A continuous variable is a quantitative variable that can theoretically take any value within a meaningful interval. It is usually obtained through measurement rather than counting. Height, elapsed time, temperature, distance, and body weight are common examples because values can exist between any two recorded measurements.
Introduction
Continuous variables are used whenever researchers measure how much, how long, how far, how fast, or to what degree something occurs. They appear throughout medicine, psychology, education, economics, engineering, environmental science, and the social sciences.
Correctly identifying a continuous variable matters because variable type influences how data are collected, summarized, graphed, modeled, and reported. However, the distinction is not always as simple as checking whether a value contains a decimal. Measurement precision, research purpose, recording method, and the theoretical nature of the characteristic must also be considered.
This article explains:
- What makes a variable continuous.
- How continuous variables differ from discrete and categorical variables.
- How to classify ambiguous examples such as age and test scores.
- How continuous data are measured, analyzed, transformed, and reported.
- Which mistakes researchers should avoid.
Key Takeaways
- A continuous variable can theoretically take any value within an interval.
- Continuous characteristics are generally measured, whereas discrete quantities are generally counted.
- Recorded measurements are limited by instrument precision even when the underlying variable is continuous.
- A continuous variable does not have to be normally distributed.
- Converting continuous measurements into categories can discard information.
- Statistical methods should be chosen according to the research question, design, distribution, and model assumptions—not variable type alone.
What Is a Continuous Variable?
A continuous variable is a numerical characteristic for which values can theoretically exist between any two possible values.
Suppose one person is 170.0 cm tall and another is 171.0 cm tall. Possible heights between them include 170.1, 170.25, 170.526, and many more. The number of decimal places that can actually be recorded depends on the measuring instrument, but the underlying characteristic is treated as varying continuously.
Penn State defines a continuous variable as a characteristic that can take any value, including values between other possible values. This definition distinguishes continuous measurements from counts that have separate permitted outcomes.
Simple definition
A continuous variable measures an amount that can be divided into progressively smaller units.
Examples include:
- 5.2 seconds
- 5.21 seconds
- 5.213 seconds
- 5.2134 seconds
The stopwatch may eventually reach its precision limit, but elapsed time itself is normally conceptualized as continuous.
Main Characteristics of Continuous Variables
A variable is usually continuous when it has the following characteristics.
It is quantitative
Its values represent numerical magnitudes. Arithmetic differences between values have an interpretable meaning.
Values can occur between recorded points
If 10.2 and 10.3 are possible values, values such as 10.21 and 10.254 may also be theoretically possible.
It is usually obtained by measurement
Continuous variables commonly require a:
- Scale
- Ruler
- Thermometer
- Sensor
- Timer
- Laboratory instrument
- Digital monitoring device
Its recorded precision is finite
No real instrument records infinitely many decimal places. A thermometer may report to the nearest 0.1°C, while another may report to the nearest 0.01°C.
This limitation does not automatically make temperature discrete. The variable is classified according to its theoretical and substantive nature, not merely the number of digits stored in a dataset.
It has a meaningful unit
Examples include centimeters, kilograms, seconds, degrees Celsius, millimeters of mercury, liters, and meters per second.
Its range may be bounded
Continuous does not mean unlimited. Human body temperature can vary continuously within a biologically plausible range. A proportion may vary continuously between 0 and 1. A test-completion time cannot be less than zero.
How to Identify a Continuous Variable
Use the following five-step process.
Step 1: Determine what is being observed
Identify the actual characteristic, not merely the column name or displayed numbers.
For example, “age” could mean:
- Exact elapsed time since birth
- Completed years
- An age category such as 18–24
- School grade level
These are not identical variables.
Step 2: Ask whether it is measured or counted
Measurements such as length, weight, time, and temperature are commonly continuous.
Counts such as the number of students, hospital visits, errors, or publications are normally discrete.
This is a useful first test, but it is not sufficient in every case.
Step 3: Ask whether an intermediate value is theoretically possible
Consider two valid values. Can the characteristic meaningfully take a value between them?
Between 60 kg and 61 kg, 60.4 kg is possible. Weight is therefore continuous.
Between three children and four children, 3.4 children is not a valid family count. Number of children is discrete.
