MS-DS Master of Data science Research & Data Analysis 2 — Questions and Answers
Question 1: A researcher wants to determine whether a new teaching method improves test scores compared to a traditional method. Which study design is most appropriate?
- Cross-sectional survey
- Randomized controlled experiment (Correct answer)
- Case study analysis
- Retrospective cohort study
Correct answer: Randomized controlled experiment
A randomized controlled experiment allows causal inference by randomly assigning participants to treatment and control groups.
Question 2: In hypothesis testing, a p-value of 0.03 with a significance level of 0.05 means:
- Fail to reject the null hypothesis
- Reject the null hypothesis (Correct answer)
- The effect size is large
- The study has 97% power
Correct answer: Reject the null hypothesis
Since 0.03 < 0.05, the result is statistically significant and we reject the null hypothesis.
Question 3: Which technique is used to assess the stability of a regression model by partitioning data into training and validation subsets multiple times?
- Bootstrap sampling
- K-fold cross-validation (Correct answer)
- Stratified random sampling
- Jackknife resampling
Correct answer: K-fold cross-validation
K-fold cross-validation splits data into k subsets and iteratively trains/validates across all folds to estimate model performance.
Question 4: A dataset has a mean of 50, median of 45, and mode of 40. This distribution is best described as:
- Symmetric
- Negatively skewed
- Positively skewed (Correct answer)
- Bimodal
Correct answer: Positively skewed
When mean > median > mode, the distribution has a longer right tail, indicating positive (right) skew.
Question 5: A data scientist uses Lasso regression instead of OLS. The primary advantage of Lasso in high-dimensional data is:
- It increases model complexity
- It performs automatic feature selection by shrinking some coefficients to zero (Correct answer)
- It guarantees no multicollinearity
- It maximizes variance explained without penalty
Correct answer: It performs automatic feature selection by shrinking some coefficients to zero
Lasso's L1 penalty shrinks some coefficients to exactly zero, effectively selecting a sparse subset of predictors.
Question 6: In a meta-analysis, publication bias most commonly results in:
- Underestimation of effect sizes due to negative study exclusion
- Overestimation of effect sizes because significant results are more likely published (Correct answer)
- More accurate pooled estimates
- Reduced heterogeneity across studies
Correct answer: Overestimation of effect sizes because significant results are more likely published
Studies with statistically significant results are more likely published, inflating the pooled effect size in meta-analyses.
Question 7: Which measure of association is most appropriate when analyzing the relationship between two continuous variables that may not have a linear relationship?
- Pearson correlation
- Spearman rank correlation (Correct answer)
- Chi-square statistic
- Point-biserial correlation
Correct answer: Spearman rank correlation
Spearman rank correlation captures monotonic relationships and is robust to non-linearity and outliers unlike Pearson correlation.
MS-DS Master of Data science Research & Data Analysis 2
A researcher wants to determine whether a new teaching method improves test scores compared to a traditional method.
Which study design is most appropriate?