The future of; 25; How Forecasters Forecast

From WMUR found in Washington D.C. the future of another in a series of discussions of alternative futures. Your moderator is Joe coach of the world future society. Mr. Coates the subject of this evening's discussion is how forecasters who are cast we have who's this evening Lieutenant Colonel Joseph Martino of the Air Force Office of Research Analysis Holloman Air Force Base Alamogordo New Mexico Colonel Martino holds a doctorate in mathematics from hio state granted 961. He's the author of numerous articles and publications in forecasting and he is the author of a forthcoming book on technological forecasting. He also edits the technological forecasting section of the r journal The Futurist. Well Geo I think that there's a widespread feeling that there's a lot of voodoo associated with forecasting and predicting and looking into the future and somehow or other I think there's a feeling that it's not quite respectable that there's a little bit of legerdemain in there. Perhaps we could begin by you telling us what forecasting is.

All right fair enough. First let me preface my remarks by stating that anything I say represents my own opinion I'm not necessarily that of the United States Air Force. Now on to for can we get a forecast you would say that. The essential notion is that a forecast is intended to tell somebody something about an event which has not yet occurred and in practice we tend to restrict ourselves to what I choose to refer to as rational methods of forecasting. These are methods which depend on the use of historical data and apply to that data. Some explicit technique that is it can be explained to people it can be demonstrated and a technique which is replicable that is that anyone can use. Provided he can learn the manipulation of your act. But in a sense there is nothing hidden or mysterious about it we try to use methods that are open to anyone who will take the time to learn to use them well before we get into what those

specific methods are geo who's interested in forecasting. Just about anybody who's going to live in the future seems to be interested in forecasting nowaday who's interested in putting money on the line for the Department of Defense. Of course with which I'm associated has put considerable money on the line for this kind of thing. The Army and Navy both have established service wide technological forecasting programs of considerable extent. The Air Force has been in the field off and on over the past 20 years and it apparently has now made a decision to go into it on a moderate scale not as extensive as the other two services yet but deeper than it has been in the past. So at any rate the Department of Defense is willing to put money into it. A number of industrial corporations have also put their money on the line so to speak. TRW and LTV are two of the biggest in the field that I am aware of.

TRW was formally Thompson remodeled and that's right. And TV was working. Ling Temko bought OK. Are there other companies or corporations that are moving into there and yes one of the most interesting recent forecasts for instance was prepared by a SmithKline and French a pharmaceutical house and this was a forecast of the future of medicine including medical education medical treatment para medical treatment and so on. Only you know the defense area are there any industries that are particularly interested in or have the history of the use of forecasting taking their own space in the story for its civilian applications has been in this for quite some time. They are concerned with developing commercial air transport aircraft. They want these to go on the market at a time when they'll meet in market demand and not be too far behind the competition. So they are very much interested in forecasting what the market will require. We don't always want to be ahead of the competition. They want to be they want don't want to be too far ahead of the competition was it. That will mean they're stretching the state of the art in Italy to too high

cost an aircraft. Are you saying that poor one aircraft manufactured to be let's say five years ahead of the other people in the industry is going to cost him sufficient money that he may in fact lose lose money if indeed he has really pushed the state of the art to produce they have to make the airplane yes. So the idea is to establish a narrow margin of superiority right enough to be ahead but not too far and forecasting. You say the major tool is a major tool for them in doing. Let's talk a little bit Colonel Martino about what some of the methods are that are used are you. Are we talking about one method or 50 or 60 methods approximately how many are they pens on the level of abstraction you choose. Fundamentally all rational forecasters must use. One of two assumptions in preparing their forecasts they must gather historical data and then apply a procedure to this and they must assume either that some historical pattern is going to repeat itself or they must assume that a current pattern is going to continue

