WEBVTT

00:00:00.000 --> 00:00:02.994 align:middle line:90%
[AUDIO LOGO]

00:00:02.994 --> 00:00:12.480 align:middle line:90%


00:00:12.480 --> 00:00:14.190 align:middle line:84%
RICHARD DE NEUFVILLE:
And so we have

00:00:14.190 --> 00:00:16.390 align:middle line:90%
basically two topics for today.

00:00:16.390 --> 00:00:21.730 align:middle line:84%
And the first one is the concept
of the value of information.

00:00:21.730 --> 00:00:27.750 align:middle line:84%
And we're going to end up with
two ways of thinking about it.

00:00:27.750 --> 00:00:32.280 align:middle line:84%
I am going to present in
some detail the procedure

00:00:32.280 --> 00:00:34.110 align:middle line:84%
for the definition
of the upper bound

00:00:34.110 --> 00:00:39.210 align:middle line:84%
on the value of information, the
EVPI, which stands for Expected

00:00:39.210 --> 00:00:41.910 align:middle line:90%
Value of Perfect Information.

00:00:41.910 --> 00:00:47.970 align:middle line:84%
At the end is a supplement for
your interest if you want to is

00:00:47.970 --> 00:00:52.950 align:middle line:84%
the calculation of the
exact value of EVSI,

00:00:52.950 --> 00:00:59.660 align:middle line:84%
which is a interesting,
probabilistic exercise.

00:00:59.660 --> 00:01:01.400 align:middle line:84%
But I personally
don't think that it's

00:01:01.400 --> 00:01:02.940 align:middle line:90%
very valuable in practice.

00:01:02.940 --> 00:01:05.450 align:middle line:84%
But it's part of
standard curricula,

00:01:05.450 --> 00:01:08.990 align:middle line:84%
and it's there if you
want to go for it.

00:01:08.990 --> 00:01:12.890 align:middle line:84%
It will not be part of a
quiz or anything like that.

00:01:12.890 --> 00:01:17.890 align:middle line:84%
I think you should
know about it,

00:01:17.890 --> 00:01:23.560 align:middle line:84%
so you can appreciate the other
aspect I will be focusing on,

00:01:23.560 --> 00:01:24.860 align:middle line:90%
so methods.

00:01:24.860 --> 00:01:26.830 align:middle line:84%
And then I want to
stand back from this

00:01:26.830 --> 00:01:29.410 align:middle line:84%
and talk about
information gathering

00:01:29.410 --> 00:01:32.470 align:middle line:84%
as a basis for having
a flexible strategy.

00:01:32.470 --> 00:01:35.490 align:middle line:90%


00:01:35.490 --> 00:01:39.990 align:middle line:84%
So there is this notion
that, before you commit

00:01:39.990 --> 00:01:44.580 align:middle line:84%
to almost anything, you can
reduce your uncertainty,

00:01:44.580 --> 00:01:49.590 align:middle line:84%
or you can hope to reduce
your uncertainty by inserting

00:01:49.590 --> 00:01:53.640 align:middle line:84%
some kind of information
gathering, which from a jargon

00:01:53.640 --> 00:01:55.080 align:middle line:90%
point of view is called a test.

00:01:55.080 --> 00:01:56.280 align:middle line:90%
It's not an examination.

00:01:56.280 --> 00:01:58.140 align:middle line:84%
But it's a short
way of referring

00:01:58.140 --> 00:02:00.510 align:middle line:84%
to whatever you do
to get information

00:02:00.510 --> 00:02:02.410 align:middle line:90%
before your decision problem.

00:02:02.410 --> 00:02:06.990 align:middle line:84%
So in this diagram at the bottom
of the page, I have given you--

00:02:06.990 --> 00:02:08.535 align:middle line:90%
you have your decision problem.

00:02:08.535 --> 00:02:11.070 align:middle line:90%


00:02:11.070 --> 00:02:12.940 align:middle line:90%
Do this thing or not.

00:02:12.940 --> 00:02:16.500 align:middle line:90%
And you say, OK.

