Flashcards
correlation, regression, causality
Cards in this deck (8)
- 1Types of Association1. One way causal (A -> B)
2. Two way causal (A <-> B)
3. Spurious association ( A <- C -> B)
[C is lurking variable]
4. Chain relation ( A -> C -> B)
[C is intervening var] - 2Controlling for Lurking/ Intervening variable1. To remove the dependence btw A & B
by fixing for lurking/intervening variable C (i.e. control for C)
2. For statistical controls
-> group the observations
for diff ranges of C
3. Experiment controls
->physically fix C - 3Simpson's ParadoxDifferent grouping of data eliminates/ changes the direction of association
[diff levels of aggregating data can lead to diff conclusions drawn]
e.g. when given 2 statements that appear to give diff conclusion, maybe aggregation of data is diff (individual vs county data) - 4Linear Regression ModelR^2 = 0 [explanatory var say nothing abt response]
R^2 = 1 [explanatory var completely determines response]
r > 0 [positive association]
r < 0 [negative association] - 5Overfitting the modelThis happens when too many random explanatory variables are added to the model
-> makes the predictions rubbish - 6Regression to the mean (when assessing performance of actions/interventions)-a form of selection bias
-exaggerates perceived effects/lead to false conclusions
-not possible to reach conlusion based on the data from 2 years alone
-if correlation is not perfect (100%), we expect improvement due to regression to mean - 7Ecological fallacyWhen draw conclusions about individuals /smaller groups from aggregate data
- 8Ommitted variable biasWhen fail to control for some variables
-e.g. difference in correlation coefficients (dec in value)
Information
- Description
- Flashcards on correlation, regression, causality, created by avery pumpkin on 06/01/2019.
- Created by
- avery pumpkin
- Views
- 4
- Rating