Step 4: Separate the underlying characteristic from its recording method
A person’s age changes continuously, but a survey may record only completed years.
Likewise, cholesterol concentration is continuous even when a laboratory reports a rounded whole-number value.
Step 5: Examine how the variable will be analyzed
The research purpose can influence how a recorded variable is represented. Exact age may be modeled continuously in regression, grouped into age bands for reporting, or used to define an eligibility category.
The researcher should state the operational decision rather than assuming that one label applies in every context.
Examples of Continuous Variables
| Field | Continuous variable | Possible unit | Why it is continuous |
|---|---|---|---|
| Medicine | Systolic blood pressure | mmHg | Pressure can vary between recorded readings |
| Public health | Body mass index | kg/m² | Values can occur throughout a numerical range |
| Psychology | Reaction time | Milliseconds | Time can be divided into smaller units |
| Education | Time spent completing an examination | Minutes or seconds | Intermediate durations are possible |
| Biology | Plant height | Centimeters | Height is measurable at different levels of precision |
| Chemistry | Solution temperature | °C or K | Intermediate temperatures are possible |
| Environmental science | Rainfall | Millimeters | Amounts can include fractional values |
| Engineering | Tensile strength | MPa | Strength is measured on a numerical continuum |
| Economics | Household income | Currency per year | Commonly modeled as continuous despite currency increments |
| Business | Delivery time | Hours or days | Duration can be measured fractionally |
| Sports science | Running speed | m/s | Speed can take intermediate values |
| Computer science | Server response time | Milliseconds | Response duration is measured continuously |
| Geography | Distance from a city center | Kilometers | Distance can be subdivided |
| Agriculture | Soil moisture | Percentage or volumetric ratio | Moisture can vary throughout a range |
| Astronomy | Distance between objects | Kilometers or light-years | The characteristic is measured, not counted |
Continuous Variable Versus Discrete Variable
A continuous variable can theoretically take any value within an interval. A discrete variable takes separate, countable values.
| Feature | Continuous variable | Discrete variable |
| Basic meaning | Measurement along a continuum | Count or separated outcome |
| Possible values | Any value within an interval | Finite or countably infinite set |
| Intermediate values | Usually possible | Not possible between adjacent permitted values |
| Common values | Integers, fractions, or decimals | Often integers, but not necessarily |
| Typical source | Measurement | Counting |
| Example | Body weight | Number of patients |
| Example | Temperature | Number of defective products |
| Example | Elapsed time | Number of website visits |
| Common graph | Histogram, box plot, density plot | Bar chart or frequency plot |
| Probability representation | Probability density function | Probability mass function |
The most reliable distinction
The best question is not “Does it contain decimals?”
Ask:
Can the underlying characteristic meaningfully take values between two possible observations?
A decimal can still represent discrete data. For example, if three successes are observed in ten trials, the proportion is 0.3. With ten fixed trials, only values 0.0, 0.1, 0.2, and so on are possible. The proportion is based on a discrete count despite being written as a decimal.
Continuous Versus Categorical Variables
A continuous variable expresses numerical magnitude. A categorical variable assigns observations to groups.
| Variable | Classification | Explanation |
| Height in centimeters | Continuous | Numerical measurement |
| Blood group | Categorical nominal | Categories have no quantitative order |
| Satisfaction: low, medium, high | Categorical ordinal | Categories are ordered |
| Temperature in Celsius | Continuous | Differences between values are meaningful |
| University department | Categorical nominal | Labels rather than measurements |
| Exact response time | Continuous | Measured duration |
Giving categories numerical codes does not make them continuous. Coding departments as 1, 2, and 3 does not create meaningful numerical distance between the departments.
Continuous Variables and Levels of Measurement
Continuous variables are commonly measured on interval or ratio scales.
Interval-scale continuous variables
An interval scale has equal differences between values, but zero is not a true absence of the characteristic.
A common example is temperature in degrees Celsius. The difference between 10°C and 20°C is comparable with the difference between 20°C and 30°C. However, 20°C is not meaningfully “twice as hot” as 10°C.
Ratio-scale continuous variables
A ratio scale has equal intervals and a meaningful zero point.