00:02:16.500 --> 00:02:20.250 align:middle line:84%
I'm going to go through
a process of collecting

00:02:20.250 --> 00:02:22.620 align:middle line:84%
some relevant information
or information I think

00:02:22.620 --> 00:02:25.930 align:middle line:84%
is relevant, get
that information,

00:02:25.930 --> 00:02:29.500 align:middle line:84%
and then consider
my decision problem.

00:02:29.500 --> 00:02:32.940 align:middle line:84%
So, in principle, that's
always available to you,

00:02:32.940 --> 00:02:37.500 align:middle line:84%
if only to say, OK, let me stop
for a minute and think about it

00:02:37.500 --> 00:02:40.680 align:middle line:84%
and remind myself from
the back of my brains.

00:02:40.680 --> 00:02:41.970 align:middle line:90%
And so what should I do?

00:02:41.970 --> 00:02:45.450 align:middle line:84%
Now, of course, at some point,
you've got to do something.

00:02:45.450 --> 00:02:46.780 align:middle line:90%
You've got to react right away.

00:02:46.780 --> 00:02:49.890 align:middle line:84%
But, in general, you have the
opportunity to do something.

00:02:49.890 --> 00:02:54.360 align:middle line:84%
And this has a have
an important effect

00:02:54.360 --> 00:03:00.190 align:middle line:84%
on your value of your
decision in your procedure.

00:03:00.190 --> 00:03:03.090 align:middle line:90%
So what does a test do?

00:03:03.090 --> 00:03:06.157 align:middle line:84%
So as a starting point,
any decision problem,

00:03:06.157 --> 00:03:08.490 align:middle line:84%
even though you might not
think about it quite this way,

00:03:08.490 --> 00:03:09.365 align:middle line:90%
you're driving a car.

00:03:09.365 --> 00:03:10.080 align:middle line:90%
You turn left.

00:03:10.080 --> 00:03:14.640 align:middle line:84%
You may not think at all in
terms of the probabilities.

00:03:14.640 --> 00:03:20.190 align:middle line:84%
But you have some uncertainties
and some assessment of them

00:03:20.190 --> 00:03:21.090 align:middle line:90%
in principle.

00:03:21.090 --> 00:03:24.240 align:middle line:84%
And these are known as
prior probabilities.

00:03:24.240 --> 00:03:25.530 align:middle line:90%
Why prior?

00:03:25.530 --> 00:03:29.670 align:middle line:84%
Because we want to
distinguish them from what

00:03:29.670 --> 00:03:31.830 align:middle line:90%
you might have after a test.

00:03:31.830 --> 00:03:35.610 align:middle line:84%
So it assumes that these are
your starting probabilities

00:03:35.610 --> 00:03:36.960 align:middle line:90%
or assessment.

00:03:36.960 --> 00:03:38.790 align:middle line:84%
For example, you're
running a business,

00:03:38.790 --> 00:03:41.040 align:middle line:84%
or your group is
running a business.

00:03:41.040 --> 00:03:45.030 align:middle line:84%
And you have some measure of
the cost of production of unit.

00:03:45.030 --> 00:03:48.570 align:middle line:84%
You have some measure of
the likelihood of sales

00:03:48.570 --> 00:03:53.230 align:middle line:84%
and whatever other features
that you might have.

00:03:53.230 --> 00:03:56.710 align:middle line:90%
So what does a test do?

00:03:56.710 --> 00:03:58.260 align:middle line:84%
Well, there are
various forms of it.

00:03:58.260 --> 00:04:01.440 align:middle line:84%
But, in general, it gets
information on these issues.

00:04:01.440 --> 00:04:04.290 align:middle line:84%
For example, you
could say, well,

00:04:04.290 --> 00:04:07.677 align:middle line:84%
I think I know how it's going
to produce in this plant.

00:04:07.677 --> 00:04:09.510 align:middle line:84%
I've run several of
them back in the States.

00:04:09.510 --> 00:04:13.620 align:middle line:84%
But now I'm doing one
in South Africa say,

00:04:13.620 --> 00:04:16.149 align:middle line:84%
and situations
might be different.

00:04:16.149 --> 00:04:17.730 align:middle line:90%
So you can run a test plant.

00:04:17.730 --> 00:04:20.459 align:middle line:84%
See how it works with
the local materials.

