Does Life Satisfaction Vary with Time and Income? Investigating the Relationship Among Free Time, Income, and Life Satisfaction

Abstract

Time and income are distinct and critical resources needed in the pursuit of happiness (life satisfaction). Income can be used to purchase market goods and services, and time can be used to spend time with friends and family, rest and sleep, and other activities. Yet little research has examined how different combinations of time and income affect life satisfaction, and if more of both is positively associated with greater levels of life satisfaction. We investigate whether life satisfaction significantly varies with time and income using data from the American Time Use Survey and its well-being module, which is a nationally representative sample of over 5000 US respondents over the age of 15. We plot a three-dimensional space exploring the relationship among time, income, and life satisfaction, finding people with similar incomes with less free time have lower levels of life satisfaction. We also identify different four subpopulations, three of which have low well-being along time and income, and one with high well-being along time and income. These sub-groups significantly differ along key characteristics. Respondents with less free time and low income—the doubly poor—are more likely to be female, less educated, and have more than two kids and young children. Those with low income but lots of time, in comparison, are more likely to be black, unemployed, and have some physical or cognitive difficult. We conclude that time provides unique insights into human well-being that income alone cannot capture and should be further incorporated into research and policy on life satisfaction.

1. Introduction

The question of whether one can buy happiness (heretofore referred to as life satisfaction) has engendered a rich scientific literature. Investigators of varying stripes have pursued the connection between money and life satisfaction (e.g., Easterlin 1974, 1995; Kahneman and Deaton 2010). If money is the currency of commerce, time is “the currency of life” (Krueger et al. 2009a). Thus, authors are beginning to consider time in addition to income as drivers of life satisfaction (Diener and Oishi 2000; Gilovich and Kumar 2015; Graham and Pettinato 2001; Hershfield et al. 2016; Mogilner 2010; Mogilner et al. 2018; Whillans et al. 2017) Little is known, however, about how life satisfaction varies with differing amounts of free time and income.

This gap is significant. Theoretically, Becker’s (1965) theory of time allocation modeled utility (i.e. life satisfaction) as a function of time and budget constraints. Gronau (1977) extended this model to explicitly include time spent working in the market, at home, and in leisure. Here, utility is maximized by allocating time to leisure, and to consumption of market goods and services and home-produced goods, which are all subject to time and budget constraints. Yet the relationship between utility and varying amounts of free time and income, and whether having more of both consistently leads to higher levels of utility, is underexplored. Time and income are “upstream” resources individuals can use to increase life satisfaction by, for instance, using it to purchase meals, housing, and market goods and services, or to spend time with friends and family, rest and sleep, and exercise (Becker 1965; Gronau 1977; Williams et al. 2015). Disentangling how time and income alone affects life satisfaction is important for developing policies targeting these resources. While there is substantial debate about using life satisfaction as a proxy for utility (Glaeser et al. 2016), researchers have increasingly done so in empirical studies investigating utility (Dolan et al. 2008; Frey and Stutzer 2001, 2002; Frijters et al. 2012; Layard et al. 2008; Merz and Rathjen 2014a, b).

To date, research on how these “upstream” resources are associated with life satisfaction has largely focused on monetary resources. For example, the Easterlin Paradox highlights how significant increases in real income have not resulted in significant changes in life satisfaction in the long run at the national level (Easterlin et al. 2010; Easterlin 1974, 1995).¹ Others have found a significant and positive relationship between income and life satisfaction (Blanchflower and Oswald 2004; Clark et al. 2005; Gerdtham and Johannesson 2001; Graham and Pettinato 2001; Lelkes 2006; Shields and Price 2005). Kahneman and Deaton (2010) found life satisfaction as measured by life evaluation continually increases across income categories, although for emotional life satisfaction (i.e., positive affect, blue affect, and stress) there is no effect after about $75,000 a year. Relative income also affects life satisfaction (Clark et al. 2008; Graham and Pettinato 2001; Knight et al. 2009; Luttmer 2005).

As scholars have unpacked the relationship between monetary resources and life satisfaction, parallel research efforts have focused on investigating the determinants of life satisfaction that go beyond income. Indeed, the psychology community has long recognized that income alone is insufficient for increasing life satisfaction (Diener et al. 1999). In the economics literature, the Easterlin Paradox has spurred great debate and research on non-monetary factors affecting life satisfaction. Some have hypothesized that once income covers basic needs, life satisfaction is driven more by non-monetary factors (Lane 2000). Factors such as age (Frijters and Beatton 2012), physical health (Clark and Oswald 1994; Gerdtham and Johannesson 2001), relative income status (Clark and Oswald 1996), employment status (Clark 2003; Winkelmann and Winkelmann 1998), and marital status (Clark and Oswald 1994; Gerlach and Stephan 1996; Theodossiou 1998; Winkelmann and Winkelmann 1998), having children (Theodossiou 1998), and other intrahousehold characteristics (Shields and Price 2005) are associated with life satisfaction. “As societies become more affluent, the non-monetary aspects of well-being take on increased salience” (Haveman 2009, p. 397).

