The role of financial scarcity in the relationship between income and health behaviors: a causal mediation analysis

Abstract

Background

Financial scarcity is an increasing problem and health behaviors may be impaired by subjective experience of stress due to financial scarcity. Financial scarcity is a stress-response to having insufficient financial means and may affect cognitive functioning. This study aimed to assess whether financial scarcity mediates the association between income and health behaviors.

Methods

We used self-reported cross-sectional data of 2379 adults from the 2021 wave of the Dutch population-based GLOBE study. The exposure was equivalent household income, divided in quartiles. The mediator was financial scarcity, measured with the validated Psychological Inventory of Financial Scarcity (range 5–25). Outcomes were body mass index (BMI), fruit consumption, vegetable consumption, leisure time physical activity, and attempts to change health behavior in the past year, respectively: to (1) eat healthier, (2) increase physical activity and (3) lose body weight. Causal mediation analysis adjusted for eight covariables was used to decompose the total effect of income on the outcomes into a direct and indirect effect via financial scarcity as the mediator.

Results

Contrasting the lowest and highest income group showed the largest effects of income on health behaviors. We found a total effect of income on BMI (-1.13, 95% CI: -1.68;-0.56) and fruit consumption (155 g per week, 95% CI: 19.61;281.40), but not for leisure time physical activity, vegetable consumption, and attempts to lose weight, eat healthier and increase physical activity. Financial scarcity mediated the effect of income on BMI (natural indirect effect: − 0.42, 95% CI: − 0.64;-0.24), and fruit consumption (natural indirect effect: 41.78, 95% CI: 3.46;79.88), and also mediated the non-significant effects of income on leisure time physical activity, and whether a respondent reported an attempt to eat healthier, lose weight, and increase physical activity. Contrasting adjacent income groups showed smaller effects.

Conclusions

Financial scarcity mediated the effect of income on BMI and all health behaviors studied except vegetable consumption. Financial scarcity may trigger mechanisms that affect health behaviors, which are currently not addressed in health behavior interventions. Reducing financial scarcity may contribute to closing income inequalities in health behavior.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-24556-5.

Background

Income insecurity is an important contributor to health inequalities in EU countries [1]. In health inequality studies, income insecurity has been conceptualized and measured in different ways, often using measures such as a shortage of income or difficulties with balancing the income and fixed household expenses [2–4]. In general, income insecurity measures do not include the experience of financial scarcity, which is a stress-response to the financial situation. It has become evident that stress, related to economic constraints, has a large impact on health behaviors [5–16]. Moreover, through the lens of the capability approach, the strain of enduring financial difficulties may reduce people’s capability to pursue health goals [17].

Previous studies suggest that financial scarcity may partly explain why higher prevalence rates of unhealthy behaviors are observed among people living on low incomes [2, 3, 15, 18–21]. Examples of these behaviors are a less healthy diet [22–25], less leisure time physical activity [26, 27] and more difficulties in adopting and maintaining a healthy lifestyle [28, 29]. Financial scarcity, in addition to income, could help understand how income affects the individual capability to live a healthy life [30, 31].

Financial scarcity can be defined as the subjective experience of having insufficient financial means to make ends meet [32]. Financial scarcity entails stress-appraisals and stress-responses elicited by financial shortcomings [33]. Moreover, the scarcity theory states that financial scarcity increases cognitive constraints. According to the scarcity theory [32] financial scarcity could put pressure on the capability of pursuing and fulfilling a healthy life via two possible pathways. First, people experiencing financial scarcity worry a lot [34]. Worrying about finances may result in less mental capacity to exert cognitive functions which are needed for health behaviors [3]. Second, financial scarcity may increase a short-term focus [35]. Living in financial scarcity often means being burdened by acute stressors, such as difficulties paying bills and debts. Financial stressors constantly require attentional focus and consume energy, which may lead to self-control failures or motivational issues in goal-directed behavior [36]. A short-term focus on pressing financial demands may impede health behaviors, since the health benefits of these behaviors merely lie in the future, making them less tangible [16, 37].

Financial scarcity is highly prevalent in the Dutch adult population and financial worries are increasing [38]. The Dutch National Monitor of Financial Scarcity reports that 18% of the population experiences high financial scarcity, 47% experiences some financial scarcity and 36% does not experience financial scarcity [39]. Experiencing financial scarcity is not exclusively found among people with low incomes. It is also prevalent among people with middle- and high-income levels [39]. Financial scarcity is more prevalent among low- than high-income groups, therefore we suggest that the experience of financial scarcity may have a role in income inequalities in health behaviors. Furthermore, since financial scarcity is also prevalent among middle- and high-income groups, it may further enhance the understanding of individual capabilities related to a healthy lifestyle in the context of financial stress, regardless of income level [16].

This study aimed to test whether financial scarcity mediates the association between income and health behaviors and between income and self-reported attempts to change health behaviors. Causal mediation analysis, based on the potential outcomes framework and the counterfactual framework, is used to estimate effects that reflect the potential outcomes of hypothetically intervening in income or financial scarcity on health behaviors [40]. These potential outcomes may shed light on what is needed to improve support for people experiencing financial scarcity to reach health behavior goals. Specifically, we explored whether changing the income, while holding the value of financial scarcity fixed at a predetermined level, would affect health behaviors and whether targeting financial scarcity, while holding income fixed, would affect health behaviors.

Methods

Data

Study population

Self-reported data were used from the most recent wave of the Dutch cohort study GLOBE in 2021. GLOBE is a Dutch acronym for ‘Health and Living Conditions of the Population of Eindhoven and surroundings’. The longitudinal study aims at explaining socioeconomic health inequalities in the Netherlands through quantitative assessment of behavioral and psychological mechanisms and factors. The study design is described on the website (https://globe-study.nl) [41]. In 2021 a representative sample of the source population of the Eindhoven region (N = 7971) was invited to complete a postal survey, which could be returned by post or online. A total of 2379 participants, aged 25 to 98, provided informed consent and participated.

