Descriptive Statistics

Descriptive statistics tell you what your response file actually contains, without drawing any inference from it. Each numeric column is summarised by its number of valid observations, its mean, median and mode, its standard deviation, its lowest and highest value, and its lower and upper quartiles. Taken together those eight figures say where a distribution gathers, how widely it spreads and between which bounds it stays. Columns holding text or category labels get a summary of their own: how many valid answers there are, how many distinct values appear, which answer occurs most often and how many times it recurs. Above all of this the summary reports the row count, the column count and the list of column names, so you can see at a glance that the file opened the way you expected. Nothing is tested here, and that is the point: this is the step where you get to know the data before any method that tests something.

• Which columns in the file I just uploaded were read as numbers, and which stayed as text?
• How far apart are the mean and the median of the satisfaction scores, and is the distribution therefore skewed?
• Does the age range of the participants really stay inside the bounds set by the sampling plan?
• On which scale item are the answers widely spread, and on which does everyone cluster at one point?

• As the first step of any analysis, right after the file opens, to confirm that each column was read as the right type.
• When you need the summary table that belongs in the sample description section of a report or thesis.
• To find out whether a scale item shows a floor or ceiling effect, which you see when the quartiles press against one end of the scale.
• When the distance between the mean and the median is the first clue you want about whether a parametric test is reasonable here.
• To catch data entry errors, since a lowest or highest value outside the scale bounds is almost always a coding mistake.

• Numeric columns receive the full summary: scale totals, durations, amounts, age and anything else measured at interval or ratio level.
• No measure of centre is computed for text and category columns; they are described by valid answers, distinct values and the most frequent answer instead.
• A mean can be computed on an ordinal scale, but it presumes equal spacing between the steps, so on a single Likert item the median and the mode are the safer readings.
• The method takes no column selection and summarises the whole dataset, which means technical columns that merely look numeric, such as a respondent id or a timestamp, also appear in the table.

Column types declared correctly

If a column that should be treated as numbers was read as text, no mean or dispersion is computed for it at all and it falls into the category summary instead. A comma as the decimal separator, or a unit written inside the cells, is the usual cause.

Missing value codes declared

Codes such as 99 or -1 keep counting as numbers until you declare them, and one such code is enough to distort both the mean and the highest value visibly.

The mean being a meaningful summary

In markedly skewed distributions, or in columns holding a few very distant values, the mean does not represent the typical participant. For such columns the median together with the quartiles gives a more honest account.

Observations from a single population

The summary is taken over the dataset as a whole. Where two separate subsamples have been merged into one file, the pooled mean can be an intermediate value that describes neither group.

• Column types declared correctly: The Variable tab shows the type and measurement level of every column and lets you change them. The method itself runs no type check and raises no warning, so it is up to you to notice whether a column landed in the part of the table you expected.
• Missing value codes declared: Missing value codes are declared on the Variable tab. The method does not infer on its own that an unusual value might be a missing code.
• The mean being a meaningful summary: No test is run to support this judgement. Having the mean and the median side by side in the summary table is what lets you compare them by eye.
• Observations from a single population: The panel does not check this assumption automatically; the researcher evaluates it.

1. 1Drag your response file onto the upload area; a CSV or XLSX file opens into the editable grid on the Data tab.
2. 2On the Variable tab review the column names, labels, types and measurement levels, and correct any column that should be numeric.
3. 3Declare missing value codes and, if you need them, value labels on the same tab, where a derived column such as a scale total can also be built.
4. 4On the Analysis tab pick descriptive statistics from the method picker, using the search box or the 'Only ones that fit my data' switch.
5. 5There is no parameter form to fill in for this method, because the summary covers the whole dataset, so run it directly.
6. 6In the results start with the bar charts of the means and of the standard deviations, then read the numeric and categorical summary tables in the collapsible sections.
7. 7Download the tables as Excel and the charts as PNG; the computation credits card names the library call and the version behind the figures.

Dataset dimensions

The row count, the column count and the list of column names. Those three pieces of information are the quickest confirmation that the file opened in full.

Measures of central tendency

For every numeric column the mean, followed immediately by the median and the mode. When all three sit close together, the distribution is roughly symmetric.

Measures of dispersion

The standard deviation together with the lowest and highest value. The first says how far answers scatter around the mean, the other two mark the bounds of the range actually used.

Quartiles

Read together, the 25th percentile, the median and the 75th percentile show the band in which the middle half of the participants falls.

Categorical column summary

For each non-numeric column, the number of valid answers, the number of distinct values, the most frequent value and how often it occurs.

Summary charts

One bar chart presents the means of the numeric columns and a second presents their standard deviations.

Standard deviation

This is not an effect size but a measure of spread expressed in the variable's own units. Divided by the mean it becomes a coefficient of variation, which is useful for comparing the spread of columns measured on different scales. Since no hypothesis is tested in this method, no standardized effect measure is produced; a measure of difference between groups requires the relevant comparison method.