Examples include:
- Length
- Mass
- Duration
- Distance
- Absolute temperature in kelvins
- Volume
A duration of 20 seconds is meaningfully twice a duration of 10 seconds.
The interval–ratio distinction comes from measurement theory and should not be confused with the continuous–discrete distinction. A variable can be continuous at either the interval or ratio level.
Types and Roles of Continuous Variables
Terms such as continuous independent variable and continuous dependent variable describe a variable’s role in a study rather than a separate mathematical type.
Continuous independent variable
A continuous independent variable is used as an explanatory, exposure, or predictor variable.
Example:
A researcher studies whether weekly study time predicts examination performance.
- Weekly study time: continuous predictor
- Examination result: outcome
Continuous dependent variable
A continuous dependent variable is the measured outcome.
Example:
A clinical trial compares the mean reduction in systolic blood pressure between two treatments.
- Treatment group: categorical independent variable
- Change in blood pressure: continuous dependent variable
Continuous control variable
A researcher may adjust for a continuous variable that could influence the relationship of interest.
For example, age, baseline blood pressure, or household income may be included as a control variable.
Continuous random variable
A continuous random variable is a probability-theory concept. It assigns numerical values to random outcomes and is represented by a continuous probability distribution.
Examples include an idealized model of:
- Waiting time
- Measurement error
- Lifespan
- Rainfall amount
- Component strength
Continuous Random Variables and Probability
A continuous random variable is usually described by a probability density function, written as (f(x)).
The probability that (X) lies between (a) and (b) is:
[
P(a \leq X \leq b)=\int_a^b f(x),dx
]
The total area under a valid probability density function is:
[
\int_{-\infty}^{\infty} f(x),dx=1
]
For an ideal continuous distribution, the probability of one exact value is zero:
[
P(X=x)=0
]
This does not mean that the value cannot be observed. It means probability is assigned to intervals rather than infinitely precise individual points.
Continuous variable versus continuous distribution
These expressions should not be used interchangeably:
- Continuous variable: the characteristic being measured.
- Observed continuous data: the values recorded from participants or objects.
- Continuous random variable: a mathematical representation of random outcomes.
- Continuous probability distribution: the probability model assigned to those outcomes.
Is a Continuous Variable Always Normally Distributed?
No. A variable can be continuous without following a normal distribution.
Possible shapes include:
- Symmetric
- Right-skewed
- Left-skewed
- Uniform
- Bimodal
- Heavy-tailed
- Bounded
- Zero-inflated
- Censored
Examples:
- Adult height may be approximately symmetric within a relatively homogeneous population.
- Household income is often right-skewed.
- Waiting time may follow an exponential or other right-skewed distribution.
- Percentages are bounded between 0 and 100.
- Laboratory measurements may be censored by a detection limit.
Variable type and distribution shape are separate characteristics.
Commonly Misclassified Variables
Is age continuous or discrete?
Age is theoretically continuous because elapsed time since birth can be measured in years, months, days, seconds, or smaller units.
However:
- Exact age is continuous.
- Age in completed years is a rounded measurement.
- Age bands such as 18–24 and 25–34 are ordinal categories.
Researchers should report how age was calculated and represented.
Is income continuous?
Income is usually analyzed as continuous because it has many ordered values and meaningful numerical differences.
Technically, a currency has a smallest transaction unit, such as a cent or penny. This creates finite measurement increments. In most economic and social-science analyses, however, the number of possible values is large enough that treating income as continuous is reasonable.
Income is frequently skewed, so the median, interquartile range, transformation, or a suitable model may be more informative than the mean alone.
Is heart rate continuous?
This depends on what is measured.
- The number of beats counted in a fixed period is a count and therefore discrete.
- An instrument-derived rate expressed in beats per minute may be modeled as approximately continuous.
- Beat-to-beat intervals are continuous time measurements.
The operational definition should determine the classification.
Are examination scores continuous?
The number of correct answers on a fixed examination is discrete because it has a limited set of possible values.
A standardized or scaled score may have many possible values and is often treated as approximately continuous. This is an analytical convention, not proof that the underlying item count is theoretically continuous.
Are Likert-scale data continuous?
A single Likert item, such as a response from 1 to 5, is ordinal rather than continuous.