A robustness check of the results included income data from the previous wave in 2014. A power size calculation was performed for mediation analysis, based on the correlations between income and financial scarcity; income and BMI; and financial scarcity and BMI, respectively, with a target power of 0.8, resulting in a minimum N of 1300.

Measures

Income

Self-reported monthly income was measured in income categories (ranging from 1: 0 to 1200 euro per month, to 5: over 4000 euro per month). The income data were equalized by dividing the income levels by the square root of the number of people dependent on the income. The equivalent household income was then divided into 4 quartiles (lowest, middle lowest, middle highest and highest quartile) [3, 4]. We used the income quartiles, because these values (and the ensuing effects) would be easier to interpret. Furthermore, the income quartiles also allowed for logical and meaningful contrasts to analyze in the counterfactual approach in the mediation analyses.

Financial scarcity

The validated Psychological Inventory of Financial Scarcity scale (PIFS) measures the subjective experience of financial scarcity [33]. The scale consists of two stress appraisals: the experience of a shortage of money and a lack of control over the financial situation; and two responses to these appraisals: financial rumination and a short-term focus [33]. This validated scale has contributed to the understanding of how financial stress affects economical decision-making behavior [42, 43] and mental health [44]. It is not commonly used in studies on health behaviors. To our knowledge only one study used the PIFS as a predictor of dietary quality [12]. Measuring financial scarcity with the validated scale may therefore provide more insight into the mechanisms between income, financial scarcity and health behaviors.

We used the 5-item shortened version. Participants were asked to what extent they agreed on a 5-point Likert scale, ranging from ‘strongly agree’ to ‘strongly disagree’, with the following statements: ‘I often don’t have enough money’, ‘I am constantly wondering whether I have enough money’, ‘I often worry about money’, ‘I experience little control over my financial situation’ and ‘I am only focusing on what I have to pay at this moment, rather than my future expenses’. The potential range of the scores was 5 to 25. The lowest score of 5 indicated high financial scarcity, ranging to the highest score of 25, which indicated the absence of financial scarcity.

Health behaviors and attempts to change health behavior

We used body mass index (BMI) as a measure indicating a healthy or unhealthy weight, and as a health behavior related outcome. Prevention of overweight or obesity is part of national and international recommendations for a healthy lifestyle [45–47]. BMI was assessed by dividing self-reported weight (kg) by body height squared (m²). Scores below 15 and above 45 can be considered rare [48]. We included participants with BMI-scores between 15 and 50, since the National Monitor of the Dutch controlled lifestyle interventions reported that 6% of all Dutch participants had BMI-scores between 45 and 50 and there were no participants with a BMI-score above 50 [49]. In total, 27 outlier scores were excluded: 1 respondent with a BMI-score below 15 and 26 participants above 50.

Consumption of fruit and consumption of vegetables were assessed by calculating the total amount of fruit and vegetables in grams participants ate on average per week, with a maximum score of 14 kg of fruit and 14 kg of vegetables per week, considering higher values as unrealistic amounts. In total 12 scores of fruit consumption and 308 scores of vegetable consumption above 14 kg per week were excluded. Leisure time physical activity was assessed with the validated Short QUestionnaire to ASsess Health-enhancing physical activity (SQUASH) [50]. The total duration of physical activity in minutes participants reported per week was calculated over three categories: walking and bicycling to work, walking and bicycling in leisure time and sport activities. We calculated physical activity by summing the activities of middle and high intensity, indicating physical activities meeting recommendations for a healthy lifestyle.

Attempts to change health behavior were assessed by asking participants whether they had made attempts to eat healthier, lose weight, and increase physical activity in the past year. Attempts were specified by asking whether the attempts were under supervision or not. Answers were categorized in ‘no’, ‘yes, under supervision’, ‘yes, without supervision’ and ‘not applicable’. For statistical purposes we dichotomized the answers into ‘no’ and ‘yes’ (using ‘1’ as an indicator for an attempt), excluding the participants who answered ‘not applicable’ from analyses on the attempts (not applicable accounted for 11.8% of the scores on attempts to increase physical activity, 12.4% of attempts to eat healthier and 16.4% of attempts to lose weight).

Potential confounders

Potential confounders were gender (male or female), age (continuous), level of education in ISCED-categories (low, medium or high), employment status (employed, not-employed, retired, or non-employed (for example students, homemakers, volunteers)), country of birth (the Netherlands or elsewhere), perception of general health condition (excellent, very good, good, moderate, bad), living together with a partner (yes or no) and having children living at home (yes or no). In the model including the outcome attempts to lose weight and eat healthier, obesity (BMI of 30 or higher, 13.1% of participants) was also included as a possible confounder because BMI was associated with attempts to lose weight and attempts to eat healthier.

Statistical procedure and analysis

Data was considered missing when the question was unanswered or when multiple answers were given. The question concerning gender allowed for the option ‘other gender’ which 7 participants chose. These 7 participants were excluded before imputation. The variables ‘income’ and ‘having children living at home’ had relatively many missing values (respectively 16.6% and 13.9%). Other variables had up to 3.8% missing values. Missing values on the exposure, mediator and confounders were imputed by multiple imputation (m = 25). Cases with missing values on the outcome variables were excluded from the analyses. The imputed dataset was pooled according to the bar procedure [51], resulting in a single datafile based on the average or mode values of the multiple imputed values, retaining the between sample variability. This allowed us to perform the mediation analysis. The bar procedure has been used in other studies [52–54].

This study reports the effects of income on seven health behaviors. Therefore, we have seven separate models. Four models include continuous outcomes (BMI, physical activity, fruit consumption and vegetable consumption). Three models include dichotomous outcomes (the three attempts to eat healthier, lose weight and increase physical activity). In the mediation models we estimated the effects of income on health behaviors, direct and indirect via financial scarcity.

