Descriptive statistics reduce a long list of numbers into a handful of values that a reader can absorb in a few seconds. If you measured the resting heart rate of 60 people, nobody wants to read 60 numbers - they want to know the typical value, how much people varied around it, and whether the distribution was lopsided. That is exactly what this page computes.

The three families of descriptive statistics

Measures of central tendency locate the middle of the data. The arithmetic mean is the balance point of the distribution; the median is the value that splits the sorted data in half; the mode is the most frequent value. When the distribution is symmetric these three sit close together. When it is skewed they separate, and the direction of that separation tells you which way the tail runs.

Measures of dispersion describe how tightly the observations cluster. The range is the crudest (max − min) and is dominated by the two most extreme points. The variance is the average squared deviation from the mean, and the standard deviation is its square root, expressed back in the original units. The interquartile range covers the middle 50% and ignores the tails entirely, which makes it the robust choice. The coefficient of variation expresses the standard deviation as a percentage of the mean, letting you compare variability across variables with different units.

Measures of shape describe the silhouette of the distribution. Skewness quantifies asymmetry: a positive value means a long right tail, a negative value a long left tail. Kurtosis quantifies tail weight relative to a normal curve: positive excess kurtosis means heavier tails and more extreme values than a normal distribution would produce.

Descriptive versus inferential

Descriptive statistics stop at your sample. Inferential statistics - t-tests, ANOVA, regression - use that sample to make claims about a wider population. The bridge between them is the standard error and the confidence interval, both of which this calculator reports, because they translate a descriptive spread into a statement of how precisely the mean has been estimated.

Options

4.1 Reading the centre: mean, median and mode together

Never read the mean alone. Read it next to the median. If the two are within roughly one-tenth of a standard deviation of each other, the distribution is close to symmetric and the mean is a fair summary of the typical case. If the mean sits clearly above the median, a minority of large values is pulling the average upward - the classic signature of right-skewed data such as income, waiting times, seed counts or parasite loads. If the mean sits below the median, a few unusually small values are dragging it down, which happens with ceiling-limited scores such as exam marks out of 100 where most students score highly and a handful fail badly.

The mode adds a third piece of information that the other two hide: whether the data have one peak or several. A dataset with two distinct modes is usually a mixture of two sub-populations that have been pooled by mistake - two measurement instruments, two sites, two age classes. If the histogram in the results section shows two humps, stop and check whether a grouping variable was omitted before reporting a single mean at all.

4.2 Reading the spread: SD, variance, IQR and CV

The standard deviation is the average distance of an observation from the mean, in the original units. For approximately normal data the empirical rule applies: about 68% of observations lie within one SD of the mean, about 95% within two SDs, and about 99.7% within three. You can sanity-check your own data against this immediately - if far more than 5% of your values sit beyond two SDs, the distribution has heavier tails than a normal curve and the SD is understating how far apart the observations really are.

The variance is the SD squared. It is the quantity that statistical models actually decompose (ANOVA partitions variance; R² is a ratio of variances), but because its units are squared it is almost never reported on its own in a results paragraph. Report the SD for readers and keep the variance for the model.

The interquartile range is the robust counterpart. Because it is built from the 25th and 75th percentiles it is completely unaffected by how extreme the most extreme values are. When your data contain genuine outliers that you do not wish to delete, the median and IQR describe the bulk of the data honestly while the mean and SD do not.

The coefficient of variation converts the SD into a percentage of the mean, which makes variability comparable across measurements on different scales. A CV below about 10% is generally considered low variability, 10–30% moderate, and above 30% high. Two caveats: the CV is meaningless when the mean is near zero (the ratio explodes), and it should not be used for interval scales such as temperature in Celsius where the zero point is arbitrary.

4.3 Reading the shape: skewness and kurtosis

Skewness measures asymmetry. Values between −0.5 and +0.5 indicate an approximately symmetric distribution. Between ±0.5 and ±1 the distribution is moderately skewed and the mean is beginning to mislead. Beyond ±1 the skew is substantial and you should either report the median instead of the mean, transform the variable (log or square root for right skew), or choose a non-parametric test. The sign tells you the direction: positive means the long tail points to the right, toward large values.

