![[Review] Mostly Harmless Econometrics: An Empiricist's Companion (Joshua D. Angrist) Summarized](https://episodes.castos.com/660078c6833215-59505987/images/2320366/c1a-085k3-pkwj6jwoao64-h7pete.jpg)
Show Notes
Mostly Harmless Econometrics: An Empiricist's Companion (Joshua D. Angrist)
- Amazon USA Store: https://www.amazon.com/dp/0691120358?tag=9natree-20
- Amazon Worldwide Store: https://global.buys.trade/Mostly-Harmless-Econometrics%3A-An-Empiricist%27s-Companion-Joshua-D-Angrist.html
- Apple Books: https://books.apple.com/us/audiobook/the-covens-secret/id1506626271?itsct=books_box_link&itscg=30200&ls=1&at=1001l3bAw&ct=9natree
- eBay: https://www.ebay.com/sch/i.html?_nkw=Mostly+Harmless+Econometrics+An+Empiricist+s+Companion+Joshua+D+Angrist+&mkcid=1&mkrid=711-53200-19255-0&siteid=0&campid=5339060787&customid=9natree&toolid=10001&mkevt=1
- Read more: https://mybook.top/read/0691120358/
#appliedeconometrics #causalinference #instrumentalvariables #differenceindifferences #regressiondiscontinuity #MostlyHarmlessEconometrics
These are takeaways from this book.
Firstly, The Credibility Mindset and the Logic of Identification, A central theme is that good econometrics starts with a research design, not a clever regression. The book pushes readers to ask what variation identifies the causal effect and whether that variation is plausibly as good as random. This mindset frames empirical work as a sequence of choices: define the causal question, specify the treatment and outcome, and then defend the comparison group. Instead of relying on broad functional form assumptions, it emphasizes threats to validity such as omitted variables, selection into treatment, reverse causality, and measurement error. Readers learn to think in terms of counterfactuals, and to judge whether an estimating strategy recovers the average causal effect for the population of interest or only for a specific subgroup. The discussion also highlights why standard regression output can be misleading if identification is weak or if standard errors fail to reflect the true uncertainty. Practical guidance includes how to interpret coefficients causally, how to use control variables responsibly, and how to diagnose when adding controls changes the estimand. The overall message is that transparent assumptions and defensible sources of variation are the foundation of trustworthy applied results.
Secondly, Regression as a Tool for Causal Comparisons, The book treats regression as a flexible way to implement causal comparisons when the identifying variation is credible. It explains how regression connects to simple differences in means, how controls can adjust for observable confounding, and why the meaning of a coefficient depends on the underlying design. A major focus is on interpreting regression as conditional comparisons: the estimated effect compares treated and untreated units that are similar on included covariates. The authors stress that including many controls is not automatically a cure for bias, because unobservables can still confound the relationship and because over controlling can remove meaningful variation. Readers are guided through typical empirical challenges such as multicollinearity, functional form choices, and the use of nonlinear models, with an emphasis on when these choices matter for causal inference. The text also underscores the importance of correct standard errors, especially under heteroskedasticity and when data are clustered in groups like schools, firms, or regions. By tying regression mechanics to identification, the book helps readers move from running regressions to explaining what their regression is actually comparing, and why that comparison should be interpreted as causal.
Thirdly, Instrumental Variables and Local Causal Effects, Instrumental variables are presented as a workhorse method for dealing with endogeneity when a valid source of exogenous variation exists. The book explains the two key requirements for an instrument: it must shift the treatment and it must affect the outcome only through that treatment. It also clarifies why these conditions are strong and why instruments must be defended using institutional knowledge and careful argument, not just statistical tests. A major contribution of the modern IV perspective is the interpretation of IV estimates as local average treatment effects under common assumptions, meaning the estimate applies to compliers whose treatment status is changed by the instrument. This local nature has important implications for external validity and policy interpretation. The authors discuss weak instruments and why they can produce misleading estimates and confidence intervals, motivating diagnostics and robust inference. They also connect IV to two stage least squares, showing how it can be understood as a particular weighting of causal effects. By combining intuition with practical warnings, the book teaches readers how to use IV responsibly, how to interpret what IV identifies, and how to communicate results without overstating what the instrument can deliver.
