Data Analysis Foundations and Theory

researched 2026-05-30T12:23:01.017Z· 0 sources · 6 concepts · skill da-1-foundations-theory

Conceptual grounding for a data analysis effort. This skill answers the

Data Analysis: Foundations & Theory

Conceptual grounding for a data analysis effort. This skill answers the [source]
before-you-pick-a-tool questions: what is data analysis, how does it relate to [source]
neighboring fields, what kind of analysis is called for, what can the data [source]
support given how it was measured, and what assumptions ride underneath. It does [source]
not execute techniques - it scopes and frames them. [source]

This hub consolidates 37 foundations sub-skills as on-demand references - match [source]
the task to the table and **Read the listed references/<name>.md before [source]
answering deep questions**. The overview below is enough for framing and [source]
scoping; load the reference when a question needs depth. [source]

Data analysis is the systematic process of inspecting, cleaning, transforming, and interpreting data to extract useful information, support conclusions, and aid decision-making. In practice it is bounded and goal-directed. [source]
Distinguish four neighboring terms - they overlap but are not synonyms: [source]
Data analysis - the act of evaluating data to answer a defined question. The process verb. [source]
Data analytics - the broader practice/field built around analyzing data; typically about the past and present, with more elementary statistics. [source]
Data science - a wider, multidisciplinary field that includes analytics but reaches into ML, forecasting, and large-scale data engineering; more oriented toward predictive models. [source]
Statistics - the mathematical discipline of collecting, describing, and drawing inferences from data under uncertainty. Data analysis uses statistics as a toolkit. [source]
Rule of thumb: data analysis is the activity; analytics is the field around it; data science extends it toward modeling and engineering; statistics supplies the inferential mathematics. State which definition you are using. [source]

Analysis breaks a whole into parts; synthesis recombines parts into a new integrated whole or recommendation. Name which mode you are in to avoid presenting raw decomposition as a conclusion. [source]

Quantitative analysis operates on numeric measurements (interval and ratio scales). [source]
Qualitative analysis operates on categorical/textual data (nominal and ordinal scales). The split is about method too, not just data type. Many efforts are mixed-method. [source]

The four form a maturity progression but are not strictly sequential per project. "Diagnostic" maps loosely onto exploratory work, but don't conflate the marketing taxonomy with Tukey's exploratory/confirmatory split. [source]

The de facto reference is CRISP-DM (six phases you can revisit): [source]
Business Understanding - define the question and success criteria. [source]
Data Understanding - collect, describe, explore, verify quality. [source]
Data Preparation - select, clean, construct, integrate, format. [source]
Modeling - choose technique, build, assess. [source]
Evaluation - check against business goal; review process. [source]
Deployment - deliver, monitor, report. [source]
The phases are iterative, not a one-way pipeline. A lighter generic framing - define → collect → clean → analyze → interpret → communicate - works for non-mining work. [source]

Exploratory Data Analysis (EDA) - Tukey's approach for summarizing a dataset's main characteristics, often with graphics, to generate hypotheses and check assumptions. Techniques: box plots, stem-and-leaf, histograms, scatter plots. [source]
Confirmatory Data Analysis (CDA) - classical hypothesis testing: pick a model before examining the data, then assess inference precision. [source]
Researchers need both. Critical pitfall: running exploratory and confirmatory analysis on the same data introduces systematic bias (double-dipping). Reserve a holdout or fresh sample for confirmation. [source]

Stevens' four levels (1946): [source]
Nominal and ordinal are categorical/qualitative; interval and ratio are continuous/quantitative. [source]
Common pitfalls: computing a mean of ordinal codes (median is safer); treating an arbitrary numeric label as quantitative; forgetting interval scales lack a true zero (ratios are meaningless). Know the controversy: Velleman & Wilkinson (1993) and Luce (1997) contested Stevens' typology. Treat the level of measurement as a useful first filter, not an iron law. [source]

Data does not interpret itself. Every analysis rides on assumptions: representative sample, measurements meaning what labels claim, model preconditions holding. Two practices: [source]
State assumptions explicitly and tie each to the analysis family and measurement level. [source]
Match method to question and to data. Misalignment - a prescriptive recommendation on descriptive data, or a mean on ordinal categories - is the most common foundational error. [source]

Term check - analysis, analytics, data science, or statistics? State the definition. [source]
Family - descriptive, diagnostic, predictive, or prescriptive? [source]
Mode - exploratory or confirmatory? Don't mix on the same data. [source]
Lifecycle - which CRISP-DM phase; what's next? [source]
Measurement - what level is each variable; which statistics are licensed? [source]
Assumptions - what must be true; have I checked? [source]
[source]