Analysis
Statistical Analysis
Exact definitions of the Overview indicators and the descriptive statistics used throughout the workspace, and how the time series is aggregated and smoothed.
Conventions
All statistics are computed in the browser from the observations in the current selection (see Data Explorer). In the formulas below, x₁ … xₙ are the selected values and n is their number.
- Dates are UTC calendar days, independent of the viewer's time zone.
- Sample statistics (variance, standard deviation) use the n − 1 denominator.
- Quantiles use linear interpolation between closest ranks (Hyndman & Fan type 7, the default in R and NumPy).
- Sums use compensated (Kahan) summation, which keeps long series accurate.
- Non-finite values are ignored.
Descriptive only
These are descriptive statistics of the selected data. They are not forecasts, and no indicator in the workspace is a significance test.
Overview indicators
The Overview section shows six indicators for the current selection.
Total observations
The number of observations in the selection, n.
Dataset coverage
Coverage measures completeness, not volume: a day with three observations counts once. Because the category filter and search reduce the selection, they can also reduce coverage. The illustrative datasets have one observation per category for every day, so their coverage is 100% unless a search excludes days.
Mean
Standard deviation
Unique values
The number of distinct values in the selection. For continuous data recorded with one decimal, this shows how many different readings occur; for discrete data, how many different integers occur.
Last updated
The latest observation date in the selection. It is a property of the data, not the time the data was loaded.
Descriptive statistics
The following definitions are used wherever the statistic appears: in the Distributions summary table, in Comparisons and in Reports.
Example: for the values 1, 2, 3, 4, the type-7 quartiles are Q1 = 1.75 and Q3 = 3.25, so IQR = 1.5. For 2, 4, 4, 4, 5, 5, 7, 9 the mean is 5, the median 4.5 and the sample variance 32/7 ≈ 4.571.
Values that cannot be computed
When the selection is too small for a statistic (mean: n ≥ 1; variance and standard deviation: n ≥ 2; skewness: n ≥ 3), the workspace shows the statistic as not available (—) instead of a number.
Time series
The Overview time series plots the selected observations over time. Three controls change it.
Aggregation
| Option | Period | Period labelled by |
|---|---|---|
| Day | One calendar day | The date |
| Week | ISO week, Monday to Sunday | The Monday |
| Month | Calendar month | The first day of the month |
| Quarter | Calendar quarter (Jan–Mar, Apr–Jun, Jul–Sep, Oct–Dec) | The first day of the quarter |
Each point is the mean of the observations that fall in the period. Periods without observations are omitted rather than plotted as zero. A partial period at the edge of the date range (for example a week cut by the range start) is averaged over the days it contains.
Moving average
The window is counted in points of the current aggregation: with weekly aggregation, a window of 4 averages four weeks. The first k − 1 points have no moving-average value because the window is not yet full. A trailing average lags turning points by design.
Zoom and tooltip
The zoom slider below the chart narrows the visible part of the series without changing the selection; indicators and other panels are not affected. To change what is analysed, use the date range. The tooltip shows the period and the value of each series at the pointer.