v = mean (ts) returns the mean of each data element over time, so v has the size of one sample; with 'time' weighting it is the time-weighted mean. v is double whatever the class of the data, and [] for a series with no samples.

v = mean (ts, name, value, …) takes these options, which all nine statistics share, the names matched in any case:

'Quality'

the quality codes of samples to leave out, as missing; a value left without samples is NaN, a sum 0. A code no sample has leaves nothing out, with a warning.

'MissingData'

'remove', the default, leaves missing values out; 'interpolate' fills those between samples linearly in time first.

'Weighting'

'none', the default, or 'time', which weights each sample by the time it covers: an interior sample by the mean of its two intervals, an end sample by its one interval, the weights scaled to a mean of 1, so that uniform time changes nothing.

NaN is missing while TreatNaNasMissing is true; when it is false, mean, median, std, var and sum give NaN where the data holds one, while min, max, mode and iqr still skip it, as in MATLAB.

MATLAB applies 'time' weighting by multiplying the data by the weights and taking the ordinary statistic of the product, which is right for mean and sum only: there the minimum of [1 2 4 8 16] at times [0 1 3 7 8] is 0.5556. Here each statistic is the time-weighted one: 1 for that minimum. MATLAB also fails on logical data, and in std, var and iqr on integer data, which here are computed in double.

See also: timeseries.median, timeseries.std, timeseries.sum

Source Code: timeseries