Life satisfaction amongst working parents: examining the case of mothers and fathers in Ireland

Utility theory, individual wellbeing and parenthood

In recent years theoretical and empirical literature in the area of economics and wellbeing has surged (Blanchflower, 2009; Ferrer-I-Carbonell, 2013; Layard, 2005; Frey and Stutzer, 2002; Kahneman et al., 1999). Utility theory provides an economic framework for the evaluation of alternative choices made by individuals and has underpinned key wellbeing research for decades (Frey and Stutzer, 2000a). Utility refers to the satisfaction that each choice provides to the decision maker and utility theory assumes that any decision is made on the basis of the utility maximisation principle. According to the utility maximisation principle the best choice is the one that provides the highest utility or satisfaction to the decision maker (Frey and Stutzer, 2000b). Assuming the individual is rational and fully informed they will choose the best alternative that maximises their utility (Aleskerov et al., 2002). Recently, instead of the decision utility approach of revealed preferences and market-based choices, economics studies tend to use individual, self-reported levels of wellbeing (for example questions about happiness or life satisfaction answered on a scale from 0 to 10) as a proxy for utility (Dolan et al., 2008; Rayo and Becker, 2007). These self-reported measures have been found to be relatively robust indicators of an individual's overall wellbeing (Dolan and White, 2007) and several empirical studies have sought to establish the economic and non-economic determinants of wellbeing using large datasets (for reviews of this research see Dolan et al., 2008; Frey and Stutzer, 2002; Mackerron, 2012)

Within the economics literature to date there has been some examination of the association between individual wellbeing and parenthood as one of those determinants. In terms of childbearing, and consistent with utility theory, it may be assumed that individuals make a rational decision to have a child because they believe it will increase their utility or individual satisfaction, or in the case of subsequent children that they have derived additional utility from having their first child and predict an increase in satisfaction from additional childbearing (Becker, 1960). Conversely, those, who after the first child have experienced negative utility or predict that any additional children will not yield them any increase in satisfaction will decide not to have any more children (Aassve et al., 2012). To that end, various research to date has examined the relationship between individual wellbeing and parenthood. Some recent literature that compares mothers and non-mothers amongst those with completed fertility (using 26-years panel data) rather than parents versus non-parents, found positive satisfaction gains amongst mothers (Baetschmann et al., 2016). Prior to this any positive effects tended to be small and insignificant (Clark and Oswald, 2002; Baetschmann et al., 2016).

Despite these recent findings, since the 1970s patterns of lower levels of wellbeing amongst parents compared to nonparents emerged (Mclanahan and Adams, 1989; Margolis and Myrskylä, 2011; Senior, 2014). A recent review (Baetschmann et al., 2016) found an abundance of research showing significant negative associations between parenthood and wellbeing in developed countries (Blanchflower, 2009; Ferrer-I-Carbonell, 2013; Clark et al., 2008b; Dolan et al., 2008). These findings contradict the beliefs that parenthood is pivotal for developing and maintaining happiness in adulthood (Margolis and Myrskylä, 2011; Hansen, 2012) and an anomaly with respect to beliefs about social roles and the positive effect of relationships on mental health (House et al., 1988), as well as going against utility theory. A reason for this could be that the emotional rewards of having children is out weighted by the emotional and financial costs associated with contemporary parenthood (Nomaguchi and Milkie, 2003; Mclanahan and Adams, 1989; Evenson and Simon, 2005; Liefbroer, 2005; Woo and Kelly Raley, 2005; Begall and Mills, 2011; Balbo et al., 2013). Other reasons proposed for the negative association of parenthood on individual wellbeing appear to be linked to theories of stress and mental health; children increase adults exposure to stressors which increase symptoms of distress, depression and anxiety, as well as other negative emotions. Such stressors include, the time and energy demands on parents, coupled with sleep deprivation (Avison et al., 2007; Nelson et al., 2014; Pollmann-schult, 2014); work-family conflicts (Nomaguchi et al., 2005; Begall and Mills, 2011; Berger, 2013; Roeters et al., 2016; Karabchuk, 2016); difficulties getting affordable, high-quality childcare and financial strains. These of course vary by family composition and environmental considerations (including socio-economic characteristics).

