Talk:Putnam model

This page lacks any quantitative or unit mentions that would empower some insight into the application. It's possible via links to find values of the Productivity Parameter and Index that Putnam recommends for typical projects, but nowhere is this tied to the units that Effort, or Size or Time are measured in. It would be grand to fix this here. If I ever find the information I'll drop back and improve it, but Putnam himself in 5 Core Metrics is wishy washy on such details!

It also notably lacks a robust mathematical relationship between Process Productivity and Productivity index. There is a slight clue provided here:

https://www.thedacs.com/techs/estimation/comparison.php

See Table 4 on that page. A reasonable fit to the data they provide yields:

ProductivityIndex = 9.5 Log(ProcessProductivity) - 26

Go figure, a pretty arbitrary dB scale from what I can tell. That's with a pretty clean fit. But if I approximate it then it looks like this is pretty close:

ProductivityIndex = ProcessProductivity in dB re 600

                 = 10 * Log(ProcessProductivity/600)

Which is still pretty wild. As in why a reference of 600? Empiricism gone wild?

I'm no expert, but I think it's one of those things that evolved over time. Putnam defines the process productivity as

$ProcessProductivity={\frac {Size}{\left[{\frac {Effort}{\beta }}\right]^{1/3}\cdot Time^{4/3}}}$

It is the Process Productivity (PP) that matters. not the Productivity Index (PI). However, the PP values are pretty unwieldy, in the same way that raw sound pressure level are unwieldy. Solution - create an (arbitrary) log scale. Is 10*log(PP/600) any more or less arbitrary than .002 dynes /cm^2 ? IMHO the 600 value takes the values for PP he was getting for the least productive projects he observed and gave them an index value near 1. From there, every 3 increase in PI equates to a doubling in productivity. When you are managing changes in productivity over time, it is just easier to work with the log values. Productivity of what, you ask? What ever units make sense for you.

Putnam appears to have done all his work using physical Lines_of_code. He has said (repeatedly) his technique could use many different units - could be SLOC, Function_points, or pages of documentation, if that is what you are estimating and creating / delivering. It is critical that the PP, Size, and ${\beta }$ all be in the same units. What he does NOT say is that to compute a PP, you need to have a ${\beta }$ term first. You can get beta values for SLOC from web site like the one you cite, but generating your own is not so easy. DSParillo (talk) 05:40, 27 September 2008 (UTC)