00:04:20.459 --> 00:04:22.845 align:middle line:84%
See how it works with
the local engineers.

00:04:22.845 --> 00:04:23.940 align:middle line:90%
It might work better.

00:04:23.940 --> 00:04:26.400 align:middle line:84%
They know things you
didn't know. whatever.

00:04:26.400 --> 00:04:29.040 align:middle line:90%
So you can run a test plant.

00:04:29.040 --> 00:04:31.710 align:middle line:90%
Get more information.

00:04:31.710 --> 00:04:34.215 align:middle line:84%
Or, for example, you can
run a market analysis.

00:04:34.215 --> 00:04:36.660 align:middle line:84%
You can say I'm going
to take my new product,

00:04:36.660 --> 00:04:40.740 align:middle line:84%
and I'm going to test it in
St Louis, see how it picks up.

00:04:40.740 --> 00:04:43.380 align:middle line:84%
Maybe I'll go for a
different demographic

00:04:43.380 --> 00:04:45.360 align:middle line:84%
and also test it
in San Francisco.

00:04:45.360 --> 00:04:46.530 align:middle line:90%
Whatever you want.

00:04:46.530 --> 00:04:50.880 align:middle line:84%
But there are things that
you can do individually

00:04:50.880 --> 00:04:53.460 align:middle line:84%
as a company, as an
organization to get more

00:04:53.460 --> 00:04:55.890 align:middle line:84%
information about
something before you

00:04:55.890 --> 00:04:59.910 align:middle line:90%
make a commit to a decision.

00:04:59.910 --> 00:05:02.450 align:middle line:90%


00:05:02.450 --> 00:05:05.290 align:middle line:84%
So the operation of
test is basically

00:05:05.290 --> 00:05:07.840 align:middle line:84%
you get new information,
or you get the information

00:05:07.840 --> 00:05:09.340 align:middle line:84%
from the thing
that you're doing--

00:05:09.340 --> 00:05:12.250 align:middle line:84%
the test plant or the
market survey or the test

00:05:12.250 --> 00:05:14.300 align:middle line:90%
marketing or whatever.

00:05:14.300 --> 00:05:18.820 align:middle line:84%
And it says, ah, I got
the idea that this--

00:05:18.820 --> 00:05:20.810 align:middle line:84%
I thought I knew
what was happening.

00:05:20.810 --> 00:05:23.350 align:middle line:84%
But gee, This test
market showed me

00:05:23.350 --> 00:05:25.960 align:middle line:90%
that the product really stinks.

00:05:25.960 --> 00:05:27.460 align:middle line:90%
So maybe it isn't so good.

00:05:27.460 --> 00:05:29.470 align:middle line:84%
So maybe my
probabilities are wrong.

00:05:29.470 --> 00:05:31.570 align:middle line:90%
Or it says, wow.

00:05:31.570 --> 00:05:34.060 align:middle line:84%
It really took off
like gangbusters.

00:05:34.060 --> 00:05:35.440 align:middle line:90%
It's going to be wonderful.

00:05:35.440 --> 00:05:38.230 align:middle line:84%
So whatever it is,
you have the chance

00:05:38.230 --> 00:05:44.080 align:middle line:84%
to give the new information to
revise your prior probabilities

00:05:44.080 --> 00:05:46.540 align:middle line:90%
known as posterior probability.

00:05:46.540 --> 00:05:49.510 align:middle line:84%
Has nothing to do with
your tush or your posterior

00:05:49.510 --> 00:05:51.520 align:middle line:84%
but just that they
come afterwards.

00:05:51.520 --> 00:05:55.750 align:middle line:84%
It's sort of a mouthful, but
it's the after probabilities.

00:05:55.750 --> 00:05:58.580 align:middle line:84%
And then you have new
values for the decision.

00:05:58.580 --> 00:06:03.280 align:middle line:84%
So, in general, then you've
had the original decision

00:06:03.280 --> 00:06:08.410 align:middle line:84%
that had a certain value
that you could calculate,

00:06:08.410 --> 00:06:09.640 align:middle line:90%
as we've done before.

00:06:09.640 --> 00:06:12.260 align:middle line:84%
And then having with
new probabilities,

00:06:12.260 --> 00:06:15.500 align:middle line:90%
you have a different value.

