in State construct concerned three groups of questions asked to the respondents in the following order:

• Imagine you found yourself in a long term, difficult life situation (e.g., due to health problems or job loss). Whom would you expect to get material support from, knowing that you will be unable to pay it back? (Group 1, G1)

• Where, in your opinion, support like that should generally come from? (Group 2, G2)

• Consider the following scenario. Your friend, who used to work illegally, got in an accident and would be unable to work for several years. Your friend has no savings. Whom should he ask for support? (Group 3, G3)

For the G1 and G2 questions, possible answers included (a) friends and acquaintances; (b) state institutions; (c) charity organizations; (d) family/children, other relatives; and (e) self-organized fundraising. In the G3 group, the available options were limited to (a)–(c).

Questions belonging to G1–G3 groups were designed to enable identification of who, in the respondent’s opinion, should act as a rescuer of the last resort while, at the same time, helping to avoid the potential problem of reluctance to declare true preferences when the respondent must relate to their situation in potentially sensitive areas. The estimation results confirm the adequacy of this type of concern, and the analysis of the constructs’ quality (CFA) unambiguously showed that answers to the questions from the first group (G1) appeared to have a distinct character; these did not ultimately form part of the analyzed latent construct. This is in line with previous findings in the literature that respondents answer questions that concern them directly because it is much easier for them to show their true attitudes when these attitudes concern other people (Lusk and Bailey Norwood 2010). Therefore, the latent construct Trust in State included only questions belonging to the G2 and G3 groups.

Initially, we intended to build two latent constructs based on these questions: Trust in state and Trust in NGOs. However, the CFA and SEM analyses indicated no reason to treat them separately (the criteria for assessing the quality of constructs and the value of regression weights in the SEM model indicated they should be modeled jointly). This fact can be explained by the specific situation of the third sector in Poland, which could be looked upon as the state’s contractor: its ability to provide long-term, large-scale support depends on state funding.¹⁵ As a result, NGOs are not perceived as a group of autonomous entities and are, de facto, identified with the state (the guarantor of its stability). To clearly show the originality of the proposed concept of Trust in State and to identify the existing dependencies, several related concepts were included in the analysis: satisfaction with the welfare state, Trust in Public Institutions, and distrust of the market.

Using the latent variable Satisfaction with the Welfare State, following Edlund and Lindh (2013), we measure the perceived public sector performance in areas typical of the welfare state. Questions Q5–Q8 were formed on the basis of Edlund and Lindh (2013), but we added an additional question on housing, an area of increasing state activity in Poland.

According to the existing literature, Trust in State may be associated with trust in institutions. Therefore, we asked about people’s confidence in the institutions that manage the public pension system, thereby constructing the Trust in Public Institutions variable. Questions Q13–Q15 were based on questions proposed in the European Social Survey.

The last analyzed factor, which could be related to the concept of Trust in State, is the respondents’ conviction that state institutions could solve various problems better than the market ones. Questions Q9–Q12 that form the Distrust in Market construct are related to the questions proposed by Edlund and Lindh (2013) but were modified to better match the Polish situation.

We decided not to operationalize or measure distrust of the government, which is understood to support the current government. Polish society is intensely politically polarized (Kinowska-Mazaraki 2021), and asking questions about current politics could evoke emotions that negatively affected the quality of other responses.

As previously pointed out, we hypothesize that Trust in State plays a significant role in shaping saving behavior, particularly in shaping retirement savings. If the state is perceived as an institution obliged to guarantee (and provide) help in critical situations, then Trust in State may reduce its propensity to save, especially for emergency and pension purposes. This paper deals with the second of these areas, for which we introduce another latent variable, Pension Saving Propensity.

We have analyzed several questions regarding the declared savings or savings behavior to select the pension saving propensity measure items. First, it may seem that it would be more informative to ask direct questions about an individual’s pension-saving products. However, this approach may not have produced a credible measure of pension saving propensity. In Poland, private savings labeled as “pension” constitute the so-called third pension pillar. We recognize them as pension savings due to the liquidity constraints employed, which limit their withdrawal before reaching old age.¹⁶ However, the participation rate in this third voluntary pillar is very low even lower than 10% for every saving vehicle.¹⁷ In our sample, 11.3% of the respondents declared that their household possessed “voluntary pension funds or insurance policies,” and 12.8% expected to receive some benefits from third-pillar instruments after retirement. However, when we asked about the reason for saving, 30.8% of all respondents declared an old-age security goal. This means that people might save for old age using instruments that are not formally a part of the third pension pillar. Therefore, we decided to use a different approach to measure the likelihood that respondents are pension-savers.