Kurtosis measures tail weight, not "peakedness" in any useful sense. This calculator reports excess kurtosis, so a perfect normal distribution scores 0. Positive excess kurtosis (leptokurtic) means more extreme values than a normal distribution would produce - outliers are expected, not anomalous, and methods that assume normality will have inflated error rates. Negative excess kurtosis (platykurtic) means the data are unusually evenly spread with thin tails, which is common for uniform-like measurements. Values within ±2 are usually treated as acceptable for parametric procedures on moderate samples; beyond that, take the deviation seriously.

Both statistics are unstable in small samples. With n below about 30, a skewness of 0.8 could easily arise from sampling noise in genuinely symmetric data. Read the histogram and Q–Q-style plot alongside the numbers rather than trusting a single coefficient.

4.4 Reading precision: SEM and the confidence interval

This is the distinction that most often goes wrong in student writing. The standard deviation answers "how much do individual observations differ from each other?" The standard error of the mean answers "how precisely have I pinned down the population mean?" They are related by SEM = SD ÷ √n, so the SEM always shrinks as you collect more data while the SD does not - a larger sample does not make people less variable, it only makes your estimate of the average better.

The confidence interval turns the SEM into a range. A 95% confidence interval means that if you repeated the whole study many times, about 95% of the intervals so constructed would contain the true population mean. It does not mean there is a 95% probability that the true mean lies inside this particular interval - the true mean is a fixed number, not a random one. In practice, use the width of the interval as your headline: a narrow interval means a precise estimate, a wide one means you cannot rule out very different population values and should be cautious about strong claims.

When comparing two clusters, the interval overlap is a useful but imperfect signal. Non-overlapping 95% intervals nearly always correspond to a significant difference at α = 0.05. Overlapping intervals, however, do not prove the absence of a difference - modest overlap is still compatible with p < 0.05. Run a formal t-test or Mann–Whitney test if the comparison matters.

4.5 Reading the outlier flags

The calculator flags any value below Q1 − 1.5×IQR or above Q3 + 1.5×IQR as a mild outlier, and beyond 3×IQR as an extreme outlier. A flag is a prompt to investigate, never an instruction to delete. Ask three questions in order: is it a recording or transcription error (a heart rate of 720 instead of 72)? Is it a measurement failure (an instrument that saturated)? Or is it a real, biologically or socially meaningful extreme value? Only the first two justify removal, and any removal must be reported in the methods along with how many points were dropped and why. If you are unsure, run the analysis both with and without the point and report whether the conclusion changes - that sensitivity check is far more persuasive than a silent deletion.

4.6 Reading a comparison between clusters

When you load more than one cluster, compare four things in this order. First the centres - how large is the raw difference in means or medians, and is that difference practically meaningful in the units of your field, regardless of statistics? Second the spreads - if one group's SD is more than roughly twice the other's, the equal-variance assumption behind a standard t-test or ANOVA is in doubt and you should use Welch's correction. Third the shapes - groups that differ in skew are not simply shifted versions of each other, and a difference in means may be masking a difference in distribution. Fourth the sample sizes - a group with n = 8 contributes far less reliable estimates than one with n = 80, and unequal n makes the analysis more sensitive to unequal variances.

Finally, translate the difference into a standardised effect size before writing your discussion. Divide the difference in means by the pooled standard deviation to obtain Cohen's d; roughly, 0.2 is small, 0.5 medium and 0.8 large. A statistically significant difference with d = 0.15 is real but probably unimportant; a non-significant difference with d = 0.7 in a small sample is a signal that the study was underpowered, not that the effect is absent.

This section turns the numbers in the results panel above into finished sentences, tables and figure captions. Work through it in order: decide what to report, choose the reporting pair, write the sentence, build the table, caption the figure, then run the final checklist. Every template below is ready to copy and adapt.

5.1 The grammar of a descriptive results paragraph

Descriptive results follow a fixed order that reviewers expect. Deviating from it makes a paragraph feel disorganised even when every number is correct.