In causal mediation analysis the contrast of interest of the independent variable can be selected. We considered the contrast between the lowest income quartile and the highest income quartile of interest, since this is the largest contrast possible, and potential outcomes could be informative since they indicate the largest potential effect possible. We also considered the smaller contrasts between income quartiles of interest, since they provide insights that may be used for more actionable policy scenarios when intervening on income or financial scarcity with the objective to increase the capability to obtain a healthy lifestyle. Therefore, the four contrasts of interest which were analyzed were the contrasts between the income quartiles in consecutive steps: the lowest income group was compared to the middle lowest income group, and consequently, the middle lowest to the middle highest and the middle highest to the highest income group; and the largest contrast possible: the lowest and the highest income group. The effects of changing income on health behaviors for all possible contrasts between income groups are reported in Additional file 1.

Figure 1 shows the presumed relationships between income (the exposure), financial scarcity (the mediator) and health behaviors (the outcomes). Not represented in the figure are the potential confounders.

Fig. 1.

Conceptual model of presumed relations between income (exposure), financial scarcity (mediator) and health behaviors (outcomes)

We first explored the associations between income and financial scarcity with descriptive statistics, Pearson correlations and analyses of variance to compare mean scores. Then, in seven separate mediation models, we tested whether income was related to health behaviors and whether financial scarcity mediated the associations between income and health behaviors.

We performed causal mediation analysis, based on the potential outcomes framework and counterfactual framework, introduced by VanderWeele en Vansteelandt [55]. Causal mediation can provide a more profound explanation of presumed causal mechanisms and is the preferred approach in models with dichotomous outcomes [56–58]. As in traditional mediation analysis, causal mediation makes it possible to separate the total effect into a direct effect of income on health behaviors, and an indirect effect, through financial scarcity. Different from traditional mediation analysis, causal mediation defines causal effects as the difference between two potential outcomes, meaning the values that would be observed under a certain exposure or intervention level [59].

Traditional mediation analysis cannot be used in models with dichotomous outcomes based on logistic regression with a log linear function, because the total effect and the indirect effect of the traditional regression approach represent effects that include the non-collapsibility effect [58, 59]. In causal mediation, the natural indirect effect (NIE) no longer represents the non-collapsibility effect which therefore no longer leads to bias. For uniformity in the interpretation, we used causal mediation analyses in all our models, also the models including the continuous outcomes, even though this was not strictly necessary1.

Mediation analysis was conducted using linear regression in the four models with continuous outcomes and with linear regression combined with log linear regression in the three models with dichotomous outcomes. Linear and logistic regression require that several assumptions about linearity, normality, independence and homoscedasticity are met. Explorations of the assumptions have been conducted, and we considered our statistical approach appropriate. In the first four models including the continuous outcomes, we tested for exposure-mediator interactions, which were non-significant. In the models with the dichotomous outcomes, a possible exposure-mediator interaction factor was included.

Causal mediation analysis requires four additional assumptions about no (unmeasured) confounding, to allow for a causal interpretation of the effects [55]. The four assumptions require that there is no unmeasured confounding of the exposure-outcome relation; mediator-outcome relation; exposure-mediator relation; and that there are no confounders of the mediator-outcome relation that are affected by the exposure [60]. The four additional assumptions about non-confounding cannot be tested statistically but must be justified through theory [60]. In all studied relationships, there can be confounding we cannot account for. Therefore, we cannot be certain about the causality of the relationships between income and health behaviors and between financial scarcity and health behaviors. In all models, we adjusted for potential confounders by including them as covariables. By including eight confounders we presumed that unmeasured confounding is limited as much as possible.

Our analysis is conducted on cross-sectional data, which can only show associations between variables. However, in all models a temporal precedence was assumed. The exposure income precedes the experience of the mediator financial scarcity [61, 62]. Financial scarcity was measured with retrospective items indicating its enduring presence. Empirical evidence indicates that financial scarcity precedes health behaviors [3, 63, 64]. Cognitive and physical mechanisms have been put forward as explanations for why people experiencing financial stress might have difficulties in maintaining a healthy diet [9, 12, 36, 65, 66]. Since financial scarcity is a representation of stress appraisals and stress responses concerning one’s financial situation, it is unlikely that the health behavior outcomes will influence the experience of financial scarcity.

A robustness analysis with the income data from the previous wave of the GLOBE study (in 2014), supports the presumption of a temporal order of income and health behaviors. All robustness analysis and results are described in Additional file 2.

We followed the AGReMA-guidelines for reporting mediation analysis [40] and the recommended steps in causal mediation analysis of Rijnhart within epidemiological research [59]. All analyses were performed with SPSS (IBM, version 28.0: ref IBM Corp. Released 2021. IBM SPSS Statistics for Windows, Version 28.0. Armonk, NY: IBM Corp). For all mediation analyses the macro of Valeri and VanderWeele [67] was used.

Interpretation of the effects

For the purpose of interpretation of the causal mediation effects reported in the current study, the effects of interest are described here. The total effect (TE) shows how much health behaviors would change on average if all in the lowest income quartile would get an income in the highest income quartile. This effect may occur through both the path of income to health behaviors directly and the path of income to health behaviors through financial scarcity. This effect may occur though the NIE and the NDE (TE = NIE + NDE).

The natural indirect effect (NIE) shows how much the outcome would change on average if we change everyone’s financial scarcity score as if in response to changing the income from everyone from the lowest income quartile to the highest income quartile, while actually not letting income change (so there is no direct effect). This can be understood as intervening on the mediator such that those in the lowest income groups would get the values of financial scarcity belonging to (‘naturally seen in’) those in the highest income quartile. The total natural indirect effect (TNIE) is a specific type of the NIE and is the effect reported in the models which include possible exposure-mediator interaction; in our study the models with dichotomous outcomes (the attempts to lose weight, increase physical activity and eat healthier).

