How Important Is Mortality in Explaining the Fertility Transition in Developing Countries?
Mateusz studied the MSc in Economics at UCL, focusing mainly on econometrics. Prior to that he studied PPE (Philosophy Politics and Economics) at Oxford, and is also a chartered accountant (qualification is called ACA and it's given by the ICAEW). Since finishing the MSc he has joined Oxera Consulting LLP as an Economic Consultant focusing on Energy and Analytics. He has also decided to take part in pro-bono economics programme, which helps charities for free to estimate their impact.
What was your thesis topic?
My thesis estimates the effect of changes in mortality rates on fertility rates at the national level, using the a cross-country panel with observations from 1970 to 2017. I find that much previous cross-country empirical research has used econometric techniques (e.g. OLS, Within-Group etc.) that are unlikely to be appropriate for modelling fertility, and likely suffer from significant omitted variable bias. The estimators I select (e.g. Arellano-Bond, Mean Group, and Pooled Mean Group) indicate that there is no relationship between mortality and fertility. I conclude that further cross-country research is of limited value because the results are highly variable (and often implausible) relative to micro-data-based research.
In what ways do you think your topic improves the world?
EA organisations like GiveWell, and international organisations like the World Bank, evaluate interventions with respect to relatively short-term impacts like whether a life is saved. They do not consider many of the long-term or indirect impacts. One long-term/indirect impact is that changes in mortality affect fertility, which in turn affects population growth. High population growth may be good or bad for humanity depending on your views on things like resource abundance. Therefore understanding the long-term/indirect impacts of development interventions seems very important. In particular, if you believe the long-termist paradigm (i.e. that most human utility is in the long-term future so the long-term is disproportionately important) then even small improvements in understanding these effects are important.
What do you think the stronger and weaker parts of your thesis are?
The main strength of this dissertation is that it performs a fairly thorough analysis/review of existing cross-country work. Section 4 provides a clear exposition of the econometric problems in this area, thereby demonstrating the inappropriateness of some previous estimation methods (e.g. OLS, Within-Group etc.). I then show that there is limited consistency in the results from cross-country methods, including my own. The conclusion that cross-country methods may be inappropriate is not ground-breaking but I have not seen it made by anyone else, so it seems an important (albeit boring) contribution.
The main weakness is that the methods used in the dissertation do not actually provide a very plausible answer (i.e. that there is no relationship between mortality and fertility). Theoretical models and micro-data-oriented methods both suggest a positive association between mortality and fertility, and I suspect the inconsistencies in the cross-country literature reflect data issues rather than genuine ambiguity in the relationships. Further, due to this being my first large coding assignment, there may also be programming errors – however looking into these is likely to be lower-impact than doing a new micro-data-based analysis because the inconsistencies in other papers still stand.
What recommendations would you make to others interested in taking a similar direction with their research?
As eluded to, the main area for future research is the micro-data-based literature. This will likely be subject to similar econometric problems to the cross-country literature in terms of mortality not being exogenous and lack of good instruments but I suspect measurement error may be lower and there is obviously no aggregation. Also datasets are much larger. There are not many papers published in this area so there are real opportunities for high impact. A lot of the (in my opinion) best micro-data-based analysis has used duration analysis so application of this technique to new datasets would be useful to confirm existing results. Further, I have not seen Arellano-Bond applied; whilst this approach does not resolve censoring problems like duration analysis does, it allows for instrumentation in an area where instruments are difficult to find. Some good datasets to use would be:
- IPUMS provides census data from a load of different countries – I think you need to query the database using SQL to get the data extract you want, but if you don’t know SQL I recommend asking someone at Effective Thesis (or sending an email to Computer Science undergrads or something) to help you;
- Whatever dataset is used in this paper – they essentially just estimate the Intention to Treat (ITT) so this could easily be expanded to do 2SLS – this is actually what I wish I had done for my dissertation; and
- General Household Surveys – which are available for download from the World Bank (e.g. this one) – you will need to download an entire series so if you are using Arellano-Bond for example you will need a country with at least 3 surveys.
I would advise against using the Demographic Health surveys because these are repeated cross-sections rather than panels.
A completely different research approach would be to build a population growth model using inputs from previous micro-research, and see how population changes in response to changes in mortality. This should be somehow microfounded.
I have been relatively brief in the above so if you have any questions feel free to reach out to me on Mateusz.firstname.lastname@example.org – I like discussing econometrics so will be happy to talk and will not consider it a burden.