Descriptive statistics are methods for organizing, summarizing, and presenting the characteristics of collected data. They include frequencies, percentages, measures of central tendency such as the mean or median, measures of variability such as the standard deviation or interquartile range, and visual displays such as tables, histograms, and box plots.
Introduction
A raw dataset may contain hundreds, thousands, or millions of observations. Reading the individual values rarely provides an immediate understanding of what the data show. Descriptive statistics reduce that complexity by answering practical questions such as:
- How many observations are available?
- What is a typical value?
- How much do values differ?
- Is the distribution symmetrical or skewed?
- Which categories are most common?
- Are there unusual observations?
- How do two groups compare descriptively?
This guide explains the major types of descriptive statistics, how to calculate and choose them, how to interpret them, and how to report them transparently in academic research.
Key takeaways
- Descriptive statistics describe the data that were observed; they do not by themselves establish causation or generalize results to an unobserved population.
- A useful summary normally combines sample size, center, spread, and an appropriate table or graph.
- The correct statistic depends on the variable’s measurement level, distribution, missing values, and research purpose.
- Means are sensitive to extreme observations, whereas medians and interquartile ranges are more resistant.
- Missing values, weighting, group definitions, and software settings should be reported rather than hidden.
- Calculation is only one step; accurate interpretation and transparent reporting are equally important.
What are descriptive statistics?
Descriptive statistics are numerical and graphical methods used to describe the main features of a dataset.
A descriptive statistic is one calculated value, such as a mean, percentage, median, or standard deviation. Descriptive statistics refers more broadly to the process of summarizing and presenting data.
They may be applied to:
- An entire population, such as every student enrolled at a university
- A sample selected from a larger population
- Survey responses
- Experimental measurements
- Observational records
- Administrative or business data
- Text, image, sensor, or digital-behaviour data after relevant features have been coded
The defining feature is not the source of the data. It is the scope of the conclusion. A descriptive statement remains limited to the observed dataset unless inferential methods or an appropriate population-level design support a broader conclusion.
Why are descriptive statistics important?
Descriptive statistics make data interpretable without requiring readers to inspect every observation. They help researchers:
- Check data quality before formal analysis.
- Describe participant or sample characteristics.
- Identify missing values, impossible values, and unusual observations.
- Compare groups at a preliminary level.
- Select appropriate inferential procedures.
- Communicate findings through tables, figures, and concise text.
- Detect patterns that deserve additional investigation.
Descriptive analysis is often the first formal stage of quantitative data analysis, but it should not be treated as an automatic button-clicking exercise. The selected summaries must match the variable and the research question.
What are the main types of descriptive statistics?
Descriptive statistics can be organized into six practical categories:
- Frequency and distribution: counts, proportions, percentages, and frequency tables
- Central tendency: mean, median, and mode
- Position: quartiles, percentiles, ranks, and the five-number summary
- Variability: range, interquartile range, variance, standard deviation, and related measures
- Distribution shape: skewness, kurtosis, modality, and graphical form
- Relationships: cross-tabulations, covariance, correlation, and grouped summaries
Some introductory sources describe only three categories—distribution, central tendency, and variability. The six-category structure is more useful in research because it separates concepts that require different decisions.
Frequency distributions, counts and percentages
A frequency distribution shows how often each value, category, or interval occurs.
Absolute frequency
The absolute frequency is the number of observations in a category.
Suppose 100 students report their preferred learning format:
| Format | Frequency |
|---|---|
| In person | 46 |
| Online | 31 |
| Hybrid | 23 |
The absolute frequencies are 46, 31, and 23.
Relative frequency and proportion
A relative frequency divides a category’s frequency by the total number of relevant observations:
[
\text{Proportion} = \frac{\text{Category frequency}}{\text{Total relevant observations}}
]
For hybrid learning:
[
\frac{23}{100}=0.23
]
Percentage
A percentage multiplies a proportion by 100:
[
\text{Percentage}=\text{Proportion}\times100
]
Therefore, 23% of the students preferred hybrid learning.
Valid percentage
A valid percentage excludes missing responses from the denominator.