In classic economic theory (Modigliani and Brumberg 1954), planning retirement is easy: individuals are rational and forward-looking planners who optimize their life-cycle consumption path while knowing their needs at different points of life. However, in reality, this is a complex decision-making process (Topa, Moriano, and Moreno 2012). Should an individual wish to save an adequate stock of assets for old age, the minimal list of decisions includes how long to work, how much to save, how to invest money, and how to decumulate the assets in the end. Consequently, financial planning has been considered one of the key dimensions of general retirement preparation (Hershey, Jacobs-Lawson, and Neukam 2002; Taylor and Geldhauser 2007). Numerous studies confirm that financial planning attitude is closely linked to retirement savings (Topa and Herrador-Alcaide 2016).

Therefore, we decided that our Pension-Saving Propensity measure should be based on questions on retirement planning. Having said that, it is yet another variable that can be measured and interpreted using various approaches. Lusardi and Mitchell (2007) use a single item, thinking about retirement. However, Clark et al. (2014) argue that thinking about retirement is a nebulous concept and advocates asking directly whether the respondents have planned for retirement. Deaves et al. (2007) and Stawski, Hershey, and Jacobs-Lawson (2007) proxy the planning behavior by employing a multi-item battery of questions and constructing a latent variable. Our approach is similar; we use a two-item construct based on the following statements:

• I have a plan to save for retirement (Q16).

• I believe that I will be able to save an amount of money large enough to increase my pension in a significant way (Q17).

The first item is similar to that used by Clark et al. (2014). The second follows Stawski, Hershey, and Jacobs-Lawson (2007), who claim that financial planning activities are expressed in behaviors like attempting to calculate the targeted amount of money.

To ensure that Pension-Saving Propensity is a construct that reflects the actual course of long-term saving processes, we decided to check its relationship with the saving goals. We believe that respondents exhibiting high Pension-Saving Propensity should also pick up long-term motives (future of children, old age, and bequest) for already accumulated savings. Therefore, to include the saving motives in our model, we designed another latent variable called Long-term Saving. Questions (Q18–20) that form this variable generally come from the Social Diagnosis¹⁸ study (Czapiński and Panek 2015), but we have supplemented the list of motives with a bequest one.

The latent variables and indicators employed in the model are listed in Table 1. The description presented includes the wording of the questions asked to the respondents, basic descriptive statistics, and the assignment of individual items to latent constructs. All questions were answered on a 7-point scale (1 meant “definitely no” to 7 meant “definitely yes”). The only exceptions were the last three questions (Q18–Q20) related to the saving goal, which took the binary value of 0 when the respondents did not declare a given saving goal and 1 when they did.

Results

General Results

In the estimation process, we calculated the parameters of several models, allowing for a variety of potential links between the designed constructs. Our central variable is Trust in State, and ultimately, we looked for the model that best describes this construct and its role in shaping the willingness to save for retirement. The graphical representation of the final model is pictured in Figure 1.

Figure 1. Relationship between latent variables.

All the defined relationships between latent constructs were significant for p ≥ .001. The assessment of the estimated model’s quality shows that it suits the empirical data well, which is confirmed by the values of the Chi²/df = 1.800 (below the threshold 3) and the RMSEA = 0.018 (with 0.90 confidence interval of 0.016 to 0.019), below the threshold of 0.08. In addition, the previously mentioned measures of the model assessment achieved satisfactory values: TLI = 0.959 and CFI = 0.970 (with cutoffs 0.95).

CFA and structural equation modeling were used to assess the relationship between the defined constructs and their validity and reliability. The estimation results have been divided into two parts for clarity. Table 2 presents the results for the latent constructs and their measurement models without considering the mutual relations between them, allowing a separate assessment of the quality of each latent construct. Table 2 shows the loading for individual items, the significance of regression weights, and statistics that assess each construct’s quality.

Table 2. Estimates and statistics describing the model quality.