Specifically, previous empirical evidence demonstrates life satisfaction amongst parents is associated with age, education, income, marital status, parents' employment status, characteristics of children (including age and number of children) (Pollmann-schult, 2014; Louis and Zhao, 2002; Evans and Kelley, 2004; Musick et al., 2016; Luppi, 2016; Aassve et al., 2012, 2016; Mikucka and Rizzi, 2016; Myrskylä and Margolis, 2014; Stanca, 2012; Nomaguchi and Milkie, 2003). For example, single parents and divorced mothers appear to suffer the most when it comes to their overall wellbeing (Frey and Stutzer, 2000a; Schoon et al., 2005). While the association varies in size, depending on personal and household characteristics, it is evident across household types, in both mothers and fathers and across dimensions of wellbeing (Glass et al., 2016). Furthermore, supports and family policies such as subsidized childcare, work flexibility (Abendroth and Den dulk, 2011), for example, can reduce the gap between parents and nonparents, improving parents' wellbeing without negatively impacting non-parents (Glass et al., 2016). More recently, Pollmann-schult (2018) investigated how the costs and rewards of parenthood are modified by socio-institutional contexts (such as family financial benefits; level of childcare provisions; flexible working arrangements etc.) thereby affecting parents' life satisfaction.

Despite the rich literature much of the economic research to date on individual wellbeing and parenthood are focused on the fertility decision, rather than examining the factors affecting different parents' wellbeing that could be addressed through targeted policies. Furthermore, many previous studies tend to compare parents to childless individuals; this can result in biases (Baetschmann et al., 2016). We are motivated by Baetschmann et al. (2016) 's critique and focus our analysis on different types of parents. In doing, we are responding to calls by Margolis and Myrskylä (2015) and Kohler and Mencarini (2016) to provide evidence on the factors influencing parents' wellbeing which can be used to inform policies for generating sustainable high satisfaction amongst parents in developed societies.

Irish context/background

Unlike other Anglo-Saxon and Nordic countries, Ireland suffered to a greater extent from the Financial Crisis, the subsequent recession, austerity budgets and public sector reforms that followed. The latter did reduce the deficit and debt but at a significant cost; public services shrank and lack of investment has resulted in social problems, including rising poverty levels, economic stress, homelessness and persistent excess budgets in key areas such as health and welfare (Maccarthaigh, 2017; Watson et al., 2017). These measures hindered progress in creating family enhancing policies; the effects of which are not yet known but we suspect could have long term consequences on dependency and replacement rates. Also, commentators suspect many of the austerity policies pursued may have an unacknowledged, yet a significantly negative impact from a gender equality perspective (Barry and Conroy, 2013).

While female labour force participation in Ireland has increased substantially since the removal of the Marriage Bar in 1973 (prohibited most women from continuing employment as soon as they married – which was on average at age 24) it is still below the EU average (80.1%) (Eurostat, 2019b; Martyn, 2019). One reason for this may be from ambiguous policy principles as to whether women are encouraged to be in the labour market or stay at home (Barry and Conroy, 2013); and a welfare system that is a mixed hybrid model of private market action and women (largely) providing unpaid care and support in the family. Even during the Celtic Tiger decent and affordable childcare, after school facilities, and elder care supports were scarce and the situation worsened post-recession with severe austerity measures that placed pressure on primary carers (mainly women) transferring paid work to unpaid (Barry and Conroy, 2013).

Like elsewhere, fertility rates in Ireland have been steadily declining since the mid-1960s. Nevertheless, Ireland currently has the third highest fertility rate in Europe (Eurostat, 2019a). Previous international research found a negative relationship between parenthood and wellbeing in Ireland and suggests that Ireland ranks second after the USA with respect to the gap between parents and non-parents (Glass et al., 2016). Moreover, earlier Irish research found a negative relationship between life satisfaction and number of children (Brereton et al., 2008). More recently however, Pollmann-schult (2018) found large positive life satisfaction gaps between fathers and childless men in Ireland. These wellbeing gaps seem to be persisting and if national policy makers wish to design policies to improve the wellbeing of parents; and avoid discouraging childbearing, further country specific research untangling the factors affecting parents' wellbeing is warranted.

Data preparation and methods

The analysis in this paper uses data from the Irish component of the European Social Survey (ESS). The ESS consists of “rounds” of data collection undertaken every 2-years with each round involving an independent cross-sectional sample in each country. In this study we utilise three rounds of data in one pooled dataset; Round 3 (2006/2007), Round 5 (2010/2011) and Round 8 (2016/2017). Pooling the data increases the sample size and these years take account of pre, during and post economic recessionary times in Ireland (Fitzgerald, 2014).

In order to represent a population it is important to take into account the survey design correctly. This is of particular importance in cross-country studies because it is usually necessary to use a different design in each nation as outlined in Lynn et al. (2007). According to Kaminska (2020) sample designs are designed to achieve a minimum effective sample size in the most cost effective way for each country, taking into account each country context. Thus, countries differ in selection probabilities, clustering and stratification (Kaminska, 2020). Although our analysis is focused on a single country we do combine three rounds from the ESS covering the aforementioned time periods, and so it is important to describe how and what estimation procedures were adopted for jointly taking into account three different sampling design plans. Correct specification of estimates requires the use of appropriate indicators of components of the sample design (Kaminska and Lynn, 2017).