For example, suppose 100 participants were enrolled but only 92 answered a question. If 46 selected “yes”:
- Percentage based on everyone enrolled: (46/100=46%)
- Valid percentage based on respondents: (46/92=50%)
Both calculations are mathematically correct, but they answer different questions. A research report should state the denominator.
Cumulative frequency
Cumulative frequency adds observations progressively across ordered categories. It is meaningful for ordered values such as age groups, income bands, test-score intervals, or satisfaction levels. It is usually not meaningful for unordered nominal categories such as nationality or blood group.
Measures of central tendency
Measures of central tendency describe a dataset’s central or typical value. The three familiar measures are the mean, median, and mode, but they are not interchangeable.
Mean
The arithmetic mean is the sum of all values divided by the number of values:
[
\bar{x}=\frac{\sum_{i=1}^{n}x_i}{n}
]
Where:
- (\bar{x}) is the sample mean
- (x_i) is each observed value
- (n) is the number of observations
For the values 2, 4, 6, and 8:
[
\bar{x}=\frac{2+4+6+8}{4}=5
]
Strengths of the mean
- Uses every value
- Is familiar and easy to use in later statistical calculations
- Works well for approximately symmetrical quantitative distributions
- Supports calculations such as variance, standard deviation, correlation, and regression
Limitations of the mean
The mean is sensitive to very high or low values.
Consider five annual salaries in thousands:
30, 32, 34, 35, 250
The mean is:
[
\frac{381}{5}=76.2
]
A value of 76.2 does not represent the salary received by most members of this small group. The extreme value of 250 pulls the mean upward.
Median
The median is the middle value after observations have been ordered.
For an odd number of observations, it is the single middle value. For an even number, the conventional median is the mean of the two middle values.
For:
30, 32, 34, 35, 250
the median is 34.
When is the median preferable?
The median is often more informative when:
- The distribution is strongly skewed
- Extreme values are present
- The variable is ordinal
- An open-ended category is used, such as “£100,000 or more”
- The research question concerns the midpoint rather than the arithmetic balance point
The median should normally be paired with the interquartile range or another quantile-based measure of spread.
Mode
The mode is the most frequently occurring value or category.
For:
2, 3, 3, 4, 6
the mode is 3.
A dataset may be:
- Unimodal: one mode
- Bimodal: two prominent modes
- Multimodal: several prominent modes
- Without a unique mode: no value occurs more often than the others
The mode is the only standard measure of central tendency that can be applied directly to nominal categories. For example, the modal transport method could be “bus.”
A mode may also reveal subgroups. A bimodal distribution of completion times might indicate two populations using different strategies or technologies.
Weighted mean
A weighted mean gives different observations or groups different importance:
[
\bar{x}_w=\frac{\sum w_i x_i}{\sum w_i}
]
Where (w_i) is the weight assigned to each value.
Weighted means are used for:
- Course grades with differently weighted assessments
- Survey estimates adjusted for sampling probabilities
- Price indices
- Portfolio returns
- Meta-analytic summaries
Researchers should not substitute a simple mean when a survey design requires weights. Weighted percentages and weighted measures of spread may also be necessary.
Geometric mean
For positive values, the geometric mean is:
[
G=\left(\prod_{i=1}^{n}x_i\right)^{1/n}
]
It is useful for multiplicative processes, growth factors, fold changes, and some positively skewed biological or financial measures.
It cannot be calculated in the ordinary way when values are zero or negative. Its interpretation and any logarithmic transformation should be explained.
Mean, median or mode: which should you choose?
| Situation | Preferred summary | Reason |
|---|---|---|
| Unordered categorical data | Mode, counts, percentages | Categories have no numerical order |
| Ordered categories | Median, mode, cumulative percentages | Order is meaningful but intervals may not be equal |
| Symmetrical quantitative data | Mean | Uses all numerical values |
| Strongly skewed quantitative data | Median | Less affected by extreme values |
| Multiplicative positive data | Geometric mean | Reflects proportional rather than additive change |
| Weighted survey or grade data | Weighted mean | Observations have unequal contribution |
| Several distinct peaks | Modes plus a graph | A single center can conceal subgroups |
No single measure is automatically “best.” The correct choice depends on what “typical” means in the research context.