ID	Standardized estimate	Estimate	Standard deviation	P
Latent concept: Trust in State
Q1	0.666	1.000
Q2	0.925	1.471	0.106	0.000
Q3	0.679	1.138	0.076	0.000
Q4	0.761	1.206	0.098	0.000
KMO = 0.690, Bartlett’s Test: p = .000 CR = 0.847, AVE = 0.585, Cronbach’s alpha = 0.812
Latent concept: Distrust in Market
Q9	0.674	1.000
Q10	0.701	0.958	0.083	0.000
Q11	0.806	1.138	0.091	0.000
Q12	0.677	0.830	0.073	0.000
KMO = 0.766, Bartlett’s Test: p = .000 CR = 0.808, AVE = 0.513, Cronbach’s alpha = 0.786
Latent concept: Trust in Public Institutions
Q13	0.778	1.000
Q14	0.967	1.206	0.051	0.000
Q15	0.866	1.051	0.047	0.000
KMO = 0.713, Bartlett’s Test: p = .000 CR = 0.906, AVE = 0.763, Cronbach’s alpha = 0.904
Latent concept: Pension Saving Propensity
Q16	0.767	1.000
Q17	0.759	0.952	0.107	0.000
KMO = 0.500, Bartlett’s Test: p = .000 CR = 0.736, AVE = 0.582, Cronbach’s alpha = 0.729
Latent concept: Long-term Saving
Q18	0.685	1.000
Q19	0.729	1.135	0.102	0.000
Q20	0.653	0.783	0.072	0.000
KMO = 0.685, Bartlett’s Test: p = .000 CR = 0.731, AVE = 0.476, Cronbach’s alpha = 0.727

The results presented in Table 2 confirm the significance of all estimated parameters and the appropriate levels of unit loadings (standardized estimates substantially exceed 0.4). Both the KMO statistics (values equal to or higher than 0.500) and the results of Bartlett’s test (p = 0.000 for all variables) confirm the adequacy of constructs obtained in CFA. Appropriately high (exceeding 0.7 for all constructs) Cronbach’s alpha values indicate internal consistency and satisfactory reliability. The values of CR (also exceeding 0.7) and AVE indicate that the indicators are adequately represented in the construct.¹⁹

As observed, one variable, Satisfaction with Welfare State, was not included in the final SEM model. Despite satisfactory properties in terms of validity and reliability and appropriately high loads for its indicators, the relationship of this construct with trust in state turned out to be statistically insignificant. A detailed discussion is presented in the next subsection. The estimation results describing the relationships between the latent variables used in the final model are presented in Table 3.

Table 3. Relationships between latent variables.

Relationship			Standardized estimate	Estimate	Standard deviation	P
Trust in State	➔	Trust in Public Institutions	0.157	0.215	0.063	0.000
Trust in State	➔	Pension-Saving Propensity	−0.203	−0.225	0.066	0.000
Trust in State	➔	Distrust in Market	0.220	0.263	0.057	0.000
Pension-Saving Propensity	➔	Long-term Saving	0.446	0.119	0.019	0.000

Among the factors that may significantly affect the observed relationships, particularly the relationship between Trust in State and Pension-Saving Propensity, two seem to be of particular importance: education and income level. In our research, we decided to use education as a key variable. Although education does not affect the level of variables that build trust in the state construct and, therefore, does not directly impact the level of trust in the state, it may condition its consequences. We believe that education is crucial in shaping the needs of future pensioners (the higher the education, the higher the needs) and the potential to finance them, as education is highly correlated with income.²⁰ An additional advantage of using education, rather than income, is that it is more reliable: respondents have no problems revealing their education level, whereas the response rate in the case of questions concerning incomes is much lower (Yan, Curtin, and Jans 2010).²¹

Instead of introducing education as a separate variable in the model, we adopted a different approach. Education level affects many latent variables used in our model, but these effects are not within the scope of our analysis. Rather, we are interested in identifying whether education level impacts the observed relationships in terms of their direction and significance. Therefore, we decided to run separate models and analysis on subgroups differing in education level. We made two subgroups: those who attained higher education and those who did not. The results for less- and better-educated respondents are presented in Tables 4 and 5.

Table 4. Relations between latent variables for subgroup of less-educated respondents.

