Steven Line wrote:

> Hi -

>

> We just started using Scrum a couple of weeks ago. To calculate our

> projected burn down we're using a 7 day linear regression algorithm.

> The resultant projected burn down line is trying to replicate the line

> of the past 7 days instead of merely project where it's currently

> headed (imagine an actual burn down that curves). If there's an

> upswing then a downswing in our actual burn down, the extrapolated burn

> down will actually jog upwards then downwards, which while

> mathematically correct for a regression, isn't communicating that the

> team is actually burning down, not backing up.

>

> What formula or algorithm do you use out there in scrum land?

>

Personally I do not think there is any formula that could safely predict a

half ways reasonable regression with the amount of data you are feeding it.

What confidence intervals do you get with the regressions you are using now?

just as everybody else I draw the ideal burn down and then all that comes

along shows pretty easily. interpreting a burn down chart just needs

experience. Over time you learn pretty quickly what cause might be behind a

particular form of a burn down chart.

However if you wish to experiment, go and install R (www.r-project.net) with

that you can pretty easily perform simple linear regression models, just as

multiple regression models, such as One-sided test, Two-sided test and F-test.

-d