Measures of position
Measures of position describe where an observation lies within an ordered distribution.
Quartiles
Quartiles divide ordered data into four parts:
- (Q_1): first quartile, near the 25th percentile
- (Q_2): second quartile, the median
- (Q_3): third quartile, near the 75th percentile
Percentiles
The (p)th percentile is a value associated with a specified cumulative position in the distribution. For example, the 90th percentile identifies a point near or below which 90% of observations fall, depending on the precise computational definition.
Percentiles are common in:
- Standardized tests
- Growth charts
- Reference ranges
- Income distributions
- Performance dashboards
Five-number summary
The five-number summary consists of:
- Minimum
- First quartile
- Median
- Third quartile
- Maximum
It provides a compact description of location and spread and forms the basis of a conventional box plot.
Why quartile results may differ
There is no single universally implemented sample-quantile algorithm. Statistical packages may interpolate between observations differently, especially in small datasets. Hyndman and Fan (1996) compared several definitions, and software such as R allows users to select among multiple quantile types.
For reproducible work:
- Record the software and version.
- State the quantile method when small-sample quartiles materially affect results.
- Avoid assuming that a hand-calculated quartile must match every program.
Measures of variability
Measures of variability describe how far observations are spread around one another or around a central value.
A measure of center without spread may be misleading. Two classes can have the same mean test score but very different consistency.
Range
The range is:
[
\text{Range}=\text{Maximum}-\text{Minimum}
]
For 3, 5, 6, 8, and 12:
[
12-3=9
]
The range is simple but depends only on two observations. It is highly sensitive to extreme values and tends to increase as sample size increases.
Interquartile range
The interquartile range is:
[
IQR=Q_3-Q_1
]
It describes the width of the middle 50% of ordered observations.
Because it ignores the most extreme quarters of the distribution, the IQR is more resistant to outliers than the full range. It is commonly reported with the median.
Variance
Variance measures squared deviations from the mean.
Population variance
[
\sigma^2=\frac{\sum_{i=1}^{N}(x_i-\mu)^2}{N}
]
Sample variance
[
s^2=\frac{\sum_{i=1}^{n}(x_i-\bar{x})^2}{n-1}
]
The (n-1) denominator is used in the conventional sample variance to correct the tendency of the unadjusted sample calculation to underestimate population variance under standard random-sampling assumptions.
Variance is fundamental in statistical theory, but its units are squared. If observations are measured in centimetres, the variance is expressed in square centimetres.
Standard deviation
The standard deviation is the square root of the variance:
[
s=\sqrt{s^2}
]
It can be interpreted as a measure of the typical spread of values around the mean, expressed in the original measurement unit.
A small standard deviation indicates that observations are relatively concentrated around the mean. A large standard deviation indicates greater dispersion.
The standard deviation should not be interpreted as a complete description of the distribution. The same SD can accompany different shapes, outliers, or clusters.
Mean absolute deviation and median absolute deviation
The term MAD is used inconsistently, so researchers should define it.
Mean absolute deviation from the mean
[
\frac{\sum |x_i-\bar{x}|}{n}
]
This calculates the average absolute distance from the mean.
Median absolute deviation from the median
[
\operatorname{median}(|x_i-\operatorname{median}(x)|)
]
This robust measure is less sensitive to extreme observations than the standard deviation. Some software reports a scaled version so that it is comparable to the SD under a normal-distribution model.
Always state whether MAD refers to a mean absolute deviation, an unscaled median absolute deviation, or a scaled robust estimate.
Coefficient of variation
The coefficient of variation is commonly expressed as:
[
CV=\frac{s}{\bar{x}}\times100%
]
It compares relative variability across measures with different units or substantially different means.
The CV is normally appropriate only when:
- The variable has a meaningful zero
- Ratio comparisons are meaningful
- The mean is positive and not close to zero
It is generally inappropriate for temperatures measured in Celsius or Fahrenheit, variables that can be negative, or datasets whose mean is near zero.