Within this literature researchers are increasingly examining the relationship between time—the other upstream resource—and life satisfaction. This literature has largely found time is a primary contributor to life satisfaction: those having preferences for free time are happier (Hershfield et al. 2016), and buying time or experiences can lead to life satisfaction (Aknin et al. 2018; Whillans et al. 2017). Some scholars have even tested interventions examining how focusing more on time rather than income might lead to greater life satisfaction (Mogilner 2010). Other research, mainly driven by Merz and Rathjen (2014a, b), has examined how different levels of time and income are associated with life satisfaction using a Constant Elasticity of Substitution function that explicitly captures tradeoffs between the two resources.

What is missing in this expanding literature is an exploration of whether different levels of both time and income are associated with life satisfaction. Previous studies have been limited by assuming a strict functional form (Merz and Rathjen 2014a, b), or by examining preferences around whether time or income leads to greater life satisfaction (Hershfield et al. 2016). Others have conceptualized a dichotomous tradeoff between preferring or obtaining more time or more income (Mogilner 2010), rather than examining the tradeoff of continuous amounts of time and income. We fill this gap by investigating two overarching questions rooted in exploratory empirical analyses. First, are more time and money associated with higher levels of life satisfaction? Utility theory suggests more of both will be associated with greater levels of life satisfaction. Second, within any level of income, does life satisfaction vary with different levels of time? If time does not matter for life satisfaction, life satisfaction will not vary within any given level of income for all levels of discretionary time (e.g., free time beyond work and household requirements). There may be diminishing returns of time or of income on life satisfaction in the presence of the other input. If life satisfaction only requires a time (income) floor, we expect little increase in life satisfaction beyond a certain level of time (income). Too much time may be detrimental to life satisfaction if a minimum income threshold is not met, such as with the long-term unemployed (Clark et al. 2001; Clark and Oswald 1994; Theodossiou 1998; Winkelmann and Winkelmann 1998). The complex interdependence of time and income may play out differently in terms of life satisfaction throughout the joint range of time and income.

We investigate these questions by using well-being, time, and income data from the American Time Use Survey (ATUS) and Current Population Survey (CPS). We find life satisfaction varies across both time and income, where the people who have more of both are happiest. However, we also find lower than expected life satisfaction beyond the usual income poverty threshold, and evidence that the amount of discretionary time plays a significant role in a person’s life satisfaction. We identify four groups with lower and higher than expected life satisfaction over varying levels of time and income. These subpopulations differ along key characteristics and have unique experiences with life satisfaction and factors that may strongly influence life satisfaction. One group represented by high income and discretionary time, the Well-off, has higher than expected well-being, and is significantly more likely to own a home or business and be married. There are three subpopulations with lower than expected life satisfaction, which we identify as the Doubly poor (low time and income), the Full-time poor (low income and average time), and the Idle poor (low income and lots of time). The Doubly poor have the least amount of discretionary time and are more likely to be young, Hispanic, female, employed, and live in large households with young children. The Idle poor are more likely to be black, be unemployed, and have some physical or cognitive difficulty. The Full-time poor and the Doubly poor are less likely to be food secure in the past month or past year, and more likely to be under-banked or utilize alternative financial services.

2. Data and Empirical Strategy

The ATUS is a nationally representative dataset on time use for adults 15 years of age and older in the United States. One adult is randomly selected from households rotating out of the Current Population Survey (CPS) to report on his or her activities for the previous day. The ATUS can be linked to the CPS and many of its supplement surveys. We use the 2012 and 2013 ATUS because they included a Well-being Module, which collected information on a respondent’s subjective well-being.

The ATUS provides rich and unique data for investigating the relationship between time, income, and life satisfaction. First, there is the Well-being Module data collected concurrently with the usual time diary. The ATUS employs a Cantril Ladder Scale question to measure life satisfaction, and we use this as our measure of life satisfaction, much like Easterlin (1974), Frey et al. (2007), Stevenson and Wolfers (2008). This question asks, “Please imagine a ladder with steps numbered from zero at the bottom to ten at the top. The top of the ladder represents the best possible life for you and the bottom of the ladder represents the worst possible life for you. If the top step is 10 and the bottom step is 0, on which step of the ladder do you feel you personally stand at the present time?” Second, the ATUS data can be linked to the many supplementary surveys that are part of the CPS. We are primarily interested in linking ATUS data with the Annual Social and Economic Supplement (ASEC, a.k.a. the March supplement), which has detailed information on household demographics and income in the past calendar year for all household members. Income and household structure are then used to calculate poverty-income-ratios (PIR), which divide household income by the appropriate poverty threshold given family composition (provided by ASEC for each respondent). PIR less than one is an indicator of income poverty. Expressing income as a PIR makes income comparable across both years and family sizes. Finally, the ATUS provides detailed time use data, which is critical for classifying time use activities into necessary and discretionary categories (Williams et al. 2015). Categorization of time use activities into such bins is standard practice by time poverty scholars (e.g., Burchardt 2008; Goodin et al. 2005; Harvey and Mukhopadhyay 2007). Here, we build on Williams et al.’s (2015) guidance on classifying activities as necessary if activities are plausibly “required to meet the basic necessities of life in a given society”, and activities that are not necessary are considered discretionary activities. This allows us to create a measure of discretionary time. The ATUS time classification system has three levels of activity categories: 18 major categories (e.g., personal care, household activities), which are then further classified into two levels. We sum time reported for the following major categories, including details from the travel category associated with each, for our discretionary time measure. These activities are:

1. Caring for non-household members

2. Eating and drinking

3. Socializing, relaxing, and leisure

4. Sports, exercise, and recreation

5. Religious and spiritual activities

6. Volunteer activities

7. Personal/social communication (phone, email)

We merged the ATUS with the ASEC, retaining adults with complete time diaries accounting for all 24-h of the prior day, a life satisfaction ladder score, and income data reported for the prior year. Because of CPS’ rolling sample, our final analytic sample consists of 5417 adults with both ATUS and ASEC data collected in the same year, out of 17,597 adults with complete ATUS data. There are no differences in observables between the analytic sample and the complete ATUS data (Appendix 1).

