How robust are intergenerational income correlations?
1:00 pm to 2:00 pm, 20 November 2019
In this seminar, Per Engzell investigates to what extent undisclosed flexibility in research design can give rise to different estimates of (trends in) intergenerational income mobility.
This event is free.
Room G02Centre for Longitudinal Studies55-59 Gordon SquareLondonWC1H 0NU
Recent work highlights how "researcher degrees of freedom" – undisclosed flexibility in research design – can give rise to varying results. The study of intergenerational income mobility is no exception, with decisions ranging over income concept, unit of observation, functional form, treatment of outliers, and so on.
Using Swedish data on the population of children born in 1958–1972, Per Engzell and his colleague Carina Mood exhaust a model space of several hundred thousands of specifications to answer three questions:
- What is the range of reasonable estimates?
- Which specification fits data best?
- How sensitive are estimated trends?
As their analysis demonstrates, most intergenerational income correlations fall in the 0.15–0.30 range, but trends vary widely by specification. Linear correlations fit somewhat better than rank correlations, while log-log correlations (and hence, elasticities) fit poorly and behave erratically over time.
Even with more robust measures of association, different income definitions follow opposite trends over time: increasing persistence in family income and women's earnings, compared to flat or decreasing trends in men's earnings.
About the Speaker
Per Engzell is a sociologist and currently a Postdoctoral Prize Research Fellow at Nuffield College, University of Oxford. His research interests are in education, social and economic mobility, and international migration, and he also takes an active interest in social science methodology.More about Per Engzell