class: center, middle, inverse, title-slide .title[ # What is Econometrics? ] .author[ ### Brantly Callaway ] .date[ ### University of Georgia ] --- # What will we learn this semester? `$$\newcommand{\E}{\mathbb{E}}$$` <style type="text/css"> .alert { font-weight:bold; color: red; } .alert-blue { font-weight: bold; color: blue; } .remark-slide-content { font-size: 23px; padding: 1em 4em 1em 4em; } .highlight-red { background-color:red; padding:0.1em 0.2em; } </style> At a very high level, our goal for this semester is to develop skills related to <span class="alert">using data</span> to <span class="alert-blue">learn something</span> -- The three main topics that we will consider this semester are: - Linear regression - We'll become experts on this topic -- - Modern approaches to <span class="alert">prediction</span> using machine learning -- - Modern approaches to <span class="alert">causal inference</span> --- # What will we learn this semester? Because this is an advanced course, I also have the goal of learning both <span class="alert">the mechanics</span> of Econometrics *and* <span class="alert-blue">how it works</span> -- <center><img src="lawnmower1a.jpeg", width="400px", height="300px"> -- <img src="lawnmower2a.jpg", width="400px", height="300px"></center> --- # What will we learn this semester? In order to get to where we want to go, we will have to learn some tools along the way: -- - Statistical programming - In this course, we will use the `R` programming language - This is an important skill in its own right -- - Crash course in Probability and Statistics - This may be a review - But it is the "language" we'll be speaking this semester -- --- # Econometrics vs. Statistics? - Certainly, there is a lot of overlap -- - Main differences: - We'll focus on economic questions - Different data availability - Influenced by economic theory --- # Data Availability in Economics In statistics classes (and, for example, lab sciences), researchers often have access to <span class="alert">experimental data</span> -- - Ex: If a chemist is interested in studying the effect of "more heat" on some chemical process, they can often directly manipulate the amount of heat -- - Ex: A doctor studying the effect of some new medicine can randomly assign some patients to take it and assign others a placebo -- In some ways, this makes statistics in some disciplines not-too-complicated --- # Data Availability in Economics In economics, we very rarely have access to experimental data -- - Ex. Hard to imagine that you could convince countries to randomly assign their: - interest rate, immigration policy, patent policy, anti-trust laws, minimum wage, etc. - though all of these are of substantial interest to economists - and we often have quite a bit of related data -- We'll have to think quite carefully about how we can use existing data to learn in these settings - ...or if we can learn anything in these settings --- # Econometrics and Economic Theory Economics, as a discipline, has a rich history in economic theory. -- - Utility / profit maximization - Supply / Demand -- - Most economic models involve <span class="alert">tradeoffs</span> (no free lunch, etc.) - Econometrics can be useful for weighing tradeoffs --- # Econometrics and Economic Theory The "theory" part of much empirical work in economics is often the <span class="alert">key step</span> -- - "Confirmation bias" -- - Patterns in data that are "surprising" often have other explanations ("Correlation is not causation") -- - Ex. Classical music and baby's IQ -- - But other explanations often "matter" for results that "make sense" too -- - Ex. Education and wages --- # Econometrics and Economic Theory At a minimum, we'll develop skills to <span class="alert">think critically</span> about empirical research and/or empirical claims -- "Fixing" these critiques can be more challenging, but we'll: 1. Learn some "tricks" -- 2. Learn to be precise about potential weaknesses of our own research --- # Economic Data One of the exciting things about studying Econometrics is that there is a lot of available data -- Examples: - GDP of different countries [[Link: World Bank Indicators]](http://databank.worldbank.org/data/reports.aspx?Code=NY.GDP.PCAP.CD&id=af3ce82b&report_name=Popular_indicators&populartype=series&ispopular=y) - Micro data from Census Bureau [[Link]](https://usa.ipums.org/usa/) - Micro data from Current Population Survey [[Link]](https://cps.ipums.org/cps/) - Micro data on demographics and health in developing countries [[Link]](https://www.idhsdata.org/idhs/) - Annual Survey of Manufactures [[Link]](https://www.census.gov/programs-surveys/asm.html) -- That being said, a lot of important contributions to our understanding of the economy have come from <span class="alert">data collection</span> - This is usually not glamorous but important --- # Goals for the Econometrics 1. Prediction / Forecasting - If bank makes a loan to a firm, how likely are they to pay it back? - What do we think that U.S. GDP will be in 2024? - Predicting Covid-19 cases across locations/time periods -- 2. Policy Analysis / Causal Inference - What was the effect of increasing the minimum wage on employment? - Did Covid-19 lockdowns affect the number of Covid-19 cases, employment, etc.? - How much does smoking affect health outcomes? -- 3. Counterfactuals - What would happen if the Fed increased the interest rate? - What would happen if patent protections were decreased to 10 years? - What would happen if Georgia increased its minimum wage? --- # Two Examples of Econometrics Prediction - 2016 Presidential Election - I like this example because it is an example of not-very-good predictions - But in hindsight, it is fairly easy to see why Causal Inference - Causal Effect of Union Membership on Wages/Employment - This is an example of "regression discontinuity" --- # Example 1: Predicting Elections **2016 Election Forecast** * NY Times: Trump 15%, Clinton 85% * Huffington Post: Trump 1.7%, Clinton 98.0% * Five Thirty Eight: Trump 28.6%, Clinton 71.4% * Princeton Election Consortium: Trump 1%, Clinton 99% [[Link: Interesting Read on Why Most Election Forecasts Did Poorly]](http://fivethirtyeight.com/features/why-fivethirtyeight-gave-trump-a-better-chance-than-almost-anyone-else/) --- # Election <center><img src="election-map.png"/></center> --- # Example 2: Union Wage Premium On average, union members earn about 15% more than non-union members with similar characteristics (education, demographic characteristics, etc.) -- This could be the causal effect ("union premium") -- But there are also other potential explanations - Perhaps, among observationally similar individuals, union members tend to have higher productivity - Unions may tend to organize at relatively successful firms ("threat effects") -- [This "debate" may seem esoteric, but there are many researchers that think very carefully about these sorts of issues] --- # Example 2: Union Wage Premium <span class="alert">Idea/Trick:</span> - Whether a company becomes unionized is typically based on a secret-ballot vote by its employees - Compare very close "successful" union elections to "unsuccessful" elections -- - By construction, these companies should be "similar" to each other --- # Dinardo and Lee (QJE, 2004) <center><img src="data:image/png;base64,#union_premium_rd.png"/></center> <!-- Comments: * The middle line is the one that you should look at (there is just not a figure with the "simple" information by itself) * Table 2 in the paper indicates that an estimate that wages increased by 1.5% due to unionization though it is not statistically different from 0. Also, if you didn't know where the cutoff was, it is not obvious that you should pick 50%, this is an argument against seeing actual wage effects here. * Alternative explanations: (i) union power is lower in recent periods than in other periods (esp. due to more international competition, probably there is less "surplus" to be "negotiated" over) or (ii) these estimates are for firms that just barely became unionized, and the effect of unionization could be different for different firms, (iii) maybe low power (once you factor in uncertainty due to sampling error, estimates ranging from negative to 5 or 6% are more or less "compatible" with the data being presented) though it does seem to rule out large effects like 15%. * Finally, this might seem like an argument against unions, but they do not find effects on employment or firm survival either (you could argue that unions might decrease both of these). Basically, they are not finding effects of unions on anything. -->