Our goal is to model the surface of life satisfaction over the space of time and income, and then examine areas of lower versus higher life satisfaction as a function of time and income. Our analysis controls for the effects of other factors, many discussed above, that influence life satisfaction to model local changes in life satisfaction associated with changes in time and income.

Our modeling, like much of the literature (Dolan et al. 2008; Kahneman and Deaton 2010), treats the 11-point Cantril Ladder Scale as continuous. We use Generalized Additive Models (Wood 2009, 2017) to combine parametric estimation of the effects of common covariates of life satisfaction with a non-parametric, flexible spline function of time and income. Our general form is:

$$\mathit{\text{Life}}\mathit{\text{satisfactio}}{n}_i={\beta }_0+{\beta }_1{P}_i+s\left(\mathit{\text{time}},\mathit{\text{income}}\right)\text{+}{\epsilon }_i$$ (1)

where s(·) is a spline function discussed further below and P is a vector of covariates commonly identified in the literature as key determinants of life satisfaction (Bhuiyan and Szulga 2017; Bjørnskov et al. 2008; Diener et al. 1999, 2018; Dolan et al. 2008). Systematic reviews on the determinants of life satisfaction have generally categorized variables into themes such as personal characteristics (e.g., age, sex), personality traits and values (e.g., extraversion), socially developed characteristics (e.g., education, health, type of work, unemployment), time use (e.g., hours worked, commuting time), environmental factors (e.g., local crime rates), and others. Although there is variation in these reviews, factors such as gender, education, age, household composition, unemployment status, and personal health are commonly found to be important variables. Our covariate selection is informed by these reviews, although we are limited to variables available in our analytic dataset. For instance, our dataset does not have information on factors such as genetics (Bartels 2015), personality traits (Costa and McCrae 1980), and perception of control (Verme 2009). We note that past reviews have also found variation between datasets, such as the survey country and how variables were operationalized (Dolan et al. 2008). As a result, P is a vector containing sex, age (and its square to account for curvilinear effects), whether the respondent reports having any physical or cognitive difficulty, education (less than high school, high school, any education beyond high school), employment status (unemployed or not in labor force, one job, more than one job), race (White, Asian or Pacific Islander, Black, and Hispanic), marital status (married, separated, or widowed), veteran status, whether a household member owns the home and whether a member owns a business, the number of adults and number of children in the household, whether there is a child under 6 in the household, whether the prior day was a weekday or a weekend or holiday.

Our goal is to investigate the relationship between time and income on life satisfaction, allowing this relationship to change throughout the income and time space. In this exploration, we are not saying that income and time “cause” the observed differences in well-being, and of course any variables not among our covariates may continue to affect well-being through income and/or time. The spline function allows us to produce a three-dimensional view of how life satisfaction changes, on average, with local changes in time and income. Our approach differs from those that assume an additive model (Dolan et al. 2008), or a strict functional form (Merz and Rathjen 2014a, b). Instead, we follow other recent exploratory analyses of life satisfaction data (e.g., Jebb et al. 2018) by using non-parametric and semi-parametric approaches to understand the relationship between income, time, and life satisfaction. These methods are appropriate given the exploratory nature of our study, and because polynomial regressions are often inflexible (Lind and Mehlum 2010; Yatchew 1998).

To create our smoothed surface, we use tensor plate splines with cubic regression basis functions to model life satisfaction net of the covariates as a function of the percent of median discretionary time (i.e. time) and PIR (i.e. income). This allows us to optimize smooths via cross-validation while penalizing for “wiggliness” (Duchamp and Stuetzle 2003; Wood 2017). We then create a predicted surface of life satisfaction over an appropriate range of time and income, with the covariates set at constant levels (means or one level of factor variables). Our resulting topographical map traces the contours of life satisfaction, conditioned on our covariates (i.e. low expected life satisfaction), over the latitude and longitude of time and income. Given the distribution of time and income (Fig. 5 in the Appendix), we employ convex hulls for our topographic maps to avoid over extrapolating in areas of the time and income space where data may be sparse.

We then continue the explorations by describing and investigating whether people in areas of low life satisfaction possess distinct traits and are at risk of factors traditionally associated with poverty (e.g., food security, health care access, use of alternative financial services). Here, we merge our analytic data with CPS supplement files (Appendix 2). These additional variables provide important information about individual and household constraints and poverty experiences, including questions on food security, health insurance coverage, and access to and use of financial services (banks or credit unions versus higher cost bank alternatives, such as check cashing stores, pawn shops, and pay-day loans). Not all individuals in our analytic sample had matching CPS supplement files. This follow-up exploration required respondents to have not only aligning ATUS and ASEC data, but also the other CPS questions completed in the same year, leading to 9.2–100% of respondents having matching data. Overall, we find almost no statistically significant differences between the analytic sample used to map life satisfaction and the sample used to explore these correlates (Table 5 in the Appendix).