A total or average score created from many related items is often analyzed as approximately continuous when the scale has sufficient range and its measurement properties support that decision. Researchers should justify the treatment and consider sensitivity analyses when the choice is consequential.
Are percentages continuous?
A percentage can be continuous or discrete depending on how it is generated.
- The percentage of an area covered by vegetation may be measured continuously.
- The percentage of six participants who responded “yes” can take only a small set of values and is based on a discrete count.
Is money continuous?
Currency is technically recorded in fixed increments. Nevertheless, expenditure, revenue, salary, and price variables are often modeled as continuous when they have many possible values.
Are dates continuous?
Calendar dates are stored in discrete units such as days or seconds, but elapsed time between events may be treated as continuous. The correct classification depends on whether the variable represents a calendar label, ordered occasion, or measured duration.
How to Measure and Operationalize a Continuous Variable
Operationalization specifies exactly how a concept will be measured.
A strong operational definition should identify:
- The construct being measured.
- The instrument or data source.
- The unit.
- The timing of measurement.
- The possible range.
- The required precision.
- Any averaging or repeated measurements.
- Rules for impossible, missing, censored, or outlying values.
Example operational definition
“Resting systolic blood pressure will be measured in millimeters of mercury using a calibrated automatic monitor. After five minutes of seated rest, three readings will be collected at one-minute intervals. The mean of the second and third readings will be used in the analysis.”
This is more reproducible than stating only that “blood pressure will be recorded.”
Instrument precision and accuracy
Precision refers to how finely a measurement is recorded. Accuracy refers to how close it is to the true value.
A highly precise instrument can still be inaccurate if it is poorly calibrated. Researchers should document calibration, validation, and quality-control procedures where relevant.
Rounding
Rounding should normally occur after necessary calculations rather than at every intermediate step. Excessive rounding can:
- Hide genuine variation.
- Produce artificial ties.
- Alter threshold classification.
- Change calculated statistics slightly.
Detection limits and censoring
A laboratory instrument may report values as “below detection limit” rather than producing an exact result. Such observations are censored, not simply missing.
Replacing every censored value with zero or half the detection limit can distort results. The treatment should match the study design and analytical method.
Summarizing Continuous Data
Continuous data are commonly summarized using measures of center and spread.
Measures of center
- Mean: Useful when the distribution is reasonably symmetric and not dominated by influential extremes.
- Median: Useful for skewed distributions and outlier-prone measurements.
- Mode: Less informative for highly precise continuous data because exact values may rarely repeat.
Measures of variability
- Standard deviation
- Variance
- Interquartile range
- Range
- Median absolute deviation
- Percentiles
A good summary matches the distribution. Mean and standard deviation often work well for approximately symmetric data. Median and interquartile range are often more informative for strongly skewed measurements.
Visualizing Continuous Variables
Histogram
A histogram divides the measurement range into intervals and displays the number or proportion of observations in each interval.
It can reveal:
- Center
- Spread
- Skewness
- Outliers
- Gaps
- Multiple modes
The apparent shape can change with bin width, so researchers should not rely on one arbitrary set of bins.
Box plot
A box plot summarizes the median, quartiles, spread, and potentially unusual observations. It is especially useful for comparing a continuous outcome across groups.
Density plot
A density plot provides a smoothed representation of a distribution. Its appearance depends on the smoothing parameter and should not be interpreted as an exact picture of the population.
Scatterplot
A scatterplot displays the relationship between two numerical variables. It can reveal:
- Direction
- Strength
- Curvature
- Clusters
- Unequal variability
- Influential observations
Line graph
A line graph is useful when a continuous measurement is observed over ordered time points. Lines should not imply unobserved continuity when measurements occur only at a few unrelated categories.
