21 January 2016  1.00pm - 2.30pm

Mastering Metrics

Jörn-Steffen Pischke, Professor, Economics Dept, London School of Economics

Steve began by considering some of the key questions that we might like to answer by using econometrics - from the benefits of going to university to the impact of monetary policy on the economy. One of the big problems in undertaking statistical research is that we don’t know the counterfactual. If someone makes a decision between two competing choices, selection bias can intervene - for example, an average person going to a hospital is actually unhealthier than someone who doesn’t go, but that is because we are looking at two different types of people!

 Another example relates to health insurance - at first glance, those with insurance seem healthier than those that don’t have it. The insured tend to be older, better educated, have slightly smaller families, are more likely to be employed and have higher family incomes. Steve cited the RAND experiment of healthcare in the US in the 1970s. This showed that people receiving more healthcare were not significantly more healthy - which seemed a disappointing result. A study by Oregon painted similar conclusions.

Why are we doing so many experiments? Development economics has been transformed by experiments over the past twenty years, not least because experiments are cheaper to run in developing economies. Experiments can also be used as benchmarks for framing one’s thinking - as an econometrician, consider what experiment you would like to run, then try to mimic it with available statistics as best as possible.

 Steve then turned his attention to education, beginning by comparing average starting salaries of LSE versus City University graduates. He then shifted focus to the US, where there is greater variation in tuition fees than in the UK. Is there a payoff from attending a private vs. public university in the US? We don’t have an experiment, of course, so we must use the data we have to undertake the best “apples-to-apples” comparison we can. We could match individuals with similar characteristics such as family income and SAT scores in order to try to compare similar types of people as to whether university choice is important. But that may not be sufficient - a better way to be sure we are comparing similar types of students is by looking at those who applied, but were rejected, from private universities.

 It is found that students receive 9% higher earnings from going to a private university based on a simple regression methodology. But when we control - as we must do - for selectivity bias, this is not the case. Just like healthcare, Steve suggests there is little indication that the more you spend on education the greater the eventual benefit - however that benefit may be measured.

 Steve’s message? Start any analysis with a good question and look for the appropriate data. Don’t be naive when thinking about causation in your empirical models. In his own words, “fruitful application does not demand extraordinary mathematical sophistication” but simply firm evidence-based foundations.