3. Results

3.1. Descriptive Statistics

Average household size, racial, and other characteristics mirror national averages (Table 1), although the average age of the analytic sample is slightly older. Approximately 83% live in metropolitan areas. Discretionary time is on average 7.7 h, and those unemployed or not in the labor force have significantly more time (an additional 3.3 h). Discretionary time does not significantly differ between respondents living in metropolitan and non-metropolitan areas. Approximately 12% face income poverty (PIR < 1). Respondents facing income poverty have significantly lower life satisfaction and discretionary time. As expected, this group has significantly less education, and is also less likely to be employed and own home and business. They are less likely to be veterans, but are significantly more likely to report having cognitive or physical difficulties.

Table 1.

Descriptive statistics

Individual characteristics	Mean	SD
Happiness scorea	7.1	2.0
Poverty income ratio	4.2	4.4
Discretionary time (hours)	7.7	4.0
Discretionary time (% of median)	106	57
Age	49	17
Female (%)	46	50
Married (%)	50	50
Black (%)	16	36
Hispanic (%)	14	34
Has cognitive or physical difficulty (%)	12	32
Veteran (%)	10	30
Metro (%)	82	38
Schooling (%)
Less than high school	11	31
High school	26	44
Beyond high school	63	48
Employment (%)
Employed	59	49
More than 1 job	5.0	21
Unemployed/NILF	36	48
Household characteristics
Household size	2.6	1.5
Adults in household	1.8	0.67
Children in household	1.7	0.99
Children under 6 (%)	17	38
Owns home (%)	71	45
Owns business (%)	13	33
n	5417

^aHappiness scores range from 0 to 10

There are also some notable differences by sex, household composition, and veteran status. Veterans are more likely to report having cognitive or physical difficulties. Respondents reporting cognitive or physical difficulties had significantly lower life satisfaction scores. Females had significantly higher life satisfaction scores, mirroring Kahneman and Deaton’s (2010) findings. Female respondents are less likely to report having some cognitive or physical difficulties, and also report living in higher income households. They are also significantly more likely to be employed compared to male respondents, although no more likely than males to work more than one job. Finally, respondents without children have up to 2 h of additional discretionary time, while female respondents have approximately 30 min less discretionary time compared to male respondents.

3.2. Exploring the Life Satisfaction Map

Figure 1 presents the topographical maps of life satisfaction over the time (vertical axis) and income (horizontal axis) dimensions, and Table 2 presents model results. We restrict our map of predicted life satisfaction to the space supported by observed levels of discretionary time and income. The top right corner of Fig. 1 is empty because no respondent has high levels of both income and time, which illustrates trade-offs between the two inputs. The smoothed well-being values range from 4.33 to 9.77, or about 54% of the well-being range of 0 to 10. Note that the choice of levels at which to set the covariates merely shifts the predicted well-being up or down.

Fig. 1.

Residual happiness, income, and time topographical maps. Figure 5 in the Appendix presents a hexplot for the distribution of time and income

Table 2.

General additive model results

Linear parameters	Estimate	SE	t	p
Intercept	6.973	(0.394)	17.714	0.000
1 child in HHD	0.038	(0.086)	0.442	0.658
2 children in HHD	0.244	(0.095)	2.561	0.010
3 + child in HHD	0.426	(0.126)	3.385	0.001
2 adults in HHD	0.095	(0.084)	1.119	0.263
3 + n adults in HHD	−0.250	(0.100)	−2.503	0.012
Child < 6 in HHD	0.128	(0.091)	1.401	0.161
Family business	−0.030	(0.085)	−0.349	0.727
Age	−0.047	(0.011)	−4.372	0.000
Age2	0.001	(0.000)	5.567	0.000
Male	−0.224	(0.059)	−3.803	0.000
More than 1 job	−0.058	(0.130)	−0.450	0.653
Unemployed/NILF	−0.108	(0.074)	−1.461	0.144
Owns home	0.099	(0.068)	1.452	0.146
Weekday	0.023	(0.217)	0.107	0.915
Weekend	0.026	(0.217)	0.118	0.906
No difficulty	0.604	(0.092)	6.555	0.000
Education: HS	0.090	(0.066)	1.356	0.175
Education: less than HS	0.219	(0.098)	2.229	0.026
White	−0.025	(0.197)	−0.127	0.899
Asian or Pacific Islander	−0.282	(0.237)	−1.189	0.235
African American	0.270	(0.207)	1.304	0.192
Hispanic	0.343	(0.084)	4.063	0.000
Married	0.386	(0.096)	4.017	0.000
Separated or divorced	−0.120	(0.095)	−1.261	0.207
Widowed	0.223	(0.131)	1.698	0.090
Veteran	−0.046	(0.100)	−0.466	0.641
Non-parametric function	EDF	Ref. DF	F	p
s (income, time)	17.250	18.830	5.605	0.000
n	5417
Adj. R2: 0.0741	Deviance explained: 8.15%

s() refers to the tensor plate spline function, with cubic regression basis functions

