引導式測驗 · 中等 等級

CUPED 變異數減少測驗

應用 CUPED 的期前調整併確定協變量時間、預測強度和隨機化如何影響有效推理。

相關引導路徑Cuped 變異數減少
問題 1 的 8

Which variable is appropriate for a standard CUPED adjustment?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. Which variable is appropriate for a standard CUPED adjustment?
    • A pre-treatment measurement correlated with the experiment outcome
    • A post-treatment variable changed by assignment
    • A future outcome observed after the test
    • A control-group label chosen after results
  2. What does theta=Cov(X,Y)/Var(X) represent in a common CUPED adjustment?
    • The treatment assignment probability
    • The linear coefficient used to remove predictable outcome variation
    • The fraction of users excluded
    • 模型的準確性
  3. What happens when the pre-period covariate has little correlation with the outcome?
    • Power is guaranteed to increase fourfold
    • The treatment effect becomes zero
    • CUPED may provide little variance reduction
    • Random assignment is invalid
  4. Why avoid using a treatment-affected covariate?
    • It prevents calculating a mean
    • It always increases sample size
    • It makes variance exactly zero
    • Conditioning on it can introduce post-treatment bias
  5. What does centering X by its mean do in Y_adj=Y-theta(X-mean X)?
    • Changes the random assignment
    • Subtracts predictive deviations while retaining the outcome's mean scale
    • Converts Y into a probability
    • Removes all sampling uncertainty
  6. What core design feature does CUPED not replace?
    • A numerical outcome
    • A pre-period covariate
    • Valid random assignment
    • An experiment window
  7. Why check baseline covariate coverage across groups?
    • It removes the need for confidence intervals
    • It determines the model's layer count
    • It guarantees a higher treatment effect
    • Differential missingness can affect the adjusted analysis
  8. How should CUPED results commonly be communicated?
    • Alongside the unadjusted estimate and the prespecified adjustment method
    • As proof every user benefited
    • Without describing the covariate
    • As a replacement for the outcome definition