Relationship			Standardized estimate	Estimate	Standard deviation	P
Trust in State	➔	Pension-Saving Propensity	−0.273	−0.321	0.069	0.000
Trust in State	➔	Trust in Public Institutions	0.128	0.181	0.064	0.005
Trust in State	➔	Distrust in Market	0.249	0.289	0.057	0.000
Pension-Saving Propensity	➔	Long-term Saving	0.406	0.091	0.015	0.000

Table 5. Relationships between latent variables for the subgroup of better-educated respondents.

Relationship			Standardized estimate	Estimate	Standard deviation	P
Trust in State	➔	Pension-Saving Propensity	−0.030	−0.027	0.076	0.719
Trust in State	➔	Trust in Public Institutions	0.208	0.263	0.084	0.002
Trust in State	➔	Distrust in Market	0.141	0.169	0.078	0.030
Pension-Saving Propensity	➔	Long-term Saving	0.540	0.207	0.040	0.000

The observed changes concerned the impact of the latent variable not just Trust in State on Pension-Saving Propensity but also Trust in Public Institutions and (to a lesser extent) Distrust in Market. The definition of individual constructs did not change for subgroups (neither the significance nor the sign of individual estimates). Therefore, full specifications were not provided.

Detailed Analysis

We begin a detailed analysis of the results by discussing the characteristics of trust in the state and its relationship with other trust-related variables, followed by a discussion of its role in shaping the propensity to save for retirement.

The primary goal of this study was to characterize a new variable called Trust in State. As shown in Table 2, it is a well-defined construct characterized by high reliability and validity. Using weights in the estimation process makes it difficult to demonstrate the imputed distribution of this latent variable (the estimation procedure is based on weighted correlation coefficients, not on individual records). However, to obtain an indirect knowledge of its distribution, it suffices to analyze the distribution of the items that define it. We look at the responses to questions Q1–Q4 and acknowledge that responses of five or higher indicate the existence of trust in the specific domain. As many as 78.8% (Q1), 58.8% (Q2), 65.5% (Q3), and 71.2% (Q4) of all respondents convey some form of Trust in State. Thus, it is a variable that takes high values for a substantial proportion of Polish society. At the same time, the answers to each of the questions defining Trust in State were diversified enough to introduce a necessary level of variability (differences within the population); the coefficient of variation exceeded 28% for all questions Q1–Q4. In addition, the domains represented by questions Q1–Q4 did not depend on basic personal characteristics. Pearson’s chi-squared test indicates their independence of sex, age (categories: age below the median and above median), and education level (categories higher, non-higher). All p-values exceeded 0.05.

By estimating several models, we managed to establish relationships between all latent constructs under study; however, in this paper, we limit ourselves to those involving Trust in State. As demonstrated in the General Results section, two trust-related concepts were found to have a significant relationship with Trust in State. The first is Distrust in Market. The standardized estimates of loads show a positive relationship between Trust in State and Distrust in Market for the whole sample (0.220, p = .000) as well as for the sub-samples of low-education (0.130, p = .045) and high-education respondents (0.206, p = .003). This relationship is in line with our intuition – the more one trusts that the state would help them, the more is their distrust (less trust) of market institutions.

A second trust-related construct, which shows a significant, positive relationship with Trust in State, is Trust in Public Institutions. Even though the sign of this relationship is consistent with intuition (when one trusts public institutions, one also trusts the state as a whole), its significance is smaller than that of the former. This is reflected in the lower value of the standardized load in the case of the whole sample (0.157, p = .000). This is yet another demonstration that trust in the state reflects a deep trust, not necessarily resulting from the person’s attitude toward the specific public institutions.

However, the strongest support for the hypothesis that Trust in State is indeed a deep-faith concept, unrelated to actual experience, comes from the analysis of the relationship between Trust in State and the variable, which was eventually not included in the final SEM model: satisfaction with welfare state. Despite satisfactory econometric properties (KMO = 0.774, Bartlett’s test: p = .000, CR = 0.930, AVE = 0.770, Cronbach’s alpha = 0.893), its relationship with Trust in State was statistically insignificant: the standardized load for Satisfaction with Welfare State was 0.045 (p = .257).²² Therefore, a deep belief that the state would serve as a rescuer of last resort is unrelated to whether an individual is satisfied with actual state support. When thinking about potential satisfaction or dissatisfaction with the state, respondents refer to their experience with the way the state handles health care pension system, kindergartens, or real estate investments. In contrast, Trust in State captures a much more basic and primary need, that is, the guarantee of not being left alone in the case of an extreme situation.

