MS-DS Master of Data science Research & Data Analysis 3 — Questions and Answers
Question 1: A researcher collects data from volunteers who self-select into a study. The primary threat to external validity is:
- Measurement error
- Self-selection bias (Correct answer)
- Regression to the mean
- Instrumentation drift
Correct answer: Self-selection bias
Self-selection bias occurs when participants who volunteer differ systematically from the target population, limiting generalizability.
Question 2: When comparing more than two group means simultaneously, a one-way ANOVA is preferred over multiple t-tests because:
- ANOVA has lower statistical power
- Multiple t-tests inflate the Type I error rate (Correct answer)
- ANOVA requires no distributional assumptions
- t-tests cannot handle continuous outcomes
Correct answer: Multiple t-tests inflate the Type I error rate
Running multiple t-tests increases the familywise error rate; ANOVA controls it by testing all groups in a single analysis.
Question 3: A confusion matrix for a binary classifier shows TP=80, FP=10, FN=20, TN=90. What is the precision?
- 0.80
- 0.89 (Correct answer)
- 0.75
- 0.88
Correct answer: 0.89
Precision = TP / (TP + FP) = 80 / (80 + 10) = 80/90 ≈ 0.889.
Question 4: In principal component analysis (PCA), the first principal component is defined as:
- The variable with the highest variance
- The linear combination of variables that explains the most variance (Correct answer)
- The eigenvector with the smallest eigenvalue
- The mean-centered version of the original data
Correct answer: The linear combination of variables that explains the most variance
PC1 is the direction (linear combination) in the feature space along which the projected data has maximum variance.
Question 5: A dataset contains 5% missing values that are Missing Not at Random (MNAR). The best imputation strategy is:
- Mean imputation
- Listwise deletion
- Multiple imputation with auxiliary variables explaining missingness (Correct answer)
- Zero imputation
Correct answer: Multiple imputation with auxiliary variables explaining missingness
MNAR requires modeling the missingness mechanism; multiple imputation with variables that predict missing data reduces bias compared to simpler methods.
Question 6: Effect size measures like Cohen's d are reported alongside p-values because:
- They replace the need for significance testing
- They quantify the practical magnitude of a difference, which p-values do not (Correct answer)
- They are required by all journals
- They eliminate sampling error
Correct answer: They quantify the practical magnitude of a difference, which p-values do not
A statistically significant result can have a trivially small effect; Cohen's d communicates how large the difference actually is in standardized units.
Question 7: In time series analysis, differencing a series is used primarily to achieve:
- Stationarity (Correct answer)
- Normality
- Homoscedasticity
- Independence of observations
Correct answer: Stationarity
Differencing removes trends and seasonality, transforming a non-stationary series into a stationary one required for ARIMA modeling.
MS-DS Master of Data science Research & Data Analysis 3
A researcher collects data from volunteers who self-select into a study.
The primary threat to external validity is: