Subjective Well-Being, Income and Relative Concerns in the UK

Abstract

We present an empirical model aimed at testing the relative income hypothesis and the effect of deprivation relative to mean income on subjective well-being. The main concern is to deal with subjective panel data in an ordered response model where error homoskedasticity is not assumed. A heteroskedastic pooled panel ordered probit model with unobserved individual-specific effects is applied to micro-data available in the British Household Panel Survey for 1996–2007. In this framework, absolute income impacts negatively on both completely satisfied and dissatisfied individuals, while relative income affects positively the most satisfied ones. Such an effect is asymmetric, impacting more severely on the relatively poor in the reference group. We argue that our results buttress the validity of the relative income hypothesis as an explanation of the happiness paradox.

Appendix: Dataset Features and Statistical Package

Quoting the official BHPS web site “The British Household Panel Survey” began in 1991 and is a multi-purpose study whose unique value resides in the fact that:

• it follows the same representative sample of individuals—the panel—over a period of years;

• it is household-based, interviewing every adult member of sampled households;

• it contains sufficient cases for meaningful analysis of certain groups such as the elderly or lone parent families.

The wave 1 panel consists of some 5,500 households and 10,300 individuals drawn from 250 areas of Great Britain” . From Wave7 (1997), there is oversampling of low income people for comparability with ECPH. “ Moreover, many more observations have been sampled for Scotland and Wales. Additional samples of 1,500 households in each of Scotland and Wales were added to the main sample in Wave9 (1999), and in 2001 a sample of 2,000 households was added in Northern Ireland, making the panel suitable for UK-wide research”.

Data in each wave are organized in different macro-groups: INDSAMP includes all sampled individuals (either respondents or not), INDALL is an individual level record for all members of the household, corresponding to the household grid, INDRESP includes responding individuals only. The same applies to household-specific data, collected into HHSSAMP, HHSAMP and HHRESP. Hence, when extracting the individual interview outcome (IVFIO) from INDSAMP/HHSAMP, we are taking more observations than those that we have in INDRESP/HHRESP. They are dropped when drop** according to IVFIO (we drop all the observations where the interview outcome was not 1, i.e. all the non-full interviews). Also, in order to maintain comparability with previous waves and random sampling, we keep only observations belonging to the original sample (MEMORIG=1 for INDRESP and HHORIG=1 for HHRESP), disregarding the data added from 1997, 1999 and 2001 mentioned before.

Here follows a list of BHPS codes for the raw variables used in our analysis, in alphabetical order:

Raw Data
age	Age from birth	Biographic	Continuos
fihhyl	Annual household labor income	Derived	Continuous
fihhynl	Annual household non-labor income	Derived	Continuous
hgemp	In paid employment—household grid	Self-reported	Binary
hllte	Health no indrance daily activities	Self-reported	Binary
hlstat	Health over last 12 months	Self-reported	1–5 ordered
lfsato	Satisfaction with life overall	Self-reported	1–7 ordered
mastat	Marital status	Biographic	5 different stati
nchild	Number of own children in household	Biographic	Continuous
race	Ethnicity	Biographic	5 different races
region	Region / metropolitan area	Biographic	18 UK sub-regions
sex	Gender	Biographic

By means of STATA, the PPOP model has been estimated using the standard command oprobit . For the HPPOP model, instead, we have made use of a STATA module by Williams (2010, 2011), known as oglm .