Flashcards
GBA 424
Cards in this deck (25)
- 1Data Sensibility1. Frequencies (weird distributions)
-----------------------------------------------------------
2. Min / Max Values outside of E(range)
-----------------------------------------------------------
3. Correlations (100%) - 2Is The Sample Representative?Chi-Square
--
Goodness of Fit to evaluate - 3If Face-Validity Check Rejects SampleIs difference economically meaningful?
--
No, doesn't matter.
-----------------------------------------------------------
Can observations within each group represent that group well?
--
Yes, Re-weight - 4TURF AnalysisTotally Unduplicated Reach and Frequency
-- - 5TURF Analysis UsesOriginally developed for media, however, useful for deciding which product(s) to launch next
- 6TURF Analysis Process1. Start from set of options
-----------------------------------------------------------
2. Add every potential combination of options one at a time
-----------------------------------------------------------
3. Calculate reach
-----------------------------------------------------------
4. Present in contrast to original - 7ReachThe set of people who get at least one exposure.
- 8FrequencyThe number of exposures each person receives
- 9Conjoint Analysis Uses1. Product Design
-----------------------------------------------------------
2. Competitive and cannibalization analysis
-----------------------------------------------------------
3. Pricing Decisions
-----------------------------------------------------------
4. Benefit Segmentation - 10Benefit Segmentation (Segments)1. Individual
-----------------------------------------------------------
2. A Priori
-----------------------------------------------------------
3. Aggregate - 11Benefit Segmentation (Analysis)A Priori Segmentation
--
When you know segments beforehand
-----------------------------------------------------------
Cluster Analysis
--
Group based on demographics or purchase history - 12Conjoint Analysis (Warnings!!!)1. Sample bias of heavy users
-----------------------------------------------------------
2. Under-estimate price sensitivity
-----------------------------------------------------------
3. Over-estimate purchase likelihood
-----------------------------------------------------------
4. Lower Prediction accuracy in contrast to actual data - 13Experiments (Analysis Design)Incremental
--
Exclusive v. Lift
-----------------------------------------------------------
Conditional Causal Effect
--
Slice & Dice, Interactions, Causal Forests
-----------------------------------------------------------
Diff v. Diff-in-Diff - 14DiffRandom Assignment
- 15Diff-in-DiffMatched Pairs
--
1. First Diff: Treatment & Control
-----------------------------------------------------------
2. Second Diff: Before & After - 16Experiments (Experimental Design)1. Use Test and Control Groups
-----------------------------------------------------------
2. Assign randomly or match groups
--
Don't forget to control contamination - 17Experiments (Warnings!!!)1. Self-Selection
--------------------------------------------------------------
2. Spillovers
--------------------------------------------------------------
3. Compliance - 18Exclusive Offers1. Obtain Ground Truth
--------------------------------------------------------------
2. Can apply any predictive algorithm - 19Inclusive Offers1. Evaluate Incremental Effect
--------------------------------------------------------------
2. Lift actions are useful for obtaining causal inference - 20Principle Component Analysis1. Starts from correlation matrix
-----------------------------------------------------------
2. Re-arrange variables to form new variables called principle components
-----------------------------------------------------------
3. Order these components in-terms of importance (most to least) - 21Factor Analysis by MLE1. Starts from correlation matrix
-----------------------------------------------------------
2. Reduce variables by forming factors - 22Matrix FactorizationImputes missing ratings which are recommendations
- 23Ground Truth (Definition)The data that is believed to be true
- 24Problems with Ground Truth1. Sample Bias
-----------------------------------------------------------
2. Insufficient Sample
Over-sample groups with poor predictive performance
-----------------------------------------------------------
3. Missing data - 25Dealing with Missing Data (Methods)1. Mean Imputation
-----------------------------------------------------------
2. Random Imputation
-----------------------------------------------------------
3. Use Predictive Model
Information
- Description
- Flashcards on GBA 424, created by Ryan Page on 29/02/2020.
- Created by
- Ryan Page
- Views
- 27
- Rating