Variables, Measurement, and Data Types

Playback speed

Swipe, scroll or use ← →

Classify by meaning

Variable type determines which summaries, displays, and models have a defensible interpretation. The digits stored in a column do not decide the type; the measurement meaning does. Categorical variables represent groups or labels, while quantitative variables have numerical magnitudes for which arithmetic comparisons are meaningful. Nominal categories have no intrinsic order, whereas ordinal categories have an order without guaranteed equal spacing between levels.

Type	What a value is	Comparison that holds	Example
Nominal	A label with no order	Equal or not equal	Account code
Ordinal	An ordered level, spacing not guaranteed	Higher or lower	Rating from 1 to 5
Discrete quantitative	A number from a finite or countable set	Differences and totals	Purchases per week
Continuous quantitative	A magnitude across an interval	Differences and totals	Transaction value

1 / 16

Classify by meaning

A customer table contains a numeric account code, a satisfaction rating from 1 to 5, a purchase count, and transaction value. The code is categorical, the ordered rating is ordinal, the count is discrete quantitative, and the monetary amount is treated as continuous quantitative at the analysis scale. A column stored as numbers can still be categorical when the numbers are identifiers or codes rather than measured magnitudes. The observational unit determines what one row represents, and each measured characteristic becomes a variable for that unit.

2 / 16
Quick check

A customer ID is stored as six digits. What determines whether it is categorical or quantitative?
1. AAny variable stored as digits is quantitative, so arithmetic on account codes or postal codes has an interpretable magnitude and average.
  
  Numeric encoding does not give labels meaningful arithmetic distance.
2. BCategorical variables represent groups or labels, while quantitative variables have numerical magnitudes for which arithmetic comparisons are meaningful.
  
  Arithmetic differences have an interpretation for measured magnitudes but not for labels such as account or postal codes.
3. CA measured amount becomes categorical whenever analysts group it for a chart, and the discarded within-group magnitudes remain fully recoverable.
  
  Binning creates a categorical representation and loses quantitative detail; it does not change the original construct.
3 / 16
Distinguish discrete and continuous

A discrete quantitative variable has a finite or countable set of possible values, while a continuous variable can take values across an interval. A column stored as numbers can still be categorical when the numbers are identifiers or codes rather than measured magnitudes. Subtracting ordinal labels, a 4-star rating minus a 3-star rating, does not return a measured distance, because the gaps between levels are not known to be equal. Order does not guarantee equal spacing between ordinal categories.

4 / 16
Distinguish discrete and continuous

First define what one row represents. For every column, ask whether arithmetic differences are meaningful, whether values are labels or ordered categories, and whether the support is countable, continuous, bounded, or otherwise constrained. Means and standard deviations require quantitative meaning; frequencies and proportions are natural summaries for categorical variables. Summaries and models must match the variable type because category labels do not support the same arithmetic as measured quantities.

5 / 16
Quick check

Which distinction separates a purchase count from an underlying transaction time?
1. AA discrete quantitative variable has a finite or countable set of possible values, while a continuous variable can take values across an interval.
  
  A purchase count takes separate whole-number values, whereas a transaction time can fall anywhere between two recorded times.
2. BA discrete variable can take every value in an interval, while a continuous variable must have a finite list of possible values.
  
  Those support definitions are reversed.
3. CA variable is continuous only if the observed sample contains infinitely many distinct recorded values.
  
  Finite samples and instrument resolution cannot display infinitely many values from an underlying continuum.
6 / 16
Respect support

The observational unit determines what one row represents, and each measured characteristic becomes a variable for that unit. Counts are typically discrete because their possible values are countable, even when a model approximates them with a continuous distribution. A common error is to assume that a continuous variable must display infinitely many distinct values in the observed sample. A tool with limited resolution, such as a ruler read to the nearest centimeter, records the same few values again and again, even though the length it measures is continuous.

7 / 16
Respect support

Computing a mean of postal codes is mathematically possible but substantively meaningless. Conversely, converting a measured quantity into broad categories discards information and changes the question. The support of a variable includes its possible values and constraints, such as nonnegative counts or proportions bounded between zero and one. Measurement resolution can make a recorded continuous quantity look discrete without changing the underlying quantity being measured.

8 / 16
Quick check

What does defining one row as one customer establish before classifying the columns?
1. AThe observational unit is whichever column has the most unique values.
  
  Uniqueness does not identify the entity measured by each row.
2. BOne row always represents one independent person, even when the table contains transactions, visits, or repeated measurements from the same person.
  
  The data-generating entity must be defined from provenance, not assumed from row count.
3. CThe observational unit determines what one row represents, and each measured characteristic becomes a variable for that unit.
  
