It would be useful for data analysis to track the standard deviation/variance during the statistics period in LTS - the mean/max/min does not tell enough of a story, since you may have single data-point outliers in otherwise nearly constant data. This information on the variation within the statistics period is currently completely lost.

Secondly, I would argue for inclusion of a difference/change column, showing by how much a sensor changed in the statistics period. This would allow the statistics cards to produce bar charts showing e.g. energy use in the period, instead of cumulative energy use we can do using the “sum” chart option.

While this might be very relevant for some measurements, these characteristics are irrelevant (or even unfitting) for others, yet again other measurements might call for different statistical characteristics. What is your actual use case?

As you mention “data analysis”: I don’t think that the majority of users would find standard deviation/variance intuitive for that matter. Trends or outliers can be inspected visually. Everything beyond visual “analysis” should be done mathematically, which brings me to…

You probably know that the statistics integration exists to derive specific characteristic values from a sensor. Why does that not fulfill your need?

I am a contributors to this integration. Happy to hear your thoughts.

My use case is at the moment quite simple. I’m not making decisions based on this (yet?), more for illustration:

I have a dashboard with a few historical plots of various weather conditions, rain, river levels or things like temp. humidity in the house. If I collapse the data so that it is grouped by e.g. day, it would be good to get a spread/variability on the plot. This is kind of what the min/max subplots do, but standard deviation is a much better quantity to use to describe variability in the time series, at least for the way that I think about the data.

If I had the standard deviation in some long-tem data, you can then ask the question to what extent is the current state outside of the long-term trend, just by eye on the plot.

I do actually calculate stdev using statistics for a few cases when I use it for decisions in triggers (e,g, in solar irradiance as a proxy for the variability of cloud cover which then decides on whether to close blinds on my windows), I have to admit to not having realised that this means I have an stdev LTS time series for that sensor, so I have the data there. But to group it in the plot would require reprocessing it (adding the variances for the data inside the bin) and it would not be possible to plot on a statistics card without making two more sensors (data - stdev, data+stdev). It seems to me like a nice feature.

The other “change” column would possibly go toward solving the issue that currently there is no way using the statistics card to replicate the style of bar chart as in the energy dashboard. If I have some accumulating consumption sensor, I would like to know what was consumed in a binning period (number of washer cycles per week, coffee cups consumed per month), rather than the global total which the “sum” statistics plot currently shows.

And since you are here: another related request would be for the statistics integration to be able to access the LTS database at least for some of the functions. I was hoping to use it to create sensors such as energy consumed in 30 days, but the most I can do is to use the recorder data. In the end, I am using the SQL integration to look back in the database, but statistics is much better suited for this in principle. It would never work for all the state characteristics, but for changes, min/max etc the data are there and the way I see it, it would be much more useful than e.g. the utility meter integration for the purpose of monitoring consumption trends.