Re: Is strong typing worth the cost?

[tmoran@acm.org]

>So we might say, "There is no rigorous, *scientific* proof of the

>OTOH, there actually *is* some half-way scientific evidence that Ada *does*
  The relevant question is "what is likely to make me money".  There's
rarely "rigorous, *scientific* proof" available (except on late night
TV), so you go with half baked evidence and statistical probabilities.
That's what "statistical decision theory" is about.  You have a set
of probabilities and costs, eg, 1% chance strong typing will double
productivity, 10% chance it will improve productivity by 20%, 69%
chance it will make no difference, and 20% chance it will decrease
productivity by 15%.  Given those numbers,

       Ratio of new   expected
 Prob  Productivity   value
+0.01  200%           2
+0.10  120%           12
+0.69  100%           69
+0.20  85%            17
                      = 100 %

the expected value of strong typing is the same productivity you had
before, ie, no improvement.

But if you spend money on an experiment and improve your estimates to,
say, no chance of a doubling of productivity, but a 15% chance of a 20%
improvement, and a 20% chance of only a 10% loss, then

       Ratio of new   expected
 Prob  Productivity   value
+0.00  200%           0
+0.15  120%           18
+0.65  100%           69
+0.20  90%            18
                      = 105%

Let's assume the experiment is 50-50 likely to give those results, or to
leave the original estimates untouched.  Then the expected value of the
experiment is a 2.5% improvement in your company's software costs - for
the foreseeable future.  You ought to be willing to spend up to the
discounted present value of that improvement for such an experiment.

If your company expects to employ 10 programmers for 10 years and has
a 10% discount rate, your future software costs are
10+.9*10+.9**2*10+...+.9**9*10 = 65 man years,
2.5% of that is 1.625 man years, which is what it's worth to your
company to perform the experiment.  If you have 100 programmers,
it's worth investing 16 man years in the experiment.
So the obvious question is "why haven't large companies with many
programmers and a long time horizon, done the experiment yet?"