Choosing a measure of spread
| Distribution or purpose | Recommended spread measure |
|---|---|
| Approximately symmetrical quantitative data | Standard deviation |
| Strongly skewed data | Interquartile range |
| Robust analysis with outliers | IQR or median absolute deviation |
| Quick description of observed extremes | Range, with caution |
| Comparison of relative variability | Coefficient of variation, when valid |
| Formal statistical modelling | Variance or standard deviation, as required |
Distribution shape
A dataset cannot be understood fully from its center and spread alone. Shape can reveal asymmetry, multiple groups, floor or ceiling effects, and unusual tails.
Skewness
Skewness describes asymmetry.
- Positive or right skew: the longer tail extends toward higher values.
- Negative or left skew: the longer tail extends toward lower values.
- Approximate symmetry: the two sides have broadly similar form.
Income, waiting time, and hospital-stay duration are often right-skewed because most values are moderate but a small number are very high.
A skewness coefficient can support interpretation, but it should be considered together with a histogram, density plot, box plot, or quantile plot. A numerical threshold should not replace substantive judgment.
Kurtosis
Kurtosis is related primarily to the weight of a distribution’s tails and its propensity to produce extreme observations relative to a reference distribution. It is often described imprecisely as “peakedness.”
Software may report either:
- Pearson kurtosis, for which a normal distribution has a value of 3, or
- Excess kurtosis, for which a normal distribution has a value of 0
Researchers should identify which definition their software uses before interpreting the number.
Modality
Modality describes the number and prominence of peaks.
A bimodal pattern may indicate:
- Two underlying populations
- Different measurement conditions
- A treatment and control mixture
- Seasonal or temporal regimes
- A coding or data-merging problem
Reporting one mean for a clearly bimodal distribution can conceal the most important feature of the data.
Univariate, bivariate and multivariate descriptive analysis
Univariate descriptive statistics
Univariate analysis describes one variable at a time.
Examples:
- Mean age
- Median income
- Percentage of participants in each region
- Distribution of test scores
Bivariate descriptive statistics
Bivariate analysis describes two variables jointly.
Examples:
- Mean test score by teaching method
- Cross-tabulation of gender and programme choice
- Scatter plot of study time and examination score
- Correlation between two quantitative variables
A correlation coefficient may be used descriptively to summarize the observed linear association. Statistical significance, confidence intervals, population generalization, and causal interpretation require additional inferential reasoning.
Multivariate descriptive statistics
Multivariate descriptive work considers several variables simultaneously. Examples include:
- Correlation matrices
- Grouped profile tables
- Multidimensional visualizations
- Cluster summaries
- Covariance matrices
As dimensionality increases, a small number of summaries can hide important patterns. Visualization and clear variable definitions become especially important.
Which descriptive statistics should be used for each variable type?
| Measurement level | Examples | Appropriate summaries | Useful graphs |
|---|---|---|---|
| Nominal | Country, diagnosis, transport type | Frequency, percentage, proportion, mode | Bar chart, dot plot |
| Binary | Yes/no, event/no event | Count, proportion, percentage, risk | Bar chart, icon array |
| Ordinal | Satisfaction level, class rank | Frequency, cumulative percentage, median, mode, quartiles | Ordered bar chart, dot plot |
| Interval | Temperature in °C, some scale scores | Mean, median, SD, variance, range | Histogram, box plot, density plot |
| Ratio | Age, income, distance, duration | Mean or median; SD, IQR, range; geometric mean or CV when suitable | Histogram, box plot, scatter plot |
| Count | Number of visits, errors, publications | Frequency, mean or median, variance, IQR, proportion of zeros | Bar plot, histogram, dot plot |
| Time series | Daily sales, monthly rainfall | Center and spread by period, change, moving summaries | Line chart |
| Two categorical variables | Treatment group and outcome category | Cross-tabulation, row/column percentages | Grouped or stacked bar chart |
| Two quantitative variables | Height and weight | Covariance, correlation, grouped summaries | Scatter plot |
A note on Likert items
A single Likert-type item produces ordered categories. Frequencies, percentages, the median, and the mode are usually straightforward summaries.
