Discussion Paper

No. 2015-63 | September 23, 2015
Issues in the Estimation of Dynamic Happiness Models: A Comment on "Does Childhood Predict Adult Life Satisfaction?"

Abstract

This short note offers methodological comments on an Economic Journal article (Frijters, P., Johnston, D.W. and Shields, M.A. (2014). Does Childhood Predict Adult Life Satisfaction? Evidence from British Cohort Surveys. Economic Journal 124(580): F688–F719). The comments consider its use of a dynamic model – the inclusion of a lagged dependent variable – and its employment of General Method of Moments (GMM) estimation.

JEL Classification:

I31

Assessment

  • Downloads: 1566

Links

Cite As

Alan T. Piper and Geoffrey T. Pugh (2015). Issues in the Estimation of Dynamic Happiness Models: A Comment on "Does Childhood Predict Adult Life Satisfaction?". Economics Discussion Papers, No 2015-63, Kiel Institute for the World Economy. http://www.economics-ejournal.org/economics/discussionpapers/2015-63


Comments and Questions


Anonymous - Referee Report
November 05, 2015 - 11:22

This paper raises two concerns on the empirical analysis of Frijters et al. (2014) in which adult life satisfaction is predicted by certain childhood and family characteristics. Piper and Pugh (2015) replicate the results in Frijters et al. (2014) by using two data sets in order to present comparisons: the ...[more]

... German Socio-Economic Panel Data (G-SOEP) and the British Household Survey (BHPS). The first concern is on the use of ordinary least squares (OLS) in the estimation of adult life satisfaction with lagged life satisfaction as an independent variable. They argue that the use of OLS causes point estimates on lagged life satisfaction to be upward biased. The second concern is about the result in section 3.5 of Frijters et al. (2014), in which an autoregressive model is used by regressing contemporaneous adult life satisfaction on the past life satisfaction. Since the number of instruments and the diagnostic test outcomes are not reported in the paper, the authors have concerns about the validity of this result, and argue that the current debate on the instrument validity used in system GMM estimation makes the lack of instrument presentation a very important problem of the paper.

Major Comments:

1) The paper should put forward a detailed discussion of why G-SOEP and BHPS are good substitutes to the British cohort surveys used in Frijters et al. (2014) and provide information on the variables that are used to replicate the results in Frijters et al. (2014). This is an important point, especially because Frijters et al. emphasize the importance of early life characteristics on adult life satisfaction and criticize the BHPS and G-SOEP for their lack of childhood characteristics and early family circumstances of the adult sample. In this respect, estimates of adult life satisfaction might be weak compared to the one in Frijters et al. (2014). The authors should discuss this point.

2) The use of OLS with lagged explanatory variables is criticized since it biases the results, and the use of system GMM is suggested. Since Frijters et al. additionally estimate adult life satisfaction with system GMM, the first concern seems to be resolved already. However, the second concern stays valid since there is no explanation in Frijters et al. (2014), either qualitative or quantitative, on the instruments used in system GMM.

3) It would enrich the paper to give examples on how the econometric issues raised about Frijters et al. apply to other contexts, in addition to studies of well-being.

Minor Comments:

1) The use of OLS estimation with lagged life satisfaction is presented in section 3.4 of Frijters et al. (2014) instead of section 3.3 as written in Piper and Pugh (2015) (page 2).

2) Results of Frijters et al. (2014) should be included in the tables for the reader to better see the differences between their estimates and the estimates in Frijters et al. (2014).


Alan Piper - Reply to reviewer
November 09, 2015 - 21:50

Please see the attached file.


Anonymous - some comments
November 10, 2015 - 11:08

This paper makes the following relevant points (i) OLS is generally not appropriate to estimate the coefficient of a lagged dependent variable and (ii) details of instrumentation and diagnostic tests should be clearly reported when using GMM to be able to assess instrument validity. Although not mentioned in this ...[more]

... note, presumably details on whether one-step versus two-step GMM is used, how standard errors are obtained (e.g. Windmeijer correction or not), as well as whether the instrument set is collapsed should also be reported.

These points apply generally to any GMM analyses but are worth repeating - especially in the context of the happiness literature - where researchers have typically ignored dynamics in the outcome variable.

A downside to this note is that it is specific in its criticism of Frijters et al., (2014) but without being able to use the same dataset as Frijters et al., (2014) this note cannot shed any light on the validity of the analyses reported in Frijters et al., (2014), despite being able to show problems arise when using other datasets. Perhaps Frijters et al., (2014) can share their data with the authors of this note so that their analyses can be subject to appropriate probing?


Alan Piper - Thanks for your comments.
December 03, 2015 - 13:18

Thank you for your comments and interest. We agree with you that whenever GMM is used there is much that needs to be reported. Regarding your last paragraph, please see our two responses elsewhere among the comments (Alan Piper - Reply to reviewer November 09, 2015 - 21:50; Alan ...[more]

... Piper - Response to Anonymous - Referee Report (13-11-2015) and rejoinder to Paul Frijters, David W. Johnston, Michael A. Shields – Reply (20-11-2015). December 03, 2015 - 13:06 . This issue is addressed in both comments.


Anonymous - Referee Report
November 13, 2015 - 11:53

This methodological note reviews the findings of a recent paper (Frijters et al., 2014) published in The Economic Journal, which studies the effects of childhood characteristics on adult life satisfaction using British (NCDS) cohort data.

Specifically, it replicates their findings with British (BHPS) and German (GSOEP) panel data, ...[more]

... pointing towards (i) potential bias in their parameter estimates when using lagged life satisfaction as an independent variable under OLS and (ii) potential variability in their findings depending on the number and types of instruments used under GMM.

Both points are very valid, and they raise questions about the accurateness of the findings in Frijters et al. (2014).

My major comment:

* The authors criticize Frijters et al. (2014) for not reporting details about the number and types of instruments, as well as diagnostics under GMM. At the same time, however, the authors do not report details, such as which variables and variable definitions are used, in their replications. Ideally, any methodological note that aims at pointing towards differences in findings should use the same variables and variable definitions, as well as models, to the extent possible. It is difficult to say whether this is the case here.

That said, it should be possible to obtain the data used in Frijters et al. (2014) for replication purposes, as The Economic Journal has a strict data policy, see http://www.res.org.uk/view/datapolicyEconomic.html.

My minor comment:

* In my opinion, this methodological note would benefit greatly from a more general discussion of the advantages and disadvantages of cohort relative to panel data for investigating the given research question.


Paul Frijters, David W. Johnston, Michael A. Shields - Reply on Alan T. Piper and Geoffrey T. Pugh´s comment on "Does Childhood Predict Adult Life Satisfaction?"
November 20, 2015 - 08:44

see attached file


Alan Piper - Response to Anonymous - Referee Report (13-11-2015) and rejoinder to Paul Frijters, David W. Johnston, Michael A. Shields – Reply (20-11-2015).
December 03, 2015 - 13:06

Please see attached file