class: center middle main-title section-title-3 # In-person<br>session 4 .class-info[ **February 6, 2025** .light[PMAP 8521: Program evaluation<br> Andrew Young School of Policy Studies ] ] --- name: outline class: title title-inv-8 # Plan for today .box-1.medium[Logic models and evaluation] -- .box-6.medium[More regression things] -- .box-2.medium[Measuring outcomes] -- .box-5.medium[DAGs] --- layout: false name: exam class: center middle section-title section-title-4 animated fadeIn # Exam 1 details --- layout: false name: exam class: center middle section-title section-title-4 animated fadeIn # Final project details --- layout: false name: exam class: center middle section-title section-title-7 animated fadeIn # Extra bonus things --- layout: false name: logic-models class: center middle section-title section-title-1 animated fadeIn # Logic models<br>and evaluation --- layout: true class: middle --- .box-1.medium[Do people really have the<br>job title "program evaluator"?] -- .box-1.medium[How much does this evaluation stuff cost?] -- .box-1.medium[Can you do scaled-down versions<br>of these evaluations?] ??? <https://www.eval.org/> --- .box-1.medium[Isn't it best to always<br>have an articulated theory?] .box-1.medium[Should implicit theory and articulated theory<br>be the same thing in most cases?] --- .box-1.medium[What if a program exists already<br>and doesn't have a logic model?] --- .box-1.medium[Why would a program aim for final outcomes<br>that can't be measured?] --- .box-1.medium[What should you do if you find that your<br>theory of change (or logic model in general)<br>is wrong in the middle of the program?<br>Is it ethical to stop or readjust?] --- .box-1.medium[How does regression<br>relate to impact evaluation?] --- layout: false name: regression-faqs class: center middle section-title section-title-6 animated fadeIn # More regression things --- layout: true class: middle --- .box-6.medium[Do we care about the actual coefficients<br>or just whether or not they're significant?] -- .box-6.medium[How does significance relate to causation?] -- .box-6.medium[If we can't use statistics to assert causation<br>how are we going to use this information<br>in program evaluation?] --- .box-6.large[What counts as a "good" R²?] --- .center[ <figure> <img src="img/02-class/nice-plot-1.png" alt="Euler diagram" title="Euler diagram" width="45%"> </figure> ] --- .center[ <figure> <img src="img/02-class/plot-diagram-prediction-1.png" alt="R2 prediction" title="R2 prediction" width="75%"> </figure> ] --- .center[ <figure> <img src="img/02-class/plot-diagram-estimation-1.png" alt="R2 estimation" title="R2 estimation" width="75%"> </figure> ] --- .center[ <figure> <img src="img/02-class/prediction-vs-estimation.jpg" alt="R2 estimation vs prediction" title="R2 estimation vs prediction" width="100%"> </figure> ] --- class: middle .box-6.large[Side-by-side regression tables] ---
(1)
(2)
(3)
(4)
(Intercept)
362.307
-5780.831***
-5736.897***
-5433.534***
(283.345)
(305.815)
(307.959)
(286.558)
bill_length_mm
87.415***
6.047
-5.201
(6.402)
(5.180)
(4.860)
flipper_length_mm
49.686***
48.145***
48.209***
(1.518)
(2.011)
(1.841)
sexmale
358.631***
(41.572)
Num.Obs.
342
342
342
333
R2
0.354
0.759
0.760
0.807
R2 Adj.
0.352
0.758
0.759
0.805
AIC
5400.0
5062.9
5063.5
4863.3
BIC
5411.5
5074.4
5078.8
4882.4
Log.Lik.
-2696.987
-2528.427
-2527.741
-2426.664
F
186.443
1070.745
536.626
457.118
RMSE
643.54
393.12
392.34
353.66
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
--- class: middle .box-6.large[See full documentation and<br>examples for `modelsummary()` [here](https://vincentarelbundock.github.io/modelsummary/articles/modelsummary.html)] --- class: middle .box-6.large[Make nicer tables [with {tinytable}](https://vincentarelbundock.github.io/tinytable/)] --- layout: false name: outcomes class: center middle section-title section-title-2 animated fadeIn # Measuring<br>outcomes --- layout: true class: title title-2 --- # The paradox of evaluation -- .box-2.medium[Evaluation is good, but expensive] .box-inv-2.sp-after.SMALL["Evaluation thinking"] -- .box-2.medium[Too much evaluation is bad] .box-inv-2.sp-after-half.SMALL[Taming programs] --- # Outcomes and programs .box-2.medium[Outcome variable] .box-inv-2.sp-after-half.SMALL[Thing you're measuring] -- .box-2.medium[Outcome change] .box-inv-2.sp-after-half.SMALL[∆ in thing you're measuring over time] -- .box-2.medium[Program effect] .box-inv-2.SMALL[∆ in thing you're measuring over time *because of* the program] --- # Outcomes and programs <figure> <img src="img/04/program-effect.png" alt="Outcomes and program effect" title="Outcomes and program effect" width="100%"> </figure> --- layout: false class: middle .box-2.large[Abstraction] --- layout: false name: dags class: center middle section-title section-title-5 animated fadeIn # DAGs --- layout: true class: middle --- .box-5.large[Causal thinking is necessary—<br>even for descriptive work!] --- .center[ <figure> <img