Let us now discuss the role that trust plays in the willingness to save for retirement. As explained before, we measured this willingness by Pension-Saving Propensity. The estimated relationship between that variable and Long-Term saving suggests a well-designed measure: the relationship is positive, significant, and independent of the education level.²³ The estimation results demonstrate that Trust in State has a significant negative influence on Pension-Saving Propensity (standardized load −0.203, p = .000). However, the submodels’ results show a sharp difference between the less-educated and the better-educated respondents. The relationship between Trust in State and Pension-Saving Propensity turns out to be significant only for less-educated respondents (standardized load −0.273, p = .000) and completely insignificant for better-educated respondents (standardized load −0.030, p = .719). We interpret this in the following way: tTrust in State is independent of education level. Less-educated and better-educated respondents believe the state would not desert them. However, depending on the attained education level, the respondents subconsciously allowed this belief to shape their lives differently. Therefore, trust in the state diminishes their propensity to save for the future for less-educated respondents. In contrast, highly educated respondents are more aware of life’s potential hazards. Presumably, they expect a higher standard of living, which may not be delivered by the state. Thus, they position the state’s role as the rescuer of last resort but do not perceive it as a guarantor of an adequate living standard after retirement.

Even though we chose to use education as the primary factor that could explain the link between Trust in State and Pension-Saving Propensity, we also estimated a model using income as an additional variable.²⁴ The willingness and the potential to save for retirement are related to the income level (Garcia and Marques 2017; Rey-Ares, Fernández-López, and Milagros 2018). At the same time, earnings (income) have long been found to depend on education level (see Card 1999). In our model, this could mean that the level of education only seemingly conditions the analyzed relationship between Trust in State and saving for retirement, and income is a significant variable. Considering income as an additional variable in the model, we wanted to ensure that Trust in State does indeed affect Pension-Saving Propensity, and their relationship is not entirely explained by the fact that both these variables depend on the level of income. Therefore, we enriched the structural model with income as an additional variable influencing both Pension-Saving Propensity and Trust in State. Such a construction means that if the income level fully explained the relationship between Pension-Saving Propensity and Trust in State, the direct link between these variables would become statistically insignificant.

It is worth emphasizing that interpretation of this impact could depend on the definition of income. The respondent’s individual income better describes their personal status and aspirations. Regardless of an individual’s family situation, personal income reflects their position in the labor market and is usually considered a benchmark for comparisons with others in the workplace. However, the income per consumption unit in the household (expressed as equivalent income) better reflects the standard of living; this category considers both the earnings (income) of other household members and the number of people in that household. Therefore, the impacts of the two income categories on saving processes do not need to be the same.

This additional estimation changed the values of the load but did not change the conclusion. For the whole sample, the standardized load value for the relationship between Trust in State and Pension-Saving Propensity turns out to be −0.192 (p = .000) for the model with equivalent household income and −0.182 (p = .000) for the model with individual income. For both income definitions, this relationship remained negative and statistically significant for the subsample of less-educated respondents and completely insignificant for better-educated respondents. It confirms the impact of the level of education, regardless of the level of income. This observation is of key importance for formulating social policies and proposing incentives to save for retirement, as an educational policy is one of the areas of social policy that can be effectively shaped.

Conclusion

The starting point of our research was the hypothesis that peoples’ trust in help from the state might be one of the important factors influencing long-term planning and saving behavior. To address it, we operationalized “trust” by looking at all potential interpretations of this concept, which led us to distinguish four latent variables: Trust in State, Trust in Public Institutions, Satisfaction with Welfare State, and Distrust in Market.

Trust in State is our key variable. We sought to demonstrate that this deep trust variable, rarely analyzed in the literature, is an important component that affects peoples’ willingness to save for the future. We showed that it is a distinct variable from other trust-related concepts. It is not a frequently discussed, purely political trust, as our Trust in State attitude is independent of current government actions. Neither it is a type of stable trust in other society members because government is recognized as trustee in this trust relationship. Furthermore, it is unrelated to individuals’ sociodemographic characteristics and significantly affects the unwillingness to save among the less educated respondents. Here, we discuss the limits and potential consequences of our findings, as well as directions for further studies.

