DSMA Admission key terms

The system of decision rights and responsibilities that determines who may access, change, and share data, and how its quality and security are maintained across an organization.

The day-to-day operational responsibility for a data domain, including maintaining definitions, resolving quality issues, and approving access requests on behalf of data owners.

A recorded trace of where a dataset originated, how it was transformed, and where it flows downstream, used to audit results and diagnose quality problems.

Documentation of the source, ownership, and collection method of a dataset, supporting reproducibility and helping analysts judge whether data is fit for a new purpose.

An ethical and legal requirement that people understand what data is collected about them, why, and how it will be used, and agree freely before collection begins.

The principle of collecting and retaining only the data actually needed for a stated purpose, reducing privacy risk and storage burden.

Removing or altering identifiers so that individuals cannot reasonably be re-identified, including from combinations of quasi-identifiers such as age, postcode, and occupation.

Systematic unfairness in model outputs that disadvantages a group, often arising from biased training data, proxy variables, or labels that encode past discrimination.

A reference listing each field in a dataset with its meaning, data type, allowed values, units, and ownership, enabling consistent interpretation by different users.

Structured information describing other data, such as its schema, creation date, source system, and access restrictions, used for discovery, governance, and reproducibility.

The population is the full set of units a question concerns; the sample is the subset actually observed, used to draw inferences about the larger population.

A systematic distortion that occurs when the selected sample is not representative of the population, so estimates deviate consistently rather than merely randomly.

The result that the sampling distribution of a sample mean approaches normality as sample size grows, regardless of the shape of the underlying population distribution.

The standard deviation of a statistic's sampling distribution, quantifying how much an estimate such as a mean would vary across repeated samples.

A range of plausible values for a population parameter, constructed so that a stated proportion of such intervals would contain the true value under repeated sampling.

The probability of observing a test statistic at least as extreme as the one obtained, assuming the null hypothesis is true; it is not the probability that the null is true.

A Type I error rejects a true null hypothesis, a false positive; a Type II error fails to reject a false null, a false negative. Their rates trade off with sample size and threshold.

The probability that a test correctly detects a real effect of a given size, equal to one minus the Type II error rate and increased by larger samples or smaller noise.

An unmeasured factor that influences both the predictor and the outcome, creating a spurious association that can be mistaken for a causal effect.

Correlation measures how two variables move together, while causation requires that changing one actually produces a change in the other; observational association alone cannot establish this.

An experiment in which units are assigned to treatments by chance, so that measured and unmeasured confounders are balanced on average and causal effects can be estimated.

A rule for updating a probability when new evidence arrives: the posterior probability is proportional to the prior probability multiplied by the likelihood of the evidence.

The probability that an event occurs given that another event is known to have occurred, written P(A|B) and generally differing from the unconditional probability of A.

A relationship in which knowing the value of one variable gives no information about another, formally P(A and B) equals P(A) times P(B).

A database storing data in tables linked by shared values; a primary key is a column or set of columns that uniquely identifies each row in a table.

A column in one table that references the primary key of another table, enforcing referential integrity and expressing relationships between entities.

An operation combining rows from two tables based on a matching condition; inner joins keep only matches, while left joins retain all rows from the first table.

A SQL clause that collapses rows sharing key values into groups, over which aggregate functions such as COUNT, SUM, AVG, MIN, and MAX are computed.

The treatment of missing or unknown values in databases and code; NULL is not equal to anything, so comparisons require IS NULL and aggregates usually ignore it.

A SQL function that computes a value across a set of related rows while keeping each row, such as running totals or rankings within partitions, without collapsing the result.

A data type defines which values and operations a variable supports; coercion is the automatic or explicit conversion between types, which can silently truncate or misorder values.

A tabular layout in which each variable is a column, each observation is a row, and each observational unit is a table, simplifying analysis and plotting.

Converting data between wide layouts, with one column per category, and long layouts, with a category column and a value column, to suit analysis or plotting tools.

Writing operations that apply to whole arrays at once instead of looping element by element, producing shorter code that runs substantially faster in numerical libraries.

Code that, together with recorded inputs and environment details, regenerates the same results when rerun, often supported by version control and fixed random seeds.

Supervised learning trains on labeled examples to predict a target; unsupervised learning finds structure such as clusters or lower dimensions in data without labels.

Fitting a model so closely to training data that it captures noise, yielding strong training performance but poor generalization to new, unseen data.

Partitioning data into a subset used to fit a model and a held-out subset used to estimate performance on unseen cases, guarding against optimistic evaluation.

A resampling procedure that rotates which fold of the data is held out, averaging validation performance across folds for a more stable estimate with limited data.

Adding a penalty on model complexity, such as L1 or L2 norms on coefficients, to reduce variance and prevent overfitting while trading off a little training accuracy.

The tension between models too simple to capture the signal, which underfit, and models so flexible they fit noise, which overfit; total error is minimized between these extremes.

A table of predicted versus actual classes showing true positives, false positives, false negatives, and true negatives, from which precision, recall, and accuracy are derived.

Precision is the share of predicted positives that are correct; recall is the share of actual positives that were found. They trade off as the decision threshold moves.

A situation where one outcome is far rarer than another, making accuracy misleading and requiring resampling, class weights, or threshold tuning.

A probabilistic classifier that applies Bayes' theorem while assuming features are conditionally independent given the class, which is rarely true yet often works well.

A feature that stands in for a protected or unmeasured attribute, such as postcode for ethnicity, letting a model discriminate indirectly even when the attribute is excluded.

Absent values in a dataset; handling depends on whether values are missing completely at random, at random conditional on observed data, or systematically, since deletion can bias results.

The proportion of a graphic's ink devoted to actual data rather than decoration, gridlines, or redundant labels; higher ratios generally yield clearer charts.

Choosing a visual form to match the question: bars for comparing categories, lines for trends over time, scatter plots for relationships, histograms for distributions.

A chart flaw in which a truncated or non-zero baseline, uneven scale, or dual axes exaggerate differences and distort the reader's perception of magnitude.

Tailoring a data presentation to what the audience already knows and needs to decide, leading with the key message and supporting it with the most relevant evidence.