Statistical Methods for Continuous Variables
No single statistical test is “the test for continuous variables.” The appropriate method depends on the outcome, predictors, design, sampling process, distribution, and research objective.
| Research question | Example | Possible method |
| Describe one continuous variable | What is the distribution of waiting time? | Mean, median, SD, IQR, histogram |
| Compare two independent groups | Do two treatments differ in mean recovery time? | Independent-samples t-test or suitable alternative |
| Compare paired measurements | Did blood pressure change after treatment? | Paired t-test or suitable paired alternative |
| Compare three or more groups | Do mean scores differ across teaching methods? | ANOVA or suitable alternative |
| Relate two numerical variables | Is study time associated with examination performance? | Correlation and regression |
| Predict a continuous outcome | Can income and education predict expenditure? | Linear or flexible regression model |
| Analyze repeated measurements | How does pain change over several visits? | Repeated-measures or mixed-effects model |
| Model time to an event | How long until equipment failure? | Survival analysis |
| Examine agreement | Do two instruments give similar measurements? | Agreement analysis, not correlation alone |
Parametric and nonparametric methods
The labels “parametric” and “nonparametric” should not be reduced to “normal versus non-normal data.”
Researchers should consider:
- Independence
- Study design
- Sample size
- Outcome distribution
- Model residuals
- Variance structure
- Functional form
- Outliers
- Clustering
- Repeated observations
- Missing-data mechanism
For a linear model, the raw predictor variables do not all need to be normally distributed. Relevant assumptions generally concern the model’s errors, functional form, dependence structure, and variance—not a blanket requirement that every continuous column be normal.
Continuous Predictors in Regression
A continuous predictor is often entered into a regression model as a linear term:
[
Y=\beta_0+\beta_1X+\varepsilon
]
Here, (\beta_1) represents the expected change in (Y) for a one-unit increase in (X), under the model.
However, using a continuous predictor does not guarantee that its relationship with the outcome is linear.
Possible alternatives include:
- Logarithmic transformations
- Polynomial terms
- Restricted cubic splines
- Fractional polynomials
- Generalized additive models
- Piecewise models based on defensible thresholds
Flexible approaches can represent curvature without discarding the original measurement scale. The choice should consider subject knowledge, sample size, interpretability, and the intended use of the model.
Transforming Continuous Variables
A transformation changes the numerical scale of a variable.
Common transformations include:
- Natural logarithm
- Base-10 logarithm
- Square root
- Reciprocal
- Logit transformation for suitable proportions
- Standardization
- Centering
Transformations may be used to:
- Represent a relationship more appropriately.
- Reduce strong right skew.
- Stabilize variance.
- improve numerical performance.
- Make coefficients easier to interpret.
- Place predictors on comparable scales.
A transformation should not be applied automatically merely because a normality test is significant. Researchers should examine the scientific meaning, distribution, model diagnostics, and analysis objective.
Standardization
A standardized score is commonly calculated as:
[
z=\frac{x-\bar{x}}{s}
]
The transformed variable has a sample mean of approximately zero and standard deviation of one.
Standardization changes the unit but does not change an underlying continuous variable into another variable type.
Should Continuous Variables Be Converted into Categories?
Usually, researchers should preserve continuous measurements whenever possible.
For example, age may be divided into “younger” and “older,” or blood pressure into “low” and “high.” Although these categories may be convenient, they can:
- Discard information.
- Reduce statistical power.
- Treat similar observations on opposite sides of a cut point as fundamentally different.
- Treat substantially different observations within one group as equivalent.
- Create an artificial threshold.
- Leave residual confounding.
- Make results dependent on an arbitrary cut point.
Altman and Royston demonstrated that dichotomizing at the median can cause information loss comparable to discarding a substantial portion of the observations.
When categories may still be justified
Categories may be appropriate when:
- A clinically or legally established threshold is central to the research question.
- Categories correspond to actual decision rules.
- Public communication requires understandable risk groups.
- Privacy or data-release rules require grouping.
- The original exact values are unavailable.
- A descriptive table needs supplementary categories.
Even then, the continuous version can often be retained for the primary analysis and categories used secondarily for presentation.
Continuous Variables in Modern Research
Modern research can collect continuous measurements at much higher frequency than traditional surveys.
Examples include:
- Wearable-device heart-rate measurements
- Smartphone accelerometer data
- Continuous glucose monitoring
- Environmental sensors
- Satellite measurements
- Industrial telemetry
- Website response times
- Eye-tracking coordinates
- Audio signals
- Neurophysiological recordings
These data create new analytical issues.