so that fundamentally all forecasters make use of one or the other of these approaches. But in detail the methods differ considerably and there are quite a few of them. What was jumping to start with anyone. Oh it probably at least in the technological forecasting field the most common when used today is known as Delphi. This is everything the Delphic oracle. That seems to be the origin of the tournaments an unfortunate origin I'm afraid because it implies more than can be delivered. There is nothing or ocular about the Delphi method. What is it. It involves getting a panel of experts if you will only treats them differently from the usual group or committee action the usual arrangement is to bring a committee face to face let them debate some issue and come up with. In this case let's say a forecast Delphi differs from this in that the committee never sees each other the committee never meets. Instead they are interrogated anonymously by a set of questionnaires and a panel director is

responsible for preparing the questionnaires and sending them out and getting them back. Consolidating the replies and making up the next questionnaire. The purpose of this anonymity and so on is anonymity from each other not each other but not meant to directly or the purpose of this anonymity is to avoid the usual difficulties of committee action which can be summed up as the notion that a committee tends to take on a life of its own a momentum of its own which may have little to do with the original purpose for which it was put together. Are there more specific good directions to do panel meetings namely that the guy that talks the most tends to brood over the man with the most prestigious tends to prevail. This is true it has been demonstrated quite clearly by numerous psychological tests. That in a committee meeting it is not the strength of the arguments that prevail but the number of times an argument is repeated. Somebody who sits there and gives the same argument time after time will eventually carry the day.

Delphi eliminates this by the anonymous reply and then the panel director processing the replies so that each argument is given only once in written form. So what is he doing now he's gotten all these replies back let's say from 20 people. And what does he do with them Joe. He must. Well first of all the replies will consist of a list of events let's say these are the events which are forecast. And there will be a date associated with each event. So he must find those events which are common to all the lists and get a spread of dates. There will be an earliest date latest date and a spread of dates in between for the panel estimates. He prepares a statistical summary of the case. Let's say the median date and a a three quarters date and a one quarter date one fourth one and let's make it more specific in that SmithKline French study you made. Let's say for example one of the questions was one we have chemicals that will allow us to grow amputated regrow amputated limbs.

So the plan would be to arrive you day would give a variety of date and then what would be done with let's say those 20 dates that you only get only get 20 dates here and then when we arrange them in order starting from the earliest to the latest then we count the first 10. So we now have a date that's exactly halfway between all of them. There are 10 higher in 10 lower that is known as the median. And this is really taken as the panel forecast. All right then we start again and count and get the date such that there are five below it and 15 above it. And another date such there 15 below and 5 above it. And these last two are known as the core tiles and the panel. These are the extremists so to speak. It would not be right the extremists nonwhite the ones that either ran but once a quarter of the way in from each end. And the panel forecast is given in terms of the medium and the two core tiles. Here is the median date that is the panel estimate. And here as the core tiles are a measure of the panel disagreement.

So in this particular case with forecasting chemicals to regrow limbs the range might be 1994 to 2050 with maybe a median date of 2025 and something like that yes and the forecast will be quoted in that fashion. Now the panel director takes these dates prepares the medium into core tiles for each event and that becomes the next questionnaire to go back to the panelists a statement not of the complete set of estimates but only these three date and then if some panelists have given reasons why they believe their date is the correct one. He will summarize these reasons he may find for instance that three or four panelists have given the same reason. So he will consolidate and summarize the reasons present presenting then a short summary of the reasons why the date should be later than most of the panel thinks and why the date should be earlier than most of the panel things. And these written reasons with the medium import tiles are then returned to all the panelists as the next questionnaire to which they are responding.

That's don't what to report. It depends for time seems to be the maximum required at some Delphi sequences have been done successfully with only two ounds And so there's no good technique for using expert opinion and kind of making it quantitative and relatively precise. That's correct making the the reasons for the range of answers rather clear cut. Has anybody actually used that as the basis of an action program has anybody ever done anything on the basis of these forecasts. Yes the two companies that I mentioned earlier TRW and LTV have both done this. TRW ran a sequence internally using its own experts. With the intent of forecasting technology relevant to the various corporate divisions over the next 15 or 20 years and the forecast is proprietary to the company they feel seriously enough about it that they don't let it out. They won't tell you what events were forecast or the dates for them but it is a basis of at least that one rather large company whose technological future.

