PCED flashcards
PCED · Build recall with a focused review session.
33 cards. Read the term and the definition here. The deck at the top of this page is optional.
- Population vs sample
- The population is every unit you want to conclude about; the sample is the subset actually measured. Statistics computed on the sample estimate population parameters.
- Simple random sampling
- Every unit in the population has an equal and independent chance of selection, typically via random number generation. It removes selection bias but may under-represent small subgroups.
- Stratified sampling
- The population is split into homogeneous strata (e.g., region, age band) and random samples are drawn within each stratum. Guarantees representation of every stratum and usually reduces variance.
- Selection bias
- A distortion where the units included in the data differ systematically from the population of interest in a way tied to the outcome being measured.
- Nominal vs ordinal data
- Nominal categories are unordered labels such as color or country. Ordinal categories have a meaningful order but unequal or unknown spacing, such as satisfaction ratings.
- Tidy data layout
- One row per observation, one column per variable, one value per cell. This layout lets libraries operate column-wise and makes grouping, joining, and plotting straightforward.
- Type coercion with errors
- Converting strings to numbers where invalid entries become missing instead of raising an exception, e.g. pandas to_numeric with errors set to coerce.
- IQR outlier rule
- Compute Q1 and Q3; the interquartile range is Q3 minus Q1. Values below Q1 minus 1.5 IQR or above Q3 plus 1.5 IQR are flagged as potential outliers.
- Missing data mechanisms
- MCAR means missingness is unrelated to any value; MAR means it is explained by observed variables; MNAR means it depends on the unobserved value itself and is hardest to correct.
- Duplicate key removal
- Identify rows sharing identical values on chosen key columns, then keep one record per key. Always specify the key subset; default full-row comparison misses near-duplicates.
- Post-cleaning validation checks
- Compare row counts before and after cleaning, count nulls per column, verify numeric min and max ranges, and confirm the intended primary key is unique.
- Sample vs population standard deviation
- Sample standard deviation divides by n minus one (ddof equals one); population standard deviation divides by n. NumPy defaults to population, so pass ddof equals one for samples.
- NumPy axis argument
- In reductions like sum or mean, axis zero collapses rows and returns one value per column; axis one collapses columns and returns one value per row.
- Boolean mask filtering
- A boolean array of the same shape selects elements where true. Combine conditions with ampersand, pipe, and tilde inside parentheses; Python and/or do not work elementwise.
- collections.Counter
- A dict subclass that counts hashable items from an iterable. most_common(k) returns the k highest-count pairs, useful for frequency summaries of categorical fields.
- csv.DictReader usage
- Reads a CSV file treating the first row as headers, yielding each record as a dict keyed by column name. Values arrive as strings and need explicit conversion.
- datetime timedelta arithmetic
- Subtracting two date or datetime objects yields a timedelta. Access days directly, use total_seconds for finer resolution, and format output with strftime.
- groupby with agg
- Splitting rows by key, applying reductions per group, and returning one row per key. transform instead broadcasts group results back to the original row count.
- np.where vectorized conditional
- Returns elements chosen from two arrays based on a boolean condition, equivalent to an elementwise if/else without a Python loop.
- Inverted pyramid summary
- Report the conclusion first, then the supporting metric, then method and caveats. Readers who stop after one sentence still get the actionable finding.
- Numerator and denominator reporting
- State counts with their base, such as 38 of 200 orders, 19 percent. This bounds the claim, makes it checkable, and prevents overstating small absolute counts.
- Chart type selection rule
- Bar for category comparison, line for trends over time, histogram for a single distribution, scatter for two-variable relationships, pie only for few parts of a whole.
- Zero baseline for bars
- Bar length encodes magnitude, so the axis must start at zero. Truncating it exaggerates differences and misleads readers about relative size.
- Confidence interval reporting
- An estimate paired with a range reflecting sampling uncertainty. A mean reported without spread or sample size cannot be judged for reliability.
- Correlation versus causation
- Two variables moving together does not establish that one causes the other. Confounders, reverse causation, and coincidence are alternative explanations to rule out.
- Purpose limitation principle
- Personal data collected for a stated purpose should not be reused for unrelated purposes without a new lawful basis. It constrains secondary analysis of existing datasets.
- Quasi-identifiers and re-identification
- Fields like age, postal code, and date are not direct identifiers but combine to single out individuals. Removing names alone does not anonymize a dataset.
- k-anonymity
- Every record is indistinguishable from at least k minus one others on the quasi-identifier columns. Higher k means stronger privacy but greater information loss.
- Small-cell suppression
- Withholding or aggregating table cells with counts below a threshold, commonly five, so published statistics cannot be traced back to specific individuals.
- Sequential versus diverging scales
- Sequential data, ordered low to high, needs a single-hue ramp. Diverging data needs two hues meeting at a meaningful midpoint such as zero or an average.
- Data-ink ratio
- The proportion of chart ink that encodes data rather than decoration. Removing gridlines, 3-D effects, and redundant labels raises the ratio and improves readability.
- Mutable default argument trap
- Default argument values are evaluated once when the function is defined. A list default is shared across calls, so use None and create a new list inside.
- Context manager for files
- The with open statement guarantees the file is closed even if an exception occurs inside the block, preventing leaked handles and unflushed writes.