The target population for each country in the ESS is defined as “all persons 15 years or older resident in private households within the borders of the nation, regardless of nationality, citizenship, language or legal status” (Lynn et al., 2007). In Ireland the sample is drawn from a database of all addresses using strict random probability methods meaning that one person is selected at random at each address in a random sample of addresses. Therefore, it is important to correct for different selection probabilities because people living on their own have greater selection probabilities that people in households with more than one adult. Failing to correct for this (by not using the correct weights) will result in samples being heavily skewed towards one-adult households and estimates consequently being biased towards the opinions of such individuals (Kaminska, 2020). Errors due to non-response and sampling error may also lead to biased estimates, however the use of post-stratification weights reduce this impact (Kaminska, 2020).

The ESS recommend the use of analysis weights (anweight) in all analysis (European Social Survey, 2020). This weight is suitable for all types of analysis, including single country studies (Kaminska, 2020). The analysis weight corrects for differential selection probabilities within a country as specified by sample design and for nonresponse, noncoverage and sampling error related to four post-stratification variables (Kaminska, 2020). It was necessary to derive the anweight for use in our analysis as the ESS does not includes it prior to Round 9 (2018). For rounds 1 to 7 the sample design indicators (indicating the probability of selection for each sampled unit at each stage of the sample design) are recorded and supplied in separate sample design data files for each country and for each round. Following Kaminska (2020) the three sample design indicators files were merged with the main data files for each of the three datasets used. After pooling the datasets, the anweight was constructed in Stata 14.2 using the post-stratified design weight and the population size weight. Furthermore, we use the svy command in Stata 14.2 to provide estimates that fully take into account aspects of the sample design.

Descriptive statistics

In total, the pooled dataset consists of 7,133 individuals in three time-points. As this paper is primarily interested in analysing the life satisfaction of working parents, we restrict the sample to those individuals aged between 20 and 50 years old who reside with their minor children (<18 years old) (as per Aassve et al. (2012) and Le moglie et al. (2019)). To do so, questions relating to parenthood and household members are utilised. With regards to parenthood, the ESS asks if the individual currently resides with children (yes/no). Other questions were asked about members of the household and the respondent's relationship to the other members of the household. This information was used to confirm that the individual residing with children was indeed a parent and not another relative. We exclude empty nesters as minor children have a greater effect on life satisfaction (Evenson and Simon, 2005). This restricted sample accounts for 26% (n = 1,830) of the total sample; of which, 65% (n = 1,194) are mothers and the remainder (35%) are fathers. Table 1 presents the descriptive statistics.

For the dependent variable we used the responses to the following question on life satisfaction: “All things considered, how satisfied are you with your life as a whole nowadays?” Respondents are asked to choose a number on a scale from 0 to 10 (where 0 means extremely dissatisfied and 10 means extremely satisfied) that best represents a subjective rating of their own life satisfaction. This results in an ordered dependent variable. The inclusion of, and wording of this question has not changed in all eight rounds of the ESS and so we have a consistent measure across all respondents in the three rounds used in this analysis. Furthermore, this measure has been used extensively in life satisfaction studies and has been shown to be a reasonably valid and reliable indicator of subjective wellbeing (Diener et al., 1999; Kahneman and Krueger, 2006). It correlates well with another wellbeing measure, “happiness”, contained in the data (r = 0.75).

The main explanatory variables pertain to the individuals' parental status. We take into consideration the number of children and their ages. As well as confirming respondents' relationships to each household member, the questionnaire also contains each household member's birth year. These data were used to create four separate variables to indicate the age of the children using the following age categories; 0–4 years, 5–12 years and 13–18 years, and the number of children in each age category. These categories were created to reflect pre-school; primary education and secondary education years.

Other explanatory variables considered, and in line with previous literature, are economic climate, feelings about income, employment status, working hours, partner is employed, two income household, age, gender, education and relationship status (living with a partner). The former is controlled for using three dummy variables for pre-recession (2006/07), crisis years (2010/11) and post-recession (2016/17). Income is captured in the ESS as a categorical variable using income deciles. Respondents are asked to indicate their total net household income from all sources. However, the income deciles are not consistent across the 3 rounds of the ESS used in this study and therefore it was not possible to accurately harmonise the data into consistent income deciles. Furthermore, the data on reported income suffers from a large degree of missing/not reported data; 455 income observations are “missing” (25% of the sample). As it was not possible to use income directly, a variable that captures how the individual feels their household is coping on the present household income is used as a proxy for income in this study (as similarly employed by Pollmann-schult (2018)). Four dummy variables are created to indicate if the household is, living comfortably on present income; coping on present income; finding things difficult on present income or finding things very difficult on present income. 99% of respondents answered this question. Most parents report that they are coping on their present income (45%) followed by those who say they are living comfortably (29%). About 25% of individuals are finding it difficult or very difficult to cope on their present income.