Researchers often combine several items into a scale score and analyze the composite approximately as an interval variable. That practice requires justification based on the scale’s construction, distribution, reliability, disciplinary conventions, and intended interpretation. The decision should not be made merely because software permits a mean.
Worked descriptive-statistics example
Suppose a researcher records weekly study hours for seven students:
1, 3, 3, 5, 6, 7, 10
Step 1: Count the observations
[
n=7
]
Step 2: Calculate the mean
[
\bar{x}=\frac{1+3+3+5+6+7+10}{7}
=\frac{35}{7}=5
]
Step 3: Find the median
The values are already ordered. The fourth value is the middle observation:
[
\text{Median}=5
]
Step 4: Find the mode
The value 3 appears twice; all others appear once:
[
\text{Mode}=3
]
Step 5: Calculate the range
[
10-1=9
]
Step 6: Calculate quartiles and IQR
Using the median-of-halves convention and excluding the overall median:
- Lower half: 1, 3, 3; therefore (Q_1=3)
- Upper half: 6, 7, 10; therefore (Q_3=7)
[
IQR=7-3=4
]
Step 7: Calculate the sample variance
The mean is 5. The squared deviations are:
| Value | Deviation from 5 | Squared deviation |
|---|---|---|
| 1 | -4 | 16 |
| 3 | -2 | 4 |
| 3 | -2 | 4 |
| 5 | 0 | 0 |
| 6 | 1 | 1 |
| 7 | 2 | 4 |
| 10 | 5 | 25 |
| Total | 54 |
[
s^2=\frac{54}{7-1}=9
]
Step 8: Calculate the sample standard deviation
[
s=\sqrt{9}=3
]
Summary
| Statistic | Result |
|---|---|
| Valid (n) | 7 |
| Mean | 5 |
| Median | 5 |
| Mode | 3 |
| Minimum | 1 |
| First quartile | 3 |
| Third quartile | 7 |
| Maximum | 10 |
| Range | 9 |
| IQR | 4 |
| Sample variance | 9 |
| Sample SD | 3 |
Interpretation
The students studied for an average of 5 hours per week, with a median of 5 hours. The sample standard deviation was 3 hours, indicating substantial variation relative to the mean. The middle 50% of observations extended from 3 to 7 hours.
This interpretation describes the seven observed students. It does not automatically establish the average study time of all students at the institution.
How to conduct descriptive statistical analysis
1. Define the unit of analysis
Determine what each row represents:
- One person
- One household
- One school
- One transaction
- One measurement occasion
- One repeated observation
A dataset with repeated measurements cannot always be summarized as though every row were an independent participant.
2. Classify each variable
Identify whether each variable is:
- Nominal
- Ordinal
- Interval
- Ratio
- Binary
- Count
- Date or time
- Free text requiring coding
This determines which calculations have a defensible interpretation.
3. Check data quality
Inspect:
- Valid sample size
- Missing values
- Duplicate records
- Impossible values
- Inconsistent units
- Coding errors
- Out-of-range values
- Unusual observations
- Unexpected categories
A negative age or a category coded sometimes as “Female,” sometimes as “F,” and sometimes as “2” can invalidate an automated summary.
4. Examine the distribution
Use numerical and visual methods together.
For a quantitative variable, inspect:
- Histogram
- Box plot
- Dot plot
- Quantile plot
- Minimum and maximum
- Median and mean
- IQR and SD
- Possible floor or ceiling effects
5. Select complementary summaries
Do not report a mean alone. Pair:
- Mean with SD
- Median with IQR
- Count with percentage
- Correlation with a scatter plot
- Category totals with the valid denominator
6. Calculate overall and relevant group summaries
A total mean can conceal group differences.
Depending on the research design, calculate summaries by:
- Treatment group
- Sex or gender category
- Study site
- Time point
- Region
- Cohort
- Exposure status
Group definitions should be determined by the research question rather than chosen after inspecting results solely to produce an interesting pattern.