1. What was measured, and in whom. Name the variable and the unit. "Resting heart rate (bpm) was recorded in 40 healthy adults."
2. How many cases contributed. Give the analysed n, not the recruited n, and account for the difference if they differ.
3. The centre. Mean or median, chosen by the rule in 5.3.
4. The spread. SD paired with the mean, IQR paired with the median. Never mix the pairs.
5. The range or interval if precision or extremes matter to the argument.
6. A distributional note only when it changes what you did next, for example a skew that forced a transformation.

Nothing in this paragraph should interpret, compare or speculate. "Scores were disappointingly low" belongs in the discussion. "Mean score was 42.1 (SD 9.8)" belongs here. Keep the tense past and the voice consistent with the rest of the manuscript.

5.2 Choosing precision and rounding

Report one decimal place more than the raw measurement, and no more. If heart rate was recorded to the nearest whole beat, the mean is reported to one decimal and the SD to one or two. Six decimal places on a variable measured with a ruler tells the reader you copied output without thinking.

Two further rules. Keep the same number of decimals for a statistic throughout the whole manuscript, including tables and figures. And round only at the reporting stage, never in the intermediate calculations, because rounding early propagates error into the SD and the confidence interval.

5.3 Decision rule: which centre and spread to report

Read your skewness value from the results panel, then apply this table. It resolves the majority of reporting questions in one step.

5.4 Sentence templates you can copy

Replace the italic placeholders with your own values. All templates follow APA 7 conventions: statistical symbols italic, no space before a percent sign, a leading zero on values that can exceed 1.

5.5 A full worked results paragraph

Here is what a complete opening paragraph looks like when the templates above are stitched together. Note that it reports, and does not interpret.

5.6 Building the descriptive table

Almost every journal expects one compact descriptive table before any inferential result. Use one row per group, put units in the column header rather than in every cell, align decimals, and define every abbreviation in a table note.

Note. n = number of participants with complete data; SD = standard deviation; CI = confidence interval; IQR = interquartile range. Variances computed with the sample divisor (n - 1).

Table rules that reviewers check

• The table must be understandable without reading the text, and the text must not simply repeat every cell.
• Number tables in the order they are first mentioned, and mention each one explicitly ("Table 1 summarises...").
• Use the same decimal precision down each column.
• Do not use vertical rules; APA and most biomedical styles use horizontal rules only.
• Put the unit once, in the header. Writing "72.4 bpm" in thirty cells is noise.
• If a cell is empty, say why in the note rather than leaving a blank the reader must guess at.

5.7 Writing figure captions

The single most common reviewer complaint about descriptive figures is unlabelled error bars. Every caption must state what the bars represent and the n behind them.

5.8 SD, SEM or confidence interval in a figure?

These three bars look similar and mean entirely different things. Choose deliberately, then say which you chose.

Because SEM bars are the smallest of the three, choosing them can make a difference look more convincing than the data justify. Reviewers and editors are alert to this. State the bar type in the caption, and if you are unsure, plot the confidence interval, which is directly interpretable, together with the raw data points.

5.9 Style guide differences at a glance

Check the target journal's author guidelines before submission; house style overrides all of the above. The universal rule is internal consistency: pick one form and use it in the text, tables and captions without variation.

5.10 Where descriptive statistics go in a thesis

In a typical thesis, descriptive results open Chapter 4 and set up everything that follows. A reliable structure is:

1. Sample description. Recruited n, analysed n, reasons for any loss, and demographic characteristics as counts and percentages.
2. Data screening. Missing data handling, outlier checks with the rule you applied, and any transformations, each with a one-sentence justification.
3. Distributional checks. Skewness and kurtosis for every continuous outcome, and the resulting decision about parametric versus non-parametric analysis. This is the paragraph that licenses the tests in the rest of the chapter.
4. Descriptive table. One table covering every outcome variable, split by group if relevant.
5. Descriptive figures. Histograms or box plots for the primary outcomes only. Move the rest to an appendix.
6. Bridge sentence. One line stating that assumptions were adequately met and naming the test that follows.