That's right. They are taking it seriously and using it. And LTV is also using it seriously but they're a little more open about it. They will let you see a copy of their forecast but not of the plans based on each corporate division in LTV is required to state in its annual plan to corporate headquarters what action it is proposing to take in response to the company's forecasts and what the SmithKline and French were because it is not available that is available in fact the world future society has distributed that to members who are subscribers to the. So I'm under supplementary program. Thank you. One. Well let's go on and look at another method you almost do something else that you want to get across about the one you know and we can go on I think another important method perhaps not as widely used Delphi is that growth curves loose I think in growth curve. Yes the they are named for the biological analogy. If you were to plot on a graph for instance the weight of a

human being from the time of birth the time of death. Generally you would find that the plot resembles an S shape the letter S. As an infant growth initial growth is fairly slow. Then there is a spurt of growth say in the 12 to 15 18 year bracket. Which then reaches a peak a plateau and saturates the weight or height remains relatively constant after that. So this describes a growth situation in the biological case and by analogy they are referred to as growth curves. This is virtually universal in the main living Kingdom That's right. No Are you saying that those biological phenomena is in some sense mimicked by many other technical situation. That's correct any measure of performance. Let's take speed of aircraft for instance this is a good one. If you start with the Wright brothers first aircraft and plop the speed of propeller driven aircraft you will find the same growth

curve behavior for the first few years there wasn't much increase in speed then as streamlining and more powerful engines came into the picture you saw dramatic increases in speed finally topping out in the high 400 and the high for hundreds. Four hundred fifty four hundred seventy five mph. And that's strictly for propeller driven aircraft. Now do you then get the same kind of curve would you have driven aircrew who would I believe except that we haven't reached the top and we're still on the rise along with rising force Well is there some fundamental reason why this occurs or is this kind of an accident or other classes of events that you can say will be growth curve cases and others that won't. Generally yes you can say that there are classes that will be M.. These are the cases where you can see some fundamental physical let's say physical or chemical limitation on performance in the case of propeller driven aircraft it was clearly the speed of sound. You are not going to get a propeller which operates efficiently at much above the speed of sound so here was a clear cut case of a physical limitation on the growth of this type

since doing so to speak right on which you can't go and this is the kind of thing we look for in attempting to predict using growth carriers. Find the limit for this technology based on some fundamental physical considerations. Has this been used in any substantial degree in practical affairs. Not to any great degree no. Primarily because in many cases it's too hard to estimate just what the upper limit will be. I have seen some attempts for instance to forecast the growth in the fish and sea of steam engines and it's very difficult really to say what the the upper limit will be if somebody gives you an upper limit you can plot a growth curve very nicely but there are some of the bait say is whether it should be 50 percent 55 percent 60 percent as an upper limit was what you're doing basic planning let's say who the NASA's space program is a decade ago. Could you have used these techniques to have anticipated what you would be what you would have available as rockets and spacecraft were being built.

I believe so. In fact the growth curve would not have worked very well because we don't seem to have reached the plateau yet in rocket thrust. But starting back in one thousand forty to a plot of maximum thrust available of rockets from that time to the present seems to show a very well behaved exponential trend which does not appear to have reached the limit yet. So we still may have more room than Iraq. That's right no. Well are there other techniques you mentioned Delphi and those biological analogy the growth curve. Another one very commonly used is trendy extrapolation. Here it is a question of getting data historical data on several different let's say manifestations of the same technology things that do the same kind of function but perhaps in a different manner. For instance one that I have used extensively in illustrating this technique is the efficiency of illumination sources. If you go back about a century to the paraffin candle and up through

