Prescriptive Analytics: A Hands-on Introduction to Getting Actionable Insight from Your Data
4 likes522 views
The document discusses prescriptive analytics, emphasizing its role in providing actionable insights from data. It contrasts prescriptive analytics with predictive analytics, highlighting the importance of understanding causality and methods to achieve accurate causal inference. Various practical solutions for data analysis and modeling, such as simulations, optimization models, and quasi-experiments, are also presented.
Prescriptive Analytics: A Hands-on Introduction to Getting Actionable Insight from Your Data
- 1. Barton Poulson @ MADS 2018. ANALYTICS. PRESCRIPTIVE
- 2. Barton Poulson @ MADS 2018. A hands-on introduction to getting actionable insight from your data.
- 6. 5WILLIAM JAMES. My thinking is first and last and always for the sake of my doing.
- 7. 6WILLIAM JAMES. In other words… thinking is for doing.
- 12. 12PREDICTIVE ANALYTICS. Models that use historical data to estimate the likelihood of a desired outcome… or maybe a current or past outcome.
- 13. 13PREDICTIVE ANALYTICS. Predictions can be very precise & accurate if data is sufficiently large & diverse.
- 17. 17PRESCRIPTIVE ANALYTICS. Models that use historical or novel data to estimate the causal relationship between an event and a desired outcome.
- 18. 18NOT JUST “IF THIS, THEN THAT.”
- 19. 19 CAUSALITY.
- 20. First. 2020 Cause & effect correlated. Second. Cause occurs before effect. Third. No alternative explanations. CAUSAL INFERENCE.
- 21. Construct validity. Measure what you mean to measure. Internal validity. Accurate causal inference. 2121 External validity. Generalize to other methods & groups. FORMS OF RESEARCH VALIDITY.
- 22. 33THREATS TO INTERNAL VALIDITY. History, maturation, testing, instrumentation, mortality, regression, selection, diffusion, compensatory equalization, rivalry, & demoralization.
- 26. 37THEORETICAL SOLUTION. Recruit many participants, randomly assign to carefully manipulated conditions. Repeat many times with different methods & groups.
- 27. 38THEORETICAL SOLUTION. RCTs aren’t always hard to do. A/B testing for websites is an easy, automated version of a randomized experiment.
- 32. 43WHAT-IF SIMULATIONS. Model an outcome and then manipulate the input values to see the effect on the outcome.
- 33. Scenarios. Sets of possible values & outcome. Data tables. Manipulate values on 1 or 2 dimensions. 4444 Goal seek. Iterate one value to reach a set goal. WHAT-IF SIMULATIONS.
- 34. 45WHAT-IF SIMULATIONS. Great for heuristic values; take the results as suggestions. Easy to do in Excel.
- 37. 48OPTIMIZATION MODELS. Models that find the distribution of time or resources among several alternatives that maximizes an outcome given constraints.
- 38. One variable. Setting price to maximize revenue. Many variables. Allocating resources to several projects in different markets. 4949OPTIMIZATION MODELS.
- 39. 50OPTIMIZATION MODELS. Can do one-variable by hand, multiple variable models in Excel using the Solver add-in, and complex models in R or Python.
- 42. .5 60CROSS-LAGGED CORRELATIONS. Analytics in 2015 Revenue in 2015 Analytics in 2018 Revenue in 2018 .2 .3
- 43. .3 61CROSS-LAGGED CORRELATIONS. Analytics in 2015 Revenue in 2015 Analytics in 2018 Revenue in 2018 .2 .3
- 44. 62CROSS-LAGGED CORRELATIONS. Can do simple models in Excel and more sophisticated models in R or Python using specialized packages.
- 47. 65QUASI-EXPERIMENTS. Estimating causal effects without random assignment to conditions.
- 48. Data can come from: ~ Observational studies. ~ Historical sources. ~ Nonrandomized experiments. 66QUASI-EXPERIMENTS.
- 49. 67QUASI-EXPERIMENTS. Regression discontinuity, interrupted time-series, propensity score matching, & block entry regression.
- 50. 68QUASI-EXPERIMENTS. Possible to do these in Excel, but much easier using specialized packages in R or Python.
- 51. 69 IN EXCEL TRY THEM & R.