WEEK 2 CHEAT SHEET

Transformations are not just about cleaning your spreadsheet. They reshape your data so you can answer the questions that matter. Almost every meaningful visualization relies on transformed data, even if your software does it automatically. The more comfortable you become with shaping data intentionally, the more control you'll have over what your charts actually reveal.

Filtering, grouping, aggregating, binning, and reshaping data each change how the story in the data shows up. Many datasets are too granular or too messy to visualize effectively without these steps. The core idea is: make the data match the question you're trying to answer.

Some of the most useful insights come from variables that didn't originally exist. Things like "days since last workout," "age group," or "approval status." These new variables often reveal patterns that raw data can't show on its own. When you think about derived variables, you are already doing real analytical thinking.

Several students worried they "didn't have enough data" because their dataset felt too qualitative. The good news: qualitative data becomes incredibly useful once it's coded. Coding lets you organize open-ended responses, categorize behaviors, group patterns, and even assign ordinal structure, without distorting the original meaning. You're not turning text into false numbers; you're giving it analytical structure.

Store dates, times, and numbers in their raw form (ratio or interval) whenever possible. You can always transform downward, to categories, bins, or labels, but you can never reconstruct lost detail. This ensures maximum flexibility when building visualizations.

Transforming downward (like turning a continuous date into a discrete weekday) helps you answer specific, categorical questions. It can expose patterns that raw values hide, such as peak days, clusters, or segment differences. The trick is knowing when a simpler form reveals more insight.

Nominal, ordinal, interval, and ratio variables each allow different mathematical and visual operations. Chart types expect certain levels of measurement, so misclassifying data can lead to misleading visuals. Knowing the level of measurement helps you select charts intentionally, not accidentally.

Not every number in a dataset is actually measuring something. One of the biggest breakthroughs this week was students realizing that many numeric-looking values are really just labels. A number is only quantitative if it reflects a measurable amount or magnitude. If it's simply naming or tagging something, it's qualitative, even if it contains digits. This matters because treating identifiers as true numeric values leads to meaningless statistics and misleading visuals.

Nominal variables represent categories without order. Things like candy types, team names, or departments. These are perfect for bar charts, frequency tables, and categorical comparisons. They help answer questions about "which groups" or "how many in each group."

Ordinal variables have an order but not consistent spacing. Examples include ratings, rankings, or bins like "low / medium / high." You can compare direction and order, but not meaningful numeric differences. These variables work well in sorted bars or comparisons that rely on sequence rather than magnitude.

Interval variables have equal spacing but no true zero. Temperature in Fahrenheit, raw dates, and many scaled scores fall here. You can add and subtract these values, but you cannot make ratio statements like "twice as much." Interval values often appear in line charts, trends, and time-series visuals.

Ratio variables have a true zero and allow meaningful "twice as much" comparisons. Examples include duration, steps, money, age, or counts. These variables unlock more mathematical operations and give you flexibility when choosing visualizations such as line charts, bar charts, comparisons, and trend analysis.

Dates are unique because they change measurement level depending on how you use them. A raw date is interval, but weekday names are nominal, and elapsed time between dates becomes ratio. Understanding this flexibility helps you choose the right chart: line charts for continuous dates, bar charts for discrete categories.

Continuous data can take any value within a range, while discrete data jumps in whole units. This matters because continuous data is suited for lines and trends, while discrete data is best shown using bars, histograms, or tables. The choice directly affects how your audience interprets the shape of the data.

You'll do this frequently: cluster numeric values into categories, rank outcomes, create low/medium/high groupings, assign experience levels. The key point: ordinal data is almost always the result of a transformation.

The level of measurement, combined with your question, determines which charts will work and which will fail. Bar charts, line charts, scatterplots, and heatmaps each expect certain data structures. Good visualization is simply matching chart logic with the nature of the data.

Changing the transformation, such as grouping daily data into weeks, or splitting users by category, can completely change what patterns appear. Slicing correctly requires understanding the story you're trying to uncover. Sometimes one small shift in how the data is aggregated reveals something you would have missed otherwise.

Visualization tools often let you choose whether to treat a variable as continuous or discrete. This choice dramatically changes the chart's design and the story it tells. Knowing the difference helps you align the chart to the actual question you're answering.

A common question this week was, "Do I have enough data?" Not meaning number of rows, but number of variables. The way to answer that is simple: look at whether your dataset allows you to add context, explain variation, or compare groups. If you can do all three, your dataset is strong. If you're missing one or two, consider adding variables that strengthen those dimensions.

Most modern tools detect levels of measurement and choose continuous or discrete settings automatically. But software is guessing. You understand your context better than the tool. Knowing the underlying concepts ensures the visualization reflects your intent, not default settings.

One of the most important habits in analytics is letting the question lead the process. Instead of scrolling through data and wondering what to chart, start with what you want to understand. That question then tells you what variables you need, how to transform them, how to classify them, and which visualization best reveals the answer. This mindset shifts you from chart-making to problem-solving.