Kvíz spojený s průvodcem · Střední úroveň
Korelace vs kauzální kvíz
Tests association versus intervention effects, confounding and reverse causation, randomized designs, and assumptions behind observational causal methods.
Otázka 1 z 8
V příkladu zmrzliny a úmrtí na utonutí, jaký je skutečný vysvětlení jejich korelace?
Náhled každé otázky8 otázky
Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.
- V příkladu zmrzliny a úmrtí na utonutí, jaký je skutečný vysvětlení jejich korelace?
- Ice cream consumption directly weakens swimmers
- A confounding variable, hot weather, independently increases both
- Drowning causes people to buy less ice cream
- The correlation is caused by reverse causation
- V příkladu policie a kriminality, jak by reverzní kauzalita mohla vytvořit pozorovanou asociaci?
- Assuming there is no relationship between A and B at all
- A statistical technique for removing confounders
- A synonym for spurious correlation
- Assuming A causes B when actually B causes A
- V nemocničním příkladu, proč by léčba mohla být spojena s horšími výsledky i když pomáhá?
- A treatment appearing linked to worse outcomes because sicker patients were more likely to receive it in the first place
- Patients choosing treatments at random
- Doctors intentionally hiding treatment outcomes
- A treatment that has no correlation with outcomes at all
- Který termín popisuje spojení mezi dvěma nesouvisejícími proměnnými, které vzniká čistě náhodou, zejména při porovnávání velkého počtu proměnných?
- Confounding
- Reverse causation
- Spurious correlation
- Structural risk
- What notation does the guide attribute to Judea Pearl's causal calculus for distinguishing an observed association from an interventional effect?
- E[X] vs. Var(X)
- P(Y|X) vs. P(Y|do(X))
- argmax vs. argmin
- L(h(x),y) vs. R(h)
- Why do randomized controlled trials help establish causation rather than mere correlation?
- Random assignment guarantees identical groups in every finite trial
- Random assignment removes the need to measure outcomes
- Random assignment proves the treatment works in every population
- Random assignment breaks systematic links between treatment selection and baseline factors in expectation, reducing confounding
- A model predicts outcomes accurately from an observed feature. What extra evidence is needed before claiming that intervening on the feature will change outcomes?
- A larger training set with the same observational pattern
- A second predictive model that agrees
- A causal design or justified assumptions that identify the intervention effect
- A higher correlation between the feature and outcome
- Which technique is mentioned as a way to statistically approximate causal effects from observational data when randomization isn't possible?
- Instrumental variables and propensity score matching
- Feature hashing
- One-vs-rest classification
- Binning continuous variables