EDF refers to the Effective degrees of freedom

We find that the surface is not uniform within any given level of income, indicating the importance of discretionary time for life satisfaction. As expected, individuals with significant income and discretionary time have high life satisfaction, which is presented by the green areas of positive life satisfaction. There is a dip in life satisfaction among those with relatively little discretionary time and high income, although this pocket of low life satisfaction is driven by five data points and should be interpreted cautiously. There is also evidence of diminishing returns: at most levels of time, additional income beyond a PIR of 15 to 25 is associated with less additional life satisfaction than similar increases at lower levels, and particularly for those with PIRs between 5 and 15 additional time is associated with crossing few contours of life satisfaction.

There are areas of low life satisfaction even at income levels beyond the standard income poverty threshold (PIR > 1.0). Specifically, at PIR of < 5 there are three areas of lower than expected life satisfaction; the lowest level of predicted life satisfaction is present where people have the most discretionary time (top left corner). The varying levels of life satisfaction for individuals with PIR < 5 but > 1 suggests that the amount of discretionary time plays a significant role in a person’s life satisfaction even when basic needs are ostensibly met.

3.3. Exploring Life Satisfaction Profiles

We extend our analysis by investigating whether different pockets in the life satisfaction surface differ along key factors associated with poverty experiences. Again, the intent is to conduct an exploratory investigation of different areas of Fig. 1 that have lower and higher than expected well-being. We first visually classify individuals into groups based on their life satisfaction levels bounded by levels of time and income. We identify five groups (Fig. 2): (1) the Idle Poor (n = 152; PIR ≤ 5.5, time ≥ 232% of median), (2) Full-time (FT) poor (n = 1221; PIR ≤ 2.85, time between 77 and 158% of median), (3) Doubly poor (n = 450; PIR ≤ 2.85, time ≤ 57% of median), (4) Well off (n = 68; PIR ≥ 10, time between 137 and 363% of median), and (5) Plateau (n = 3562; everyone else).

Fig. 2.

Profiles in residual happiness, income, and time topographical maps. Figure 5 in the Appendix presents a hexplot for the distribution of time and income

3.3.1. Validating Profiles: Differences in Life Satisfaction

First, we check our visual identification of the boxes in Fig. 2 and the resulting profiles. We include indicator variables for each profile type (excluding plateau) in a linear regression with the covariates P to examine whether they are significant predictors of life satisfaction. Model results are summarized in Table 3 (full regression results in Table 7 in the Appendix). The four group indicators significantly increase prediction of life satisfaction: Adjusted R² increases to 6.70% from 5.81% in a model with only the covariates (not shown), while the Bayesian Information Criterion decreases to 22,950 from 22,971. All coefficients for the profiles are statistically significant, with the coefficients in the expected direction.

Table 3.

Validation of profile group differences on life satisfaction versus observations not in any group, net of covariates (not shown)

Group	Estimate	SE	t	p
Well-off	0.528	(0.242)	2.178	0.029
Idle poor	−0.293	(0.170)	−1.724	0.085
Doubly poor	−0.431	(0.103)	−4.163	< 0.001
Full-time poor	−0.446	(0.070)	−6.326	< 0.001
n	5,417
Joint F(4, 5386)			12.441	< 0.001

3.3.2. Comparing Profiles on Common Correlates of Poverty

We present predicted differences for each of the so-called poor groups (Full-time poor, Doubly poor, and Idle poor) and the Well-off group versus the reference Plateau group, with 95% confidence intervals. Here, we regress each variable of interest on the indicators for the identified groups (with all other observations as the reference), and then simulate the confidence intervals. Figure 3 thus plots the mean of each variable of interest for each group with the simulated confidence intervals, allowing comparison of the groups for each variable. Again, not all variables are available for the entire sample. Our results suggest the so-called poor groups face significantly different constraints and poverty experiences from each other (Fig. 3).

Fig. 3.

Characteristics and poverty experiences of profiles. The Plateau group serves as the referent group

The poor groups all significantly differ from each other on the probability of being unemployed or being out of the labor force, as well as reporting some disability status and being a veteran. They also significantly differ in age and household composition and size. The Doubly poor are significantly more likely to be female and have more than two kids and young children in their household, live in larger households, are more likely to be employed, are less likely to have a disability or be a veteran, and are the youngest compared to the other poor groups.

Altogether, the poor groups also are significantly different from the Plateau or Well-off groups across most dimensions. The poor groups are significantly less likely to be married, own a home or business, or live in metropolitan areas compared to the Plateau and Well-off groups. All three poor groups are less likely to have health insurance than the Plateau group. The Doubly poor and Full-time poor (but not the smaller Idle poor group, for whom differences are less precisely estimated) are more likely to have used alternative financial services and less likely to be fully banked than the Plateau group.

