I'm a complete noob when it comes to statistics, so please excuse my lack of standard terminology. I think my question might be related to normalization, but please re-tag if I'm wrong.

I'm a programmer by trade, and I've been tasked with creating a chart which shows how the number of errors in a data set change from month to month. Unfortunately we don't have a statistician in-house to help me figure it out...

The data is as follows: Each month, a number of items are entered into a database. These items are checked for errors, and the errors are also entered into the database. Each item can have multiple errors of different types, and the number of items added each month can vary a lot.

How can I show how the error frequency changes from month to month?

It doesn't "feel" right to show a percentage of errors/number of items since there can be multiple errors for each item. The chart also needs to be easy to understand even by people with no background in statistics, so something as close as possible to the real values or percentages would be good.

An example of the data with total number of errors per month:

Month  Errors   Items 
March     208    2027
April     276    1304
May       609    1721
June      167    1561
July      268     513

The errors for March could be broken down like this:

Type        Errors
Type 1           2
Type 2          43
Type 3          93
Type 4           0
Type 5           1
Type 6           0
Type 7           4
Type 8           0
Type 9          22
Type 10          0
Type 11         43
  • $\begingroup$ @Lizzan Are you interested in displaying each error type or just the overall error rate? $\endgroup$ – chl Mar 16 '11 at 9:33
  • $\begingroup$ @chl Both, in two different charts. $\endgroup$ – Lizzan Mar 16 '11 at 10:26
  • $\begingroup$ @Lizzan I think sth like Cleveland's dotplot can do well; can you provide a snapshot of what your data look like and how they are arranged (e.g., just some records with corresponding error type)? $\endgroup$ – chl Mar 16 '11 at 11:01
  • $\begingroup$ @chl I added some example data, I hope it's what you're looking for. $\endgroup$ – Lizzan Mar 16 '11 at 11:56
  • $\begingroup$ Is each item monolithic, or does it consist of several sub-items, each of which is either right or wrong? An example of monolithic is a word (like "century"), which may have several spelling errors (e.g. "sentery" has two). An example of non-monolithic is something like purchasing data, an item is "name:computer, price:$500". In this case, you don't care how many spelling error there are, you just count how many sub-items are wrong. $\endgroup$ – SheldonCooper Mar 19 '11 at 11:05

If you only have two charts I would show the following pair. You are right to want to avoid percentages if some items have several errors, but you can show average errors per item which amounts to the same thing without the misleading impression, and could go over 1 without any need for an explanation of how a percentage goes over 100%.

You want to emphasize that: errors peaked in May; the error rate peaked in July; March and June were "good months" with high items and low errors and July was a "bad month"; that type 3 errors are most common; and (if true) the types of error vary between months. I think these do that.

enter image description here

  • $\begingroup$ Thanks @Henry, these are really useful! I'm not actually limited to only two charts, it was just how I had thought of it previously. If you have other ideas on how to show the data they would be very welcome! $\endgroup$ – Lizzan Mar 19 '11 at 16:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.