  Defining the entity represented by a row determines what each column measures and whether repeated rows are repeated units.
9 / 16
Keep your progress in the app

That’s 3 of 6 quick checks. In the app they stay answered, and every lesson remembers where you left off.

Keep reading here
Match summaries to type

Means and standard deviations require quantitative meaning; frequencies and proportions are natural summaries for categorical variables. Summaries and models must match the variable type because category labels do not support the same arithmetic as measured quantities. A type error occurs when an analyst claims that variable type can be chosen after modeling according to whichever result is most convenient. Type follows the construct and measurement process, not a preferred result.

10 / 16
Quick check

Which summaries naturally match a product category and which match a measured weight?
1. AMeans and standard deviations are the natural summaries for unordered category labels.
  
  Arithmetic center and distance have no meaning for nominal labels.
2. BMeans and standard deviations require quantitative meaning; frequencies and proportions are natural summaries for categorical variables.
  
  Category labels support counts and proportions, while a mean or standard deviation requires meaningful numerical distance.
3. CFrequencies are sufficient for every quantitative variable, even when magnitude and dispersion are the substantive targets of the analysis.
  
  Counts of repeated values can omit the quantitative comparisons the question asks about.
11 / 16
Diagnose coded variables

The support of a variable includes its possible values and constraints, such as nonnegative counts or proportions bounded between zero and one. Computing a mean of postal codes is mathematically possible but substantively meaningless. Conversely, converting a measured quantity into broad categories discards information and changes the question. Type follows the construct and measurement process, not a preferred result. Support rules out impossible values and is therefore an early constraint on both summaries and probability models.

12 / 16
Quick check

Defect totals are nonnegative integer counts, yet a model offered as an exact description of them assigns probability to negative and fractional outcomes. Which property rules it out as an exact model?
1. AA proportion may validly be modeled below zero or above one if the fitted curve is smooth, and a count may take negative fitted values without qualification.
  
  Support constraints are substantive; impossible values signal an unsuitable model or approximation.
2. BSupport is determined only by the values observed in one sample, so unobserved but possible values are excluded from the variable definition.
  
  Support concerns possible values under the measurement process, not just realized sample values.
3. CThe support of a variable includes its possible values and constraints, such as nonnegative counts or proportions bounded between zero and one.
  
  Support rules out impossible values and is therefore an early constraint on both summaries and probability models.
13 / 16
Key takeaways

Measurement resolution can make a recorded continuous quantity look discrete without changing the underlying quantity being measured. First define what one row represents. For every column, ask whether arithmetic differences are meaningful, whether values are labels or ordered categories, and whether the support is countable, continuous, bounded, or otherwise constrained. Storage format does not give identifiers or category codes quantitative meaning.

14 / 16
Key takeaways
Variables, Measurement, and Data Types review lens:
- Meaning: Categorical variables represent groups or labels, while quantitative variables have numerical magnitudes for which arithmetic comparisons are meaningful.
- Support: A column stored as numbers can still be categorical when the numbers are identifiers or codes rather than measured magnitudes.
- Summary: Means and standard deviations require quantitative meaning; frequencies and proportions are natural summaries for categorical variables.
- Resolution: Measurement resolution can make a recorded continuous quantity look discrete without changing the underlying quantity being measured.
15 / 16
Quick check

Temperature is physically continuous but stored to the nearest degree, so a finite file contains repeated integers. How should the underlying variable be classified?
1. AMeasurement resolution can make a recorded continuous quantity look discrete without changing the underlying quantity being measured.
  
  Finite recording precision explains repeated values without changing the continuous construct that the instrument measures.
2. BRounding a continuous measure to two decimals changes the underlying construct into a count.
  
  Resolution changes recorded precision, not the kind of quantity.
3. CRepeated rounded values prove that the variable is nominal.
  
  Ties can arise from measurement resolution and do not remove quantitative meaning.
16 / 16
6 quick checks · then the test

In the app, finishing the quick checks opens this lesson’s 10-question test, and the ones you miss come back exactly when you’re about to forget them.

Take this lesson’s test → Next lesson: Sampling Bias, Precision, and Independence

← Previous

Next card →

The whole course, on your phone

Lessons you can read, audio you can listen to on the way to work, and practice that remembers what you got wrong.

“This app is the ‘One Lake’ for preparing the certification exams. Really worth it.”

Abdelhak Y. Google Play review

Variables, Measurement, and Data Types

Classify by meaning

Distinguish discrete and continuous

Respect support

Keep your progress in the app

Match summaries to type

Diagnose coded variables

Key takeaways

6 quick checks · then the test

The whole course, on your phone