As expected, the Doubly poor and Idle poor have the most unique time allocation patterns (Fig. 4). We see that the Doubly poor spend considerably more hours on weekends and weekdays caring for or helping household members, conducting household activities, and working than other groups. The contrast is stark when examining time spent socializing, relaxing, and in leisure on the weekday, where the Doubly poor allocate approximately one-third the time of the next lowest group (Plateau). The Idle poor reported having the lowest level of life satisfaction, and they spent the most amount of time during the weekday socializing, relaxing, and in leisure. The Full-time poor spent less time working than Doubly poor or Plateau groups, but spent more time on household activities than the Plateau group on weekdays. The Well off, as expected, spent the second most time socializing, relaxing, and in leisure on the weekdays, and spent considerably less time working on weekends and weekdays.

Fig. 4.

Time use by group for weekdays and weekends/holidays with 95% confidence intervals. a weekdays, b Weekends/holidays

4. Discussion

Exploration of how life satisfaction data can inform policy is still in its nascence (Allin 2014) despite its growing prominence in policy discussions (Cameron 2010; Schulte et al. 2015). We are not the first to think about how life satisfaction data can inform long existing topics, such as poverty, and be useful for public policy (Hicks et al. 2013). A vast literature, for example, has searched for an alternative to, or complement of, Gross Domestic Product (Allin 2014; Fleurbaey 2009; Jones and Klenow 2016; Stieglitz et al. 2010), one that captures the non-monetary components of progress, such as leisure.

Our study empirically investigates the relationship of time and income on life satisfaction. We find that more time and income is associated with greater levels of life satisfaction, although there are indications of diminishing returns. There appears to be no evidence of a unidimensional floor above which all people have approximately equal levels of life satisfaction. Otherwise, we find that time matters for life satisfaction, especially for those below PIR < 5 and those with PIR > 15 where we see a “hill” of higher life satisfaction. For a given subset of the population having higher income beyond the poverty line is not associated with higher levels of subjective well-being. For this group, unless there is an increase in discretionary time subjective well-being does not increase. It is possible, for instance, that personality traits may be an important factor that explains this variation (Lu and Hu 2005), as the way in which we experience subjective well-being via increases in income or time may vary by personality. Our results also indicate that income poverty alone (PIR > 1) may be a poor indicator for subjective well-being: there are a substantial number of people with PIR > 1 with well-being levels similar to those below 1, even with significant discretionary time.

We identify three groups who experience lower than expected life satisfaction beyond accepted thresholds of income poverty (PIR < 1). It is not hard to imagine how lack of time can lead to lower levels of life satisfaction and lead to poverty traps. Indeed, recent evidence found lacking money and time leads to poor cognitive function, further perpetuating the cycle of poverty (Mani et al. 2013). Being income poor can have significant time costs (Romich 2006), as, for example, utilizing anti-poverty services and programs can require office visits and administrative paperwork. Less discretionary time means less time for rest, exercise, and other activities necessary to maintain a healthy life. For those in income poverty, it can mean less time to gain education or more marketable skills. These areas of lower than expected well-being may merit further exploration and policy considerations, and may be one of the areas where subjective well-being data can be combined with income and time use data to provide policy insights (Frey and Stutzer 2002).

Our findings also support much of the burgeoning literature on income, time, and life satisfaction, and we believe our analysis complements existing findings. For example, like the Idle poor, numerous studies found negative correlations between unemployment and life satisfaction (Alesina et al. 2004; Clark et al. 2005; Clark and Oswald 1994; Di Tella et al. 2001; Dolan et al. 2008; Frey and Stutzer 2001; Gandelman and Hernandez-Murillo 2009; Theodossiou 1998; Vinet and Zhedanov 2010; Winkelmann and Winkelmann 1998). Some among our Idle poor group have lower life satisfaction than people with slightly more free time but less income. Working many hours is not always detrimental to life satisfaction (Graham 2003; Juster 1985; Scitovsky 1976), and our predictions show that those with discretionary time below the median had relatively high life satisfaction as long as they had relatively high income.

We identified several areas for further exploration. First, while we attempted to explore life satisfaction, data limitations prevented us from netting out all the key variables identified in the literature, such as personality traits (Ferrer-i-Carbonell and Gowdy 2007; Helliwell 2006; Lu and Hu 2005). Limitations with the ATUS data did not allow contextual data, such as air pollution and crime rates, to be fully explored. Spatial variation in access to public services, community investments, pollution, and other factors may play significant roles in life satisfaction, income, and time. The effects of all these omitted variables on life satisfaction may influence the measured local slopes along time and income. Understanding whether these contextual variables affect one or all three factors is critical for formulating effective policies. Second, identifying a time and income poverty threshold may be critical for informing policies. A key challenge to using life satisfaction and time use data in poverty analysis is its scarcity. Life satisfaction data are not readily available in most national surveys, and further efforts to collect these data are critical. Future work should also explore how the surface of life satisfaction over time and income differs along other drivers of well-being (e.g., different topographical maps for males versus females, renters versus homeowners, etc.) and the trade-off between time and income (i.e., how much income would be needed for more time?), which may provide policy-relevant insights into topics such as work-life balance. Some work has explored this (Merz and Rathjen 2014b), and this topic warrants further in-depth analysis.

