Social Comparisons on Subjective Well-Being: The Role of Social and Cultural Capital

This four effects, standard in literature to interpret the effect of income comparisons, will be explained in detail in Sect. 2.

In particular, while life satisfaction coincides with an economic point of view on well-being and represents a possibility to satisfy own preferences (Diener 1984), happiness reflects the degree to which an individual judges the overall quality of his or her own life as a whole favorably (Headey and Wooden 2004).

D’Ambrosio and Frick (2012) refer to this relationship as “the absolute component”.

Similarly, the notion of “societal expectations” appears in Zuzanek (2012). This notion refers to a generalized longing for better socio-economic conditions compared with the past rather than with other income groups or countries. In D’Ambrosio and Frick (2012), the notion of dynamic absolute component also captures the idea of the influence of own income history.

The usual criteria to construct the reference groups is the geographical area where the respondent lives (Blanchflower and Oswald 2004; Luttmer 2005; Graham and Felton 2006), the individual’s age cohort (McBride 2001), educational level, age, and employment status (van de Stadt et al. 1985) or region, age, and educational level (Ferrer-i-Carbonell 2005).

Note that age, education, and other factors used in the literature to build reference groups are basic determinants of income.

To avoid confusion between this concept of satisfaction, the original denomination of Hey and Lambert (1980), and the one coming from usual measures of subjective well-being, in this paper we adopt the term “affluence” for comparisons with individuals of lower income (Deaton 2003).

McBride (2001) and Distante (2013) also use the idea of a ratio but with respect to the average income of the reference group, that is, under the mean dependence framework. The findings suggest that the impact of mean income is more severe for the poor.

In a similar way, Budría and Ferrer-i-Carbonell (2012) examined whether the rank effect differs between individuals endowed with different sets of non-cognitive skills. They found a large heterogeneity in the importance of income comparisons.

As Ferrer-i-Carbonell (2005) and Ferrer-i-Carbonell and Frijters (2004) pointed out, an ordered logit model with individual fixed effects should yield similar results as the approach used in this paper.

As Norton et al. (2004) have shown, the interpretation of interaction terms in linear regression models does not extend to nonlinear regression models, and the computation of the marginal effects and statistical significance of the parameters in the latter case involve an additional difficulty.

The data employed in this paper were extracted using the Add-On package PanelWhiz or Stata^®. PanelWhiz (http://​www.​PanelWhiz.​eu) was written by Dr. John P. Haisken-DeNew (john@PanelWhiz.eu). See Haisken-DeNew and Hahn (2010) for details. The Panel Whiz generated a DO file to retrieve the data used here, and is available from the authors upon request. Any data or computational errors in this paper are our own.

As pointed out by D’Ambrosio and Frick (2007, 2012) monthly income might be an inferior measure of economic well-being when compared to annual income since it tends to neglect certain irregular income components, but to better fit the time-dependent measures of subjective well-being, the more regular income components are required.

We have dropped from the presentation the index of Ferrer-i-Carbonell (2005) under mean dependence framework since it does not allow to incorporate proximity considerations. However, our results displays no effect of comparisons on social well-being, contrary to Ferrer-i-Carbonell (2005), but in our model there are some other variables such as social capital, cultural capital and interactions that she did not incorporate. Results are available upon request from authors.

We performed a sensitivity analysis concerning the choice of our parameters. Specifically, we calculated the social comparison measures for 15 values of the width of the interval (h), 25 values for the aversion parameter (β) and three values for the habituation discount δ. We then estimated the full model without interactions and with all the measures and calculated the residuals. We have chosen the value of parameters that minimize the mean error and the maximum error. It is important to note that the magnitude of the differences are of order 10⁻³ and 10⁻² for the case of the mean and maximum error, respectively, in the case of the indexes with width of the interval (h); and of order 10⁻⁵ and 10⁻³ for the case of the mean and maximum error, respectively, in the case of the indexes with the aversion parameter (β). That is, the error that one can incur with any other parameter is small. In terms of the habituation discount, the error is of order 10⁻⁶ and 10⁻⁴, respectively. The influence on the estimated parameter will be discussed below. The full set of results for all variables and parameter configuration is available from the authors upon request.

As seen in Bartolini et al. (2013) we have chosen the reported frequency ‘‘at least once a month’’ as a threshold because it well captures the sample variation. However we have checked the effect of changing frequency of visits and results do not change.

Principal components analysis is a statistical procedure using a set of variables to extract the few orthogonal linear combinations of the variables that capture the common information in the most satisfactory way. Consistently, the first principal component of a set of variables is the linear index of all the variables that capture the largest amount of information common to all the variables. Thus, our independent variables that use information coming from different domains are additive and linear (Van Praag and Ferrer-i-Carbonell 2004).

The years for which we actually have observations are 1994, 1996, 1997, 1999, 2001, 2005, and 2007.

We have also tried with hot deck and unconditional mean imputation methods and results do not change.

This decision means that the period under consideration starts in 1998. Moreover, we have dropped individuals that did not report income in four periods in a row. For value of δ as pointed out in footnote 14 we have simulated errors with the other values for parameters.

Those are the parameters that minimize the errors, see footnote 14.

Related literature has found a U-shaped dependence of well-being with respect to age. We have included age squared in our specification and alternatively two dummies for young and old people, and we find this result.

We have performed the specific test for this statement.

This signal effect also appears in social psychology research (Festinger 1954; Taylor and Lobel 1989). Hirschman and Rothschild (1973) suggested that individuals can derive positive utility from observing other people’s progression and comparing it with the positive signal (they call this effect the tunnel effect).

Note that the index in the first column assigns equal importance to all comparisons, in columns 2–3 the inclusion of a linear decreasing weighting function gives more importance to closer individuals and Paul’s index (columns 4–5) attaches even more importance to the closer individuals due to the nonlinear weights.

In dealing with job satisfaction, Bygren (2004) found that individuals primarily consider more general comparisons (others in their occupation or labor market), while comparisons with co-workers and their own past seem to be of minor importance.