7. Interpret within the observed data
State what the summaries show without implying:
- Causation
- Statistical significance
- Population generalization
- Clinical importance
- Prediction accuracy
Those conclusions require additional design and analysis.
8. Document the procedure
Record:
- Software and version
- Functions or menu procedures
- Missing-value rules
- Weighting
- Quantile method where relevant
- Transformations
- Rounding
- Exclusion decisions
- Code or syntax
Reproducible analysis allows another researcher to understand how the reported table was produced.
Descriptive statistics versus inferential statistics
Descriptive statistics summarize observed data. Inferential statistics use sample information, probability models, and study-design assumptions to estimate, test, or predict characteristics beyond the observed sample.
| Feature | Descriptive statistics | Inferential statistics |
|---|---|---|
| Main purpose | Describe collected data | Draw conclusions beyond collected data |
| Typical outputs | Mean, median, percentage, SD, IQR, chart | Confidence interval, p-value, model estimate, prediction |
| Probability model required? | Not always | Generally yes |
| Example | Mean score of 80 observed students | Estimate of the population mean |
| Hypothesis test? | No | Often |
| Generalization? | Limited to observed data unless the dataset is the target population | May support population inference under appropriate assumptions |
| Causation? | Cannot establish it | Still requires a suitable design; a p-value alone does not establish causation |
The two branches are complementary. Researchers normally describe their data before conducting inferential analysis.
Tables and graphs for descriptive statistics
Frequency table
Best for categorical and discrete variables. Include:
- Category labels
- Counts
- Percentages
- Missing category when relevant
- Total
Bar chart
Best for categorical variables. Bars should usually be separated because categories are distinct.
Histogram
Best for showing the distribution of a quantitative variable. Adjacent bars represent intervals on a continuous or ordered numerical scale.
Histogram appearance depends on bin width. A poor bin choice can hide or exaggerate patterns.
Box plot
A box plot summarizes the median, quartiles, and potential outlying observations. It is useful for comparing several groups, but it does not show every aspect of distribution shape.
Dot plot or strip plot
Useful for small or moderate datasets because individual observations remain visible.
Scatter plot
Best for examining the relationship between two quantitative variables. It can reveal:
- Direction
- Form
- Strength
- Clusters
- Outliers
- Nonlinear relationships
- Changing variability
A correlation coefficient should not replace the scatter plot.
Line chart
Appropriate when observations follow a meaningful sequence, particularly time. A line should not connect unrelated nominal categories.
How to report descriptive statistics in research
A clear report identifies:
- The variable
- The number of valid observations
- The chosen center and spread
- Units
- Group definitions
- Missing values
- Weighting, when used
- Relevant range or distribution information
Mean and standard deviation example
Participants had a mean age of 34.6 years (SD = 8.2; range = 19–58; valid n = 214).
Median and IQR example
Hospital stay had a median duration of 4 days (IQR = 2–8; range = 1–41; valid n = 187).
Categorical example
Of the 220 participants, 126 (57.3%) selected the online option, 79 (35.9%) selected the in-person option, and 15 (6.8%) did not answer the question.
Grouped example
Mean examination scores were 76.4 (SD = 8.1; n = 54) in the intervention group and 72.8 (SD = 9.6; n = 51) in the comparison group.
This is a descriptive comparison. A statement that the intervention “improved” scores requires appropriate design and inferential evidence.
Example descriptive-statistics table
| Variable | Total | Group A | Group B |
|---|---|---|---|
| Valid participants, n | 205 | 102 | 103 |
| Age, mean (SD), years | 34.8 (8.6) | 34.2 (8.1) | 35.4 (9.0) |
| Study time, median (IQR), hours | 6 (4–9) | 7 (5–10) | 5 (3–8) |
| Completed programme, n (%) | 168 (82.0) | 89 (87.3) | 79 (76.7) |
| Missing study-time value, n (%) | 7 (3.4) | 2 (2.0) | 5 (4.9) |
Add a table note explaining:
- What percentages use as the denominator
- Any weighting
- Abbreviations
- Whether SD is sample SD
- How missing values were handled
Reporting guidelines such as STROBE request transparent participant characteristics and missing-data information for observational studies, while SAMPL provides broader statistical-reporting guidance (Lang & Altman, 2013; von Elm et al., 2007).