src="img/04-class/supercentenarians.png" alt="Supercentenarians" title="Supercentenarians" width="46%"> </figure> ] ??? Necessity of causal thinking: Mention the McElreath tweet on birth certificate introduction and death ages: <https://twitter.com/rlmcelreath/status/1427564280744976384> <https://www.biorxiv.org/content/10.1101/704080v2> --- layout: false .box-5.less-medium["Every time I get a haircut, I become more mature!"] <figure> <img src="img/04-class/ben-hair.jpg" alt="Benjamin haircut" title="Benjamin haircut" width="100%"> </figure> --- .box-5.less-medium["Every time I get a haircut, I become more mature!"] <img src="04-class_files/figure-html/unnamed-chunk-2-1.png" width="70%" style="display: block; margin: auto;" /> .center[$$E[\text{Maturity} \mid \operatorname{do} (\text{Get haircut})]$$] --- .box-5.less-medium[Getting older opens a backdoor path] <img src="04-class_files/figure-html/unnamed-chunk-3-1.png" width="70%" style="display: block; margin: auto;" /> --- .box-5.medium[Ice cream causes crime] -- .center[ <figure> <img src="img/05-class/eat-ice-cream-do-crime.jpg" alt="Eat ice cream, do crime" title="Eat ice cream, do crime" width="100%"> </figure> ] -- .box-inv-5[Summer weather opens a backdoor path] ??? - <https://twitter.com/EpiEllie/status/1080680304581451778> - <https://www.redbubble.com/i/sticker/Eat-Ice-Cream-Do-Crime-by-epiellie/150075475.EJUG5> --- class: middle .box-5.medium[But what does that mean,<br>"opening a backdoor path"?] .box-5.medium[How does statistical association<br>get passed through paths?] --- .box-5.less-medium.sp-after[How do I know which of these is which?] .center[ <figure> <img src="img/05-class/dag-associations.png" alt="DAG associations" title="DAG associations" width="100%"> </figure> ] --- class: middle .pull-left[ <figure> <img src="img/04-class/slider-switch-plain-80.jpg" alt="Switch and slider" title="Switch and slider" width="100%"> </figure> ] .pull-right[ <img src="04-class_files/figure-html/confounding-dag-alone-1.png" width="100%" style="display: block; margin: auto;" /> ] --- class: middle .pull-left[ <figure> <img src="img/04-class/slider-switch-plain-80.jpg" alt="Switch and slider" title="Switch and slider" width="100%"> </figure> ] .pull-right[ <img src="04-class_files/figure-html/mediating-dag-alone-1.png" width="100%" style="display: block; margin: auto;" /> ] --- class: middle .pull-left[ <figure> <img src="img/04-class/slider-switch-plain-80.jpg" alt="Switch and slider" title="Switch and slider" width="100%"> </figure> ] .pull-right[ <img src="04-class_files/figure-html/colliding-dag-alone-1.png" width="100%" style="display: block; margin: auto;" /> ] --- .center[ <video controls loop> <source src="img/05-class/video-confounding-unblocked.mp4" type="video/mp4"> </video> ] --- .center[ <video controls loop> <source src="img/05-class/video-confounding-blocked.mp4" type="video/mp4"> </video> ] --- .center[ <video controls loop> <source src="img/05-class/video-mediating.mp4" type="video/mp4"> </video> ] --- layout: true class: middle --- .box-5.medium[d-separation] .box-inv-5[Except for the one arrow between X and Y,<br>no statistical association can flow between X and Y] .box-inv-5[This is **identification**—<br>all alternative stories are ruled out<br>and the relationship is isolated] --- .box-5.large[How exactly do colliders<br>mess up your results?] .box-5.medium[It looks like you can<br>still get the effect of X on Y] --- <img src="04-class_files/figure-html/nba-dag-1.png" width="65%" style="display: block; margin: auto;" /> --- .center[ <figure> <img src="img/04-class/facebook.png" alt="Facebook collider" title="Facebook collider" width="55%"> </figure> ] ??? <https://www.businessinsider.com/facebook-sent-incomplete-misinformation-data-flawed-researchers-2021-9> <https://www.nytimes.com/live/2020/2020-election-misinformation-distortions#facebook-sent-flawed-data-to-misinformation-researchers> --- layout: false class: title title-5 # Does niceness improve appearance? .pull-left[ <img src="04-class_files/figure-html/unnamed-chunk-4-1.png" width="100%" style="display: block; margin: auto;" /> ] -- .pull-right[ <img src="04-class_files/figure-html/unnamed-chunk-5-1.png" width="100%" style="display: block; margin: auto;" /> ] --- class: title title-5 # Collider distorts the true effect! .pull-left[ <img src="04-class_files/figure-html/unnamed-chunk-6-1.png" width="100%" style="display: block; margin: auto;" /> ] -- .pull-right[ <img src="04-class_files/figure-html/unnamed-chunk-7-1.png" width="100%" style="display: block; margin: auto;" /> ] --- .box-5.large[Effect of race on police use of force<br>using administrative data] --- .box-5.medium[Effect of race on police use of force<br>using administrative data] .pull-left[ <figure> <img src="img/05-class/klm-dag.png" alt="Use of force" title="Use of force" width="100%"> </figure> ] .pull-right[ <figure> <img src="img/05-class/klm.png" alt="Use of force" title="Use of force" width="100%"> </figure> ]