Suggestions for analyzing data for being unusual ('good', 'neutral', or 'bad' day)

[chadh712]

Sorry, I wasn't sure which forum to post this in.

I have a set of data where a group of people individually selects each day whether it was a 'good', 'neutral', or 'bad' day. I am trying to come up with a way to look at each individual's data and determine whether their incoming daily response is outside what their 'normal' response pattern is. I have been exploring using a weighted average approach where each day is scored as a weighted average of the previous XX days responses. I then tried to put this onto a control chart but the responses across individuals can be so different that a formula that works great for one individual doesn't work at all for another. The biggest issue I am dealing with is what one person thinks is a 'good' day is another person's 'neutral' day. Some individuals select 'neutral' as a baseline and go up or down from there while others select 'good' as a baseline and go down from there.

Does anyone have any suggestions on how I can take this data and determine on an individual level, based on their historic responses, whether their most recent responses are unusual?

 

[JeffM]

Frankly, I do not understand your question at all. If you have individually differentiated patterns, why would combine patterns that you know are not similar? You might combine them to see if you found the responses to be highly similar. But if there is not a standard response, then putting the data of different individuals together loses information.

 

[chadh712]

Thanks for the response and sorry if I my initial post didn't make sense.

I am not trying to put the data together, I want to look at the data based on each individual. I basically want to take an individuals historic responses and as new responses come in determine whether the new responses are better or worse. For example, over the last 2 months an individual said they had a 'good' day 95% of the time, a 'neutral' day 5% of the time, and a 'bad' day 0% of the time. Now, over the next two weeks the percent of responses has shifted to 93%, 6%, 1%. My logic tells me in this instance I can analyze this based on percent of good days.

But now I have another individual who had 0% 'good' days, 99.9% 'neutral' days, and 0.1% 'bad' days. I can't look at this person based on percent of good days, I would think I need to look at them as percent of neutral days. Multiply this by 1000 individuals and it brings me to the problem of not having time to do this type of individualized analysis every day.

What I would really like is a way to account for these different types of responses patterns and setup a formula (probably through something like Excel) and easily run through every individual, every day. Essentially giving me an output of this individual's recent responses are not normal or yes they are.

I hope that helps and doesn't just confuse more.