We have demonstrated that a substantial proportion of Polish society places confidence in the state. Interestingly, when listening to people’s open claims, one can judge the opposite. People claim not to count on the state regarding pensions and understand that they have to count on themselves. This is especially true of young people who declare that they do not count on any public pension at all. Surprisingly, and unfortunately, such open declarations do not lead to an increased saving behavior on their part. Our research provides one potential explanation: people say they do not trust the state, but, at a deep, potentially subconscious level, they do. They have lived their entire lives on the welfare support system, and they cannot even imagine living without one. Our estimations proved that Trust in State is also common in young people.

Although we put a lot of effort into reaching a representative sample of adult poles, our study has two limitations. First, when defining Trust in State, we designed it as a variable that measures people’s confidence that the state will not desert them. However, those left out (e.g., homeless people) were also not part of the questionnaire; as they have no permanent address, the pollster could not reach them. Those people might reveal a lower value of Trust in State, but as this group is not very large in Poland, we believe this effect plays a minor role in the study.

Second, the questionnaire was administered to Polish citizens, all of whom had lived their entire lives under welfare state conditions. It would be interesting to compare the results with research in countries where people have different experiences with state support.

As populations age, public retirement systems will no longer be as generous as before. Governments worldwide are looking for tools to encourage people to save for their future retirement voluntarily. This is also true of Poland, where a recent plan encouraged people to take part in Employee Capital Plans. This is a new third-pillar saving scheme, with state and employer subsidies and default enrollment, which should work as a participation booster (Chetty et al. 2014). Unfortunately, this plan seems to be failing. Instead of the 75% participation rate the government expected, it has reached only one-third of it. The government recognizes that low participation could result from distrust of government and a fear of nationalization and has taken steps to provide guarantees²⁵ intended to mitigate this distrust. However, our study demonstrates that there is yet another reason why people do not save, which, contrarily, is their Trust in State, especially among those relatively less educated who allow their latent trust to drive retirement-saving decision-making.

Our findings are pessimistic. As Trust in State is profound and subconscious, it would be challenging to diminish this trust let alone eradicate it. Nevertheless, the differences observed for the education level suggest that one could count on adequate education that teaches how the economy works and that the state is not omnipotent. However, this is a very long process. At this moment, it is futile to assume that (quasi)voluntary saving programs, with tax deductions, would substantially increase the propensity to save, especially among the relatively less educated who are also poorer. Therefore, a more promising strategy seems to be promoting retirement later rather than saving more. The government should create incentives to work longer, which can be a clear policy goal, but unfortunately, it requires strong political courage.

Compliance With Ethical Standards

The research has been accepted by SWPS University of Social Sciences and Humanities Research Ethics Committee (Dec. no. 02/P/09/2019). Survey respondents have been informed about the scope of the survey, voluntary participation, and the anonymization of the collected data.

Acknowledgments

We thank Agata Gąsiorowska for insightful comments on the measurement scales utilized in our research. All errors are our own.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are openly available at the Harvard Dataverse (https://doi.org/10.7910/DVN/IBLHDB).

Additional information

Funding

This work was supported by the National Science Centre (Poland) under Grant DEC-2017/25/B/HS4/00186.

Notes on contributors

Marek Kośny

Marek Kośny is an Associate Professor in the Faculty of Economics and Finance at Wroclaw University of Economics and Business. In his research, he concentrates on economic security and various aspects of distributional analysis. His recent publications include papers in Social Indicators Research, Post-Communist Economies, and the Research on Economic Inequality Series.

Radosław Kurach

Radosław Kurach is an Assistant Professor in the Faculty of Economics and Finance at Wroclaw University of Economics and Business. In his research, he concentrates on pension system design with a special focus on capital. His recent publications include papers in the Journal of Pension Economics and Finance, Applied Economics, and Applied Economics Letters.

Paweł Kuśmierczyk

Paweł Kuśmierczyk is an Associate Professor in the Faculty of Economics and Finance at Wroclaw University of Economics and Business. In his research, wok he concentrates on experimental economics and pension system design. His recent publications include papers in the Journal of Pension Economics and Finance, Applied Economics, and Applied Economics Letters.

Notes

1. The first pension of those who retire in a given year over the (economy-wide) average wage at retirement (Commission 2021).

2. The social minimum is one of the poverty thresholds officially used in Poland. It defines the income level that should enable a person to reproduce life forces and maintain social bonds.