5.11 Common mistakes to avoid

5.12 Final pre-submission checklist

• Every descriptive statistic is accompanied by its n.
• Centre and spread are correctly paired throughout: mean with SD, median with IQR.
• The choice of centre is justified by the skewness value, and that value is reported when it drove the decision.
• Decimal places are consistent within each statistic across text, tables and figures.
• Units appear in table headers and figure axis labels, not repeated in every cell.
• Every figure caption states what the error bars represent and the n behind them.
• Missing data, exclusions and any transformations are documented with counts and reasons.
• The outlier rule is named, and any removals are reported with a sensitivity check.
• The software, the variance divisor and the quartile method are stated in the analysis paragraph.
• Notation matches the target journal's style guide and is internally consistent.
• The results section reports and does not interpret.
• Numbers in the text match the numbers in the tables and figures exactly.

1. Choose an input method. Paste numbers on the first tab, upload a CSV or Excel file on the second, or load a ready-made example on the third.
2. Type or paste your values. Commas are the default separator, so 52, 48, 55, 61, 47 works exactly as shown in the placeholder. Spaces, tabs, semicolons and line breaks are accepted too.
3. Rename the cluster. Click the name box above each textarea and type a meaningful label such as “Treatment” or “Site A”. The name appears in the results table, the charts and the exported file.
4. Add more clusters with the ＋ Add Cluster button to compare two, three or more groups side by side. Remove one with its ✕ button.
5. Uploading a CSV? After the file loads, the columns appear as clickable buttons. Click every column that should become its own cluster - each selected column is loaded as a separate group, using the column header as the group name. Click again to deselect.
6. Set the options. Choose sample or population (sample is correct for almost all research), the confidence level, decimal places and the quartile method.
7. Press Calculate. Summary cards, a full statistics table, outlier flags, an automatic interpretation and four charts appear immediately.
8. Export. Download the results as a text or CSV file, or copy them straight into your manuscript.

8.1 Interpreting skewness

8.2 Interpreting excess kurtosis

8.3 Interpreting the coefficient of variation

8.4 Critical t-values for the confidence interval

8.5 Which measure of centre to use

Eight complete worked examples, each with the data entered, the figure the calculator produces, the full statistics, how to read them, and a ready-to-paste sentence for a results section. Every number below was computed with this tool using the sample divisor (n − 1) and linear-interpolation quartiles.

1. Define the unit of observation before you measure anything - one row must equal one independent case, not one repeated reading from the same case.
2. Specify the measurement scale and precision (nearest gram, nearest whole percent) and keep it constant across all groups and all observers.
3. Record raw values, never pre-rounded or pre-averaged ones. Rounding at collection destroys information that the SD and skewness depend on.
4. Use one column per group in your spreadsheet if you plan to upload the file here - each selected column becomes its own cluster.
5. Leave missing values genuinely blank. Never code a missing value as 0, −99 or 999; those codes silently corrupt every statistic on this page.
6. Log any instrument recalibration, observer change or protocol change with the date, so that a bimodal distribution can be traced back to its cause.
7. Aim for at least n = 30 per group if you intend to interpret skewness and kurtosis; below that they are dominated by sampling noise.
8. Keep the original data file untouched and do all cleaning in a copy, documenting each change.

Use it when…

• You need a complete numerical summary of one or more continuous variables.
• You are writing the descriptive paragraph or first table of a results section.
• You want to check normality, skew and outliers before choosing a statistical test.
• You need to compare the centre and spread of several groups quickly.
• You are teaching or learning what mean, SD, IQR, skewness and kurtosis actually do.

Do not use it when…

• Your variable is nominal (blood group, species name) - report counts and percentages instead.
• You need a formal hypothesis test - use a t-test, ANOVA, Mann–Whitney or chi-square tool.
• Your observations are not independent (repeated measures on the same subject) without accounting for that structure.
• Your data are censored or truncated (survival times with dropouts) - descriptive means are biased there.
• You are summarising growth rates or ratios, where the geometric mean is the appropriate centre.

