Survey data (clean)
Clean survey/operational-outcomes data across several groups and strata: observation_id, group, outcome, exposure, age_band, sampling_stratum, weight, response_complete. Used for the Guided Lab.
DS 104 Practice Center
Use a synthetic survey and operational-outcomes dataset to practice distribution profiling, robust summaries, outlier investigation, sampling bias judgment, and correlation-vs-causation rewrites.
Info
DS 104 practice data is synthetic teaching data. All individuals, groups, and values are illustrative and do not represent any real person or organization.
Warning
Use only the synthetic files provided for practice. Unless authorized, never paste confidential institutional or company data, personal data, or sensitive data into an AI tool; confirm your institution's rules before using real survey or operations data.
Download the practice files
Survey data (practice, with challenges)
The same schema with intentional statistical-judgment challenges: a group whose outcome column is clearly right-skewed with a long tail, a handful of observations flagged for outlier investigation, and a sampling stratum that is clearly under-sampled relative to the stated target population. Used for the Practice Center and the Independent Lab.
資料字典（繁體中文）
Column type, unit, grain, allowed values and provenance for the survey data file, in Traditional Chinese.
Data dictionary (English)
English data dictionary; columns and units match the Chinese version.
Guided Lab starter notebook
A starter notebook with guidance and TODOs that runs top to bottom from a fresh kernel, producing an EDA profile.
Independent Lab starter notebook
A starter notebook and checklist template for the EDA Report.
Practice tasks
Practice A: Distribution Profile
At least one group in the practice data has an outcome column that is clearly right-skewed with a long tail. Profile its distribution and judge whether the mean is a fair summary for that group.
Files used: Survey data (practice, with challenges), Data dictionary (English)
Steps
- Load DS104_Survey_Practice.csv and inspect the outcome column grouped by group.
- Plot a histogram or density plot for each group and find the one that is clearly right-skewed with a long tail.
- Compute the mean and the median for that group and compare their difference and direction.
- Write one sentence on how reporting only the mean would mislead readers about that group.
What you produce
- A distribution plot for each group
- A mean-vs-median comparison for the skewed group
- A one-sentence statement of the risk of reporting only the mean
Practice B: Robust Summary Comparison
For the same skewed group, compare "mean + standard deviation" against "median + IQR" and decide which summary is more robust here.
Files used: Survey data (practice, with challenges), Data dictionary (English)
Steps
- Pick the skewed group identified in Practice A.
- Compute the mean, standard deviation, median, and IQR for that group.
- Remove or adjust one extreme value by hand, recompute the four statistics, and compare how much each one moved.
- State which pair of summary statistics you would recommend for an audience unfamiliar with statistics, and why.
What you produce
- A before/after table of the four statistics
- A written robustness comparison
- A recommended summary pair for the audience, with reasoning
Practice C: Outlier Investigation
A handful of observations in the practice data are flagged for outlier investigation. Investigate their likely source before deciding whether to remove them.
Files used: Survey data (practice, with challenges), Data dictionary (English)
Steps
- Use a box plot or z-scores to find outlier observations in the outcome or exposure column.
- Check those observations' other fields (for example age_band, sampling_stratum, response_complete) for a shared pattern.
- For each outlier, write down whether it looks like a measurement error, a data-entry error, or a real but rare signal.
- State the analysis risk of deleting outliers before their source is confirmed.
What you produce
- A list of outlier observations and how they were found
- A likely-source judgment for each outlier
- A written statement of the risk of premature deletion
Practice D: Sampling Bias Scenario
One sampling_stratum in the practice data is clearly under-sampled relative to the target population described by the course scenario. State how this would affect inference.
Files used: Survey data (practice, with challenges), Data dictionary (English)
Steps
- Compute the observation count and share for each sampling_stratum.
- Identify the stratum whose share is clearly below its expected share of the target population.
- Write down what would go wrong if this under-sampling were ignored and the overall result were generalized to the target population.
- Propose at least one fix or disclosure for this limitation (for example weighting, stratified reporting, or an explicit limitation statement).
What you produce
- A table of observation counts and shares by stratum
- A written inference-risk statement for the under-sampled stratum
- At least one proposed fix or disclosure
Practice E: Correlation vs Causation Rewrite
Take a causal claim about the exposure-outcome relationship and rewrite it as an evidence statement that stays inside what the data can support, naming a possible confounder.
Files used: Survey data (practice, with challenges), Data dictionary (English)
Steps
- Compute the correlation between exposure and outcome and plot a scatterplot.
- Read (or write yourself) an overstated claim that treats this correlation as causation.
- Rewrite the claim so it separates the observed correlation, a plausible interpretation, and what remains unknown.
- Name at least one variable that could affect both exposure and outcome, and explain how it could distort the observed relationship.
What you produce
- The correlation coefficient and scatterplot
- The rewritten evidence statement
- At least one named confounder with its possible effect explained
The done mark is stored in this browser only (and synced once you join a cohort). Keep your work in your own files or notebook.