kerosene lamp the acetylene lamp Edison is first carbon filament lamp and so on. You get a succession of different technical devices for carrying out the same function that is producing illumination. And you can plot these on a graph and show that they form a very well-behaved continuation of a previous trend each one of them seems to fall very nicely. One of the two things you're plotting though let's say efficiency of elimination against time. So your time will be horizontal I guess 11 simultaneously alumina in your vertical right at the time the light source in question was invented. Let's say you put a dot on the graph and what its efficiency was and then you can practically with a ruler put a straight line through lease. So in 1920 if you were planning the lighting of office buildings for one thousand twenty five you could have anticipated the year of your illumination available. That's right and in fact the current the extrapolation seems to be good enough. That in 1940 it should have been possible to predict something like the gallium

arsenide diode although I don't believe anyone actually did it at that time. Well who is this. Is this a technique that's also used by economic forecasters or demographic orchestras it seems to me what you're saying is you take your data and you draw a line through it and you kind of just keep that line going past your available data. That's correct this has been done in the past by economic forecasters and for short term forecasts they still do a fair amount of it. They do not like to use it for longer term forecasts However demographic forecasters have had some of them gotten their fingers burned with it in the past generally because they have underestimated the actual population growth so that they too have had their fingers burned in a sense so they are not too happy with it. So I take it that in the short term whatever that may be with regard to extrapolation in the short term trend extrapolation might be quite powerful tech That's right. Over the longer term the forces which produce the trend tend to change so therefore the

trend itself will change. Well are there any methods which get more into so speak the guts of the matter would you have a more scientific or fundamental basis to. Yes there are other ways. I like to lump together as mathematical models. If you can think of an equation where the variables in the equation would represent. Things of importance such as research and development investment dollars or the number of scientists and engineers available leaf square feet of laboratory floor space available and so on one could list a number of variables which would reasonably have some impact on the growth of technology. Then the question of being one of writing an equation which had numerical coefficients by which you multiply this year's R&D investment by last year's graduate and other term involving last year's graduating class of Ph.D.s. And the number of square feet of floor space that we built in laboratories this year and so on and below it are you saying a room or two noted scientific output is in fact so closely

determined by these kinds of investments that you can make reliable forecasts on them. You saying and think that you can essentially buy development if you are willing to put enough money and floor space and so forth into it. Within reason yes. I have to qualify that because there are certain things that. You can't get any faster by pouring in more money it still takes 9 months to produce a baby regardless of the number of people put to the job. Three pregnant women do it in three months. That's right. But within limits. Yes it's quite possible to set a goal. Decide how much money it's going to cost and how many people and how much floor space and go at it. This is essentially what was done with a very famous 1961 forecast. We will place a man on the moon before the end of this decade. Well let's seems to be a different sort of thing there that you seem to have introduced a fact to which you haven't had in the other forecasting techniques. You seem to introduce now the notion of a goal not necessarily one of the uses of the mathematical model would be to say if you want this goal here is what it will cost you

to achieve it. But there is no implication in the use of the model or the equation that you should want the goal I think you will want it. It's simply a statement that should you choose to achieve this goal. This is what it's going to cost you so it's still a normatively neutral technique that's right. Well so far you've mentioned for Delphi growth curves trend extrapolation and mathematical models are there others that are of some importance. There are others that are some importance simulation has not been extensively used in Main yet technological forecasting field however it seems to be a fairly powerful tool in political forecasts. What is simulation. This is. Getting a situation either with people or with computers acting like people. Where in essence you run the situation through through a play or something of the sort. A human involved simulation let's say could have a highly structured play acting who's That's not what this sort of thing in fact this sort of thing is done in schools of business as a means of training people.

Well you'll be president I'll be vice president so you'll be treasurer and then we have a you have your company. I don't want to have your company and you're allowed you're given play money if you will to to allocate to various activities in your corporation that's a marketing R&D production and so on. It's fun to think that you can get paid to play monopoly and that time is speeded up in the sense that 15 minutes of play may represent six months of actual time so you get very rapid feedback on whether your decisions were good bad or indifferent and this turns out to be a fairly powerful training tool. Who's using the Harvard Business School has used it to some extent the Industrial College of the armed forces here in Washington are using it very extensively. So there are a number of schools which are using this approach in the school of course they are not looking to it as a means of forecasting but more as a training tool that there are people who are looking at it strictly as a forecasting tool or as a means of gaining insight. This is particularly in the political field where you have multi nation games with the players. A small group representing various