• Measurement scale. Mean, SD, skewness and kurtosis require interval or ratio data. Applying them to ordinal codes produces numbers that look valid but mean nothing.
• Independence. Each value must come from a separate, independent case. Repeated measurements on the same subject inflate n and shrink the SEM artificially.
• Complete cases. Non-numeric and blank entries are silently ignored; the reported n is the number of usable values, which may be smaller than your row count.
• The confidence interval assumes approximate normality of the sampling distribution of the mean. With n ≥ 30 the Central Limit Theorem usually makes this safe even for skewed data; with small, heavily skewed samples the interval is unreliable.
• Descriptive only. Nothing on this page tests a hypothesis. Overlapping confidence intervals are not proof of no difference.
• Quartile definitions differ. Excel, SPSS, R and Minitab use several conventions; Q1, Q3 and therefore the IQR and outlier fences can differ slightly between packages. State the method you used.
• Shape statistics need data. Skewness and kurtosis are noisy below n ≈ 30 and should be read alongside a histogram, not in isolation.
• Outlier flags are heuristic. Tukey's 1.5×IQR rule flags roughly 0.7% of values even in perfectly normal data - a flag is not evidence of an error.

Descriptive statistics are the foundation on which every later analysis rests, and they are the part of a results section that readers actually read closely. A well-chosen summary tells a reader, in one line, where the typical case sits, how much the cases differ from one another, and whether the distribution behaves the way the subsequent statistical tests assume. A poorly chosen summary - a mean for strongly skewed income data, an SEM bar labelled as an SD, a table with no sample sizes - quietly undermines everything that follows it, and reviewers are quick to notice.

The practical workflow this page is built around is straightforward. Start by computing the full set of statistics rather than only the two you plan to report, because the ones you did not intend to use are exactly the ones that reveal problems. Compare the mean against the median to detect skew. Read the skewness and kurtosis coefficients as confirmation of what the histogram already shows, not as a substitute for looking at it. Check the outlier flags and trace each one back to the original data sheet before deciding anything. Only then choose your reporting pair: mean with SD for symmetric data, median with IQR when the distribution is skewed or outlier-prone.

The second habit worth building is separating variability from precision. The standard deviation belongs to your participants and does not change as you collect more of them; the standard error and the confidence interval belong to your estimate and shrink as the sample grows. Confusing the two produces figures that overstate how confident the data allow you to be. When the estimate of the mean is itself the finding of the study, report the confidence interval - it communicates both the value and its uncertainty in one expression and lets readers judge the result for themselves.

When several groups are involved, resist the urge to jump straight to a p-value. Compare centres, spreads, shapes and sample sizes first. Unequal spreads point you toward Welch's correction; different shapes warn that a difference in means is not the whole story; small and unequal group sizes tell you how much weight the eventual test can bear. Add a standardised effect size such as Cohen's d, because a difference can be statistically detectable and practically trivial, or practically important and statistically undetectable in a small study - and only the effect size distinguishes the two cases.

Finally, document everything. State the number of observations, the software or method used for quartiles, whether the sample or population divisor was applied, how missing values were handled, and whether any observations were excluded and why. Descriptive statistics are simple to compute and easy to compute inconsistently; a reader who knows exactly what you did can reproduce your table and trust your conclusions. Run your numbers through the calculator above, copy the interpretation it generates, adapt the reporting templates in section 5 to your own variables, and your descriptive results section will be complete, defensible and finished in minutes rather than an afternoon.

APA: StatsUnlock. (2026). Descriptive statistics calculator [Online tool]. https://statsunlock.com/descriptive-statistics-calculator

MLA: “Descriptive Statistics Calculator.” StatsUnlock, 2026, statsunlock.com/descriptive-statistics-calculator.

BibTeX:

@misc{statsunlock_descriptive_2026,
  title  = {Descriptive Statistics Calculator},
  author = {{StatsUnlock}},
  year   = {2026},
  url    = {https://statsunlock.com/descriptive-statistics-calculator}
}

• Standard Deviation - sample and population SD with steps.
• Mean, Median & Mode Calculator - central tendency with worked steps.
• Quartiles & IQR Calculator - five-number summary and box plot.
• Skewness & Kurtosis Calculator - distribution shape diagnostics.
• Independent-Samples t-Test Calculator - test the difference you just described.
• Confidence Interval Calculator - intervals for means and proportions.