With regards to employment status, the ESS asks respondents to report the main activity they have undertaken in the past 7 days. From this information we constructed 3 categorical variables to indicate if an individual is in employment/paid work, unemployed (both looking and not looking for a job) and out of the labour force (in education, permanently sick/disabled, retired, community/military service, housework, other). 60% of parents report that they are in employment/paid work, 30% are out of the labour force and 8% report being unemployed. Respondents were also asked to indicate the total hours normally worked per week in their main job including overtime. The mean hours worked is 37.17 h per week (SD 13.74). Due to their relevance to this study we include; a variable to control for partner employment (62% report their partner is in paid employment) and partner and own employment (2 income household comprise 39% of the sample of parents).

Other individual information pertains to age, education and relationship status. The mean age of parents is 39.5 years (SD 7.1). Respondents' highest level of education was gathered using the ISCED educational coding system. These were subsumed into three categories as follows: ISCED 1 and ISCED 2 representing primary/lower secondary education (22%); ISCED 3 and ISCED 4 representing upper secondary and post-secondary/non-tertiary education (32%) and ISCED 5 representing tertiary education (45%). With regards to relationship status, the ESS holds data on whether the respondent lives with husband/wife/partner at house or not. This variable is chosen over data on “marital status” as some of the categories differ between rounds. 87% of parents live with a partner and the remainder do not.

Of particular importance to our analysis is the effect of employment on the life satisfaction of parents at various stages of parenthood. To that end we include 3 interaction variables that capture parental employment status and having any child(ren) in the different age categories. Here, 21% of the sample are working parents with a child aged between 0 and 4 years 31% of working parents' children lie between the ages of 5 and 12 years and 23% of working parents have child(ren) aged between 13 and 18 years old.

46% of mothers are in paid work, 45% are out of the labour force and 6% are unemployed. While 81% of fathers work, 11% are unemployed, 7% are out of the labour force. On average, per week, mothers work fewer hours than fathers. Mothers on average work 32 h per week (SD 12.7) whereas for fathers the mean is 43.7 h per week (SD 12.3).

For the multivariate regression we employ an ordered probit technique, given the ordered nature of the dependent variable, to estimate the effect of parenthood on life satisfaction in Ireland. The ordered probit model is an extension of the binary probit model but was designed to deal with discrete dependent variables with ordered multinomial outcomes. The model can be expressed in terms of an underlying latent variable y*, which itself is not observed (Jones, 2007). This unobservable could be interpreted as the individual's underlying or “true” life satisfaction. However, subjective life satisfaction can be observed by capturing individuals' answers to a question on life satisfaction. Numerical values on a scale from 0 to 10 are frequently used and although the actual numbers are arbitrary, they do represent a ranking (Jones, 2007). Thus, the higher the value of y* (unobserved, underlying state of life satisfaction), the more likely the individual is to report a higher category of self-assessed life satisfaction (observed dependent variable) on an ordinal scale.

Equation (1) shows the model employed in this research:

The mean of y* depends on the explanatory variables contained in the vectors, and therefore the whole distribution shifts when the value of one such variable changes, in a direction dictated by the sign of the corresponding β coefficient. The interpretation of the β coefficients is in terms of the underlying latent variable model. For example, a negative β coefficient means that the corresponding variable decreases an individual's life satisfaction and a positive β coefficient signaling the opposite effect.

The range of values y* are divided into intervals based on the number of categories in the dependent variable. Threshold values are reported as cut points which represent the point at which an individual moves from reporting one category to another. The model does not also include a constant term, as the upper bound of the first interval is set to zero (Jones, 2007).

To run the ordered probit regression, explanatory variables are introduced into the model by making the latent variable y* a linear function of the explanatory variables and a normally distributed error term μ is added. So then, the probability of an individual reporting a particular value of y = j, is given by the difference between the probability of having a value of y* less than μ_j and the probability of the respondent having a value of y* less than μ_j-1 (Jones, 2007). Using these probabilities then it is feasible to use maximum likelihood estimation to calculate the probabilities of the models, which include the coefficients (βs) on the explanatory variables and the unknown cut-point values (Jones, 2007). As we have 11 categories (0–10, where 0 means extremely dissatisfied with life and 10 means extremely satisfied with life) so the range of values for y* should be divided into 11 intervals, each one corresponding to a different category of life satisfaction.

In total three ordered probit regressions are estimated. According to the ESS omitting weights can cause estimates to be biased (Kaminska, 2020). Therefore, analysis weights are used in all regression analyses presented hereafter. Model 1 estimates the factors associated with parents' life satisfaction (n = 1,828). Mothers' and fathers' life satisfaction models are then estimated separately in Model 2 and Model 3 (n = 1,193 and n = 635). In all models we control for year, income proxy, individual and partner employment status, age, education and if the individual lives with a partner. We also control for a number of parental characteristics; age of and number of children and the age of the child if the parent is working.