officials of a of a nation. You mean you play out a conference table that's writing or not necessarily a conference table but some international incident. Let's say one group has been intensively studying the causes of the First World War. This is so you know so far back in time that we can be objective about it. Nobody is alive today who had a finger in the pie so we can afford to be objective. But they have set up mythical countries mythical names and so on but unknown to the players they actually represent very closely the characteristics of some of the major European powers in the early part of the second decade of this century. And they find that these players going through the roles that they've been instructed to play but without recognizing what they're actually doing virtually always come out the world war one at about the same time and place. So this appears to be a very powerful forecasting tool in the political field. It's interesting the other this seems to be rather close to an illogical

reasoning. Sure. Yes. Is this a separate technique from and a logical reason. Well one can say that any of these rational methods are an illogical in a sense but we usually try to keep that term this think. For methods which are. Less rigorous let's say than the mathematical model but more rigorous than Delphi. Or we examine a model historical situation one which actually took place in the past and another situation that we're trying to forecast and we attempt to compare that to in the sense that this current situation which is to be forecast does or does not as the case may be resemble the model situation in certain details. Now you run into a problem here if you insist on too much similarity. If the analogy must be too close. You run out of an illogical situation is because no two historical situations are ever identical in all details. So you run out of test cases like the right way to do it. Perhaps we ought to switch gears for a moment to another aspect to poor kids doing what's

what's an appropriate time scale for forecasting or you are you talking about doing this interview one year or century or how this depends on the use to which the forecasters put the usual notion of a forecast is not something made for its own sake. But made for the purpose of aiding some decision maker. He has a decision to make. And the question of the time span of the forecast involves how long it will take him to act and how long it will take in his action to have effect. For instance if he is currently concerned with the performance of a new technological product which he is about ready to market then he may be concerned with a one year or two year forecast. They are indeed director of research and development director of a major corporation whoever. Might be concerned with something over they have a 5 to 10 year forecast he wants to know what research and development programs to initiate this year. In order to have something which will be appropriate at the time of the program is complete. So since we are

talking probably of a five or more year span. For the lifetime of an R&D program this means he must forecast. The situation at the completion of his program. So the basic argument here for the research planner since you know who is going to take him three five years to launch a large program he has to have a pretty clear idea why he's getting there that's right for you some boozed and commits that will erect. And are there any killings were going out in the. Yes there are very definitely and this gets into the question of technological policy. We may well want to look at the impact of some technology say from a standpoint of pollution or what have you any kind of social impact. And decide whether we want to commit ourselves to that technology and make this decision early enough in the game that there's still time to stop. So had we looked at the automobile 30 or 40 years ago for instance we might have made a different decision on the propulsion mechanism we might

have gone for the electric car even though economically it was a little less desirable. Well you have been there perhaps a hidden assumption and called for a word. Maybe we didn't make the decision. It could have been that there wasn't a decision that wasn't a decision made you're correct and so that presumably you're only going to forecast when you have a mechanism for control and this helps guide the control. Could you talk a bit about the issue very briefly of prediction and accuracy. Can we discriminate or distinguish between core casting and predicting if by predicting You mean a single point forecast. A statement that this is it. Yes very definitely. And most forecasters in the business try to stay away from that they want to provide the user with not just a statement of what's going to happen but how much deviation from this to expect so that forecasters really aren't looking for a single event to occur but rather for a range of Medicare always a range of options.

Well thank you Colonel Dr. Martino this has been a very instructive discussion of how one CT casts the future. Those of you who are interested in learning more about the future and the world future society are invited to write for a free copy of our journal The Futurist. You may write to me Joe code some care of the station or to the world future society Post Office Box 1 9 2 8 5. Post Office Box 1 9 2 8 5 Twentieth Street Station in Washington D.C. This is the national educational radio network.