Ten questions drawn at random from a pool of 30, with the answer order shuffled each time. Click an answer to see whether it is right and why, then press New random questions for a fresh set.

1. Altman, D. G., & Bland, J. M. (2005). Standard deviations and standard errors. BMJ, 331(7521), 903. https://doi.org/10.1136/bmj.331.7521.903
2. Tukey, J. W. (1977). Exploratory Data Analysis. Addison-Wesley. https://archive.org/details/exploratorydataa0000tuke
3. Hyndman, R. J., & Fan, Y. (1996). Sample quantiles in statistical packages. The American Statistician, 50(4), 361–365. https://doi.org/10.1080/00031305.1996.10473566
4. Joanes, D. N., & Gill, C. A. (1998). Comparing measures of sample skewness and kurtosis. Journal of the Royal Statistical Society: Series D, 47(1), 183–189. https://doi.org/10.1111/1467-9884.00122
5. Cumming, G., Fidler, F., & Vaux, D. L. (2007). Error bars in experimental biology. Journal of Cell Biology, 177(1), 7–11. https://doi.org/10.1083/jcb.200611141
6. Krzywinski, M., & Altman, N. (2013). Error bars. Nature Methods, 10(10), 921–922. https://doi.org/10.1038/nmeth.2659
7. Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Routledge. https://doi.org/10.4324/9780203771587
8. Bland, J. M., & Altman, D. G. (1996). Measurement error. BMJ, 313(7059), 744. https://doi.org/10.1136/bmj.313.7059.744
9. Wilcox, R. R., & Keselman, H. J. (2003). Modern robust data analysis methods. Psychological Methods, 8(3), 254–274. https://doi.org/10.1037/1082-989X.8.3.254
10. Leys, C., Ley, C., Klein, O., Bernard, P., & Licata, L. (2013). Detecting outliers: Do not use standard deviation around the mean. Journal of Experimental Social Psychology, 49(4), 764–766. https://doi.org/10.1016/j.jesp.2013.03.013
11. Weissgerber, T. L., Milic, N. M., Winham, S. J., & Garovic, V. D. (2015). Beyond bar and line graphs. PLOS Biology, 13(4), e1002128. https://doi.org/10.1371/journal.pbio.1002128
12. Lang, T. A., & Altman, D. G. (2015). Basic statistical reporting for articles: The SAMPL guidelines. International Journal of Nursing Studies, 52(1), 5–9. https://doi.org/10.1016/j.ijnurstu.2014.09.006
13. Ghasemi, A., & Zahediasl, S. (2012). Normality tests for statistical analysis. International Journal of Endocrinology and Metabolism, 10(2), 486–489. https://doi.org/10.5812/ijem.3505
14. Kwak, S. G., & Kim, J. H. (2017). Central limit theorem: The cornerstone of modern statistics. Korean Journal of Anesthesiology, 70(2), 144–156. https://doi.org/10.4097/kjae.2017.70.2.144
15. Sullivan, G. M., & Feinn, R. (2012). Using effect size - or why the p value is not enough. Journal of Graduate Medical Education, 4(3), 279–282. https://doi.org/10.4300/JGME-D-12-00156.1
16. Mishra, P., et al. (2019). Descriptive statistics and normality tests for statistical data. Annals of Cardiac Anaesthesia, 22(1), 67–72. https://doi.org/10.4103/aca.ACA_157_18
17. Marmolejo-Ramos, F., & Tian, T. S. (2010). The shifting boxplot. International Journal of Psychological Research, 3(1), 37–45. https://doi.org/10.21500/20112084.823
18. Hoaglin, D. C., Iglewicz, B., & Tukey, J. W. (1986). Performance of some resistant rules for outlier labeling. Journal of the American Statistical Association, 81(396), 991–999. https://doi.org/10.1080/01621459.1986.10478363
19. Delacre, M., Lakens, D., & Leys, C. (2017). Why psychologists should by default use Welch's t-test. International Review of Social Psychology, 30(1), 92–101. https://doi.org/10.5334/irsp.82
20. American Psychological Association. (2020). Publication Manual of the APA (7th ed.). https://apastyle.apa.org/products/publication-manual-7th-edition