3. The selection criterion was the fulfillment of the conditions for certification (ESOMAR and PKJPA, Polish certificates of quality in questionnaire research) and ensuring adequate quality of gathering data and strict verification of this process (post hoc). The sampling frame employed in the study is based on the TERYT database, used by Statistics Poland to conduct representative surveys. It resulted in an appropriate quality of the obtained data, including its representativeness of the adult Polish population.

4. This statistic measures the proportion of variance that might be common variance. The minimum value allowing for factor extraction was set at 0.5 (Kaiser 1974).

5. In Bartlett’s test of sphericity, we verify the null hypothesis stating that there are no significant correlations between the variables. In the case of a significant test result, this hypothesis is rejected; it is recognized that there are correlations between the variables, and defining latent constructs is justified. Therefore, we expect p-values close to 0.

6. With kappa = 4 as suggested by Hendrickson and White (1964).

7. CR is a measure of internal consistency reliability.

8. This is due to the relatively small number of items in individual constructs. See Peterson (1994) for a broader discussion.

9. Values of 0.95 and higher suggest redundancy and potential reduction in construct validity (Hair et al. 2019).

10. With complex models taking into account many items and latent variables, allowing for statistically significant correlations of some error terms is usually necessary; it is not possible to include all aspects of the analyzed relationships in the model. The results of the estimation of the model, in which all the error terms’ correlations were considered insignificantly different from zero, are (in the considered case) identical in terms of sign and statistical significance of regression weights, but they slightly differ in the values of loading. However, the model does not meet some of the quality of fit requirements.

11. McDonald and Bollen (1990) define the range (−1; 1) for both statistics but without strict justification for these values.

12. With the exception of variable Q24, the deviations were not large.

13. See Schreiber et al. (2006) for details of the cutoff criteria for fit indexes.

14. The literature gives rise to a measurement problem. Vickerstaff et al. (2012, 23) point out that even the wording of trust scales may lead to measurement errors. These scales usually use terms that are universally recognized as capturing trust so that trusting behavior gains a positive moral and social value, pushing up self-reported levels of trust.

15. In the entire third sector, public sources accounted for almost 43% of revenue in 2018 and donations and private contributions, less than 20% (Goś-Wójcicka 2020, 70).

16. Withdrawal is possible before reaching retirement age but results in loss of tax benefits.

17. These voluntary savings schemes are: Employee Pension Plans (PPE), Individual Retirement Accounts (IKE), Individual Retirement Protection Account (IKZE). Please see:

https://www.knf.gov.pl/knf/pl/komponenty/img/Oprac_IKE_IKZE_2019_67694.pdf; https://www.knf.gov.pl/knf/pl/komponenty/img/RAPORT_PPE_w_2017.pdf

18. The largest panel study on the quality of life of the Poles.

19. The only exception here is the Long-term Saving construct, for which the average variance extracted is a little below the assumed threshold of 0.5, meaning that the proposed items explain slightly below 50% of the variance of this latent variable; however, this is not a significant violation.

20. See OECD data, https://stats.oecd.org/Index.aspx?DataSetCode=EAG_EARNINGS

21. Having said that, as we wanted to make sure that our findings on the relationship between Trust in State and Pension-Saving Propensity are reliable, we did analyze separately the model with Income used as variable. The main results turned out to be robust. Some of the details are discussed in the next subsection.

22. For the subgroup of better-educated respondents, the standardized load was 0.120 (p = 0.070) and for the subgroup of other respondents: 0.003 (p = 0.947).

23. Standardized load for the whole sample is 0.446 (p = 0.000); for the less-educated and better-educated respondents, it is 0.406 (p = 0.000), and 0.540 (p = 0.000), respectively.

24. In the model, we added an observable variable “Income” and defined its relations to latent variables: Trust in State and Pension-Saving Propensity. As an income measure, we used equivalent monthly household income, applying the square root of the number of people in the household as the equivalence scale. Income ranges declared by the respondents were replaced by the arithmetic mean of their lower and upper bounds. The exception was the first class (below 2000–replaced with 2000) and the last class (above 15 000–replaced with 15 000). However, changing income definition to individual income does not alter the conclusions. Results of all these additional models are available upon request.

25. It recommended a direct guarantee in the Polish Constitution on the privacy of third-pillar assets.