Interquartile range

Subtracting the lower quartile from the upper one leaves a width that extreme values do not disturb. The panel does not return this difference as a field of its own, so read the two percentiles from the table and compute it yourself. In skewed distributions it describes the data better than the standard deviation does.

This method returns no confidence interval of any kind. No standard error or interval is produced for the mean, no p value appears and no test is run; every figure returned describes the sample you have. For an interval around the population mean you need to run the one-sample t-test.

The numbers below are a representative example, not data from a real study or a real user.

Research question

In an illustrative subscriber satisfaction study, did the response file open as expected and how are the overall satisfaction scores distributed?

Variables

Overall satisfaction score, range 0-100 (continuous) · Participant age in years (continuous) · City of residence (categorical)

Example result

The file opened with 428 rows and 12 columns. Overall satisfaction has 415 valid values, with a mean of 72.4, a median of 75.0, a mode of 80.0, a standard deviation of 16.8, a minimum of 12, a lower quartile of 62.0, an upper quartile of 85.0 and a maximum of 100. Age has 428 valid values, a mean of 37.2 and a maximum of 69. The city column holds 18 distinct values across 428 valid answers, and its most frequent answer recurs 131 times.

Interpretation

With the median sitting 2.6 points above the mean, the satisfaction distribution carries a mild tail towards the low end, meaning a small group of low scorers is pulling the mean down. The middle half of the participants falls between 62 and 85, a width that shows how different the answers behind a single average satisfaction figure can be. Thirteen answers are absent from the satisfaction column, which is why its observation count falls below the row count of the file. A mode of 80.0 suggests answers were rounded towards multiples of ten; if another value occurs equally often, it will not appear in the table. These figures are illustrative and are not the outcome of a real study.

Research question: What are the mean, spread and quartiles of every measured variable in the survey?

Computation credits: scipy 1.18.0 · statsmodels 0.14.6 · scikit-learn 1.9.0 · numpy 2.5.1 · pandas 3.0.5 · semopy 2.3.11 · 89fc29a · Data seed: 20260914

Overall satisfaction averaged 72.4 (SD = 16.8), with a median of 75.0 and an interquartile range of 62.0 to 85.0 (n = 415).

An example sentence close to APA style; the numbers are representative.

• Because no test is carried out, this method supports no claim about a difference or a relationship; the figures belong to the sample in hand and nothing beyond it.
• Only the first modal value is reported. If two values tie for most frequent, the second never appears in the table and a bimodal distribution goes unnoticed.
• The summary proceeds column by column and says nothing about association between columns, for which you move on to a correlation matrix or a cross-tabulation.
• For category columns only the most frequent value is reported rather than the full frequency distribution, so percentages for every category require the frequency distribution method.
• Since the method accepts no column selection, meaningless numeric columns such as a respondent id are summarised too, and leaving those rows uninterpreted is the reader's job.
• In a weighted sample the summary is computed unweighted, so any correction for representativeness has to be handled separately.

• Frequency DistributionWhy: When categorical or short-scale columns call for counts and percentages across every category rather than just the most frequent value.
• Normality TestsWhy: When departure from normality should be decided by a test rather than by eye.
• Outlier DetectionWhy: When the minimum and maximum look suspicious and extreme observations need to be identified by a criterion.
• Missing Value AnalysisWhy: When observation counts differ across columns and you want to see where the gaps are and how large they are.

• Frequency Distribution
• Normality Tests
• Outlier Detection
• Missing Value Analysis

Can I summarise only selected columns?

There is no column selection parameter in this method: the summary covers the entire dataset and every column enters the table. The practical way to drop columns from a report is to export the table to Excel and filter there, or to remove the unnecessary columns before you upload the file.

How large does the gap between mean and median have to be to matter?

There is no fixed threshold, and the gap has to be read against the standard deviation. If the distance amounts to a small fraction of the standard deviation, the distribution can pass as roughly symmetric; once it reaches something like a quarter of it, the skew is pronounced enough to affect reporting and the median deserves a place in the write-up.

Why does the observation count vary from column to column?

Each column's count comes from the valid values in that column, so blank cells and cells flagged with a missing code lower it. In surveys with skippable questions that is entirely ordinary. To see where the difference comes from, run the missing value analysis.

Can I use this table directly in a thesis or a paper?

You can export it to Excel and move it into your own format, though publication conventions usually expect one row per variable and a consistent number of decimal places. You will be expected to round the figures the panel produces, state the observation count separately for each variable, and note the theoretical range of the scale underneath the table.

• Tukey, J. W. (1977). Exploratory Data Analysis
• Wilkinson, L., & the Task Force on Statistical Inference (1999). Statistical Methods in Psychology Journals: Guidelines and Explanations
• Altman, D. G., & Bland, J. M. (2005). Standard deviations and standard errors
• pandas.DataFrame.describe documentation