Further, we utilized discretionary time per Williams et al. (2015), and this does not capture different dimensions of time that likely influence subjective well-being (Reisch 2001), such as available time blocks during the day, synchronicity of time with other people, and autonomy over time. Future analysis should also explore differences in results when using different measure of life satisfaction. Recent advances have refined the relevant constructs, and measurement of life satisfaction have greatly improved thanks in no small part to the work of Angus Deaton, Daniel Kahneman, and Alan Krueger (Kahneman and Deaton 2010; Kahneman and Krueger 2006; Krueger et al. 2009b). But a standard definition and set of variables is still missing (Schulte et al. 2015). As a result, surveys that include a question or module to evaluate life satisfaction have done so in different ways (e.g., Cantril ladder questions, other satisfaction with life questions, emotional quality/hedonic well-being), making comparisons difficult. Finkelstein et al. (2013) used a question from the Health and Retirement Study asking respondents, “Much of the time during the past week I was happy. (Would you say yes or no?)” Di Tella et al. (2001) used a question from the Euro-Barometer survey asking, “On the whole, are you very satisfied, fairly satisfied, not very satisfied or not at all satisfied with the life you lead?” (4 point scale: 1 not at all satisfied to 4 very satisfied).

Finally, following Frey and Stutzer (2002) we believe life satisfaction data can provide insights for poverty analysis, although further work is needed to practically inform policy (Bache et al. 2016). Some work has peripherally explored this, but this has been limited to high income countries. Plug and Van Praag (1995) estimated family equivalence scales with respect to life satisfaction, or the additional income different size families would need to maintain the same level of life satisfaction. A large literature examines the effects of unemployment on life satisfaction (Alesina et al. 2004; Blanchflower and Oswald 2004; Clark et al. 2005; Clark and Oswald 1994; Di Tella et al. 2001; Dolan et al. 2008; Frey and Stutzer 2001; Gandelman and Hernandez-Murillo 2009; Theodossiou 1998; Winkelmann and Winkelmann 1998), but this is indirectly related to poverty and the emphasis is still on income (or lack thereof).

We believe that including time, along with income and life satisfaction data, can be a fruitful area for exploration. The idea that time is a critical non-monetary economic resource is not new (Folbre 2009; Harvey and Mukhopadhyay 2007; Krueger et al. 2009a; Williams et al. 2015). Its strong theoretical foundations (Becker 1965; Gronau 1977) and uniqueness from income make it an interesting “upstream” resource. Every individual has a fixed endowment of time (e.g., 24 h in a day), and it is not possible to earn more than 24 h in a day. However, people may hire others to free up time, such as with daycare, or purchase time-saving technology, such as dishwashers. The way people allocate their time has direct consequences for their own life satisfaction. This may include allocating time to rest, spend time with family, work, exercise, or anything else. Unlike income, more freely allocatable time is not always better, which is supported by our results.

As mentioned above, Merz and Rathjen (2014a, b) are among the few scholars that have explored the relationship between time, income, and life satisfaction for poverty analysis. But like studies that investigate the relationship between income and life satisfaction, this has been limited to high income countries. Future work should explore this relationship in lower income countries where social norms, constraints, and human-environment relationships are different. For instance, a large body of work has written about how women disproportionately bear the brunt of collecting water and firewood, but no work has linked whether this affects their life satisfaction. Expanding the analysis to lower income countries is likely to highlight different individual, household, and contextual factors affecting time and income poverty as it relates to life satisfaction. If we are to increase life satisfaction in society it is clear there is a need to look beyond just income constraints.

Appendix 1: Data Notes

Merging Datasets

A potential concern with the rolling sample design is losing representativeness when merging ATUS with CPS supplements and other datasets. Here, we present checks on whether data are still nationally representative when ATUS data are merged with CPS supplement files as is typical in survey attrition (see ATUS Extract Builder website for details).

CPS participant households are randomly selected to initiate the CPS sequence, which consists of four monthly interviews, then eight months off, then four more monthly interviews. Households that participate in the final interview of this sequence are then recruited to participate in ATUS 2–5 months later. In order to have income data reasonably correspond to when well-being was assessed, the ATUS and ASEC had to be in the same year. This means the March Supplement had to come during the second 4 months of the rotation. Thus, for a given year, anyone rotating out of the CPS (completing their last possible interview) in January, February, or after June would not have an ASEC. Our analytic sample merges ATUS data with ASEC data, and we find no significant differences between the original ATUS sample and matched ATUS-ASEC sample across all model covariates (Table 4). This is similar across both dependent variables and covariates.

Table 4.

Comparison of original ATUS sample and matched ATUS-ASEC sample

	ATUS		Matched ATUS-ASEC
	Mean	SD	Mean	SD
Subjective well-being score	7.1	2.0	7.1	2.0
Discretionary time (% of median)	106	57	107	57
Poverty income ratio	–	–	4.2	4.4
Age	49	17	49	17
Female (%)	55	50	54	50
Any difficulty (%)	12	33	12	32
Veteran (%)	09	29	10	30
Unemployed/NILF (%)	37	48	36	48
More than 1 job (%)	5.0	22	5.0	21
Metro (%)	83	37	83	37
2 adults in HHD (%)	52	50	52	50
3 + adults in HHD (%)	15	36	15	36
Child < 6 in HHD (%)	18	39	17	38
1 child in HHD (%)	55	50	54	50
2 children in HHD (%)	17	38	17	37
3 + children in HHD (%)	16	36	16	36
Owns home (%)	70	46	71	45
Family business (%)	13	33	13	33
n	17,597		5417

Table 5.