The natural direct effect (NDE) shows how much the outcome would change on average if we change everyone’s income from the lowest income quartile to the highest income quartile, while not letting financial scarcity change in response to the change in income. This can be understood as changing income levels in a situation where all would have the same value for the mediator, where there is no relationship between income and financial scarcity, and as a result all changes in the outcome are a direct consequence of the changing income. The pure natural direct effect (PNDE) is a specific type of the NDE and is the effect reported in the models which include possible exposure-mediator interaction; in our study the models with dichotomous outcomes (the attempts to lose weight, increase physical activity and eat healthier).

The controlled direct effect (CDE) shows how much the outcome would change on average if we set everyone’s financial scarcity score on 25, indicating the absence of financial scarcity and we change the income from low to high. The CDE informs about the hypothetical question: If we were able to reduce financial scarcity to a minimum for everyone, would increasing the income still have an effect on health behaviors? The CDE are reported when discussing the results of the models with the dichotomous outcomes (the attempts to lose weight, increase physical activity and eat healthier), since the CDE is deviant from the NDE in the models which include possible exposure-mediator interaction. The CDE and NDE are identical in the models with continuous outcomes (BMI, fruit consumption, vegetable consumption and physical activity).

Results

Descriptive results

Participant characteristics

The study included 2379 participants. Of the study sample, 54.0% were females. The mean age was 54.3 years (SD 18.1). The mean BMI was 25.7 (SD 4.9). Participants on average reported being physically active slightly over 8 h per week. The mean amount of fruits and vegetables consumed was 1 kg per week for fruits (SD 0.9) and 1.2 kg per week for vegetables (SD 1.4). The majority of the participants reported having made attempts to eat healthier (74.7%), attempts to increase physical activity (75.7%) and attempts to lose weight (55.8%).

Table 1 shows the overall descriptive non-imputed statistics for each variable in percentages and mean scores of the total study population.

Table 1.

Characteristics of the study population, before imputation

	Total study population (N = 2379)
	Valid percentages or Mean (SD)	Number
Descriptives
Gender (0.5% missing)
Female	54.0%	1279
Male	46.0%	1089
Age (1.0% missing)	54.3 (18.1)
25–34	19.1%	450
35–44	16.7%	393
45–54	14.5%	342
55–64	14.9%	351
65–74	17.7%	418
74-eldest	17.0%	401
Education (2.6% missing)
Low (ISCED 1–4)	23.2%	537
Middle (ISCED 5–6)	23.5%	544
High (ISCED 7–8)	53.3%	1235
Country of birth (1.5% missing)
Netherlands	88.2%	2067
Other than the Netherlands	11.8%	277
Employment (3.8% missing)
Employed	58.5%	1339
Unemployed	26.8%	614
Retired	8.0%	183
Non-employed	6.6%	152
Perceived health (sore range 1–5, mean + SD, 3.2% missing)	2.7 (0.9)
Excellent	10.5%	241
Very good	27.0%	622
Good	45.0%	1036
Moderate	15.4%	354
Bad	2.2%	50
Having children living at home (13.9% missing)
Yes	33.1%	678
No	66.9%	1370
Living together with partner (1.7% missing)
Yes	69.8%	1632
No	30.2%	706
Financial scarcity (9.7% missing)
Financial scarcity	20.4 (4.2)
No financial scarcity (score 21–25)	49.4%	1061
Some financial scarcity (11–20)	47.8%	1027
High financial scarcity (5–10)	2.8%	60
Health behaviors
BMI (2.1% missing)	25.7 (4.9)	2316
Physical activity (in minutes per week, 15.8% 0-scores, 0% missing)	484.6 (676.0)	2379
Fruit consumption (in kilograms per week, 2.4% 0-scores, 19.8% missing)	1.01 (0.9)	1908
Vegetable consumption (in kilograms per week, 0.4% 0-scores, 32.2% missing)	1.24 (1.4)	1614
Attempt to eat healthier (19.0% not applicable or missing)
Yes	74.4%	1433
No	25.6%	494
Attempt to increase physical activity (17.2% not applicable or missing)
Yes	75.7%	1491
No	24.3%	479
Attempt to lose weight (22.2% not applicable or missing)
Yes	55.8%	1032
No	44.2%	818

Income and financial scarcity were moderately associated (r =.39, p <.001). Of all participants, 49.4% reported having no financial scarcity (score 21–25), 47.8% reported having some financial scarcity (score 11–20) and 2.8% reported having high financial scarcity (score 5–10). Of all households in the lowest household income quartile 8.5% reported high financial scarcity, the average score in this income group was 17.9 (SD 4.8). In the highest household income quartile 0.8% reported high financial scarcity, the average score in this income group was 21.7 (SD 3.4). Of all participants with high financial scarcity, 75.9% fell in the lowest income quartile. Mean financial scarcity scores varied significantly between income quartiles (F = 117.9, p <.001). A considerable proportion of the participants reported some financial scarcity, ranging from 64.9% in the lowest income quartile to 35.2% in the highest income quartile.

Results of mediation analysis

The effects reported for all health behavior outcomes are the total, natural indirect and natural direct effects. The controlled direct effects are reported solely for the attempts to change health behaviors, as these effects are deviant from the natural direct effects in the models with the dichotomous outcome variables [57–59].

Table 2 shows the results of the mediation analysis of the models with BMI, fruit consumption, vegetable consumption and physical activity (the continuous outcomes). The number of participants included in the mediation analyses varied by outcome variable (specific numbers are reported in Tables 2 and 3).

Table 2.

