Moving Averages

#1
Hi all,

I'm trying to do a quick and rough calculation but have zero desire to pump more meaningless 'statistics' into the world. So wanted to get a sense check here, please.

I'm measuring something once a day, every day. I take a 7-day moving average up to and including yesterday. I take today's measure. I want to know if today's number is any indication that things are starting to change.

My immediate thought was merely to compare day 8's value with the SD of the 7 days and use the 68 95 99 heuristic as a yellow, orange, red warning system.

Then I confused myself with questions any elementary stats student could answer as I have long-since forgotten the basics (how embarrassing!): is this really a two-sample test in which case I need a reference distribution under the null? Does this mean one of my samples (day 8) is of size 1? What is the implication of that? Can the empirical rule be applied to non-symmetric distributions?

For context, this is something I was playing with at work, regarding staff not working due to covid reasons. It's not required or official or etc. - I just pull the number of absentees out of the database and send it off. I was just curious for my own benefit to do something statistical more interesting and useful to amuse myself. And that is when I realised every new statistical thing I learn pushes out an old thing! :)

P.S. I've done basic time series before but it's been years and, since this is not official work stuff, and I'm pressed for time, I am not in a position to dust off details and build a decent predictive model.

Cheers all!
 

Miner

TS Contributor
#3
You will probably be most interested in the Individuals-Moving Range control chart or the Exponentially Weighted Moving Average (EWMA) control chart. Most of the other types of variable control charts are based on subgroups. Since you are measuring 1x per day, there is no basis for a rational subgroup.