Additional data

Source	Description	Unit	% Merged
CPS food security	Food secure in the past month or year	Individual	27
CPS ASEC	Health insurance coverage	Individual	100
CPS unbanked supplement	Information on households not utilizing banking services, or using alternative banking services.	Household	9.2

Table 5 presents the percent of the analytic sample that matched with CPS supplement data. Because the health insurance coverage variable was included in the ASEC the entire sample has health insurance information. The low matching rates for food security and banking are due to the timing of these supplements, for the same reason discussed above in reference to matching ATUS with ASEC. The food security supplement is conducted in December, and thus can only be matched with households rotating out the following March. The banking supplement was conducted in June of 2013, and so can only be matched for households that for whom that was their final month of participation. There are almost no statistically significant differences across all matched subsamples and the analytic sample for model covariates (Table 6). This is due to the ATUS reporting (i.e. in the public microdata only residents of larger counties are assigned a county identifier) and the nature of some of the secondary datasets.

Table 6.

Sample comparisons with additional variables

			Matched samples
			CPS		CPS
	ATUS		Food security		Banking
	Mean	SD	Mean	SD	Mean	SD
Subjective well-being score	7.1	2.0	7.1	2.0	7.0	2.1
Discretionary time (% of median)	107	57	106	56	102	54
Poverty income ratio	4.2	4.4	4.2	4.1	4.6	4.8
Age	49	17	49	17	50	17
Female (%)	54	50	55	50	57	50
Any difficulty (%)	12	32	11	31	12	33
Veteran (%)	10	30	10	31	8.0	28
Unemployed/NILF (%)	36	48	36	48	37	48
More than 1 job (%)	5.0	21	4.0	20	4.0	19
Metro (%)	83	37	83	37	82	38
2 adults in HHD (%)	52	50	52	50	53	50
3 + adults in HHD (%)	15	36	15	36	15	36
Child < 6 in HHD (%)	17	38	18	38	16	36
1 child in HHD (%)	54	50	17	38	15	36
2 children in HHD (%)	17	37	15	35	17	37
3 + children in HHD (%)	16	36	7.0	26	6.0	24
Owns home (%)	71	45	73	45	72	45
Family business (%)	13	33	13	33	14	35
n	5417		872		496

Appendix 2: Regressions with and Without Profile Dummies

Table 7 presents regressions with and without the profile dummies. Note that model fit improves compared to the OLS without profile dummies.

Table 7.

Regression results with and without profiles

Variable	No profiles		With profiles
	Coefficient	SE	Coefficient	SE
Number of children in household (vs. 0): 1	−0.042	(0.087)	0.000	(0.086)
Number of children in household (vs. 0): 2	0.138	(0.095)	0.190**	(0.095)
Number of children in household (vs. 0): 3 +	0.209*	(0.124)	0.335***	(0.125)
Number of adults in household (vs. 1): 2	0.136	(0.085)	0.122	(0.085)
Number of adults in household (vs. 1): 3 +	−0.186*	(0.100)	−0.212***	(0.100)
Child under 6 in household	0.094	(0.092)	0.103	(0.091)
Family business	0.021	(0.085)	0.010	(0.085)
Age	−0.044***	(0.011)	−0.046***	(0.011)
Age-squared	0.001***	(0.000)	0.001***	(0.000)
Male	−0.190***	(0.059)	−0.205***	(0.059)
Employment (vs. employed with 1 job): more than 1 job	−0.105	(0.131)	−0.097	(0.130)
Unemployed/NILF	−0.188***	(0.069)	−0.145**	(0.070)
Own home	0.213***	(0.067)	0.157**	(0.067)
Diary day was weekday	0.070	(0.218)	−0.012	(0.218)
Diary was weekend	0.065	(0.218)	0.012	(0.217)
No difficulty	0.679***	(0.092)	0.638***	(0.092)
Education (vs. beyond high school): high school	−0.065	(0.064)	0.001	(0.065)
Education (vs. beyond high school): less than high school	0.022	(0.096)	0.111	(0.096)
White	0.035	(0.198)	0.021	(0.197)
Asian-Pacific Islander	−0.172	(0.239)	−0.211	(0.238)
Black	0.276	(0.208)	0.288***	(0.208)
Hispanic	0.262***	(0.084)	0.312***	(0.084)
Married	0.497***	(0.096)	0.437	(0.095)
Separate/divorced	−0.129	(0.095)	−0.120	(0.095)
Widowed	0.227*	(0.132)	0.216	(0.131)
Veteran	−0.0569	0.1001	−0.048	(0.100)
Well-off	–	–	0.528**	(0.242)
Idle poor	–	–	−0.293*	(0.170)
Doubly poor	–	–	−0.431***	(0.103)
Full-time poor	–	–	−0.446***	(0.070)
(Intercept)	6.713***	(0.396)	7.024***	(0.396)
R2	0.0627		0.0721
Adj. R2	0.0581		0.0670
n	5417		5417

^***p < 0.01;

^**p < 0.05;

^*p < 0.10

Appendix 3: Other Figures

See Fig. 5.

Fig. 5.

Hexplot of income and time