Effects of income on BMI, physical activity, fruit and vegetable consumption

Analyzed contrasts of quartiles of equalized household income in quartiles (EIQ)	Total effect (TE)	Natural direct effect (NDE)	Natural indirect effect (NIE)
BMI (n = 2316)
Lowest EIQ – middle lowest EIQ	− 0.37 (−0.56;−0.18)	− 0.23 (−0.42;−0.04)	− 0.14 (−0.20;−0.08)
Middle lowest EIQ – middle highest EIQ	− 0.38 (−0.57;−0.18)	− 0.24 (−0.44;−0.04)	− 0.14 (−0.21;−0.08)
Middle highest EIQ – highest EIQ	− 0.38 (−0.56;−0.18)	− 0.23 (−0.43;−0.04)	− 0.14 (−0.20;−0.08)
Lowest EIQ – highest EIQ	−1.13 (−1.68;−0.56)	− 0.71 (−1.30;−0.14)	− 0.42 (−0.64;−0.24)
Leisure time physical activity, in minutes per week (n = 2379)
Lowest EIQ – middle lowest EIQ	5.86 (−29.28;36.88)	17.80 (−16.34;47.98)	−11.95 (−19.99;−4.45)
Middle lowest EIQ – middle highest EIQ	6.45 (−29.73;38.87)	18.32 (−16.19;48.88)	−11.88 (−19.75;−4.48)
Middle highest EIQ – highest EIQ	6.34 (−30.40;37.87)	18.38 (−16.94;49.21)	−12.04 (−20.67;−3.74)
Lowest EIQ – highest EIQ	19.50 (−93.15;108.05)	55.87 (−46.63;144.53)	−36.37 (−59.88;−14.91)
Fruit consumption, in grammes per week (n = 1908)
Lowest EIQ – middle lowest EIQ	50.36 (9.25;90.66)	36.20 (−7.02;77.20)	14.16 (1.74;28.87)
Middle lowest EIQ – middle highest EIQ	50.94 (9.50;93.31)	37.29 (−7.95;78.00)	13.65 (1.24;26.88)
Middle highest EIQ – highest EIQ	51.96 (8.31;91.34)	37.93 (−6.75;77.90)	14.02 (1.63;27.17)
Lowest EIQ – highest EIQ	155.19 (19.61;281.40)	113.41 (−18.49;232.25)	41.78 (3.46;79.88)
Vegetable consumption, in grammes per week (n = 1614)
Lowest EIQ – middle lowest EIQ	−3.38 (−87.31;79.67)	1.00 (−87.95;90.73)	−4.38 (−25.95;15.03)
Middle lowest EIQ – middle highest EIQ	−4.33 (−94.76;78.51)	0.23 (−91.91;90.52)	−4.56 (−25.43;15.57)
Middle highest EIQ – highest EIQ	−5.07 (−88.06;76.68)	− 0.57 (−84.52;83.18)	−4.50 (−26.63;15.96)
Lowest EIQ – highest EIQ	0.00 (−0.03;0.03)a	− 0.00 (−0.03;0.03)a	0.00 (−0.00;0.01)a

*Causal mediation analysis, controlling for all confounders. Unadjusted effects. Between brackets bootstrap confidence intervals 95%. Significant results are bold. TE = NDE + NIE 

^a Effects are 10 to the power of 4.

Table 3.

Effects of income on attempts to eat healthier, increase physical activity and lose weight

Analyzed contrasts of quartiles of equalized household income in quartiles (EIQ)	Total effect (TE)	Pure natural direct effect (PNDE)	Total natural indirect effect (TNIE)	Controlled direct effect (CDE)
Attempt to eat healthiera (n = 1897)
Lowest EIQ – middle lowest EIQ	1.04 (0.96;1.13)	0.99 (0.91;1.08)	1.05 (1.02;1.08)	1.01 (0.91;1.12)
Middle lowest EIQ – middle highest EIQ	1.05 (0.95;1.17)	1.00 (0.91;1.10)	1.05 (1.02;1.09)	1.01 (0.91;1.13)
Middle highest EIQ – highest EIQ	1.07 (0.95;1.21)	1.01 (0.92;1.12)	1.06 (1.01;1.12)	1.01 (0.91;1.12)
Lowest EIQ – highest EIQ	1.19 (0.88;1.57)	0.99 (0.75;1.28)	1.20 (1.06;1.39)	1.04 (0.74;1.42)
Attempt to increase physical activity (n = 1970)
Lowest EIQ – middle lowest EIQ	1.07 (0.98;1.17)	1.05 (0.96;1.16)	1.02 (1.00; 1.06)	1.09 (0.97;1.23)
Middle lowest EIQ – middle highest EIQ	1.11 (1.00;1.24)	1.07 (0.96;1.19)	1.04 (1.01;1.08)	1.10 (0.97;1.23)
Middle highest EIQ – highest EIQ	1.15 (1.01;1.32)	1.09 (0.97;1.23)	1.06 (1.00;1.11)	1.09 (0.97;1.23)
Lowest EIQ – highest EIQ	1.38 (0.99;1.93)	1.17 (0.86;1.59)	1.18 (1.01;1.37)	1.31 (0.89;1.88)
Attempt to lose weighta (n = 1822)
Lowest EIQ – middle lowest EIQ	1.04 (0.98;1.11)	1.01 (0.95;1.08)	1.02 (1.00;1.04)	1.03 (0.95;1.11)
Middle lowest EIQ – middle highest EIQ	1.05 (0.97;1.13)	1.02 (0.95;1.08)	1.03 (1.01;1.05)	1.02 (0.95;1.11)
Middle highest EIQ – highest EIQ	1.06 (0.97;1.16)	1.02 (0.95;1.10)	1.04 (1.01;1.07)	1.02 (0.95;1.11)
Lowest EIQ – highest EIQ	1.15 (0.94;1.42)	1.04 (0.85;1.25)	1.11 (1.02;1.20)	1.08 (0.85;1.36)

*Causal mediation analysis, controlling for all confounders. Unadjusted effects. Between brackets bootstrap confidence intervals 95%. Significant results are bold.

^a Controlled for all confounders plus obesity.

Table 3 shows the results of the mediation analysis of financial scarcity on the relations between income and attempts to eat healthier, increase physical activity and lose weight (the dichotomous outcomes).

