Conferences

The paper explores the use of quantile regression method to cross-national social research. Drawing on data from the European Social Survey, the paper shows how – compared to conventional linear (multilevel) models – quantile regression can extract more information from the same data.

The method complements standard (multilevel) regression analysis in situations where the latter fails to pick up and adequately describe the effect of the explanatory variables on the distribution on the dependent variable.

Conventional regression estimates the effect of an explanatory variable on the mean of the distribution of the dependent variable. Summarizing the relationship between explanatory and response variable in this way is useful and valid if the significance, direction and strength of relationship is not systematically different at the lower, medium or upper end of the distribution – or if such differences do exist, they can be deemed unimportant for the substantive research question.

However, this is often not the case. A range of research questions could be better addressed and important cross-national differences captured more adequately by considering differences in the effect explanatory variables on the variance or the skew of the dependent variable (e.g. effects on inequality and heterogeneity in Y), regression effects on locations on the distribution of the dependent variable Y other than the mean (e.g. the 10% most deprived respondents rather than the mean respondent), and changes in the size and direction of the regression effect as we move along the entire distribution from lowest through to the highest percentile (e.g. the size of the gender gap in income might become smaller as we move from the least, to average and top earning men and women within one country, but not within another country).

Whilst standard regression limits the researcher to the estimation of regression effects on the mean of the response variable, quantile regression allows estimating the effects of explanatory variables on all these other features of the distribution of the response variable.

The paper gives a non-technical introduction to the method. It suggests four diagnostic questions that help the applied researcher identifying when quantile regression is a useful alternative to conventional linear regression. Key differences between quantile and linear regression in terms of data requirements, model assumptions, the practicalities of model estimation in STATA, and presentation and interpretation of the results are illustrated using data from the “subjective well-being” module fielded in Round 3 of the ESS.

The findings from the quantile regression analysis challenge our current understanding of the relationship between life satisfaction and socio-economic circumstances, affect experience and social functioning, and offer further explanation of European national differences in subjective well-being.