Repeated observations
Thousands of measurements from one participant are not thousands of independent participants. Models may need to account for clustering, serial correlation, and individual-level variation.
Sampling frequency
A sensor recording every second provides a different representation from one recording every hour. The chosen frequency affects detectable patterns, storage requirements, and noise.
Measurement drift
Sensors can become less accurate over time. Calibration records and quality-control checks should be retained.
Large numbers of observations
A very large dataset does not correct biased measurement, poor operationalization, confounding, or nonrepresentative sampling.
Missing segments
Wearable or sensor data may be missing because of device removal, battery failure, connection loss, or participant behavior. These causes may not be random.
Digital Tools and Artificial Intelligence
Continuous variables can be analyzed using R, Python, SPSS, Stata, SAS, jamovi, JASP, MATLAB, spreadsheets, and specialized scientific software.
These tools can help researchers:
- Generate descriptive statistics.
- Produce histograms and box plots.
- Detect impossible values.
- Fit statistical models.
- Check residuals.
- Apply transformations.
- Model nonlinear relationships.
- Conduct sensitivity analyses.
- Create reproducible reports.
Appropriate uses of artificial intelligence
AI tools can assist with:
- Drafting data dictionaries.
- Explaining statistical output.
- Suggesting exploratory plots.
- Producing starter code.
- Identifying inconsistent units.
- Documenting variable transformations.
- Translating code between statistical languages.
Necessary safeguards
AI should not be allowed to make unsupervised analytical decisions. Researchers should verify:
- Whether the variable was classified correctly.
- Whether the suggested model matches the study design.
- Whether generated code runs as intended.
- Whether missing-value codes were interpreted correctly.
- Whether units were converted accurately.
- Whether references and statistical claims are genuine.
- Whether confidential data were exposed to an unauthorized service.
A statistically valid-looking output can still be wrong when the input data, assumptions, or research design have been misunderstood.
Advantages of Continuous Variables
They preserve detailed information
Continuous measurements retain more variation than broad categories.
They support flexible analysis
They can be summarized, transformed, plotted, and modeled using many statistical techniques.
They can improve statistical efficiency
Preserving the original scale generally uses more information than dividing values into arbitrary groups.
They represent gradual change
Continuous variables can reflect dose–response patterns, growth, decline, and nonlinear relationships.
They allow meaningful effect estimates
Researchers can report changes per unit, per standard deviation, or across a meaningful range.
Limitations and Practical Challenges
Measurement error
Every observed value may differ from the unknown true value.
Instrument limitations
Resolution, calibration, detection limits, and environmental conditions can affect measurements.
Outliers
Extreme observations may represent genuine cases, data-entry errors, unit errors, or device failures.
Skewed distributions
Some continuous variables cannot be adequately summarized by the mean and standard deviation alone.
Nonlinear effects
A one-unit increase may not have the same meaning throughout the variable’s range.
Missing data
Missing measurements can introduce bias when the reason for missingness is associated with the variable or outcome.
False precision
Reporting excessive decimal places can imply greater accuracy than the measurement process supports.
Common Mistakes
Mistake 1: Assuming every decimal variable is continuous
Decimals may result from proportions, averages, or coded categories.
Mistake 2: Assuming every whole-number column is discrete
A continuous measurement may have been rounded to whole units.
Mistake 3: Treating continuous as synonymous with normal
Continuity describes possible values. Normality describes a distributional shape.
Mistake 4: Categorizing without justification
Convenient cut points can reduce information and create artificial differences.
Mistake 5: Ignoring units
A dataset containing kilograms in some rows and pounds in others can produce plausible-looking but invalid results.
Mistake 6: Testing only the raw variable for normality
Model assumptions should be assessed in relation to the chosen method, especially its residuals and dependence structure.
Mistake 7: Assuming correlation shows agreement
Two instruments can be highly correlated while consistently giving different measurements.
Mistake 8: Treating repeated measurements as independent
Measurements from the same participant, school, hospital, or device may be correlated.
Mistake 9: Assuming a linear effect
Researchers should inspect whether the association changes across the predictor’s range.
Mistake 10: Deleting outliers automatically
An unusual value should be investigated before exclusion. A valid extreme case may contain important information.
