"Kent Paul Dolan" <xanthian@well.com> wrote in message
news:f6a580bb517ec9eced2eaa01cf9c9525.48257@mygate.mailgate.org...
> "M�rio Amado Alves" <maa@liacc.up.pt> wrote:
>
> > You're both right. I also did a study correlating program text metrics
> > and actual development cost in programmers*month. SLOC had the best
> > results, notably beating Halstead. It was a small project (10k SLOCs if
> > I remember correctly), but it was unbiased i.e. the programmers didn't
> > know about the metrics in advance.
>
>
> Well, but SLOC is predicting by hindsight, and it has been fairly well
> known since the 1960's that individual programmer's code output is
> fairly constant when measured in keystrokes, so that about as many lines
> per day of assembly as of APL can be delivered fully burdened.
>
> Thus, one would expect that SLOC or something like it  fairly equivalent
> to keystrokes would be the best hindsight predictor of
> technologist-months of effort involved in completing a project or a
> component.
>
> But I thought (with no expertise here) that the goal of stuff like
> Halstead's metric was that you could estimate it at specification time,
> and do foresight forecasting with that estimate.  I'd shudder to think
> what the accuracy of foresight guesses at SLOC from similar projects
> might accomplish in setting milestone dates and such.

Your shudders would be misplaced.  Such estimates are actually pretty good,
IF you go to the necessary level of detail.  Discard your preconceptions and
read Barry Boehm's "Software Engineering Economics".