Fourthly, Difference in Differences and Panel Data Strategies, Difference in differences is introduced as a design that leverages before and after changes in outcomes for treated and comparison groups. The identifying assumption is parallel trends: absent the intervention, the treated group would have followed the same trend as the control group. The book explains how this design can be implemented with regression using group and time fixed effects, and how the key estimate is driven by changes over time rather than levels. Readers learn how panel data can strengthen designs by controlling for time invariant unobserved differences, while also bringing new concerns such as serial correlation and dynamic treatment effects. The authors emphasize correct inference, often requiring clustered standard errors at the group level because outcomes within a group are correlated over time. They also discuss practical design checks like plotting pre trends, using placebo tests, and considering alternative control groups. While the text is known for promoting transparent and simple strategies, it also cautions that difference in differences can fail when treatment timing is correlated with outcome trends or when spillovers contaminate controls. Overall, it equips readers to implement policy evaluation designs that resemble natural experiments and to defend the assumptions behind them.
Lastly, Regression Discontinuity and Other Quasi Experimental Designs, Regression discontinuity designs are presented as a powerful approach when treatment assignment changes sharply at a cutoff, such as eligibility thresholds or test score rules. The identifying logic is that units just above and just below the cutoff are comparable, so any discontinuity in outcomes at the threshold can be attributed to the treatment, provided manipulation around the cutoff is limited and potential outcomes vary smoothly. The book covers practical implementation issues such as bandwidth choice, local functional forms, and the use of polynomial controls, often favoring simple local comparisons to reduce sensitivity. It also highlights diagnostic tools like checking for bunching at the threshold and testing continuity of covariates. The broader theme is that many credible studies rely on quasi experimental variation: cutoffs, policy rules, lotteries, and administrative quirks that approximate random assignment. The authors encourage researchers to prioritize designs with clear identifying variation and to present evidence that the design assumptions hold. Readers come away understanding why RD estimates are typically local, how to interpret them for populations near the cutoff, and how RD fits into a broader toolkit of methods aimed at credible causal inference.
- Amazon USA Store: https://www.amazon.com/dp/0691120358?tag=9natree-20
- Amazon Worldwide Store: https://global.buys.trade/Mostly-Harmless-Econometrics%3A-An-Empiricist%27s-Companion-Joshua-D-Angrist.html
- Apple Books: https://books.apple.com/us/audiobook/the-covens-secret/id1506626271?itsct=books_box_link&itscg=30200&ls=1&at=1001l3bAw&ct=9natree
- eBay: https://www.ebay.com/sch/i.html?_nkw=Mostly+Harmless+Econometrics+An+Empiricist+s+Companion+Joshua+D+Angrist+&mkcid=1&mkrid=711-53200-19255-0&siteid=0&campid=5339060787&customid=9natree&toolid=10001&mkevt=1
- Read more: https://mybook.top/read/0691120358/
#appliedeconometrics #causalinference #instrumentalvariables #differenceindifferences #regressiondiscontinuity #MostlyHarmlessEconometrics
These are takeaways from this book.
Firstly, The Credibility Mindset and the Logic of Identification, A central theme is that good econometrics starts with a research design, not a clever regression. The book pushes readers to ask what variation identifies the causal effect and whether that variation is plausibly as good as random. This mindset frames empirical work as a sequence of choices: define the causal question, specify the treatment and outcome, and then defend the comparison group. Instead of relying on broad functional form assumptions, it emphasizes threats to validity such as omitted variables, selection into treatment, reverse causality, and measurement error. Readers learn to think in terms of counterfactuals, and to judge whether an estimating strategy recovers the average causal effect for the population of interest or only for a specific subgroup. The discussion also highlights why standard regression output can be misleading if identification is weak or if standard errors fail to reflect the true uncertainty. Practical guidance includes how to interpret coefficients causally, how to use control variables responsibly, and how to diagnose when adding controls changes the estimand. The overall message is that transparent assumptions and defensible sources of variation are the foundation of trustworthy applied results.
