applied micro slides

Transcript

1 How Give an Applied Micro Talk to Unauthoritative Notes Jesse M. Shapiro Chicago Bo oth and NBER

2 Motivation audience do es not care ab out your topic Your You have 1-2 slides to change their minds Make them count anecdotes facts p olicy questions

3 Question State research question a Policy/counterfactual what would happ en if...? question: Estimate of an imp ortant deep parameter: how forward-lo oking are consumers? revenue of an imp ortant theoretical prediction: do es Test equivalence hold in... Or, b etter yet, all three!

4 Not Research Questions the research are motivated by economics and Applied questions not economics literature. Applied research questions are not What happ ens if we apply the X mo del to industry Y? Z What we change assumption if of the X mo del? happ ens What happ ens when I re-estimate so-and-so's mo del on some other data?

5 This Pap er what pap er do es and why Outline your Convey you have something to add why Revisit the consumption CAPM using new high-quality consumption data. one Revisit the consumption CAPM b ecause no NOT has estimated it in a few years.

6 Preview of Findings the leave is ab out to Assume audience Make they walk out with something sure Be tangible but terse magic Just of your metho dology so results don't feel like enough Not so much that you crowd out the ndings

7 Go o d Level Detail of reduces adoption mandatory State leave of women's wages maternity by 5% No eect for women past fertile age No for men eect Implies approximately $0.75 of every $1 sp ent on maternity b enets are paid by mother

8 Not Detail Enough Incidence of wages leave p olicies mostly on maternity

9 To o Much Detail state-years Across observations ranges from 543 (Delaware in of to 17,645 Numb er 1976) York in 2005) (New wage for women is $17.49/hour (2008 Mean median is dollars); $15.12/hour wage is men is $25.16/hour (2008 dollars), median Mean for $22.99/hour Average annual change in women's wages is 1.34% Regression mo del with state and year xed eects Weighted by of observations in state-year numb er outliers Tukey's Exclude metho d using standard Cluster (Bertrand Duo Mullainathan 2003) errors Eect of maternity leave adoption is estimated to b e -$1.70 hour (SE = $0.30) for women p er $-0.21 p er hour (SE = $0.20) for men Compare to cost of $2.25 p er hour of providing leave

10 Data

11 Goals State the source of each variable clearly wondering confusion Prevent one should b e later: no Where did that come from? Is that measured at the state level or the county level? Anticipate over pure measurement and address them now concerns Are data sources reliable? Do the concepts you measure approximate those in your mo del?

12 Credit Be to get credit for sure data Novel ways of measuring something New sources New variation of But no one cares that This dataset to ok a long time to download the There a lot of dierent ways to weight are data and I had to read a manual

13 Your Underwear one to see your underwear No wants no one wants to know how you pro cessed the data And to I by state, county, year, and gender collapse make the dataset First easier to lo ok at Then I divide all variables by the 2008 CPI Then I observations with missing wages remove that I Then with wages remove are greater than observations $100/hour... Then I collapse by state, year and gender Try this: Average wage by state, year and gender, excluding outliers (>$100/hour in 2008 dollars)

14 Mo del

15 Be Explicit Panel data mo del with year and state xed eects Identication comes from exogenous law changes

16 Be More Explicit data with year and state xed eects Panel mo del ε = α y + δ + + β x t it it it i with y t = average wage in state i , year it = state xed eect α i δ eect = year xed t = error term ε it changes comes from exogenous law Identication

17 Be Even Explicit More data with year and state xed eects mo del Panel ε = α + + δ + y β x t i it it it with t = average y in state i , year wage it α eect = state xed i δ = year xed eect t = error term ε it Identication: E ( ε , | x 0 ,α ) = ,δ t it it i i.e. law eects are exogenous conditional on xed changes

18 Dene Bottom Line Your Let γ b e the average cost (p er hour) of providing maternity leave maternity Dene as the fraction of β/γ leave costs paid by the worker

19 Even If Better You can lay out explicit economic assumptions that justify your econometric assumptions Your mo del connects directly questions well-dened p olicy or welfare to

20 Pause To the most imp ortant Discuss of your mo deling approach vulnerabilities Why you think your mo del is a go o d approximation plausibility What will do to assess you of your assumptions / sensitivity

21 Do Pause To Not Try to anticipate every p ossible criticism Talk ab out the other mo dels you have tried Discuss ne p oints that no one will think of anyway

22 By Now audience The in your question b elieves what you measure and how understands your what will do with you data and why understands Otherwise the audience is lost no will b e able to appreciate your ndings one the talk is already over, you just don't know it No pressure though

23 Interlude: Slides

24 Principles for Design Slide reader of Unlike audience can't skip or browse pap er, So every word is precious Slides b e clear should Slides detail b e sparse: no extraneous should

25 Content While are talking, some p eople are not listening you they at lo oking Instead your slides are the Make tell the story with your voice slides (Can you hear me now?) don't Don't on a slide you anything plan to talk ab out put Amount of space you devote should corresp ond to the emphasis you intend

26 Unless You Really Want to Stress Something Pacing No pauses

27 You Really Want to Stress Something Pacing No pauses Unless

28 Really Want to Stress Something Pacing No pauses Unless You

29 Want to Stress Something Pacing No pauses Unless You Really

30 to Stress Something Pacing No pauses Unless You Really Want

31 Stress Something Pacing No pauses Unless You Really Want to

32 Something Pacing No pauses Unless You Really Want to Stress

33 Pacing No pauses Unless You Really Want to Stress Something

34 Do cumentation Your is a complete description of pap er you did and what you what learned Slides cannot b e completethere is not enough time Leave to the pap er do cumentation Use story talk to tell your your

35 Scaling A minute talk is not a 90 30 talk where you talk three times minute faster (Hat-tip to Matthew Rabin for teaching me this one.) the Cho ose and detail for emphasis amount of time you have

36 Results

37 Figures is gures wherever p ossible to tell the story of what Use in the data More honest More complete More interesting More p ersuasive

38 Tables tables to summarize key magnitudes Use Not to Show co ecient on every control variable (unless these tell an imp ortant story) Show every robustness check you did (can summarize these in bullets) Always b e telling your story

39 Bottom Line a line Have b ottom single A or (ideally) quantitative take-away qualitative Measurement error in consumption data explains 27% of equity premium puzzle Not just another description of what you did Estimated the consumption CAPM with high-quality data

40 Conclusion You hard on your research worked hard Work communicating it on Make sure the audience Cares ab out your research question Understands you answer it how Knows why they should b elieve you Walks out of the ro om knowing what you learned

41 lot A you whenever talks can Give One Thing More And Practice

42 can you whenever Give talks More One Thing And Practice A lot

43 And More Thing One Practice A lot Give talks whenever you can

Related documents