Total effects of income on health behaviors

Income was negatively associated with BMI (lowest EIQ compared to middle lowest EIQ: total effect (TE): − 0.37, 95% CI: − 0.56;−0.18) and positively associated with fruit consumption (lowest EIQ compared to middle lowest EIQ: TE: 50.36 g per week, 95% CI: 9.25;90.66). Income was not significantly associated with physical activity (lowest EIQ compared to middle lowest EIQ: TE: 5.86 min per week, 95% CI: −29.28;36.88), vegetable consumption (lowest EIQ compared to middle lowest EIQ: TE: −3.38 g per week, 95% CI: − 0.87.31;79.67), attempts to lose weight (lowest EIQ compared to middle lowest EIQ: TE: 1.04, 95% CI: 0.98;1.11), attempts to eat healthier (lowest EIQ compared to middle lowest EIQ: TE: 1.04, 95% CI: 0.96;1.13) and attempts to increase physical activity (lowest EIQ compared to middle lowest EIQ: TE: 1.07, 95% CI: 0.98;1.17). Total effects of income on health behaviors were consistent in size when comparing other income groups.

The largest possible contrast between the lowest and the highest income groups showed similar effects and larger effect sizes. Income was negatively associated with BMI (lowest EIQ compared to highest EIQ: total effect (TE): −1.13, 95% CI: −1.68;−0.56) and positively associated with fruit consumption (lowest EIQ compared to highest EIQ: TE: 155.19 g per week, 95% CI: 19.61;281.40). Income was not significantly associated with physical activity (lowest EIQ compared to highest EIQ: TE: 19.50 min per week, 95% CI: −93.15;108.05), vegetable consumption (lowest EIQ compared to highest EIQ: TE: 0.00 g per week, 95% CI: − 0.03;0.03), attempts to lose weight (lowest EIQ compared to highest EIQ: TE: 1.15, 95%BCI: 0.94;1.42), attempts to eat healthier (lowest EIQ compared to highest EIQ: TE: 1.19, 95% CI: 0.88;1.57) and attempts to increase physical activity (lowest EIQ compared to highest EIQ: TE: 1.38, 95% CI: 0.99;1.93).

Indirect effects of financial scarcity on health behaviors

Natural indirect effects (NIE) were found on the associations between income and BMI (lowest EIQ compared to middle lowest EIQ: NIE: − 0.14, 95% CI: − 0.20;−0.08), fruit consumption (lowest EIQ compared to middle lowest EIQ: NIE: 14.16, 95% CI: 1.74;28.87) and physical activity (lowest EIQ compared to middle lowest EIQ: NIE: −11.95, 95% CI: −19.99;−4.45). Non-significant was the mediation of financial scarcity on the association between income and vegetable consumption (lowest EIQ compared to middle lowest EIQ: NIE: −4.38, 95% CI: −25.95;15.03). Financial scarcity significantly mediated the associations between income and attempts to eat healthier (lowest EIQ compared to middle lowest EIQ: TNIE: 1.05, 95% CI: 1.02;1.08), attempts to lose weight (lowest EIQ compared to middle lowest EIQ: TNIE: 1.02, 95% CI: 1.00;1.04). Financial scarcity did not significantly mediate the association between income and attempts to increase physical activity (lowest EIQ compared to middle lowest EIQ: TNIE: 1.02, 95% CI: 1.00;1.06). Natural indirect effects of financial scarcity on the association between income on health behaviors were consistent in size when comparing other income groups. The indirect effects of financial scarcity on the association between income and attempts to increase physical activity were significant when comparing other income groups.

The largest possible contrast between the lowest and the highest income groups showed similar effects and larger effect sizes. Natural indirect effects (NIE) were found on the associations between income and BMI (lowest EIQ compared to highest EIQ: NIE: − 0.42, 95% CI: − 0.64;−0.24), fruit consumption (lowest EIQ compared to highest EIQ: NIE: 41.78, 95% CI: 3.46;79.88) and physical activity (lowest EIQ compared to highest EIQ: NIE: −36.37, 95% CI: −59.88;−14.91). Non-significant was the mediation of financial scarcity on the association between income and vegetable consumption (lowest EIQ compared to highest EIQ: NIE: 0.00, 95% CI: − 0.00;0.01). Financial scarcity significantly mediated the associations between income and attempts to eat healthier (lowest EIQ compared to highest EIQ: TNIE: 1.20, 95% CI: 1.06;1.39), attempts to lose weight (lowest EIQ compared to highest EIQ: TNIE: 1.11, 95% CI: 1.02;1.20), and attempts to increase physical activity (lowest EIQ compared to highest EIQ: TNIE: 1.18, 95% CI: 1.01;1.37).

Direct effects of income on health behaviors

Income had a significant natural direct effect (NDE) on BMI (lowest EIQ compared to middle lowest EIQ: NDE: − 0.23, 95% CI: − 0.42;−0.04). The NDE of income on physical activity (lowest EIQ compared to middle lowest EIQ: NDE: 17.80, 95% CI: −16.34;47.98) and fruit consumption (lowest EIQ compared to middle lowest EIQ: NDE: 36.20, 95% CI: −7.02;77.20) were larger than the natural indirect effects. However, the natural direct effects were non-significant. The NDE of income on vegetable consumption was non-significant (lowest EIQ compared to middle lowest EIQ: NDE: 1.00, 95% CI:−87.95;90.73).

The controlled direct effect (CDE) of income on attempts to increase physical activity (lowest EIQ compared to middle lowest EIQ: CDE: 1.09, 95% CI: 0.97;1.23) was larger than the indirect effect but was non-significant. The controlled direct effects of income on attempts to eat healthier (lowest EIQ compared to middle lowest EIQ: CDE: 1.01, 95% CI: 0.91;1.12) and attempts to lose weight (lowest EIQ compared to middle lowest EIQ: CDE: 1.03, 95% CI: 0.95;1.11) were non-significant.