Secondly, Regression as a Tool for Causal Comparisons, The book treats regression as a flexible way to implement causal comparisons when the identifying variation is credible. It explains how regression connects to simple differences in means, how controls can adjust for observable confounding, and why the meaning of a coefficient depends on the underlying design. A major focus is on interpreting regression as conditional comparisons: the estimated effect compares treated and untreated units that are similar on included covariates. The authors stress that including many controls is not automatically a cure for bias, because unobservables can still confound the relationship and because over controlling can remove meaningful variation. Readers are guided through typical empirical challenges such as multicollinearity, functional form choices, and the use of nonlinear models, with an emphasis on when these choices matter for causal inference. The text also underscores the importance of correct standard errors, especially under heteroskedasticity and when data are clustered in groups like schools, firms, or regions. By tying regression mechanics to identification, the book helps readers move from running regressions to explaining what their regression is actually comparing, and why that comparison should be interpreted as causal.
Thirdly, Instrumental Variables and Local Causal Effects, Instrumental variables are presented as a workhorse method for dealing with endogeneity when a valid source of exogenous variation exists. The book explains the two key requirements for an instrument: it must shift the treatment and it must affect the outcome only through that treatment. It also clarifies why these conditions are strong and why instruments must be defended using institutional knowledge and careful argument, not just statistical tests. A major contribution of the modern IV perspective is the interpretation of IV estimates as local average treatment effects under common assumptions, meaning the estimate applies to compliers whose treatment status is changed by the instrument. This local nature has important implications for external validity and policy interpretation. The authors discuss weak instruments and why they can produce misleading estimates and confidence intervals, motivating diagnostics and robust inference. They also connect IV to two stage least squares, showing how it can be understood as a particular weighting of causal effects. By combining intuition with practical warnings, the book teaches readers how to use IV responsibly, how to interpret what IV identifies, and how to communicate results without overstating what the instrument can deliver.
Fourthly, Difference in Differences and Panel Data Strategies, Difference in differences is introduced as a design that leverages before and after changes in outcomes for treated and comparison groups. The identifying assumption is parallel trends: absent the intervention, the treated group would have followed the same trend as the control group. The book explains how this design can be implemented with regression using group and time fixed effects, and how the key estimate is driven by changes over time rather than levels. Readers learn how panel data can strengthen designs by controlling for time invariant unobserved differences, while also bringing new concerns such as serial correlation and dynamic treatment effects. The authors emphasize correct inference, often requiring clustered standard errors at the group level because outcomes within a group are correlated over time. They also discuss practical design checks like plotting pre trends, using placebo tests, and considering alternative control groups. While the text is known for promoting transparent and simple strategies, it also cautions that difference in differences can fail when treatment timing is correlated with outcome trends or when spillovers contaminate controls. Overall, it equips readers to implement policy evaluation designs that resemble natural experiments and to defend the assumptions behind them.
Lastly, Regression Discontinuity and Other Quasi Experimental Designs, Regression discontinuity designs are presented as a powerful approach when treatment assignment changes sharply at a cutoff, such as eligibility thresholds or test score rules. The identifying logic is that units just above and just below the cutoff are comparable, so any discontinuity in outcomes at the threshold can be attributed to the treatment, provided manipulation around the cutoff is limited and potential outcomes vary smoothly. The book covers practical implementation issues such as bandwidth choice, local functional forms, and the use of polynomial controls, often favoring simple local comparisons to reduce sensitivity. It also highlights diagnostic tools like checking for bunching at the threshold and testing continuity of covariates. The broader theme is that many credible studies rely on quasi experimental variation: cutoffs, policy rules, lotteries, and administrative quirks that approximate random assignment. The authors encourage researchers to prioritize designs with clear identifying variation and to present evidence that the design assumptions hold. Readers come away understanding why RD estimates are typically local, how to interpret them for populations near the cutoff, and how RD fits into a broader toolkit of methods aimed at credible causal inference.