Worked Research Example
Suppose a researcher investigates whether weekly study time is associated with final examination performance among university students.
Variables
- Weekly study time in hours: continuous predictor
- Final examination percentage: approximately continuous outcome
- Academic program: categorical control variable
- Prior grade-point average: continuous control variable
Step 1: Define the variables
Study time should specify:
- Whether it is self-reported or digitally tracked.
- The reference period.
- Whether lectures are included.
- The maximum plausible value.
- How incomplete weeks are handled.
Step 2: Inspect the data
The researcher should examine:
- Missing values.
- Minimum and maximum.
- Histograms.
- Scatterplots.
- Suspicious values such as 200 study hours per week.
- Whether study time is strongly right-skewed.
Step 3: Examine the relationship
A scatterplot with a smooth trend can show whether the relationship is approximately linear or whether gains level off at higher study times.
Step 4: Select the model
A linear regression may be suitable if its functional-form and residual assumptions are reasonable. A spline or another flexible term may be preferable if the relationship is curved.
Step 5: Interpret meaningfully
Instead of reporting only a coefficient per one minute or an arbitrary “high versus low” comparison, the researcher could report the expected difference associated with an additional five hours of weekly study.
Step 6: Report limitations
Self-reported study time may contain recall or social-desirability error. The observed association does not by itself prove that additional study causes higher scores.
How to Report Continuous Variables
A research report should normally state:
- The variable’s operational definition.
- The unit.
- The instrument or source.
- The timing and number of measurements.
- The precision or rounding rule where relevant.
- The descriptive statistics used.
- Any transformation.
- Any categorization and its justification.
- The model used.
- The effect scale and confidence interval.
- Relevant missing-data and sensitivity analyses.
Example methods statement
“Reaction time was recorded in milliseconds using a computerized task. Trials below 100 ms or above 3,000 ms were flagged for review according to the prespecified protocol. Participant-level median reaction time was used because individual trial times were right-skewed.”
Example results statement
“Median reaction time was 612 ms (interquartile range 548–704 ms). The adjusted model estimated a 28-ms decrease in reaction time per additional practice session, with uncertainty reported using a 95% confidence interval.”
The number of decimal places should reflect the accuracy of the instrument and the practical meaning of the result.
Conclusion
A continuous variable is a quantitative characteristic that can theoretically take intermediate values throughout an interval. Correct classification requires more than looking for decimals: researchers must consider the underlying characteristic, measurement process, precision, and analytical purpose.
Continuous variables preserve valuable information and support detailed statistical analysis, but they require careful measurement, visualization, model checking, and reporting. Researchers should avoid assuming normality, imposing arbitrary categories, or treating rounded observations as proof that a characteristic is discrete.
References
- Altman, D. G., & Royston, P. (2006). The cost of dichotomising continuous variables. BMJ, 332(7549), 1080. doi:10.1136/bmj.332.7549.1080
- Centers for Disease Control and Prevention, National Center for Health Statistics. (n.d.). NHANES tutorials: Datasets and documentation module.
- National Institute of Standards and Technology. (n.d.). Histogram. In NIST/SEMATECH e-Handbook of Statistical Methods.
- Pennsylvania State University, Department of Statistics. (n.d.). Describing data and ethics. STAT 800: Applied Research Methods.
- Royston, P., Altman, D. G., & Sauerbrei, W. (2006). Dichotomizing continuous predictors in multiple regression: A bad idea. Statistics in Medicine, 25(1), 127–141. doi:10.1002/sim.2331
- Sauerbrei, W., Perperoglou, A., Schmid, M., Abrahamowicz, M., Becher, H., Binder, H., Dunkler, D., Harrell, F. E., Royston, P., & Heinze, G. (2020). State of the art in selection of variables and functional forms in multivariable analysis—Outstanding issues. Diagnostic and Prognostic Research, 4, Article 3. doi:10.1186/s41512-020-00074-3
- Shreffler, J., & Huecker, M. R. (2023). Types of variables and commonly used statistical designs. In StatPearls. StatPearls Publishing.
- Statistics Canada. (2021). Types of variables. In Statistics: Power from data!
- Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103(2684), 677–680. doi:10.1126/science.103.2684.677
