Google Data Analytics cheat sheet

The data analysis process has six phases: Ask, Prepare, Process, Analyze, Share, Act.

Data is a collection of facts; information is data that has been processed and given context; knowledge is the understanding drawn from that information.

Quantitative data is numeric and can be counted or measured; qualitative data describes qualities and is usually non-numeric.

Structured data fits a predefined model of rows and columns (spreadsheets, relational tables); unstructured data (text, images, audio) has no fixed schema.

Four types of analytics: descriptive (what happened), diagnostic (why it happened), predictive (what is likely to happen), prescriptive (what action to take).

Data-driven decision-making relies on facts and evidence; data-inspired decision-making blends data with intuition and experience.

Data ethics covers consent, privacy, anonymity, and protection from bias introduced during collection, sampling, or interpretation.

A data ecosystem includes data sources, producers, storage, and consumers, plus the applications that act on the data.

Core analyst tools: spreadsheets, SQL, R or Python, data visualization software, and statistical methods.

SMART questions are Specific, Measurable, Action-oriented, Relevant, and Time-bound.

The business task is the question or problem the analysis must answer; the problem statement describes the current condition that needs to change.

Open-ended questions invite elaboration and context; closed-ended questions return short, fixed answers such as yes/no or a number.

Stakeholders typically include the project sponsor, the decision-maker, subject matter experts, and the end users of the analysis.

A scope of work (SOW) records deliverables, milestones, timeline, and required resources.

Structured thinking decomposes the problem into manageable parts and maps each part to available data before analysis starts.

Lagging indicators measure results after the fact (revenue, churn); leading indicators predict future change (site visits, sign-ups).

Quantitative data answers 'how many' or 'how much'; qualitative data answers 'why' or 'how' and is gathered through observation or interviews.

A KPI is a measurable value tied to the business objective and used to judge whether the analysis succeeded.

Data type pairs: nominal (unordered categories) vs ordinal (ordered categories); discrete (countable whole numbers) vs continuous (any value in a range).

Primary data is collected firsthand by the analyst (surveys, interviews, sensors); secondary data was collected by someone else (public datasets, vendor reports).

Wide format puts each variable in its own column; long format stores one observation per row with a key-value structure.

Open data is publicly available and free to use and share, usually with attribution; licensed data carries usage restrictions.

Data integrity means accuracy, completeness, consistency, and validity of data throughout its lifecycle.

Data privacy protects individuals' personal information; data security protects data from unauthorized access and loss.

Metadata is data about data (source, format, creation date); a data dictionary lists each field's name, type, format, and meaning.

In relational databases, a primary key uniquely identifies each row and a foreign key references a primary key in another table.

A random sample selects records by chance; a convenience sample takes whatever is easily accessible and tends to introduce bias.

Dirty data is inaccurate, incomplete, duplicate, or inconsistently formatted; clean data is complete, correct, and standardized.

Cleaning workflow: identify the issue, clean it, verify the result, and document the change.

TRIM removes leading, trailing, and repeated internal spaces; LEFT, RIGHT, MID, SPLIT, and CONCATENATE/TEXTJOIN reshape text fields.

VLOOKUP and IFERROR are used to match records and to replace error values with a defined result.

In SQL, CAST or SAFECAST converts data types, COALESCE replaces nulls, and TRIM/UPPER/LOWER standardize text.

A null means no value was recorded; zero is a real recorded value and the two must not be conflated.

Data validation confirms records meet rules for type, range, and format before analysis.

Useful spreadsheet checks: COUNTBLANK, COUNTIF, ISERROR, and conditional formatting to surface outliers and missing values.

Verification includes comparing row counts and re-checking a sample of cleaned records against the original source.

Aggregation summarizes values by group using SUM, AVERAGE, COUNT, MIN, and MAX.

A pivot table groups rows and columns and applies an aggregation, letting one summary table answer many comparison questions.

Sorting reorders rows; filtering hides rows that fail a condition but does not delete them.

JOIN types: INNER returns only matching rows; LEFT returns all left rows plus matches; RIGHT mirrors LEFT; FULL OUTER returns all rows from both tables.

Cleaning changes stored values; formatting only changes display and does not affect the underlying number.

A subquery runs inside another query; a CTE (WITH clause) names a temporary result set for reuse in the main query.

Summary statistics: mean (average), median (middle value), mode (most frequent), range, standard deviation, and quartiles.

Correlation measures the strength of association between two variables; only controlled experiments support causal claims.

Final validation checks completeness, accuracy, and consistency of the dataset before results are reported.

Chart selection: bar for comparison across categories, line for trends over time, pie for parts of a whole with few slices, scatter for relationships between two numeric variables.

Pre-attentive attributes (color, size, shape, position, orientation) are processed instantly by the brain and direct attention to key data points.

Bar charts should start at zero; truncating the axis exaggerates small differences and misleads viewers.

Five types of change shown in visualizations: divergence, convergence, cyclical, static, and spontaneous.

A dashboard monitors live, updating data; a static report captures a single moment in time.

Accessibility requires sufficient color contrast, alt text for images, and never encoding meaning by color alone.

A data story combines the headline finding, the supporting visual, context or a baseline, and a clear recommended action.

Design hygiene: remove clutter, limit the color palette, label axes and units, and keep scales consistent across compared charts.

Core Python types: int, float, str, bool; core containers: list (ordered, mutable), tuple (ordered, immutable), dict (key-value pairs).

pandas provides the DataFrame (two-dimensional labeled table) and the Series (one-dimensional labeled array).

Load and inspect: pd.read_csv(), df.head(), df.tail(), df.info(), df.shape, df.describe(), df.columns.

Clean with df.dropna(), df.fillna(), df.drop_duplicates(), df.rename(), and df.astype() to convert data types.

Select with df['col'] or df[['a','b']]; filter with boolean masks, df.query(), and df.loc[rows, cols].

Aggregate with df.groupby('col').agg({'value':'mean'}) to produce one result per group.

Combine tables with pd.merge(left, right, on='key', how='inner'), where how sets which rows are retained.

NumPy supplies vectorized numeric arrays; matplotlib and seaborn produce charts from DataFrames.

Jupyter notebooks interleave code cells, outputs, and markdown so the analysis reads as a narrative.

A capstone case study runs the full process (Ask, Prepare, Process, Analyze, Share, Act) on one business question from start to finish.

The business task and stakeholder requirements are defined before any data is collected or opened.

Data sources, limitations, and assumptions are recorded in a README or data dictionary so the work is auditable.

Metrics are chosen to measure the stated business objective directly, not merely because data is available.

Typical deliverables: cleaned dataset, analysis file or notebook, visualizations, slide deck, and a written summary.

Validation before sharing includes recalculating key figures, checking row counts, and confirming results against the source.

A final presentation states the headline finding, the evidence behind it, and the recommended next step.

Ethics in a case study: minimize and mask personally identifiable information (PII), cite data sources, and disclose limitations.

A reproducible case study shares code and cleaning steps so another analyst can rerun the analysis and reach the same results.