引導式測驗 · 中等 等級
模型實驗測驗的統計效能與樣本量
計算樣本大小近似值並確定效應大小、變異數、效能和實驗設計如何影響所需流量。
問題 1 的 8
在所述的雙臂範例中,每組大約需要多少樣本量?
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。
- 在所述的雙臂範例中,每組大約需要多少樣本量?
- 392
- 196
- 784
- 80
- 如果在其他假設固定的情況下目標 MDE 減半,近似樣本量會如何改變?
- It becomes half as large
- It becomes about four times larger
- It is unchanged
- It becomes one quarter as large
- 在指定的替代方案和假設下,80% 的功效意味著什麼?
- An 80% chance every user benefits
- An 80% chance the null is false
- An 80% chance the test rejects the null if that effect is present
- An 80% confidence interval coverage
- MDE在規劃中代表什麼?
- The maximum possible variance
- The effect guaranteed by the model
- The observed effect after launch
- The smallest effect chosen as important to detect
- 為什麼聚類隨機化通常需要更多觀察?
- Each cluster has a separate p-value automatically
- Outcomes within a cluster are correlated and provide less independent information
- Clusters eliminate variance
- The treatment becomes continuous
- Why plan for multiple metrics or variants?
- Metrics cannot be correlated
- Power is always exactly zero for multiple metrics
- Many comparisons can increase false-positive risk
- More variants reduce all uncertainty
- Why is repeatedly checking fixed-horizon p-values and stopping at significance risky?
- It makes random assignment invalid
- It always lowers power
- It guarantees the effect is practical
- It can inflate false-positive rates unless a valid sequential method is used
- What should accompany a reported power calculation?
- Its assumptions about variance, test, allocation and effect size
- Only the chosen sample count
- The model's training accuracy
- A guarantee of causal generalization