Descriptive statistics in modern research
Descriptive statistics remain essential even when research uses machine learning, high-dimensional data, automated sensors, electronic records, or large administrative databases.
Participant and sample description
Researchers use descriptive tables to show who was included, what measurements were available, and how groups differed at baseline or observation.
Data-quality assessment
Summary statistics can reveal:
- Values outside possible limits
- Unit-conversion errors
- Unexpected zeros
- Duplicated measurements
- Sudden changes in sensor readings
- Imputation results that violate logical constraints
Exploratory data analysis
Exploratory data analysis combines summaries with graphical inspection. A mean and SD may reduce data to two numbers, but a histogram or scatter plot can reveal clusters, nonlinearity, outliers, and changing variability that the numbers conceal.
Model preparation
Before regression, classification, factor analysis, or machine learning, descriptive analysis helps identify:
- Highly skewed predictors
- Rare categories
- Near-zero variance
- Class imbalance
- Missingness patterns
- Strong correlations
- Data leakage risks
- Scale differences
Monitoring and reproducibility
In collaborative and computational research, analysts may generate descriptive summaries automatically after each data update. Automated reports are useful only when the underlying definitions, validation rules, and software versions are controlled.
Digital tools for descriptive statistics
Microsoft Excel
Excel’s Analysis ToolPak can produce a table of univariate statistics. Individual worksheet functions include:
AVERAGEMEDIANMODE.SNGLSTDEV.SSTDEV.PVAR.SVAR.PQUARTILE.INCQUARTILE.EXCCOUNTCOUNTIF
Researchers must choose between sample and population functions deliberately. STDEV.S and STDEV.P are not interchangeable.
IBM SPSS Statistics
The DESCRIPTIVES procedure efficiently calculates statistics such as mean, SD, minimum, maximum, variance, skewness, and kurtosis for numeric variables. Median and mode are obtained through procedures such as FREQUENCIES or EXAMINE, not through the basic DESCRIPTIVES command.
Useful SPSS routes include:
- Analyze → Descriptive Statistics → Frequencies
- Analyze → Descriptive Statistics → Descriptives
- Analyze → Descriptive Statistics → Explore
- Analyze → Tables for grouped publication tables, where licensed
Saving syntax is preferable to relying only on menu clicks because syntax documents the exact procedure.
R
Base R functions include:
mean(x, na.rm = TRUE)
median(x, na.rm = TRUE)
sd(x, na.rm = TRUE)
var(x, na.rm = TRUE)
range(x, na.rm = TRUE)
quantile(x, na.rm = TRUE)
summary(x)
table(x, useNA = "ifany")
prop.table(table(x))
The type argument of quantile() controls the sample-quantile algorithm. Scripts should make non-default choices explicit.
Python
Common pandas and NumPy operations include:
series.mean()
series.median()
series.mode()
series.std()
series.var()
series.quantile([0.25, 0.50, 0.75])
series.value_counts(dropna=False)
dataframe.describe(include="all")
Check whether functions use sample or population denominators and how they handle missing values. For example, a default standard-deviation calculation may use a different degrees-of-freedom setting from another library.
Other useful tools
- JASP and jamovi for graphical interfaces and reproducible output
- Stata, SAS, and Minitab for established statistical workflows
- Google Sheets for accessible basic calculations
- R Markdown, Quarto, and Jupyter notebooks for analysis combined with documentation
How can artificial intelligence assist?
