Normalizing Continuous Data with "0" values

B

Beggin

I have what I believe to be a simple question. A few years ago I used to work with creating control charts for processes that contained a large number of "0" values. Essentially I am developing charts that display the amount of time it takes for a call center to answer a phone call. (obviously many of the calls are answered in zero seconds).

In order to take this non-normal data and put it into a control chart I realize I must transform the data. Unfortunately I can not remember how we used to do this. I have attached the data set (sample) so everyone can see what I'm working with. In the spreadsheet, the column "Que Conv" represents the number of seconds it took to have each phone call answered. (in seconds) The data is a representative sample of the total population.
 

Attachments

  • How to Normalize.xls
    76.5 KB · Views: 108

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
A am not sure how much more information you would get from "normalizing" data - usually it just masks information and creates more work. I am thinking you would get sufficient information from a Moving R chart (see attached). It would be interesting to see what others suggest, and their rationale for the additional effort if they choose to normalize.
 

Attachments

  • moving r.doc
    137 KB · Views: 97
A

Allattar

Data sets like that are awkward, and not uncommon.
It would be interesting to ask, why do some values come out as zero and others much higher?
What is the measure from?
Am I right in thinking that it isn't really a value of 0 but rather a time somewhere between 0 and 0.5 or 0 and 1? It is just an artifact of the measurement system that it reports seconds.
In which case those numbers aren't really zero's, but the limitations of the measurement system make it difficult to identify which distribution they really come from.

Also this is probably some of the most extreme data you can put in a control chart. So roughly speaking the chance of breaking test 1 goes from 0.27% for normal distributions to about 4%.
As long as you are aware of that you should be ok.
Still I think my main question is, how/why are most calls answered in 0 seconds, eg what is the time metric measuring, what causes them to take longer?
 

Miner

Forum Moderator
Leader
Admin
Is this an automated answering system? And, does the caller immediately talk to a representative?

I cannot believe that the time to talk to a representative is often zero. Maybe a change in metric will better represent the customer experience and provide better data for analysis.
 
Top Bottom