00:06:15.500 --> 00:06:18.550 align:middle line:84%
So the consequence
of the test is

00:06:18.550 --> 00:06:21.580 align:middle line:84%
that you have an expected
value after the test.

00:06:21.580 --> 00:06:25.240 align:middle line:84%
And, in general, the
expected value of the test

00:06:25.240 --> 00:06:28.690 align:middle line:84%
is greater than
without the test.

00:06:28.690 --> 00:06:31.140 align:middle line:90%
Why?

00:06:31.140 --> 00:06:33.240 align:middle line:84%
Because you can
avoid bad choices.

00:06:33.240 --> 00:06:38.700 align:middle line:84%
You say, oh, well, I found out
that my sales estimate wasn't

00:06:38.700 --> 00:06:40.150 align:middle line:90%
as good as I expected.

00:06:40.150 --> 00:06:43.560 align:middle line:84%
So really, I shouldn't be so
aggressive in the production.

00:06:43.560 --> 00:06:46.740 align:middle line:84%
And I won't have all this
stock that I didn't use,

00:06:46.740 --> 00:06:49.810 align:middle line:90%
so I don't lose that money.

00:06:49.810 --> 00:06:53.670 align:middle line:84%
On the other hand, you can see,
oh, it sold out really well.

00:06:53.670 --> 00:06:56.160 align:middle line:84%
Maybe I should
increase my production

00:06:56.160 --> 00:06:58.380 align:middle line:84%
and do much better
than I thought.

00:06:58.380 --> 00:07:04.350 align:middle line:84%
So the general idea that the
expected value after the test

00:07:04.350 --> 00:07:07.360 align:middle line:84%
is greater than the expected
value without the test,

00:07:07.360 --> 00:07:10.920 align:middle line:84%
in other words, that
you can do better

00:07:10.920 --> 00:07:18.630 align:middle line:84%
by having some proper amount of
information available to you.

00:07:18.630 --> 00:07:22.120 align:middle line:90%
Now, we'll take that as a given.

00:07:22.120 --> 00:07:23.730 align:middle line:84%
And that's a useful
starting point.

00:07:23.730 --> 00:07:27.870 align:middle line:84%
The question really is,
since the test has a cost,

00:07:27.870 --> 00:07:29.970 align:middle line:90%
is it worthwhile?

00:07:29.970 --> 00:07:32.670 align:middle line:84%
I mean, it has a cost
because, if you're

00:07:32.670 --> 00:07:36.840 align:middle line:84%
going to test marketing
or set up a pilot plan,

00:07:36.840 --> 00:07:38.410 align:middle line:84%
you have to spend
some money on it.

00:07:38.410 --> 00:07:40.260 align:middle line:84%
You also have to
spend some time.

00:07:40.260 --> 00:07:42.150 align:middle line:84%
The time you spend
on that exercise

00:07:42.150 --> 00:07:44.790 align:middle line:84%
may mean that you
missed a market

00:07:44.790 --> 00:07:48.630 align:middle line:84%
that you could have been making
money instead of just delaying.

00:07:48.630 --> 00:07:52.560 align:middle line:84%
So the ultimate
question is, what is

00:07:52.560 --> 00:07:55.120 align:middle line:90%
the value of the information?

00:07:55.120 --> 00:07:57.510 align:middle line:90%
And does this value--

00:07:57.510 --> 00:08:00.510 align:middle line:84%
is it greater than
the cost of the test?

00:08:00.510 --> 00:08:04.530 align:middle line:84%
Or more precisely, is
it significantly greater

00:08:04.530 --> 00:08:05.880 align:middle line:90%
than the cost of the test?

00:08:05.880 --> 00:08:08.190 align:middle line:90%
Is it really worthwhile to do?

00:08:08.190 --> 00:08:11.970 align:middle line:84%
If it's $10 better, doesn't
seem like such a good deal.

00:08:11.970 --> 00:08:14.530 align:middle line:84%
But is it really a
significant value?

00:08:14.530 --> 00:08:16.820 align:middle line:90%
That's the idea.

00:08:16.820 --> 00:08:23.000 align:middle line:90%