The largest possible contrast between the lowest and the highest income groups showed similar effects and larger effect sizes. The natural direct effects (NDE) of income on BMI (lowest EIQ compared to highest EIQ: NDE: − 0.71, 95% CI: −1.30;−0.14), physical activity (lowest EIQ compared to highest EIQ: NDE: 55.87, 95% CI: −46.63;144.53) and fruit consumption (lowest EIQ compared to highest EIQ: NDE: 113.41, 95% CI: −18.49;232.25) were of the same size or larger than the natural indirect effects. However, only the natural direct effect of income on BMI was significant. The controlled direct effect (CDE) of income on attempts to increase physical activity (lowest EIQ compared to highest EIQ: CDE: 1.31, 95% CI: 0.89;1.88) was larger than the indirect effect but was non-significant. The controlled direct effects of income on attempts to eat healthier (lowest EIQ compared to highest EIQ: CDE: 1.04, 95% CI: 0.74;1.42) and attempts to lose weight (lowest EIQ compared to highest EIQ: CDE: 1.08, 95% CI: 0.85;1.36) were non-significant.

Discussion

Income was negatively associated with BMI and positively associated with fruit consumption, while no significant total effects were found for other health behaviors. Income only had a natural direct effect on BMI. The absence of natural direct effects on the other health behaviors suggests that, when financial scarcity was held constant, changing income alone did not affect fruit consumption, vegetable consumption and physical activity. Financial scarcity mediated the relationships between income and BMI, fruit consumption, physical activity, but not vegetable consumption.

Notably, the majority of the participants reported having made attempts to eat healthier, lose weight and increase physical activity. Income itself had no direct effect on these attempts. Overall, between different income groups financial scarcity mediated the non-significant associations between income and the attempts to eat healthier, lose weight and increase physical activity. The natural indirect effects exceeded 1, indicating that reductions in financial scarcity, rather than income per se, were associated with a greater likelihood of engaging in health behavior changes.

Interpretation of results

The effect sizes were small but consistent across the different income contrasts that were compared. In additional mediation analyses all possible contrasts were compared, and results were robust, which supports our findings. Furthermore, the effects of changing income from the lowest income group to the middle highest and highest income group gradually increased in size as we expected (see Additional file 1). Since non-confounding cannot be ruled out, the effects may not represent a causal relationship.

Financial scarcity could be a potentially relevant factor in explaining how income affects health behaviors and attempts to change health behavior, as has been suggested in other studies [2, 12, 63, 68, 69]. The absence of direct effects of income on fruit consumption, vegetable consumption and physical activity may suggest that an increase in income does not have a considerable effect on health behaviors, when financial scarcity does not change accordingly. These results are in line with previous findings which suggest that prolonged financial stress predicts health behaviors independent of income [3, 70].

Contrary to previous findings [71, 72], our findings suggested that a decrease in financial scarcity would lead to less physical activity when income was held constant. In this study physical activity was not limited to sport activities but included work related transport walking and bicycling and leisure time walking and bicycling. Additional mediation analyses in which physical activity was subdivided between time spent on sport activities, work related transport walking and bicycling and leisure time walking and bicycling showed significant indirect effects of financial scarcity on the relationship between income and leisure time walking and bicycling and work-related walking and bicycling [see Additional file 3]. A possible explanation could be that people experiencing financial stress spend relatively more time walking and bicycling as a coping mechanism, an explanation that was also given in a literature review in which 18% of the prospective studies showed a positive relationship between stress and physical activity [71].

Income was not associated with vegetable consumption. Even though this result was contrary to our expectations, this finding is not exceptional [73]. A possible explanation could be that the measurement of the vegetable consumption in spoons has validity issues [74]. Vegetable consumption was measured by asking participants about the amount of vegetables by the number of serving spoons they used on a day they consumed vegetables. People could consistently overestimate the number of serving spoons. The relatively large number of values we excluded from the data-analyses (308, versus 12 in the variable fruit consumption) because they were considered unrealistic, could be an indication of overestimation resulting in measurement errors.

Limitations

This study is subject to limitations which could have introduced sources of bias affecting the outcomes. We name four possible sources of bias: (1) categorization of the income variable; (2) the skewed distribution of financial scarcity may have limited the power to detect effects; (3) we cannot rule out the possibility of reverse causality and the effect of unmeasured confounding; and (4) the study population does not fully represent the Dutch population, limiting its generalizability.

First, by categorizing the income variable, the statistical power to detect associations with financial scarcity and health behaviors was reduced. The cutoff points of quartiles do not reflect meaningful distinctions between income categories, such as below or above poverty level, which makes the interpretation less intuitive. Furthermore, we carried out additional robustness analyses using non-imputed income data (non-imputed equalized household income in quartiles). We have included the results in an additional file (nr. 2). We found one deviance in the effects, as compared to the results using the imputed income data. The indirect effects of financial scarcity on the associations between income and attempts to lose weight were non-significant.

Second, within income quartile variation was not visible, which may have biased the results. Financial scarcity was negatively skewed. It is possible that across the range of financial scarcity, the effect of experiencing high financial scarcity on the outcomes is larger than experiencing low financial scarcity, representing a floor effect. Bootstrapping confidence intervals were used to establish the most accurate estimates.

Third, cross-sectional data is not preferred in mediation analysis because it does not rule out the possibility of reverse causation. However, we considered the current approach the best appropriate option, since the temporal precedence of our model could be established. We rerun the models in separate analyses with income data from the previous wave of GLOBE from 2014. The indirect effects of financial scarcity on the association between income and all health behaviors were comparable, except for a non-significant indirect effect of financial scarcity on the association between income and attempts to lose weight. Confounding cannot be ruled out entirely. Variables which were not included in the data collection, such as chronic stress, may affect the experience of financial scarcity as well as health behavior outcomes. When variables such as chronic stress are measured and included as confounders, the indirect effect might be attenuated. Other variables were included in the models, such as general health, but could be defined as a confounder or a collider, since the perception of general health could be the consequence of having a low income and experiencing financial scarcity. Controlling for education may explain the absence of total effects of income on health behaviors by ruling out possible factors associated with education such as health literacy. We rerun the models without controlling for general health and education, which did not lead to other results [see Additional file 2].