Generative AI can help researchers:
- Draft R, Python, SPSS, or spreadsheet formulas
- Explain unfamiliar output
- Suggest validation checks
- Convert a reporting requirement into a table template
- Comment code
- Produce test data for debugging
- Translate technical explanations into plain language
However, AI-generated analysis should not be accepted without verification. Common risks include:
- Selecting an invalid statistic for the measurement level
- Confusing sample and population SD
- Ignoring missing values
- Inventing variable names or software options
- Misinterpreting skewness or kurtosis
- Treating correlation as causation
- Producing code that runs but answers the wrong question
- Disclosing confidential data to an unapproved service
The researcher remains responsible for the definitions, calculations, interpretation, privacy, and final report. Sensitive or identifiable data should not be uploaded to public AI systems without appropriate authorization and safeguards.
Advantages of descriptive statistics
Descriptive statistics:
- Reduce large datasets to understandable summaries
- Support data checking
- Make group patterns visible
- Help select later analyses
- Communicate results efficiently
- Work with both samples and complete populations
- Can summarize quantitative and categorical variables
- Provide a foundation for visualization and exploratory analysis
- Can be reproduced automatically in code-based workflows
Limitations of descriptive statistics
Descriptive statistics cannot by themselves:
- Establish causation
- Determine whether an observed difference is compatible with random sampling variation
- Generalize from a convenience sample to a population
- Correct a biased or poorly measured dataset
- Recover information lost through excessive aggregation
- Explain why a pattern occurred
- Guarantee that the chosen summary represents every subgroup
- Replace examination of the underlying distribution
A concise summary inevitably discards information. The aim is not to report every available number but to preserve the features needed for accurate understanding.
Common mistakes
Reporting only the mean
A mean without an SD, IQR, range, or graph says little about variability.
Using the mean for category codes
Codes such as 1 = rural, 2 = suburban, and 3 = urban are labels. Their numerical mean has no substantive meaning.
Treating missing values as zeros
A missing income is not an income of zero. The coding and denominator must distinguish the two.
Removing outliers automatically
An unusual value may be an error, a legitimate rare case, or evidence of an important subgroup. Investigate its source and perform a sensitivity analysis when appropriate.
Confusing SD and standard error
The SD describes variability among observations. The standard error describes uncertainty in an estimated statistic under a sampling model. They answer different questions.
Describing every symmetrical distribution as normal
Symmetry is one feature. Many non-normal distributions are symmetrical.
Treating a percentage as self-explanatory
Always identify the numerator, denominator, group, period, and missing-data rule.
Over-rounding or reporting excessive precision
The reported precision should reflect measurement quality and practical interpretation. A mean age of 34.638417 years is rarely more informative than 34.6 years.
Assuming software outputs are identical
Packages may differ in default denominators, quantile algorithms, missing-data handling, category inclusion, and weighting.
Making inferential claims from descriptive differences
A higher observed mean does not by itself prove an effect, a population difference, or a causal relationship.
Practical reporting checklist
Before publishing descriptive statistics, ask:
- Is the unit of analysis clear?
- Is each variable classified correctly?
- Are valid and missing observations reported?
- Does the measure of center match the distribution and measurement level?
- Is center paired with spread?
- Are percentages accompanied by denominators?
- Are relevant groups summarized separately?
- Have outliers and impossible values been checked?
- Is an appropriate graph included?
- Are software, weighting, transformations, and non-default settings documented?
- Does the interpretation remain within the observed data?
- Could another researcher reproduce the table?
Descriptive-statistics reporting template
Data were summarized using [software and version]. Categorical variables were reported as frequencies and percentages, with denominators based on [all enrolled participants/valid responses/weighted estimates]. Approximately symmetrical quantitative variables were summarized using means and standard deviations. Skewed or outlier-prone variables were summarized using medians and interquartile ranges. Missing observations were reported for each variable. Descriptive summaries were calculated overall and by [predefined groups]. [Weights, transformations, or non-default quantile methods] were applied as specified.
Adapt the template to the research design and journal guidelines rather than using it unchanged.
Conclusion
Descriptive statistics transform raw observations into an understandable account of what a dataset contains. A sound descriptive analysis does more than calculate an average: it identifies valid sample sizes, matches statistics to variable types, combines center with spread, examines distribution shape, reports missingness, and presents appropriate tables or graphs.
The most useful summary is not the one with the most numbers. It is the one that preserves the characteristics readers need to interpret the data accurately.