Fourth, the results cannot be generalized to the general Dutch population. Overall, the study population is, on average, older than the Dutch population (respectively 54,3 and 42,6 years [75]). Furthermore, a relatively large part of the study population had a high educational level. This could explain that a relatively small part of the study population experienced high financial scarcity (2.8%) in comparison to the Dutch population (18%). Among the Dutch population, the prevalence of high financial scarcity is lower among the 55 + population (12%) than among the younger population [76]. Other research using the PIFS also shows that financial scarcity is negatively associated with age, education and income [33]. A possible explanation for the underrepresentation of people experiencing high financial scarcity is that people experiencing high financial scarcity are less likely to participate in a questionnaire study. If, in our study population, financial scarcity scores were more in line with the prevalence in the Dutch population, the indirect effects could have been larger.

Implications for future research and practice

This study focused on effects as the result of variations between income quartiles. Future research could use non-categorized income data, could explore effects as the result of more intuitive contrasts between income groups, such as incomes below and above poverty level and could explore how financial scarcity may differentially affect health behaviors within income groups. This information, especially when combined, may have greater practical value for guiding interventions and shaping policy.

The results of this study also show that health behavior interventions for populations with low incomes may become more effective when also considering financial scarcity as a disruptive factor in acquiring and maintaining a healthy lifestyle. Intervening on financial scarcity may therefore be helpful. Further research could contribute to the understanding of the cognitive mechanisms through which income and financial scarcity may affect the capability to engage in health behaviors. Empirical research that tests the assumptions made in scarcity theory about the associations between financial scarcity and economical decision-making is desirable [77] and we suggest these mechanisms should also be examined in the field of health behavior research. Experiencing financial stress may disrupt the cognitive capacity required for health behaviors [36]. The capability of living a healthy life may be attenuated in situations when budgets are limited and when there are many distracting temptations in the living environment. Possible factors to be explored are the effects of financial scarcity on sense of control and time orientation [33, 42], which are known to affect health behaviors [78–81]. Furthermore, we suggest that chronic stress should be measured to disentangle the possible interaction between and sum of effects of all sources of financial stress on health behaviors.

Conclusions

We investigated the effects of income on BMI, fruit consumption, vegetable consumption, physical activity and attempts to eat healthier, lose weight and increase physical activity. Causal mediation analysis tested whether these effects were mediated by financial scarcity. Income was associated with BMI and fruit consumption, but not with the other outcomes. Comparing different income groups, financial scarcity mediated the associations between income and BMI, fruit consumption, physical activity and attempts to eat healthier, to lose weight and increase physical activity, but not vegetable consumption. These results suggest that an increase in income does not have a considerable effect on health behaviors, when financial scarcity does not change accordingly. Financial scarcity may trigger mechanisms currently overseen that affect the capability to engage in health behaviors, through stress-appraisals and stress-responses. Intervening in income in conjunction with reducing financial scarcity could potentially contribute to health behavior. Further insights into these mechanisms may contribute to reducing income inequalities in health behavior.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AGReMA

A Guideline for Reporting Mediation Analysis

BMI

Body Mass Index

CDE

Controlled Direct Effect

GLOBE

Dutch Acronym for Health and Living Conditions of the Population of Eindhoven and surroundings

ISCED

International Standard Classification of Education

NDE

Natural Direct Effect

NIE

Natural Indirect Effect

PIFS

Psychological Inventory of Financial Scarcity

PNDE

Pure Natural Direct Effect

SPSS

Statistical Package for Social Sciences

SQUASH

Short QUestionnaire to ASsess Health-enhancing physical activity

TE

Total Effect

TNIE

Total Natural Indirect Effect

Authors’ contributions

AvdV conceptualized the study, developed the causal model and study design, performed the mediation analyses, analyzed and interpreted the results, wrote the draft of the manuscript and approved the final version of the manuscript. FJvL and CBMK led the data collection and provided the data used in the study and provided access to the secured research environment where the GLOBE-data is hosted. FJvL and TM made contributions to the causal model and study design. TM reviewed the operationalization of the variables. FJvL, CBMK, and TM contributed to the interpretation of the findings and critically revised the manuscript. All authors read and approved of the final manuscript.

Funding

AvdV was supported with an internal University of Applied Sciences Utrecht subsidy for PhD-research. The GLOBE 2021 data collection and CBMK were supported by the Innovational Research Incentives Scheme (Vl.Vidi.198.001), financed by the Netherlands Organization for Scientific Research (NWO). The funding bodies had no role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript.

Data availability

The dataset generated and analyzed during the current study is not publicly available due privacy regulations, but code files are available upon request by contacting the GLOBE study project leader Frank J. van Lenthe.The following additional files are provided:•Additional file 1: ‘Financial scarcity and health behavior GLOBE_Additional file 1.pdf’•Additional file 2: ‘Financial scarcity and health behavior GLOBE_Additional file 2.pdf’•Additional file 3: ‘Financial scarcity and health behavior GLOBE_Additional file 3.pdf’.

Declarations

Ethics approval and consent to participate

The 2021 wave of the GLOBE study was approved by the Ethics Committee of the Faculty of Social and Behavioural Sciences of Utrecht University (number 21–0355), and the same committee also approved the specific analyses done for this paper (number 25–0254). The use of personal data in the GLOBE study is in compliance with the Dutch Personal Data Protection Act and the Municipal Database Act; the study is registered with the Dutch Data Protection Authority (number 1248943). All study participants provided active informed consent. The research has been performed in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.